Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Development and validation of logistic regression models for thrombocytopenia and investigation of heparin-induced… Verma, Arun Kumar 2005

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
831-ubc_2005-104511.pdf [ 12.41MB ]
Metadata
JSON: 831-1.0092279.json
JSON-LD: 831-1.0092279-ld.json
RDF/XML (Pretty): 831-1.0092279-rdf.xml
RDF/JSON: 831-1.0092279-rdf.json
Turtle: 831-1.0092279-turtle.txt
N-Triples: 831-1.0092279-rdf-ntriples.txt
Original Record: 831-1.0092279-source.json
Full Text
831-1.0092279-fulltext.txt
Citation
831-1.0092279.ris

Full Text

D E V E L O P M E N T A N D VALIDATION OF LOGISTIC REGRESSION MODELS FOR THROMBOCYTOPENIA A N D INVESTIGATION OF HEPARIN-IN DUCED THROMBOCYTOPENIA IN CRITICALLY ILL PATIENTS by A R U N K U M A R V E R M A B.Sc, The University of British Columbia, 1986 B.Sc, The University of British Columbia, 1995 M.Sc , The University of British Columbia, 2000 A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY in THE F A C U L T Y OF G R A D U A T E STUDIES Pharmaceutical Sciences THE UNIVERSITY OF BRITISH C O L U M B I A April 2005 © Arun Kumar Verma, 2005 II ABSTRACT PURPOSE Thrombocytopenia, which is common in critically ill patients, can increase bleeding risk or be a clinical sign of heparin-induced thrombocytopenia (HIT). This investigation: 1) estimated the incidence of, identified explanatory variables at and after admission, and evaluated the performance of predictive models for thrombocytopenia, and 2) estimated the incidence of HIT, and evaluated the predictive performance of a heparin-PF4 enzyme-linked immunosorbent assay (ELISA) for diagnosing HIT. METHODS Logistic regression was used to identify predictors of thrombocytopenia (platelet count < 150 x 109/L; < 100 x 109/L) for 792 patients admitted to a community hospital combined intensive and coronary care unit (ICU/CCU). ICU/CCU and ICU admission and post-admission models were developed and validated internally using bootstrap re-sampling techniques. Admission models were validated externally using data from 572 ICU patients admitted to a different hospital. HIT diagnosis was based on clinical criteria and a positive 14C-serotonin release assay (SRA). Specificity and predictive values for the ELISA were estimated in patients who met the clinical criteria for HIT. RESULTS One hundred and twenty-two (17.3%) ICU/CCU patients developed thrombocytopenia (two consecutive counts < 150 x 109/L). Specific predictors were consistently identified for the admission models (APACHE II score, admission diagnosis, and admission platelet count) and post-admission models (admission diagnosis, A P A C H E II score, admission platelet count, fresh frozen plasma and packed red blood cell transfusions). These models demonstrated excellent discriminating ability that was supported by internal validation. On external validation, the admission models demonstrated acceptable to excellent discriminating ability, but there was a tendency to over-predict at higher probabilities. HIT incidence was 0.39% (95% CI, 0.01% to 2.1%). Positive and negative predictive values (PPV and NPV) ofthe ELISA were 10% and 100%, respectively. CONCLUSIONS Thrombocytopenia, in critically i l l patients, appears to be multi-factorial. Logistic regression models could be used to identify patients at higher risk of bleeding, and to rule out HIT prior to diagnostic testing in a subset of ICU/CCU patients who meet the clinical criteria for this syndrome. ELISA testing could then be used to rule out HIT in the remaining patients who meet the clinical criteria, thus reducing the proportion of patients requiring a change in therapy. IV TABLE OF CONTENTS Page A B S T R A C T ii T A B L E OF CONTENTS iv LIST OF TABLES xii LIST OF FIGURES xv LIST OF APPENDICES xvii LIST OF ABBREVIATIONS xviii A C K N O W L E D G E M E N T S xxi 1. INTRODUCTION 1 1.1 B A C K G R O U N D 1 1.1.1 Platelets and hemostasis 1 1.1.2 Definition and causes of thrombocytopenia 2 1.1.2.1 Causes of thrombocytopenia 3 1.1.2.1.1 Decreased platelet production 3 1.1.2.1.2 Altered sequestration and dilution of platelets 4 1.1.2.1.3 Increased destruction of platelets 5 1.1.2.1.3.1 Non-immune mediated platelet 5 destruction 1.1.2.1.3.2 Immune-mediated platelet 7 destruction 1.2 THROMBOCYTOPENIA A N D THE PROPENSITY TO B L E E D 8 PART A D E V E L O P M E N T A N D V A L I D A T I O N OF LOGISTIC REGRESSION 11 MODELS FOR THROMBOCYTOPENIA 1.3 THROMBOCYTOPENIA IN CRITICALLY ILL PATIENTS 11 1.3.1 Studies investigating thrombocytopenia in ICU patients 11 V 1.3.1.1 Limitations of the studies performed to date 16 1.3.2 Studies investigating thrombocytopenia in coronary care patients 18 1.3.3 Part A Rationale and study goal 22 PART B HEPARIN A N D THROMBOCYTOPENIA 23 1.4 HEPARIN-LNDUCED THROMBOCYTOPENIA 23 1.4.1 Therapeutic indications for heparin therapy in critically i l l patients 23 1.4.2 Thrombocytopenia resulting from heparin therapy 24 1.4.2.1 Heparin-associated thrombocytopenia (HAT) 26 1.4.2.2 Heparin-induced thrombocytopenia (HIT) 27 1.4.2.2.1 Clinical presentation of HIT 27 1.4.2.2.2 Incidence of HIT 29 1.4.2.2.3 Pathophysiology of HIT 32 1.4.2.2.4 Clinical outcomes associated with HIT 34 1.4.2.2.5 Laboratory tests for HIT 36 1.4.2.2.6 Diagnosis of HIT 40 1.4.2.2.7 Management of patients with HIT 41 1.4.3 Part B Rationale and study goal 44 1.5 PART A and B STUDY OBJECTIVES 45 2. METHODS 46 2.1 OVERVIEW OF RESEARCH 46 PART A D E V E L O P M E N T A N D V A L I D A T I O N OF LOGISTIC REGRESSION 47 MODELS FOR THROMBOCYTOPENIA 2.2 M O D E L DEVELOPMENT FOR POTENTIAL RISK INDICATORS 47 ASSOCIATED WITH THROMBOCYTOPENIA 2.2.1 Study design 47 vi 2.2.2 Study setting 47 2.2.3 Patient selection 47 2.2.4 Ethics approval 48 2.2.5 Sample size for risk indicators associated with the development of 48 thrombocytopenia 2.2.6 Data Collection 49 2.2.6.1 Data management 49 2.2.7 Criteria for thrombocytopenia 51 2.2.8 Determination of platelet count 51 2.2.9 Datasets and models developed at L G H 52 2.2.10 Demographic and patient characteristics 54 2.2.11 Risk indicators for thrombocytopenia 54 2.2.11.1 Admission (baseline) risk indicators 55 2.2.11.2 Post-admission risk indicators 60 2.3 DESCRIPTION OF D A T A OBTAINED F R O M ICU PATIENTS A T ST. 63 P A U L ' S HOSPITAL FOR E X T E R N A L VALIDATION 2.3.1 Study setting 63 2.3.2 Patient selection 64 2.3.3 Ethics approval 64 2.3.4 Data Collection 64 2.3.4.1 Data management 65 2.3.5 Criteria for thrombocytopenia 6 5 2.3.6 Determination of platelet count 65 2.3.7 Demographic and patient characteristics 65 2.3.8 Risk indicators for thrombocytopenia 66 vi 2.3.9 Datasets used for external validation 66 2.4 STATISTICAL ANALYSIS 67 2.4.1 Data management 67 2.4.2 Potential risk indicators for thrombocytopenia 67 2.4.2.1 Descriptive analysis 67 2.4.2.2 Logistic regression 67 2.4.3 Evaluation (Validation) of logistic regression models 77 2.4.3.1 Internal validation 77 2.4.3.1.1 Bootstrap procedure 77 2.4.3.1.2 Cox proportional hazards modeling 79 2.4.3.2 External validation 79 PART B HEPARIN A N D THROMBOCYTOPENIA 80 2.5 HEPARJN-INDUCED THROMBOCYTOPENIA IN CRITICAL C A R E 80 PATIENTS 2.5.1 Incidence of immune-mediated heparin-induced thrombocytopenia 80 (HIT) 2.5.1.1 Study design 80 2.5.1.2 Study setting 80 2.5.1.3 Patient selection 80 2.5.1.4 Ethics approval 81 2.5.1.5 Sample size for the incidence of HIT 81 2.5.1.6 Data collection 81 2.5.1.6.1 Data management 81 2.5.1.7 Admission diagnostic categories 82 2.5.1.8 Definitions 82 vii 2.5.1.8.1 Patients at risk for HIT 82 2.5.1.8.2 Clinical criteria for HIT 82 2.5.1.8.3 Diagnosis of HIT among patients at risk 83 2.5.1.9 Diagnostic testing for HIT 83 2.5.1.9.1 Sample selection for the SRA and the heparin- 83 platelet factor 4 enzyme-linked immunosorbent assay (heparin-PF4 ELISA) 2.5.1.9.1.1 SRA assay 84 2.5.1.9.1.2 Heparin-PF4 ELISA 84 2.5.1.9.2 Occurrence of false positives with the SRA and 85 the heparin-PF4 ELISA 2.5.1.9.2.1 Sample selection for false positives 85 2.6 STATISTICAL ANALYSIS 86 2.6.1 Incidence of HIT 86 2.6.2 Predictive performance of the heparin-PF4 ELISA 86 2.6.3 False positive rates of the heparin-PF4 ELISA and SRA 87 3. RESULTS 88 PART A D E V E L O P M E N T A N D V A L I D A T I O N OF LOGISTIC REGRESSION 88 MODELS FOR THROMBOCYTOPENIA 3.1 DEMOGRAPHIC CHARACTERISTICS A N D CLINICAL COURSE IN THE 88 ICU/CCU 3.1.1 Demographic characteristics of the L G H ICU/CCU patients included in 88 the > 150 x 109/L and > 100 x 109/L datasets 3.1.1.1 Severity of illness 88 3.2 ADMISSION DIAGNOSIS A N D CLINICAL COURSE 91 3.2.1 Admission diagnoses 91 3.2.2 Admission platelet count 91 3.2.3 Clinical course in the ICU/CCU 93 ix 3.2.3.1 Incidence of thrombocytopenia 93 3.2.3.2 Duration of ICU/CCU stay 96 3.3 LOGISTIC REGRESSION A N A L Y S I S 96 3.3.1 Multivariate ICU/CCU < 150 x 109/L admission (model 1) and 99 exploratory post-admission (model 1 PA) models 3.3.2 Multivariate ICU < 150 x 109/L admission (model 2) and exploratory 117 post-admission (model 2PA) models 3.3.3 Multivariate ICU/CCU < 100 x 109/L admission (model 3) and 123 exploratory post-admission (model 3 PA) models 3.3.4 Multivariate ICU < 100 x 109/L admission (model 4) and exploratory 128 post-admission (model 4 PA) models 3.4 M O D E L E V A L U A T I O N 136 3.4.1 Internal validation ofthe admission and exploratory post-admission 136 models 3.4.2 Comparison of logistic regression and Cox proportional-hazards 136 regression for the admission and exploratory post-admission models generated at L G H 3.4.3 External validation of the admission models generated at L G H 147 3.4.3.1 Demographic characteristics of the SPH ICU datasets 147 3.4.3.2 Admission diagnoses for SPH and L G H critical care patients 150 3.4.3.3 Incidence of thrombocytopenia among the SPH ICU patients 150 3.4.3.4 External validation: L G H ICU/CCU and ICU admission 153 models applied to the two SPH ICU validation datasets PART B HEPARIN A N D THROMBOCYTOPENIA 157 3.5 HEPARIN-LNDUCED THROMBOCYTOPENIA IN CRITICAL C A R E 157 PATIENTS 3.5.1 Description of the ICU/CCU patients in the HIT component of this 157 study 3.5.2 Demographic characteristics of patients at risk and who met the clinical 157 criteria for HIT X 3.5.3 Admission diagnoses 157 3.5.4 Heparin administration 161 3.5.5 Timing and change in the platelet count decline among patients 161 meeting the clinical criteria for HIT 3.5.6 Laboratory analysis of clinically suspected HIT samples 161 3.5.7 Incidence of HIT 164 3.5.8 Diagnostic testing with the heparin-PF4 ELISA 164 3.5.9 Clinical course of the HIT patient 166 3.5.10 Clinical course of a thrombocytopenic patient who did not develop 168 HIT 4. DISCUSSION 171 PART A DEVELOPMENT A N D V A L I D A T I O N OF LOGISTIC REGRESSION 171 MODELS FOR THROMBOCYTOPENIA 4.1 INCIDENCE OF THROMBOCYTOPENIA 171 4.2 M U L T I V A R I A T E LOGISTIC REGRESSION MODELING 172 4.2.1 Admission models 172 4.2.2 Exploratory post-admission models 180 4.3 M O D E L E V A L U A T I O N 194 4.3.1 Internal validation 195 4.3.2 External validation 196 4.3.3 Model evaluation summary and conclusions 199 4.4 STUDY LIMITATIONS 200 4.5 POTENTIAL CLINICAL APPLICATIONS OF LOGISTIC REGRESSION 204 MODELING FOR THROMBOCYTOPENIA PART B HEPARIN A N D THROMBOCYTOPENIA 209 4.6 HEPARIN-INDUCED THROMBOCYTOPENIA IN CRITICAL C A R E 209 PATIENTS xi 4.5 CONCLUSIONS A N D IMPLICATIONS FOR FUTURE R E S E A R C H 217 5. REFERENCES 219 6. APPENDICES 245 LIST OF T A B L E S Table Page 1 Studies investigating the incidence and risk factors associated with the 13 development of thrombocytopenia in critically i l l patients 2 Studies investigating the incidence and risk factors associated with the 20 development of thrombocytopenia in critically i l l cardiac patients 3 Sample size estimation 50 4 Reasons for excluding patients in the > 150 x 109/L and > 100 x 109/L datasets 89 5 Demographic characteristics for the ICU/CCU patients in the > 150 x 109/L 90 and > 100 x 109/L datasets 6 Admission diagnoses for the ICU/CCU patients in the > 150 x 109/L and > 100 92 x 109/L datasets 7 Incidence of thrombocytopenia among patients with intensive and coronary 94 care admission diagnosis included in the two datasets 8 Frequency of thrombocytopenia based on different criteria among patients 95 included in the two datasets 9A Model 1: ICU/CCU < 150 x 109/L admission model 100 9B M o d e l l : Logistic regression statistics 100 10A Model 1PA: ICU/CCU < 150 x 109/L exploratory post-admission model 113 10B Model 1 PA: Logistic regression statistics 113 11A Model 2: ICU < 150 x 109/L admission model 118 1 IB Model 2: Logistic regression statistics 118 12A Model 2PA: ICU < 150 x 109/L exploratory post-admission model 121 12B Model 2PA: Logistic regression statistics 121 13 A Model 3: ICU/CCU < 100 x 109/L admission model 124 13B Model 3: Logistic regression statistics 124 14A Model 3PA: ICU/CCU < 100 x 109/L exploratory post-admission model 127 14B Model 3PA: Logistic regression statistics 127 xi 15A Model 4: ICU < 100 x 109/L admission model 131 15B Model 4: Logistic regression statistics 131 16A Model 4PA: ICU < 100 x 109/L exploratory post-admission model 134 16B Model 4PA: Logistic regression statistics 134 17 Predictive performance and the estimated optimism of the 8 models developed 137 at L G H 18A Comparison of logistic regression and Cox proportional-hazards for Model 1 139 18B Comparison of logistic regression and Cox proportional-hazards for Model 140 1PA 18C Comparison of logistic regression and Cox proportional-hazards for Model 2 141 18D Comparison of logistic regression and Cox proportional-hazards for Model 142 2PA 18E Comparison of logistic regression and Cox proportional-hazards for Model 3 143 18F Comparison of logistic regression and Cox proportional-hazards for Model 144 3PA 18G Comparison of logistic regression and Cox proportional-hazards for Model 4 145 18H Comparison of logistic regression and Cox proportional-hazards for Model 146 4PA 19 Comparison of the demographic characteristics for the SPH ICU patients and 148 L G H ICU/CCU and L G H ICU patients included in the > 150 x 109/L datasets 20 Comparison of the demographic characteristics for the SPH ICU patients and 149 L G H ICU/CCU and L G H ICU patients included in the > 100 x 109/L datasets 21 Admission diagnoses for the critical care patients included in the SPH and 151 L G H > 150 x 109 datasets 22 Admission diagnoses for the critical care patients included in the SPH and 152 L G H > 100 x 109 datasets 23 Predictive performance of the L G H ICU/CCU and ICU admission models 154 when applied to the two SPH validation datasets 24 Description of the patients in the heparin-induced thrombocytopenia 158 component of this study xiv 25 Demographic characteristics of the 267 patients at risk and the 40 patients who 159 met the clinical criteria for HIT 26 Admission diagnoses of the 267 patients at risk and the 40 patients who met 160 the clinical criteria for HIT 27 Heparin administration for the 267 patients at risk and the 40 patients who met 162 the clinical criteria for HIT 28 Change in platelet counts among patients meeting the clinical criteria for HIT 163 29 Estimated incidence of HIT 165 LIST OF FIGURES xv Figure Page 1 Distribution of time to the onset of thrombocytopenia for patients included in 97 the > 150 x 109/L (N = 122) and > 100 x 109/L (N = 84) datasets 2 A Distribution of length of stay in the ICU/CCU for thrombocytopenic (N = 122) 98 and non-thrombocytopenic (N = 585) patients in the < 150 x 10 9/L study sample 2B Distribution of length of stay in the ICU/CCU for thrombocytopenic (N = 84) 98 and non-thrombocytopenic (N = 708) patients in the < 100 x 109/L study sample 3 Effect of the individual risk indicators in the ICU/CCU < 150 x 109/L 102 admission model (Model 1) on the predicted probability of developing thrombocytopenia 4 Calibration curve of the ICU/CCU < 150 x 109/L admission model (Model 1) 104 5 Area under the receiver operating characteristic curve of the ICU/CCU < 150 x 105 109/L admission model (Model 1) 6 Trade-off between sensitivity and specificity of ICU/CCU < 150 x 109/L 106 admission model (Model 1) with increasing cut-off probabilities 7 Estimated beta coefficients and means of the quartiles of admission platelet 107 count in assessing linearity in the logit 8 Scatter plot of studentized residuals and predicted probability for the ICU/CCU 109 < 150 x 10 9/L admission model 9 Scatter plot of Cook's Distance and predicted probability for the ICU/CCU < 111 150 x 109/L admission model 10 Effect ofthe individual risk indicators in the ICU/CCU < 150 x 10 9/L 115 exploratory post-admission model (Model 1PA) on the predicted probability of developing thrombocytopenia 11 Calibration curve for the ICU/CCU < 150 x 109/L exploratory post-admission 116 model (Model 1PA) 12 Calibration curve for the ICU < 150 x 109/L admission model (Model 2) 119 13 Calibration curve for the ICU < 150 x 109/L exploratory post-admission model 122 (Model 2PA) xvi 14 Calibration curve for the ICU/CCU < 100 x 109/L admission model (Model 3) 125 15 Calibration curve for the ICU/CCU < 100 x 109/L exploratory post-admission 129 model (Model 3PA) 16 Calibration curve for the ICU < 100 x 109/L admission model (Model 4) 132 17 Calibration curve for the ICU < 100 x 109/L exploratory post-admission model 135 (Model 4PA) 18A Calibration curve of the L G H ICU/CCU < 150 x 109/L admission models 155 applied to the SPH > 150 x 109/L ICU dataset 18B Calibration curve of the L G H ICU < 150 x 109/L admission models applied 155 to the SPH > 150 x 109/L ICU dataset 19A Calibration curve of the L G H ICU < 100 x 109/L admission models applied 156 to the SPH > 100 x 109/L ICU dataset 19B Calibration curve of the L G H ICU < 100 x 109/L admission models applied 156 to the SPH > 100 x 109/L ICU dataset 20 Platelet count profile of HIT patient 167 21 Platelet count profile of patient meeting the clinical criteria for HIT and having 170 a negative ELISA and SRA test xvi LIST O F APPENDICES Appendix Page 1 Lions Gate Hospital Healthcare Research Committee - Certificate of 245 Approval 2 Clinical Screening Committee for Research Involving Human Subjects 246 (UBC)- Certificate of Approval for L G H 3 Data Collection Worksheets 247 4 A P A C H E II score Form 251 5 Clinical Screening Committee for Research Involving Human Subjects 252 (UBC)- Certificate of Approval for SPH 6 Logistic Regression 253 7 Syntax for Generating Bootstrap Samples with Replacement 256 8 Macro for the Bootstrap Procedure 257 9 Univariate Analysis for Model 1 PA (ICU/CCU < 150 x 109/L Exploratory 260 Post-Admission Model) LIST O F ABBREVIATIONS xvii A L K Alkaline Phosphatase A L T Alanine aminotransferase A M I Acute myocardial infarction A P A C H E II Acute Physiology and Chronic Health Evaluation II APS Acute Physiology Score ARDS Acute respiratory distress syndrome A S A Acetyl salicylic acid AST Aspartate aminotransferase ATP Adenosine triphosphate ATIII Antithrombin III C A B G Coronary artery bypass graft ecu Coronary Care Unit CI Confidence interval COPD Chronic obstructive pulmonary disease C K - M B Creatine kinase M B isoenzyme CPR Cardiopulmonary resuscitation C V Coefficient of variation DIC Disseminated intravascular coagulation DVT Deep vein thrombosis E C G Electrocardiogram ELISA Enzyme linked immunosorbant assay EPIC Evaluation of c7E3 for the prevention of ischemic complications FFP Fresh frozen plasma Gl Gastrointestinal GP Glycoprotein H2 Histamine-2 receptor HIV Human immunodeficiency virus HAT Heparin-associated thrombocytopenia HIPA Heparin-induced platelet activation assay HIT Heparin-induced thrombocytopenia HF Heart failure HIV Human immunodeficiency virus ICU Intensive Care Unit ICU/CCU Intensive Care Unit/Coronary Care Unit IgA Immunoglobin A IgG Immunoglobin G IgM Immunoglobin M IL Interleukin LNR International normalized ratio ITP Idiopathic thrombocytopenic purpura L Litres L G H Lions Gate Hospital L M W H Low-molecular-weight heparin NA Not applicable NAP Neutrophil-activating-peptide NPV Negative predictive value NSAIDS Non-steroidal anti-inflammatory drugs OASIS-2 Organization to Assess Strategies for Ischemic Synd OR Odds Ratio PAT Platelet aggregation test PTCA Percutaneous transluminal coronary angioplasty PF4 Platelet factor 4 PPV Positive predictive value PRBC Packed red blood cells PURSUIT Platelet Glycoprotein Ilb/IIIa in Unstable Angina: Receptor Suppression Using Integrilin Therapy SD Standard deviation SE Standard error SIRS Systemic inflammatory response syndrome SOFA Sequential Organ Failure Assessment SRA Serotonin release assay TCP Thrombocytopenia TAMI Thrombolysis and Angioplasty in Myocardial Infarction TIMI Thrombolysis in Myocardial Infarction TTP Thrombotic thrombocytopenic purpura xxi ACKNOWLEDGEMENTS I would like to thank my two supervisors, Dr. Marc Levine and Dr. Stephen Shalansky, for their supervision and mentoring throughout my PhD research. Steve had the original idea for this research and provided key and timely comments. Marc encouraged and pushed me to become a quality researcher, improved my critical thinking skills by asking thought-provoking questions, and helped me to improve my scientific writing. I would also like to thank my research committee members, Dr. Wayne Riggs (chair), Dr. Peter Dodek, Dr. John Spinelli, and Dr. James McCormack, for their insightful comments and guidance during the present research. I would especially like to thank Dr. John Spinelli for his statistical and methodological advice throughout this project and for his help in writing the macro for the bootstrap procedure in SPSS®, and Dr. Peter Dodek for answering any questions regarding the clinical aspects and implications of my findings. I sincerely appreciated the assistance and cooperation provided by the nurses, physicians, and other members of the health care team in the ICU/CCU at L G H . I would also like to thank the pharmacy staff at L G H for their support during the data collection stage of this research, and for providing me a place to conduct my research. In addition, the support and help provided by the laboratory and medical records staff at L G H and SPH, and Lisa Brewer, an ICU research nurse at SPH, was greatly appreciated. Financial support was provided by a grant from the Lions Gate Healthcare Research Foundation, the Paetzold, Berthier, and Rick Hansen University Graduate Fellowships from the University of British Columbia. My sincerest thanks and love go to my entire family (my father, 2 brothers and their wives, my cousin Geeta, and my nephews and nieces), and especially to my late mother whose constant support, love, encouragement, and care allowed me to be where I am today. Her dedication to me following my accident has encouraged me to complete my Master's and PhD research. This thesis, and all subsequent research projects, and future endeavors I may embark on are solely due to her. INTRODUCTION 1.1 BACKGROUND 1.1.1 Platelets and hemostasis Blood coagulation and fibrinolysis result from a complex balance of cellular factors and plasma proteins that maintain the fluidity of blood, while allowing clot formation to occur when needed (Goyette, 1997). One important constituent of the hemostatic system is platelets. Platelets are small anucleate cytoplasmic fragments that arise from the fragmentation of megakaryocytes. They adhere to subendothelial surfaces of damaged blood vessels, aggregate with one another, and amplify the cascade leading to thrombin generation (Ni and Freedman, 2003; Goyette, 1997; Ware and Coller, 1995; Coller, 1990). These actions promote normal hemostasis by facilitating platelet plug formation (primary hemostasis) and then reinforcing the plug via thrombin-mediated conversion of fibrinogen to fibrin strands (secondary hemostasis) (Brass 2003). Platelet count is one important indicator of the primary hemostatic system (Handin, 2001). It is an important laboratory value in the diagnosis of bleeding disorders because it reflects the balance between platelet production by bone marrow megakaryocytes and platelet destruction or sequestration (Bessman, 1989). It is useful because it correlates well with the tendency to bleed, and it is readily available. The platelet count is normally in the range of 150 -450 x 109/L of blood, depending in part on the counting method used (Davis, 1998; Goyette, 1997; Williams et al, 1995; Bithell, 1993; Bessman, 1989; Hamilton, 1986). Normally, two thirds of the circulating platelet pool is present within the intravascular system (blood vessels) while the other third is present (sequestered) within the spleen (Goyette, 1997; Bithell, 1993; Hamilton, 1986). The platelets within these two compartments are freely exchangeable. 1 Platelets have a life span of 7 to 12 days, and the normal rate of turnover is 35 x 109/L/day (Bithell, 1993; Hamilton, 1986). 1.1.2 Definition and causes of thrombocytopenia Thrombocytopenia can be generally defined as a decrease in the absolute number of circulating platelets below the reference range (150 x 109/L) (Warkentin and Kelton, 2000; Davis, 1998; Bessman, 1989; Lind, 1995; Sultan, 1985). It is a clinical sign and not a diagnosis (Warkentin and Kelton, 1994). Patients with thrombocytopenia generally demonstrate a prolonged bleeding time and normal partial thromboplastin and prothrombin time. Thrombocytopenia is clinically important because it is the most common cause of bleeding, and severe thrombocytopenia is often associated with spontaneous bleeding in hospitalized patients (Warkentin and Kelton, 1994) (see Section 1.2). The criterion for thrombocytopenia used in the clinical and research literature varies widely, depending on the reason for identification of patients with thrombocytopenia. As stated above, some authors have defined thrombocytopenia as a platelet count < 150 x 109/L, based on the frequency distribution of the platelet counts determined in normal healthy individuals (Warkentin et ah, 2003). Other authors have used a criterion for thrombocytopenia of < 100 x 109/L, as this is the count at which the risk of induced bleeding is suggested to increase (i.e. considered to be clinically important) (Wittels et al., 1990). Furthermore, some have used a relative decrease in the platelet count as a criterion for thrombocytopenia. For example, immune thrombocytopenia related to heparin therapy is defined as a single platelet count below the reference range, or a 50% decrease in platelets from baseline (Warkentin et al, 2003; Warkentin et al, 1998). Usually, investigators do not justify their choice for the threshold used to define thrombocytopenia. Some investigators have classified thrombocytopenia as mild, moderate, or severe (Bonfiglio et al., 1995; Hanes et al., 1997; Baughman et al, 1993), presumably based on 2 a perceived increase in patients' propensity to bleed. Clinical criteria used to define thrombocytopenia have also included either a single or two consecutive platelet counts below the threshold. Studies analyzing risk factors for the development of thrombocytopenia in critically i l l patients (Baughman et al, 1993; Bonfiglio et al, 1995; Hanes et al, 1997; Cawley et al, 1999; Stephan et al, 1999; Strauss et al, 2002) have used one platelet count below the threshold to define thrombocytopenia, whereas, in a preliminary study to the present investigation involving critically i l l intensive and coronary care patients, Shalansky et al (2002) required patients have two consecutive platelet counts below the threshold. Similarly, in a study investigating immune thrombocytopenia due to heparin therapy in post-operative orthopedic patients, Warkentin et al (1995) also required two consecutive platelet counts below the threshold. 1.1.2.1 Causes of thrombocytopenia Thrombocytopenia generally occurs as a result of the following mechanisms: (1) decreased bone marrow production of megakaryocytes (marrow infiltration with tumor, cytotoxic drugs), (2) increased sequestration of circulating platelets by the spleen (splenic hypertrophy from portal hypertension) or dilution of the platelet count by multiple blood transfusions, and (3) decreased platelet survival time due to increased destruction of circulating platelets (non-immune destruction from sepsis or vasculititis or immune destruction from medication associated antibodies) (Warkentin and Kelton, 1994; Bithell, 1993a; DeSancho and Rand, 2001; Wittels etal, 1990). 1.1.2.1.1 Decreased platelet production Decreased platelet production can result from a reduction in megakaryocytes in the bone marrow (Akca et al, 2002; Patnode and Gandhi, 2000; Warkentin and Kelton, 1994; Bithell, 3 1993a; Bogdonoff et al, 1990). This can be due to congenital hypoplasia of the megakaryocytes (Fanconi's syndrome and viral infections, such as rubella), acquired hypoplasia of the megakaryocytes from the action of chemicals, drugs (thiazides, alcohol, diethylstilbestrol), antineoplastic chemotherapy, radiation, infectious agents, or infiltration of the bone marrow by malignant cells. This generally results in a decreased number of megakaryocytes (Patnode and Gandhi, 2000; Warkentin and Kelton, 1994; Bogdonoff et al, 1990). Chronic vitamin B i 2 or folic acid deficiencies are seen in some hospitalized patients and are suggested to be associated with decreased platelet production (Burnstein, 2000; Bogdonoff et al., 1990; Wittels et al., 1990). 1.1.2.1.2 Altered sequestration and dilution of platelets Increased platelet sequestration can result in thrombocytopenia due to an abnormal distribution of platelets (Warkentin and Kelton, 2000; Warkentin and Kelton, 1994; Bithell, 1993a; Wittels et ah, 1990). The spleen normally pools approximately one-third of the circulating platelets. Patients with splenomegaly or hypersplenism usually sequester an increased percentage of circulating platelets in the spleen, thus resulting in thrombocytopenia. In hypersplenism, the platelet count is usually 50 - 150 x 109/L, and rarely decreases to < 20 x 109/L (Warkentin and Kelton, 1994; Bithell, 1993a). Hemodilution can result in decreased circulating red cells, white cells, and platelets following the administration of blood products, colloids, or crystalloids (Warkentin and Kelton, 2000; Bogdonoff et al., 1990). Several authors have noted that massive transfusion of blood products can lead to dilution of the platelet count (Bucur et al, 2000; Warkentin and Kelton, 1994; Bogdonoff et al, 1990; Reed et al, 1986; Noe et al, 1982; Counts et al, 1979; Murphy and Gardner, 1969). However, it is not clear how many units of blood products transfused within 24 hours constitute massive transfusion (Riska et al, 1988; Marik, 2001). The decline in 4 platelet count can be attributed to dilution of platelets in the circulation by blood products containing low concentrations of viable platelets (Bogdonoff et al, 1990) or sequestration of platelets by the spleen following blood transfusions (Bareford et al., 1987). 1.1.2.1.3 Increased destruction of platelets Destruction of platelets is the most common reason for thrombocytopenia, especially in critically i l l patients (Bogdonoff et al., 1990). Thrombocytopenia caused by increased platelet destruction occurs when the rate of platelet destruction exceeds the ability of the bone marrow to produce platelets. Excessive platelet destruction leading to thrombocytopenia can either be non-immune or immune-mediated (Drews, 2003; Warkentin and Kelton, 2000; Handin, 2001a; Bogdonoff et al, 1990; Wittels etal, 1990). 1.1.2.1.3.1 Non-immune-mediated platelet destruction Non-immune-mediated platelet destruction can result from abnormal blood vessels, fibrin thrombi, and intravascular prostheses or surface interactions, all of which can shorten platelet survival (Warkentin and Kelton, 2000; Handin, 2001a; Bogdonoff et al, 1990). There are several potential causes of non-immune platelet destruction in critically i l l patients. Surface-mediated platelet destruction can involve abnormal or injured vasculature or tissue, or a foreign body (Warkentin and Kelton, 2000; Bogdonoff et al, 1990). In most of these cases, thrombocytopenia results from local platelet destruction, however systemic destruction can also occur. Swan-Ganz (pulmonary artery) catheters have been reported to be associated with a decline in the platelet count (Bonfiglio et al, 1995; Bogdonoff et al, 1990; Kim et al, 1980; Miller et al, 1984; Layon, 1999; McNulty et al, 1998; Vicente Rull et al, 1984), as have central and arterial lines (Cawley et al, 1999). The decline in the platelet count experienced by some patients following insertion of these catheters is likely the result of local thrombogenesis, and 5 hence, platelet destruction (Bogdonoff et al, 1990). In addition, pulmonary artery catheters are associated with heparin use, as heparin is bonded to their surface and administered in low doses to keep these catheters patent. It has been suggested that heparin could be displaced from the catheter surface and contribute to the development of heparin-induced thrombocytopenia (HIT) (Lee and Warkentin, 2001; Laster and Silver, 1988). However, there is very little prospective evidence of this reported in the literature. Abnormalities in platelet survival have been reported in patients with valvular or arterial prostheses and those with abnormal cardiac valves and surfaces, as a result of platelet aggregation to these foreign surfaces (Bogdonoff et al., 1990). Other foreign surfaces that might be associated with the development of thrombocytopenia include: intra-aortic balloon counterpulsation, extracorporeal circuits (Wittels et al., 1990), dialysis membranes (Bogdonoff et al., 1990), and devices used in artificial heart implantation (Wazny and Ariano, 2000). Respiratory failure, especially acute respiratory distress syndrome (ARDS), has also been reported to be associated with non-immune mediated platelet destruction (Akca et al., 2002; Bogdonoff et al, 1990; Hechtman et al, 1978; Heffner et al, 1987; Schneider et al, 1980; Bone et al, 1976). However, the site and mechanism of platelet destruction remain unclear (Bogdonoff et al, 1990). Interestingly, patients with ARDS frequently display overt evidence of disseminated intravascular coagulation (DIC) (Bogdonoff et al, 1990; Bone et al, 1976). DIC is an acquired disorder, which is characterized by the widespread activation of the coagulation cascade and can cause non-immune platelet destruction (Akca et al, 2002; Marik 2001; Bogdonoff et al, 1990; Levi and ten Cate, 1999). Infections are also associated with non-immune mediated platelet destruction. Bacterial, viral, fungal, protozoal, and rickettsial infections are associated with thrombocytopenia as a result of intravascular coagulation that causes platelet destruction (George and El-Harake, 1995; Bogdonoff et al, 1990). Most patients with bacteremia become thrombocytopenic, which in 6 some cases can be severe (George and El-Harake, 1995). Infectious disease, particularly septicemia, is the most common clinical condition associated with DIC, and bacterial infection is most frequently associated with development of this syndrome. Platelets are usually affected early in septicemia (Bogdonoff et al., 1990), and therefore, thrombocytopenia may be an early warning sign of sepsis. There are also disorders that are associated with platelet destruction by uncertain mechanisms (Warkentin and Kelton, 2000). One example is thrombotic thrombocytopenic purpura (TTP), which is a fulminant, often lethal disorder that may be initiated by endothelial injury and subsequent release of von Willebrand factor and other procoagulant substances from endothelial cells, resulting in platelet aggregation (Marik, 2001; Handin, 2001a). Thrombocytopenia is an essential feature of the condition and is typically severe (< 20 x 109/L) (George and El-Harake, 1995). 1.1.2.1.3.2 Immune-mediated platelet destruction Immune-mediated platelet destruction results from interactions of platelets with antibodies, immune complexes, or complement, resulting in clearance of the coated platelets by mononuclear phagocytes in the spleen or other tissues (Bogdonoff et al., 1990; Handin, 2001a). The immunologic thrombocytopenias can be classified based on the pathologic mechanism, the causative agent, or the duration of illness. The most common causes of immune-mediated thrombocytopenia are viral and bacterial infections, idiopathic thrombocytopenic purpura (ITP), and medications (George et al., 1995). An immune mechanism is likely responsible for the platelet destruction seen during some bacterial and viral infections (Bogdonoff et al., 1990). Some patients with gram-positive or gram-negative septicemia, or viral infections (Wazny and Ariano, 2000) have been reported to have elevated levels of platelet-associated IgG. Thrombocytopenia can present in patients 7 diagnosed with human immunodeficiency virus (HIV) infection (George et al., 1995). Many medications have been reported to result in immunologic destruction of platelets (Patnode and Gandhi, 2000; George et al., 1995; Wazny and Ariano, 2000; Bogdonoff et al., 1990), however, the majority of these have been implicated in less than a dozen cases each. There are five specific medications or medication classes that have been reported to be associated with 60% of all reported cases (Bogdonoff et al., 1990; Wittels et al., 1990): quinidine, quinine, gold salts, sulfonamides or sulfonamide derivatives, and heparin. Other medications, such as vancomycin, phenytoin, piperacillin, imipenem-cilastatin, and ranitidine, have been reported to be associated with the development of thrombocytopenia in critically i l l patients (Bonfiglio et al., 1995; Cawley et al., 1999; Wazny and Ariano, 2000). In most cases, the thrombocytopenia is self-limiting, provided that the drug is discontinued. Circulating immunoglobulins are thought to be the cause of the platelet destruction (George et al., 1995). However, it should be borne in mind that a causal relationship has not been conclusively established, as other factors such as sepsis or Swan-Ganz catheters may be responsible. 1.2 THROMBOCYTOPENIA AND THE PROPENSITY TO BLEED Because thrombocytopenia increases patients' propensity to bleed, platelet transfusions are given prophylactically or therapeutically to thrombocytopenic patients and to patients undergoing invasive procedures (Poon, 2003; Tinmouth and Freedman, 2003; Kickler, 2000), and they are standard treatment for acute bleeding in patients with severe thrombocytopenia (Poon, 2003). Researchers have not been able to identify a distinct threshold for increased bleeding risk, as the threshold platelet count at which bleeding may occur varies, depending on the clinical condition of the patient (Gernsheimer, 2003), and contributing risk factors, such as sepsis, uremia, or recent trauma or surgery may increase a patient's risk of bleeding (DeSancho and Rand, 2001). In critical care patients, there is evidence that the propensity to bleed increases 8 as the platelet count declines below the normal range (Strauss et al., 2002; Vanderschueren et al., 2000; Chakraverty et al., 1996). For example, Strauss et al (2002) prospectively studied 145 consecutive patients admitted to a medical ICU at a university hospital in order to assess the prevalence, risk factors, and outcomes of thrombocytopenia, and to investigate the factors associated with bleeding. Major bleeding episodes occurred in 21 of the 64 patients (32.8%) who developed thrombocytopenia (platelet count < 150 x 109/L), and thrombocytopenia preceded bleeding in all but 4 of the 21 patients. They reported that, compared with patients without thrombocytopenia, the risk of bleeding was more than three times as high in those with platelet counts of < 150 x 109/L and six times as high in those with platelet counts of < 50.0 x 109/L. Thrombocytopenia in critically i l l cardiac patients with acute coronary syndrome (acute myocardial infarction and unstable angina) has also been reported to be associated with a higher incidence of bleeding events (McClure et al., 1999; Harrington et al., 1994; Berkowitz et al.-, 1998; Eikelboom et al, 2001; Sane et al, 2000). For example, McClure et al (1999) reported that the incidence of a bleeding event and PRBC or platelet transfusions was higher in thrombocytopenic (platelet count < 100 x 109/L or 50% of baseline) than non-thrombocytopenic patients, and a multivariate logistic regression model for predictors of moderate/severe bleeding demonstrated that thrombocytopenia was independently associated with an increased bleeding risk (OR 2.0 [95%CI: 1.6 to 2.6]). When bleeding occurs, it is usually mucocutaneous (Warkentin and Kelton, 1994; DeSancho and Rand, 2001; Wittels et al., 1990). Clinical signs of small vessel bleeding include petechiae, purpura, and ecchymoses. A more serious problem is indicated by mucous membrane bleeding, gingival bleeding, gastrointestinal or urinary bleeding, and epistaxis (Warkentin and Kelton, 2000; Davis, 1998; Hamilton, 1986). There is considerable interest in defining the lowest platelet concentration at which 9 bleeding is unlikely, thus minimizing the use of prophylactic platelet transfusions. Despite the lack of a clear threshold, it is usually recommended that in the absence of trauma, surgery, or bleeding, patients with a platelet count < 20 x 109/L should be administered platelet transfusions (Tinmouth and Freedman, 2003; Kickler, 2000; Bogdonoff et al., 1990; Warkentin and Kelton, 1994), although some physicians now use a lower threshold (British Committee for Standards in Haematology, Blood Transfusion Task Force, 2003; Warkentin and Kelton, 2000). However, in thrombocytopenic patients experiencing active mucocutaneous or gastrointestinal bleeding, undergoing major surgery or invasive procedures (e.g. central venous catheterization, bronchial or endoscopic biopsy, thoracostomy tube placement, or abdominal paracentesis), or having received platelet aggregation inhibitors, it is generally recommended that platelet transfusions be administered to keep the platelet count above 50 x 109/L (DeLoughery, 2003; Gernsheimer, 2003; Drews, 2003; Marik, 2001; DeSancho and Rand, 2001; Wittels et al., 1990). Current guidelines recommend administration of platelet transfusions at low thresholds, but there is evidence from the above studies in critically ill patients that the propensity to bleed increases as the platelet count declines below the normal range. It has been suggested that higher platelet count thresholds may be indicated in conditions that adversely affect platelet function (sepsis, tissue trauma, renal failure, malignancy, extracorporeal circulation, and certain medications) (Contreras, 1998; Gernsheimer, 2003). However, to date little prospective research has been done to evaluate the effect of platelet transfusions at higher thresholds, especially in critically i l l patients. A better understanding of risk indicators for thrombocytopenia in critically il l patients could facilitate prospective research and decision making. 10 PART A: DEVELOPMENT AND VALIDATION OF LOGISTIC REGRESSION MODELS FOR THROMBOCYTOPENIA 1.3 THROMBOCYTOPENIA IN CRITICALLY ILL PATIENTS Thrombocytopenia is a common complication in critically ill patients and it can present a challenging clinical problem when it is severe, putting patients at risk for bleeding. In more rare cases the decline in platelet count may be due to HIT, an immune-mediated adverse effect, which can lead to limb- and life-threatening thromboembolic complications. Critically i l l patients are at risk for developing thrombocytopenia due to the severity of their illness on admission to an intensive care setting, in addition to risk indicators that they may be exposed to during the course of their illness. Most of the research examining risk factors for thrombocytopenia in critically i l l patients performed to date has involved ICU patients in academic settings, and little work has been performed on the development of thrombocytopenia among patients in coronary care units. While there have been cardiac patients in some of these studies, they usually have comprised a small percentage of the total patient sample (Baughman et al, 1993; Bonfiglio et al, 1995). Cardiac patients are important to study because most receive heparin for a sufficient period to be at risk for developing HIT. Many community hospitals, such as the one in the present study, Lions Gate Hospital (LGH), North Vancouver, B.C., have a combined ICU/CCU and, to date, no investigator has developed and evaluated models of thrombocytopenia in such settings. 1.3.1 Studies investigating thrombocytopenia in ICU patients Previous investigators (Baughman et al, 1993; Bonfiglio et al, 1995; Hanes et al, 1997; Cawley et al, 1999; Stephan et al, 1999; Strauss et al, 2002) assessed the incidence of and risk factors associated with thrombocytopenia in this population. Details regarding study design, criterion for thrombocytopenia, and results are summarized in Table 1. 11 The incidence of thrombocytopenia in critically i l l patients reported in previous studies varied depending on the clinical setting and criterion. According to a platelet count criterion < 100 x 109/L, the incidence of thrombocytopenia has been reported to be in the range of 13% (Cawley et al, 1999) and 41% (Hanes et al., 1997), and in the only study that used a platelet count criterion < 150 x 109/L for thrombocytopenia, Strauss et al (2002) reported an incidence of 44%. Several risk factors were identified in these previous studies (Baughman et al, 1993; Bonfiglio et al, 1995; Hanes et al, 1997; Cawley et al, 1999; Stephan et al, 1999; Strauss et al, 2002); however, ones that were identified in more than one study included sepsis, liver function abnormalities (elevated creatinine), disease severity (i.e. Acute Physiology and Chronic Health Evaluation II (APACHE II), Sequential Organ Failure Assessment (SOFA), or trauma scores), and admission (baseline) admission platelet count (Table 1). In addition, there are two other studies in which the authors reported the incidence, but did not perform multivariate regression analysis for thrombocytopenia in adult ICU patients (Chakraverty et al, 1996; Vanderschueren et al, 2000). Chakraverty et al (1996) performed a prospective, observational study to investigate the incidence and etiology of coagulation disturbances. They reported that 85 of 224 (38.0%) patients developed thrombocytopenia (a single platelet count < 100 x 109/L); 28 (12.5%) and 7 (3.1%) of 224 patients developed severe thrombocytopenia (platelets < 50 x 109/L) and very severe thrombocytopenia (platelets < 20 x 109/L), respectively. The authors suggested that for 54% of patients, the cause of thrombocytopenia (platelets < 100 x 109/L) was unknown; however, they noted that dilution following massive blood transfusions (> 6 units of red cells per 24 hours) and DIC were the likely cause of the thrombocytopenia in 16% and 9% of these patients, respectively. They did not state how they attributed the cause of the thrombocytopenia. Vanderschueren et al (2000) performed a prospective, observational study to investigate the incidence and prognosis of 12 Table 1 Studies investigating the incidence and risk factors associated with the development of thrombocytopenia in critically ill patients Reference Study Design Criterion for Thrombocytopenia Frequency of Thrombocytopenia Risk Factors Baughman al (1993) • retrospective • over 3 separate months • 162 patients 9 medical ICU • linear regression • mild: < 100 x 109/L • severe: < 50 x 109/L • mild: 38/162 (23%) » severe: 17/162 (10%) • sepsis • antineoplastic chemotherapy • elevated creatinine • elevated bilirubin Bonfiglio et al (1995) • retrospective • 18 months consecutive admissions screened » patients in ICU at least 72 hours • 314 patients • mixed medical/ surgical ICU • linear regression • simple: <200 x 109/L • significant: < 100 x 109/L • severe: <20 x 109/L » simple: 41.8%) • significant: 25.2% • severe: 2.0% • baseline platelet count • hemodynamic stability • inotropic agents • length of ICU stay • length of Ff2-antagonist treatment • liver function abnormalities Stephan et al (1999) © prospective • continuous admissions for 6 months • 147 patients • surgical ICU • logistic regression • < 1 0 0 x l 0 9 / L • 52/147(35%) • sepsis » episodes of bleeding or transfusions • A P A C H E II score > 15 • Admission platelet count < 185 x 109/L * While the present study used the term risk indicators for independent variables (see Section 2.2.11), the studies referred to in Table 1 used the term risk factors for independent variables. 13 Table 1 Continued Reference Study Design Criterion for Thrombocytopenia Frequency of Thrombocytopenia Risk Factors* Hanes et a/(1997) • prospective • observational » followed for up to 14 days o in ICU at least 48 hours » 63 patients • trauma ICU « logistic regression • significant: < 100 x 109/L » moderate: < 50 x 109/L • severe: < 20 x 109/L • significant: 26/63 (41%) • moderate: 2/63 (3.2%) • severe: 0 9 age • higher trauma scores • non-head injuries Cawley et al (1999) « retrospective • over 3 month period • in ICU at least 24 hours • 193 patients • surgical-trauma ICU • linear regression • < 1 0 0 x l 0 9 / L • 25/193 (13%) • central or arterial line Strauss et al (2002) o prospective • observational » in ICU at least 48 hours • over 13 month period • 145 patients • medical ICU ® logistic regression • < 1 5 0 x l 0 9 / L o platelet counts < 150 x 109/L were checked within 24 hours • mild: < 150 x 109/L • moderate: < 100 x 109/L » severe: < 50 x 109/L • very severe: < 20 x 109/L • 64/145 (44%) • mild: 33/64 (52%) » moderate: 19/64 (30%) • severe: 9/64 (14%) • very severe: 3/64 (4%) • DIC • CPR as an admission category » higher initial SOFA score • higher initial platelet count (lower risk of thrombocytopenia) * While the present study used the term risk indicators for independent variables (see Section 2.2.11), the studies referred to in Table 1 used the term risk factors for independent variables. 14 Table 1 Continued Reference Study Design Criterion for Thrombocytopenia Frequency of Thrombocytopenia Risk Factors* Shalansky et al (2002)' Verma (2000) 1 • prospective ® observational • over a 1 -year period • patients with at least 2 platelet counts 12 hours apart • 362 patients » combined ICU/CCU • logistic regression • 2 consecutive platelet counts < 150 x 109/L • 68/362(18.8%) • Admission platelet count2'3 • Age 2 • A P A C H E II score2 « Swan-Ganz catheter • A S A 3 • FFP transfusion • PRBC transfusion3 • 2,3 • sepsis' 2 3 • gastrointestinal ' • G l bleed2 • respiratory non-surgery2'3 • musculoskeletal/con-nective tissue2'3 * While the present study used the term risk indicators for independent variables (see Section 2.2.11), the studies referred to in Table 1 used the term risk factors for independent variables. 1 This study represented preliminary modeling on 1 year's data from L G H . 2 Risk indicators independently associated with thrombocytopenia in the baseline (admission) model. 3 Risk indicators independently associated with thrombocytopenia in the ICU/CCU (post-admission) model. 15 thrombocytopenia among 329 medical ICU patients of a university hospital and combined medical-surgical ICU patients of a regional hospital. One hundred and thirty-six (41%) patients were reported to develop thrombocytopenia (platelet count < 150 x 109/L), however thrombocytopenia was already present at the time of ICU admission in 89 of these patients. As judged clinically, the authors noted that the most common causes of thrombocytopenia were sepsis, infection, DIC, liver disease/hypersplenism, primary hematologic disorder, and medications (cytostatic drugs and heparin were most frequently implicated). Furthermore, following logistic regression analysis, the investigators noted that thrombocytopenia was independently associated with ICU mortality. It is important to note that the investigators of all the studies described above evaluated the association of risk factors present at admission together with risk factors that patients were exposed to during their stay in the ICU (i.e. the investigators developed post-admission models). None of the investigators developed an admission model to examine the impact of risk factors present at admission. In a preliminary study to the present investigation (Shalansky et al., 2002; Verma, 2000), it was observed that thrombocytopenia (two consecutive platelet counts < 150 x 109/L) developed in 68 of 362 (18.8%; 95% CI: 14.9% - 22.8%) patients (Table 1). Admission (baseline) and post-admission (ICU/CCU) models for thrombocytopenia were developed, and risk indicators independently associated with thrombocytopenia at admission included A P A C H E II score, lower age, lower admission platelet count, and five admission diagnoses. Nine risk indicators were independently associated with thrombocytopenia post-admission. 1.3.1.1 Limitations ofthe studies published to date It is apparent from the studies published by other investigators that thrombocytopenia occurs commonly in critically i l l patients, though all studies performed to date have some 16 limitations. First, the findings of Baughman et al (1993), Bonfiglio et al (1995), and Cawley et al (1999) are affected by the retrospective nature of the studies. In general, retrospective studies are limited by information obtained solely from patients' medical charts, which can be inaccurate, and the data were recorded for clinical reasons and not for study-related analyses (Kraemer et al, 1997). Second, the previous studies (Baughman et al, 1993; Bonfiglio et al, 1995; Hanes et al, 1997; Cawley et al., 1999; Stephan et al, 1999; Strauss et al, 2002) involved small numbers of patients, which reduced the power to detect specific risk factors as being associated with the development of thrombocytopenia, and in fact, the models may have been overfitted. Third, not all suspected risk factors that are potentially associated with thrombocytopenia in critically i l l ICU patients were included in the analyses (Baughman et al, 1993; Hanes et al, 1997; Cawley et al, 1999; Strauss et al, 2002), suggesting that some missing risk factors may have accounted for part of the variability in the development of thrombocytopenia. Fourth, the published reports of the studies conducted to date by other investigators all lack clarity in the statistical modeling process. The development of thrombocytopenia is usually viewed as a binary outcome, however, only three studies (Hanes et al, 1997; Stephan et al, 1999; Strauss et al, 2002) used logistic regression analysis to identify independent risk factors associated with the development of thrombocytopenia. The other three studies (Baughman et al, 1993; Bonfiglio et al, 1995; Cawley et al, 1999) used linear regression analysis, but it is unclear how these investigators performed this analysis and interpreted the findings. Fifth, no uniform criteria were used in defining thrombocytopenia in the published reports by other investigators. Finally, while some of these investigators developed multivariate models to explain thrombocytopenia, none has performed validation studies to assess the predictive performance of their models on a new data set. Validation of a model should be performed before a model can be used to predict the probability of an outcome in new patients (Scalon et al, 1996). 17 The preliminary study to the present investigation (Shalansky et al., 2002; Verma, 2000) addressed some of the limitations identified in the previous studies conducted by other investigators; however, it also had a few shortcomings. First, even though it was the largest to date, it also involved small numbers of patients, which reduced the power to detect specific risk indicators (risk markers) associated with the development of thrombocytopenia. Second, it treated diagnoses as separate, individual variables, rather than as a categorical variable (a variable with multiple mutually exclusive categories (e.g. individual diagnoses)), which might have resulted in identifying spurious associations for thrombocytopenia. Since each diagnosis was mutually exclusive, diagnosis should have been coded as a categorical variable. Third, the models reported in the preliminary study were generated in a combined sample of ICU/CCU patients, and thus, it was difficult to compare the results of this study with those published by other investigators involving ICU patients. Fourth, the criterion of thrombocytopenia used in this study (two consecutive platelet counts < 150 x 109/L) was different from the other published studies, with the exception of the study by Strauss et al (2002). Hence, it was difficult to compare the results of the preliminary study to those reported in the other studies. Finally, the preliminary study (Shalansky et al., 2002; Verma, 2000) suggested what the risk indicators (risk markers) associated with thrombocytopenia might be, but no attempt was made to validate the model on a new data set. 1.3.2 Studies investigating thrombocytopenia in coronary care patients Information regarding the incidence or risk indicators associated with thrombocytopenia in critically i l l cardiac patients has been derived from post hoc evaluations of large cardiac clinical trials. The data suggest that thrombocytopenia can develop during an acute coronary syndrome (acute myocardial infarction (AMI) and unstable angina), and is associated with adverse outcomes (mortality, ischemic events, and/or bleeding) (Williams et al., 2003; McClure * 18 et al., 1999; Berkowitz et al, 1998; Eikelboom et al, 2001). Table 2 summarizes the details regarding study design, criterion for thrombocytopenia, and the incidence of and risk factors associated with thrombocytopenia in clinical trials involving patients with acute coronary syndromes (McClure et al., 1999; Eikelboom et al., 2001; Bovill et al., 1991; Harrington et ah, 1994; Berkowitz et ah, 1998; Sane et ah, 2000). The incidence of thrombocytopenia (platelet count < 100 x 109/L) in critically i l l cardiac patients has been reported to be in the range of 1% (Eikelboom et ah, 2001) and 7.3% (Harrington et ah, 1994). Several risk factors were identified; however, ones that were identified in more than one study included a lower admission (baseline) admission platelet count and increasing age (Table 2). Harrington et al (1994) did not investigate risk indicators for the development of thrombocytopenia, but noted that patients who developed thrombocytopenia had: a lower median acute ejection fraction; a higher likelihood of three-vessel coronary artery disease; a higher incidence of HF, recurrent ischemia, and complete heart block; undergone bypass surgery, balloon pump insertion, and endotracheal intubation. While these post hoc analyses have provided information on the incidence of and risk factors for thrombocytopenia from a substantial number of patients, further research is warranted to determine the extent to which these findings can be generalized to the C C U population. Based on an understanding of the limitations of the previous studies including the preliminary study to the present investigation, further prospective research is required to develop and validate logistic regression models for thrombocytopenia in a community-based setting involving both critically i l l intensive and coronary care patients. 19 Table 2 Studies investigating the incidence and risk factors associated with the development of thrombocytopenia in critically ill cardiac patients Reference Study Design Criterion for Thrombocytopenia Frequency of Thrombocytopenia Risk Factors McClure et al (1999) • post hoc • PURSUIT trial « eptiftbatide vs. placebo • 9217 patients • logistic regression • < 100 x 10 9 /Lor<50%of baseline • < 100 x 109/L • < 50 x 109/L (severe) • < 20 x 109/L (profound) • 633/9217(6.9%) • 454/9217(4.9%) • 41/9217(0.44%) • 7/9217 (0.076%) 9 C A B G • moderate-severe bleeds 9 intra-aortic balloon pump 9 females 9 PTCA 9 increasing age * lower baseline platelet count Harrington et al (1994) • post hoc » T A M I and urokinase trials • 874 patients • no regression analysis • < 100 x 10 9 /Lor<50%of baseline • < 1 0 0 x l 0 9 / L e 143/874 (16.4%) • 64/874 (7.3%) 9 N / A Berkowitzef al (1998) • Post hoc • EPIC trial • 2099 patients • logistic regression • < 1 0 0 x l 0 9 / L • < 5 0 x l 0 9 / L • acute top** < 100 x 109/L • acute severe top < 50 x 109/L • acute profound tcp < 20 x 109/L • 81/2099(3.9%) • 19/2099 (0.91%) • 31/2099(1.5%) » 7/2099 (0.33%) 9 2/2099 (0.095%) 9 lower baseline platelet count o lower weight 9 increased age • abciximab * While the present study used the term risk indicators for independent variables (see Section 2.2.11), the studies referred to in Table 2 used the term risk factors for independent variables. ** Acute thrombocytopenia (tcp): a platelet count below the stated threshold within 24 hours of initiation of the study treatment. C A B G (coronary artery bypass graft); PTCA (percutaneous transluminal coronary angioplasty); N / A (not applicable); tcp (thrombocytopenia) 20 Table 2 Continued Reference Study Design Criterion for Thrombocytopenia Frequency of Thrombocytopenia Risk Factors* Eikelbdom et al (2001) « post hoc « OASIS-2 trial • heparin vs. hirudin • 8913 patients • logistic regression • < 100 x 109/L • < 50 x 109/L (severe) • 87/8913 (1.0%) • 6/8913 (0.067%) • lower baseline platelet count Bovill et al (1991) o post hoc • TIMI phase II trials » 3339 patients » no regression analysis • < 1 5 0 x l 0 9 / L • < 100 x 109/L • 10.3% • 1.6% • N / A Sane et al (2000) o Post hoc « various cardiac trials ® 8555 patients » no regression analysis • < 100 x 10 9/Lwith25% decrease from baseline count • 261/8555 (3.1%) • N / A * While the present study used the term risk indicators for independent variables (see Section 2.1.10), the studies referred to in Table 2 used the term risk factors for independent variables. N / A (not applicable) 21 1.3.3 Part A: Rationale and study goal There are many potential causes for thrombocytopenia in critically i l l patients, and thrombocytopenia is associated with an increased propensity to bleed (Strauss et al., 2002) and mortality (Vanderschueren et al., 2000). There are no large prospective studies that have identified the most important variables for the development of thrombocytopenia. Knowledge of risk indicators for thrombocytopenia and the ability to predict thrombocytopenia in individual patients may be clinically useful with respect to initiation of procedures and/or therapies that may result in decreased platelet count, and contribute to the development of guidelines for initiation of platelet transfusions in those patients at risk for bleeding. The overall goal of Part A of the thesis was to develop and evaluate logistic regression models for the development of thrombocytopenia in a cohort of ICU/CCU patients. 22 PART B: HEPARIN AND THROMBOCYTOPENIA 1.4 HEPARIN-INDUCED THROMBOCYTOPENIA 1.4.1 Therapeutic indications for heparin therapy in critically ill patients Unfractionated heparin (heparin) is an anticoagulant, which is composed of elongated polysulfated glycosaminoglycan polymers, also called mucopolysaccharides (Hirsh et al, 1998; Freedman, 1992; Chong, 1992). It produces its anticoagulant effect by binding to and causing a conformational change in the serine protease inhibitor anti-thrombin III (ATIII), resulting in a multi-fold increase in the ability of ATIII to inactivate thrombin (factor Ha), activated factor X , and activated factors (IX, XI, XII) leading to decreased coagulatibility (Freedman, 1992; Hirsh and Raschke, 2004; Hirsh and Fuster, 1994; Hirsh, 1991). Heparin is the anticoagulant of choice in the hospital setting when a rapid effect is required, because it is administered intravenously or subcutaneously (Freedman, 1992; Hirsh and Raschke, 2004; Hirsh and Fuster, 1994). Heparin was first used to prevent venous thromboembolism in surgical patients in 1937 (Brieger et al, 1998), and since that time it has been used for the prevention and early treatment of arterial and venous thromboembolism (Geerts et al, 2004; Buller et al, 2004; Chong, 1992; Hirsh, 1991) such as: venous thrombosis and pulmonary embolism (Geerts et al, 2004; Buller et al, 2004; Hirsh and Fuster, 1994; Hirsh, 1991), cardiac surgery using cardiac bypass (Brieger et al, 1998; Freedman, 1992; Hirsh et al, 2001; Walls et al, 1992; Chong, 1992); during and after coronary angioplasty, and in patients with coronary stents (Brieger et al, 1998; Freedman, 1992; Popma et al, 2004; Bauer et al, 1997); for extracorpeal procedures such as haemofiltration or haemodialysis (Geerts et al, 2004; Chong, 1992); for the early treatment of acute myocardial infarction and unstable angina (Menon et al., 2004; Harrington et al, 2004; Brieger et al, 1998; Trujillo and Nolan, 1998; The RISC Group, 1990; Theroux et al, 1988; Granger et al, 1996; 23 Ryan et al, 1996); for selected patients with disseminated intravascular coagulation (Hirsh et al, 2001; Hirsh, 1991); and for maintaining patency of arterial and venous lines. Heparin is usually administered at higher (therapeutic) doses for the treatment of venous thrombosis or acute myocardial ischemia, and at lower doses for primary prophylaxis (Hirsh and Raschke, 2004). Low molecular-weight heparins (LMWH) are also used for many of the indications described above (Geerts et al, 2004; Buller et al, 2004; Popma et al, 2004; Menon et al, 2004), but unfractionated heparin is still commonly used in the critical care setting due to reversibility of its anticoagulant effect. Heparin is frequently administered to critically i l l patients (Baughman et al, 1993; Stephan et al, 1999). In 2 recently published reports, Strauss et al (2002) and Shalansky et al (2002) noted that heparin was administered to 87% and 86% of their patients, respectively. Critically i l l patients are usually exposed to heparin in the following ways: full dose anticoagulation (treatment of venous thrombosis, pulmonary embolism, atrial fibrillation with embolization, acute DIC (Susla et al, 2001)); administration of 5,000 or 7,500 Units heparin subcutaneously every 12 hours (primary prophylaxis) (Saint and Matthay, 1998); or administration of small amounts used to flush intravascular catheters (Baughman et al, 1993). Patients may also be exposed to small amounts of heparin coated onto catheters, such as pulmonary artery catheters (Lee and Warkentin, 2001). Furthermore, heparin plays an important role in the therapy of coronary care patients (Hirsh et al, 2001; Nathan and Schaer, 2001; Breiger et al, 1998), as it is used in the early management and treatment of unstable angina and myocardial infarction, and during invasive cardiac procedures. 1.4.2 Thrombocytopenia resulting from heparin therapy Heparin has been noted to decrease platelet counts by two mechanisms (Spinier and Dager, 2003; Warkentin et al, 1998). First, it is suggested that some patients may develop an 24 early, benign, reversible non-immune decrease in the platelet count during the first few days of therapy (Gollub and Ulin, 1962; Davey and Lander, 1968; Saffle et al, 1980; Schwartz et al, 1985). This phenomenon is referred to as heparin-associated thrombocytopenia (HAT) or type I HIT (Warkentin and Barkin, 1999; Warkentin et al, 1998). In contrast, patients developing a delayed (late) thrombocytopenia, usually after 5 or more days of heparin therapy, manifest an increased risk for limb- and life-threatening thromboembolic complications. These patients have a drug-induced immune-thrombocytopenia, and this particular disorder that is known as HIT or type II HIT (Hirsh et al, 2001; Warkentin et al, 1998; Chong, 1995). Heparin has been suggested to be a risk factor for thrombocytopenia in reports summarizing potential causes of thrombocytopenia in critically i l l patients (Bogdonoff et al, 1990; Wittels et al, 1990). In the critical care setting, several studies have referred to heparin as a possible risk factor for the development of thrombocytopenia. Baughman et al (1993) noted that heparin was associated with thrombocytopenia following univariate analysis, but not following multivariate linear regression analysis. Bonfiglio et al (1995) reported that heparin exposure/pulmonary artery catheter use was correlated strongly with thrombocytopenia (chi square p < 0.0001) following multivariate analysis. In contrast, Cawley et al (1999) reported that heparin exposure (as drug treatment, prophylaxis, or a flush to indwelling catheters) was not associated with thrombocytopenia. However, these investigators indicated that they might have underestimated the total number of exposed patients due to inconsistencies in nursing practice. In four other studies, investigators did not find heparin to be an independent risk indicator for thrombocytopenia (Hanes et al, 1997; Stephan et al, 1999; Strauss et al, 2002; Shalansky et al, 2002). Therefore, despite the concern in the literature regarding an effect of heparin for thrombocytopenia, the studies done to date do not provide evidence that it is independently associated with thrombocytopenia in critically i l l patients. 25 14.2.1 Heparin-associated thrombocytopenia (HAT) Heparin-associated thrombocytopenia (HAT) is thought to be a non-immune mediated process (AbuRahma et al, 1991; Greinacher 1995; Chong 1995; Chong and Castaldi, 1986), and is associated with a decline in the platelet count within the first few days of heparin therapy (Greinacher, 1995; Ortel, 1998; Home, 2001). The platelet count seldom drops below 100 x 109/L and often returns to normal levels despite continued heparin administration, while patients usually remain asymptomatic. HAT has been estimated to occur in as many as 10% - 30% of patients (Lee and Warkentin, 2001; Wazny and Ariano, 2000; Ansell et al, 1980; Greinacher 1995), and has been suggested to be associated with large intravenous doses of heparin (Chong and Eisbacher, 1998). The mechanism of HAT has been suggested to be related to the mild platelet pro-aggregating effect of the heparin molecule (Chong and Castaldi, 1986; Chong, 1995; Greinacher, 1995). When heparin is administered, it may induce the formation of tiny platelet aggregates and enhance the platelet-aggregating effect of other platelet-aggregating agents, such as adenosine diphosphate, epinephrine, bacteria and/or bacterial products, and immune complexes (Chong, 1995; Greinacher, 1995; Chong and Castaldi, 1986). The decline in platelet count could lead to thrombocytopenia in patients who already have a low platelet count. Warkentin et al (1995) reported that 28% of post-operative hip surgery patients developed early thrombocytopenia (a platelet count < 150 x 109/L) within the first 2 days, and platelet counts rose > 150 x 109/L within 3 days in all patients, despite continued heparin administration. Heparin is used frequently in critically ill patients, and these patients are exposed to many other potential risk factors for thrombocytopenia (Lee and Warkentin, 2001; Warkentin et al, 1998; Strauss et al, 2002). Therefore, it is not clear whether the proposed pharmacologic effect of heparin is responsible for the observed cases of thrombocytopenia. For example, Bonfiglio et al (1995) noted that the combined variable pulmonary artery catheter and heparin was associated with thrombocytopenia, but they could not distinguish whether thrombocytopenia was caused by 26 pulmonary artery catheters or by heparin exposure. Furthermore, Greinacher (1995) suggested that the transient effect of heparin on the platelet count could be explained by the treatment of the underlying disease. As the patient's condition improves, platelets are not likely to be activated by the disease process, and the pro-aggregatory effect of heparin itself may not cause a drop in the platelet count. Thus, it is possible that many cases of thrombocytopenia referred to as HAT (or Type I HIT) are actually due to other causes. 1.4.2.2 Heparin-induced thrombocytopenia (HIT) HIT appears to be relatively uncommon (< 5%), but an important immune-mediated adverse reaction to heparin. Paradoxically, patients who develop HIT are at an increased risk for thrombosis, which can be limb-and life-threatening (Warkentin, 2001; Hirsh et al, 2001; Warkentin et al., 1998; King and Kelton, 1984). The incidence of HIT in most clinical settings is, as yet, unknown (Lee and Warkentin, 2001), and diagnosis of this immune-mediated disorder is challenging. 1.4.2.2.1 Clinical presentation of HIT HIT is generally characterized by the development of thrombocytopenia or a relative platelet count decline, 5 or more days after initiation of heparin therapy (Warkentin, 2001; Warkentin and Kelton, 2001; Chong, 1995; Greinacher, 1995). Thrombocytopenia is the most common clinical effect of HIT, but the definition of thrombocytopenia has evolved over time. A platelet count < 150 x 109/L was used in earlier studies investigating the incidence of HIT (Warkentin et al, 1995; Cummins, 1995; Nguyen et al, 1995; Ansel et al, 1980; Green et al, 1984). Using this standard criterion, absolute thrombocytopenia has been reported to occur in approximately 85% - 90% of HIT patients (Warkentin, 2001). In a "toward consensus" review of HIT, Warkentin et al (1998) suggested that HIT should be suspected when the platelet count 27 decreases by 50% or more, or to below 100 x 109/L occurring on or after day 5 of heparin use. The addition of a relative drop in the platelet count to the definition of HIT was based on previous reports of new thrombotic events and other complications of HIT that occurred even when the platelet count remained above 150 x 109/L (Warkentin, 1996; Hach-Wunderle et al., 1994; Hach-Wunderle et al, 1997; Phelan, 1983). HIT patients who do not meet the standard criterion of thrombocytopenia (i.e. drop in the platelet count < 150 x 109/L on or after 5 days of heparin use) usually have a 30% - 50% (Warkentin, 2001; Greinacher, 1995) or more decline in the platelet count after 5 or more days of heparin administration. This was also reported in a published abstract by Warkentin et al (1995a) who observed that a decline > 50% in the platelet count correlated best with HIT-IgG antibody formation in postoperative orthopedic patients at risk for HIT. However, it is not known whether a decline of > 50% in the platelet count is appropriate for other clinical settings (Warkentin, 2001). It has been suggested that clinicians should have a high index of suspicion when a patient experiences an unexpected large-percentage decline in the platelet count, regardless of whether an arbitrary threshold for "thrombocytopenia" is crossed (Warkentin, 2001). Among patients experiencing HIT, the median platelet count nadir is 50 to 60 x 109/L (Warkentin, 2001; Chong, 1995). In most cases, the platelet count normalizes 7 to 10 days (Greinacher, 1995) following cessation of all heparin therapy, but it may take up to a month to do so in some patients. The characteristic delayed onset, usually occurs between 5 and 10 days after initiation of heparin therapy (Warkentin, 2001; Warkentin and Kelton, 2001; Warkentin et al, 1995), with diminishing risk after day 10 (Warkentin, 2001; Warkentin and Kelton, 2001). However, some patients can develop "rapid-onset HIT", defined as an unexpected decrease in the platelet count that can occur within hours of re-exposure to heparin i f the previous exposure was recent, generally within the previous 100 days (Warkentin, 2001; Warkentin and Kelton, 2001) (see Section 1.4.2.2.3). Rarely, patients develop thrombocytopenia and thrombosis several days after 28 discontinuation of heparin (termed delayed-onset HIT and thrombosis) (Warkentin and Kelton, 2001a; Rice et al, 2002). These patients appear to have very high titres of platelet-activating antibodies and in vitro platelet activation has been observed in some of these patient's sera in the absence of heparin (Warkentin and Kelton, 2001a). 1.4.2.2.2 Incidence of HIT While a number of studies to estimate the incidence of HIT have been reported (Girolami et al, 2003; Warkentin et al, 2000; Rao et al, 1989; Pouplard et al, 1999; Kappers-Klunne et al, 1997; Warkentin, et al, 1995; Ansell et al, 1980; Green et al, 1984; Powers et al, 1984), there are inconsistencies among them in the criteria used to define patients at risk, the clinical signs of HIT, and the laboratory tests used for diagnosis. In earlier studies, the incidence of HIT was reported to be between 0% and 30% in different patient groups (Chong 1992; Schmitt and Adelman 1993; Freedman 1992; Chong 1995; Rao et al, 1989; Ansell et al, 1985; Green et al, 1984; King and Kelton, 1984; Bell et al, 1976), however, few well-designed studies with adequate sample sizes were conducted. The earlier studies used a variety of different clinical criteria for HIT, which resulted in different incidences being reported. In addition, few investigators ruled out other possible causes for the thrombocytopenia. In a critical review and pooled analysis, Schmitt and Adelman (1993) reported that the incidence of HIT in a variety of patient populations was < 3% with intravenous heparin and 0% with subcutaneous heparin. Among other factors, they noted that different criteria used to identify possible HIT patients could produce different estimates of the incidence of this syndrome. In the studies reviewed, most investigators had used a platelet count of either < 150 x 109/L or < 100 x 109/L to identify those with possible HIT. A few investigators suggested that a decline in baseline platelet count (e.g. > 40%) might be a predictor of the more severe HIT-related complications, even if the platelet count remained above 100 x 109/L (Schmitt and Adelman, 1993; Warkentin and Kelton, 29 1991; Ramirez-Lassepas et al., 1984; Miller, 1989). Furthermore, many of these investigators either did not perform laboratory testing or used relatively insensitive and non-specific assays (Lee and Warkentin, 2001). In contrast, the incidence of HIT in more recent studies has been reported to be between 0% and 5% in patients from different clinical settings (Lee and Warkentin, 2001). Warkentin et al (1995) reported a 2.7% incidence of HIT in 332 post-operative hip surgery patients. Patients were diagnosed with HIT if they experienced at least one platelet count below 150 x 109/L after 5 days of treatment with heparin and a positive 14C-serotonin release assay (SRA). Subsequently, Warkentin et al (2000) re-analyzed data from a subset of these patients, using revised clinical criteria for HIT: > 50% decline in the platelet count from the post-operative peak, occurring between days 5 to 14 after surgery, where no other cause for thrombocytopenia was apparent. With these revised criteria, they reported a 4.9% incidence of HIT in these post-operative orthopedic patients. However, it is not clear from this report whether all patients were included who were at risk for HIT, and thus this result may have overestimated the incidence of this syndrome. Pouplard et al (1999) reported a 3.8% incidence of HIT in 157 patients who received unfractionated heparin following heart surgery. Patients were diagnosed with HIT if they experienced at least one platelet count below 100 x 109/L or a decrease in the platelet count of > 40% 8 - 1 0 days after surgery and a positive SRA and heparin-platelet factor (PF) 4 enzyme-linked immunosorbant assay (heparin-PF4 ELISA). However, their estimate may be inaccurate because the investigators did not report the use of a priori clinical diagnostic criteria to identify patients with possible HIT. In a recently published paper, Harbrecht et al (2004) reported a 2.5% incidence of HIT in 200 neurologic patients who received heparin for at least 5 days. According to their study, HIT was clinically suspected when an otherwise unexplained thrombocytopenia of < 120 x 109/L or a platelet count decrease of more than 50% or a new thrombotic event occurred. Patients were diagnosed as having HIT if they met the clinical 30 criteria and tested positive by the heparin-PF4 ELISA. Lower incidences of HIT have been reported by Rao et al (1989) in 193 medical patients (0%), by Kappers-Klunne et al (1997) in 358 cardiac patients (0.3%), by Girolami et al (2003) in 598 medical patients (0.8%), and by Warkentin et al (2000) in 100 cardiac surgery patients (1%). The latter 4 studies suggest that the incidence of HIT is < 1% in groups of medical and surgical patients. Prior to the present investigation, there were no published studies investigating the incidence of HIT in critically i l l patients. However, in a published abstract (van Eps et ai, 2001), the investigators performed a retrospective study and reported that the incidence of HIT in 64 critically i l l patients with multiple organ dysfunction syndrome was at least as high as 4.7%. However, from the abstract it is not clear what clinical criteria they used to identify patients with possible HIT, whether their reported incidence of HIT is for heparin or for heparin and L M W H , and how they selected patients to perform the platelet aggregation test (PAT) for presence of heparin-dependent antibodies. Furthermore, the investigators did not state whether they performed the one- or two-point assay. The two-point assay increases the diagnostic specificity of the PAT (Warkentin and Greinacher, 2001; Warkentin, 2000). Most of the studies noted above were small and thus, the calculated 95% CIs were wide, with limits ranging from a low of 0% to high of 9%. One major limitation of these studies is that they failed to document the number of patients at risk for HIT in their estimation of the incidence. Patients are generally at risk for HIT on or after the fifth day because the antibodies responsible for HIT are usually not detectable earlier (Warkentin, 2001; Warkentin and Kelton, 2001; Warkentin et al, 1998). However, in patients exposed to heparin within the previous 100 days, these antibodies may develop earlier than 5 days after starting heparin and place the patient at risk for developing HIT (Warkentin, 2001; Warkentin and Kelton, 2001). Moreover, there is evidence that the PF4-heparin ELISA and PAT methods have lower specificity than the SRA, and the HIT incidence may have been overestimated in some studies (see Section 1.4.2.2.5). 31 Prior to the present investigation, no studies had been published related to the incidence of HIT in critically i l l patients. The results of the present study have been published, indicating that in ICU/CCU patients the incidence of HIT was 0.39% (95% CI, 0.01% to 2.1%) (Verma et al, 2003). Details of these findings are presented in the results and discussion sections of this thesis. 1.4.2.2.3 Pathophysiology of HIT Roberts et al (1964) first speculated that the etiology of HIT may be immunological in nature. Laboratory evidence implicating an immune mechanism was first provided by Rhodes et al (1973), and since then others have provided further evidence that HIT is immune-mediated (King and Kelton, 1984; Babcock et al., 1976; Trowbridge et al, 1978; Cines et al, 1980; Chong, 1981; Ansell et al, 1980). However, it was not until 1992 that the target antigen, platelet factor 4 (PF4) complexed to heparin, was identified by Amiral et al (1992), and an immune mechanism for HIT was provided (Visentin et al, 1994; Greinacher et al, 1994; Kelton et al, 1994; Amiral etal, 1995). Briefly, PF4, an endogenous tetrameric protein found in platelet a granules, is normally released into blood in small quantities from circulating platelets (Chong, 2003; Aster, 1995). In addition, PF4 can be released from platelets because of weak, non-immunologically mediated platelet activation caused by heparin (Home, 2001). Heparin then binds to PF4 and causes a conformational change in the protein exposing several antigenic epitopes, which triggers an immune reaction and the production of HIT-IgG, and occasionally IgM and IgA, antibodies that are normally present in serum (Suh et al, 1998; Ziporen et al, 1998; Aster; 1995). The interaction of heparin and PF4 to form heparin-PF4 complexes occurs in the presence of stoichiometric concentrations of these 2 entities (27 international units of heparin per milligram of PF4 (Greinacher et al, 1994; Amiral et al, 1995). The HIT-IgG antibodies react with 32 heparin-PF4 complexes on the platelet surface to form immune complexes (Newman and Chong, 2000; Greinacher, 1995; Amiral et al, 1996; Visentin et al, 1994). The immune complex binds via the F c portion of the antibody to Fc-receptors (also called FcyIIa or CD-32 receptors) on the membranes of circulating platelets (Newman and Chong, 2000; Visentin et al, 1994; Kelton et al, 1994). These complexes can cross-link FcyIIa receptors, resulting in platelet activation and aggregation, and activation of blood coagulation pathways (Warkentin et al, 1998; Warkentin, 1999). The release of more PF4 from a granules of activated platelets results in additional heparin binding, which combines with antibody forming more immune complexes. This establishes a cycle of platelet activation and complex formation. This in vivo activation of platelets by IgG antibodies is believed to be the primary explanation for thrombocytopenia in patients who develop HIT (Warkentin, 2002). There is a subgroup of patients who present with clinical signs of HIT, but have no antibody that is specific for the heparin/PF4 antigen complex (Greinacher et al, 1994). Amiral et al (1996a) found that 2 chemokines, interleukin-8 (IL-8) and neutrophil-activating-peptide-2 (NAP-2) are structurally related to PF4. They act as antigens, possessing a heparin binding site, which can bind with some heparin-dependent antibodies (Hirsh et al, 1998), but do so independently of heparin. The thrombotic complications that occur in patients with HIT are likely related to events influenced by the heparin-dependent antibodies. First, in vivo platelet activation releases additional PF4 (Aster, 1995), and this endogenous protein, when in excess of the amount that is bound by available heparin, can bind to glycosaminoglycans (e.g. heparan sulfate) on the surface of endothelial cells. HIT-IgG antibodies can bind to these PF4-glycosaminoglycan complexes resulting in damage to the endothelium causing release of tissue factor and other proaggregatory stimuli, thus resulting in thrombosis (Greinacher et al, 1994; Visentin et al, 1994). Second, platelet activation by HIT-IgG causes alterations to the platelet membrane exposing procoagulant 33 factors (Denomme, 2001), and the release procoagulant platelet derived microparticles, which in turn promote thrombin generation, contributing to a hypercoagulable state (Warkentin et al., 1994). Third, there is evidence that antibodies can bind to heparin-PF4 complexes on monocytes, resulting in the synthesis of tissue factor, which promotes thrombin generation (Pouplard et al., 2001; Arepally and Mayer, 2001). Finally, excess PF4 released from activated platelets binds to heparin and reduces its anticoagulant effect (heparin resistance) (Warkentin et al., 1998). Collectively, these events result in a procoagulant state, which is demonstrated by markedly elevated levels of thrombin-antithrombin complexes in HIT patients (Warkentin et al., 1997; Greinacher et al., 2000). 1.4.2.2.4 Clinical outcomes associated with H I T HIT is different than the other immune-mediated thrombocytopenias in that bleeding is uncommon, despite parenteral anticoagulation and a low platelet count (Chong, 1995; Greinacher, 1995). Patients are at increased risk of developing thromboembolic complications, referred to as "white clot syndrome" (AbuRahma et al, 1991; Towne et al, 1979), which is a result of a platelet rich thrombus that is distinct from the thrombus the heparin was intended to treat or prevent (Warkentin et al, 1998; Aster, 1995). The clot mostly consists of fibrin and platelet aggregates and, thus, has a characteristic white appearance. The new thrombus may occur prior to the onset of thrombocytopenia, while the platelet count is decreasing (Hirsh et al, 2001; Warkentin et al, 1995). Among patients who develop HIT, the frequency of thrombotic events is reported to be 50% to 75% (Hirsh et al, 2001). In a large, controlled study of 665 postoperative orthopedic patients, Warkentin et al (1995) reported that 8 of the 9 heparin treated patients who developed HIT had one or more thrombotic events, as compared to 117 of 656 heparin and L M W H treated patients without HIT (88.9% vs. 17.8%; odds ratio, 36.9; 95% confidence interval, 4.8 to 1638; p 34 < 0.001). The authors concluded that HIT is independently associated with thrombosis, even in patients (post-operative orthopedic patients) at high "baseline" risk for developing thrombosis. Similarly, in a retrospective study involving 108 patients with HIT (thrombocytopenia and a positive PAT), Nand et al (1997) reported that 32 of 108 (29%) patients diagnosed with HIT developed thrombotic complications. The researchers reported that the HIT patients who developed thrombotic complications, as compared those who did not, had more severe thrombocytopenia, developed thrombocytopenia earlier, and were older. Multivariate analysis demonstrated that severity of thrombocytopenia and an early decline in platelet count were independent risk factors for development of thrombotic complications. HIT patients may develop arterial and/or venous thrombosis, however, venous thrombotic complications, especially proximal deep vein thrombosis (DVT), and pulmonary embolism, are most common (Warkentin and Kelton, 1996, Nand et al, 1997; Warkentin et al., 1995). Other reported venous thrombotic events include adrenal hemorrhagic infarction (Warkentin and Kelton, 1996; Ernest and Fisher, 1991), cerebral vein thrombosis (Warkentin et al., 1995), and warfarin-associated venous limb gangrene (Warkentin et al., 1997). The latter event usually is characterized by the progression of DVT to peripheral necrosis in patients who develop a supratherapeutic INR (> 4.0) during treatment with warfarin, resulting from a severe reduction in protein C. Arterial thrombosis most commonly affects the large lower limbs and cerebral vessels (Warkentin, 2001), and complications include acute limb ischemia with absent pulses, thrombotic stroke, and myocardial infarction (Warkentin et al., 1998). The thrombotic events have been suggested to occur at sites of preexisting pathology, such as diseased or injured vessels (e.g. intravascular catheter use) (Boshkov et ai., 1993; Singer et al., 1993; Makhoul et al., 1986; Hong et al, 2003). Without treatment, the mortality in HIT patients with new thromboembolic complications is about 20% to 30%, with morbidity caused equally by arterial and venous thrombosis (Chong, 1995; Warkentin et al., 1997). 35 Other important sequelae associated with HIT have been reported (Warkentin, 2001; Warkentin et al., 1998). Acute systemic reactions can be a life-threatening complication (Warkentin et al., 1992a). This can occur within thirty minutes following administration of intravenous heparin bolus to sensitized patients (Warkentin et al., 1998), and clinical signs and symptoms include fever and chills, hypertension, flushing, tachycardia, dyspnea and transient global amnesia (Warkentin et al., 1994a; Warkentin, 2001; Warkentin, 2003). Some patients may present with localized skin reactions, such as erythematous plaques or skin necrosis, five or more days after initiating subcutaneous heparin therapy (Warkentin, 1996; Boshkov et al., 1993; Warkentin, 1997; Platell and Tan, 1986). Patients may not exhibit thrombocytopenia, but a mild to moderate decrease in the platelet count is observed when the skin lesions appear (Warkentin, 1996; Platell and Tan, 1986). These patients are generally at risk for thrombosis or other adverse events associated with HIT (Warkentin, 1996). Hemorrhage can occur in HIT patients, but it is less common than the thromboembolic complications (Warkentin and Barkin, 1999; Brieger et al., 1998). However, Walls et al (1992) reported that patients undergoing cardiopulmonary bypass surgery are more likely to have hemorrhagic manifestations following the development of HIT. Disseminated intravascular coagulation (DIC) has been reported (Boskov et al, 1993; Chong, 1992; Chong and Berndt, 1989; Bell et al, 1976; Klein and Bell, 1974) and is estimated to occur in approximately 5 to 10% of patients who develop HIT (Warkentin, 2001). 1.4.2.2.5 Laboratory tests for HIT When HIT is suspected, a laboratory test should be performed to support or "confirm" the diagnosis. The goal of a laboratory test should be to identify patients with high probability of HIT, while minimizing the number of false negatives. Support of a clinical diagnosis of HIT has important implications concerning decisions about continuing or discontinuing heparin for the 36 acute event of HIT, as well as for future heparin therapy. Two main classes of assays have been developed to indirectly or directly detect heparin-dependent antibodies: activation (functional) and antigen assays (Warkentin and Greinacher, 2001; Warkentin and Heddle, 2003). Activation assays imply the presence of heparin-dependent antibodies by detecting substances released from platelets or by detecting platelet aggregation following heparin-dependent platelet activation. On the other hand, antigen assays directly detect heparin-dependent antibodies based upon their reactivity with PF4 complexed to heparin or other polyanions. Activation assays A variety of platelet activation endpoints, using washed platelets, can be used, including release of radioactive serotonin (Sheridan et al., 1986; Warkentin, 2000; Warkentin et al., 1992), visual evidence of platelet aggregation (Greinacher et al., 1994a; Greinacher et al., 1991), luminography to detect release of adenosine triphosphate (ATP) (Stewart et al, 1995), or generation of platelet-derived microparticles, which are detected by flow cytometry (Warkentin et al., 1994; Lee et al., 1996). However, investigators consider the SRA is to be the reference standard (Warkentin et ah, 1998; Warkentin and Greinacher, 2001; Morewood, 2000), in which release of 14C-serotonin is used to assess platelet activation by heparin-dependent antibodies (Sheridan et al., 1986). Based on best available evidence, the sensitivity of the SRA has been estimated to be 90% - 98% in post-operative orthopedic and cardiac, and general medical patients with clinical evidence of HIT (Warkentin, 2002a; Walenga et al, 1999; Warkentin et al, 2000; Warkentin and Greinacher, 2001; Pouplard et al, 1999a). The apparent specificity of this assay is also high, and has been estimated to be 80% - 97% in similar patient groups (Warkentin and Greinacher, 2001; Warkentin et al, 2000; Chong et al, 1993; Sheridan et al, 1986). Another activation test, the heparin-induced platelet activation assay (HIPA) 37 (Greinacher et al., 1994a; Greinacher et al., 1991), has been reported to have sensitivity and specificity similar to the SRA (Leo and Winteroll, 2003; Warkentin, 2002); however, it is only used in reference laboratories in Europe. The PAT is another activation test (Greinacher et al., 1994a; Chong et al., 1993), but its sensitivity appears to be lower than the SRA (Chong et al, 1993; Greinacher et al, 1994a; Warkentin, 2002; Leo and Winteroll, 2003; Favaloro et al., 1992). It does not appear that the HIPA and PAT have been rigorously tested against the SRA for the diagnosis of HIT. The disadvantages of the SRA include: requirement of technical expertise, as this test is technically demanding and labour-intensive and is performed by relatively few reference laboratories (Warkentin and Greinacher, 2001; Eichler et al., 1998); requirement for radioactive markers; expensive to perform (Chong et al., 1993; Greinacher et al., 1991); 5% of patient samples give an indeterminate result (i.e. platelet activation occurs at all heparin test concentrations) due to IgG immune complexes generated ex vivo during the heat inactivation process (Warkentin and Greinacher, 2001); and it cannot be used to make rapid clinical decisions. Antigen assays Identification of the target antigen (heparin-PF4) for heparin-dependent antibodies resulted in development of PF4-heparin ELISA. Two commercial PF4-dependent antigen assays are available that detect antibodies against PF4 bound either to heparin (Asserachrom®, Stago, France) or to polyanion polyvinyl sulfonate (GTI-PF4®, GTI, Brooksfield, WI) (Warkentin and Greinacher, 2001; Warkentin and Heddle, 2003). The former assay uses recombinant PF4, whereas the latter assay uses PF4 from outdated platelets (Warkentin and Heddle, 2003). Both assays can detect IgG, IgA, and IgM antibodies. Prior to the present investigation, use of the PF4-heparin ELISA in diagnosing HIT had 38 not been studied in detail. Based on the limited evidence available, the sensitivity of the heparin-PF4 ELISA appears to be similar to that of the SRA, > 90% (Warkentin and Greinacher, 2001; Warkentin and Heddle, 2003; Warkentin, 2002; Warkentin et al, 1998; Arepally et al, 1995); however, the heparin-PF4 ELISA appears to have a lower specificity than the SRA, with estimates reported to be 50% - 93% (Warkentin, 2002; Warkentin et al, 1998; Arepally et al, 1995; Bauer et al, 1997; Warkentin et al, 2000; Pouplard et al, 1999a). The wide range reported may reflect the fact that heparin-dependent antibodies occur more often than HIT, and the frequency of heparin-dependent antibody formation varies among different patient populations (Warkentin et al, 1998; Lee and Warkentin, 2001; Warkentin et al, 2000). Other antigen assays that are performed at specific research laboratories include in-house PF4-heparin ELISA, the fluid-phase ELISA, and the particle gel immunoassay (Warkentin and Greinacher, 2001; Warkentin and Heddle, 2003). There are several advantages of the PF4-heparin ELISA: 1) they are commercially available, 2) they are relatively easy to perform, 3) they appear to demonstrate similar sensitivities to the SRA, 4) they can provide test results more rapidly than with the SRA, and 5) they can be performed in most laboratories with plate readers for ELISA (Warkentin and Heddle, 2003; Visentin et al, 1994; Arepally et al, 1995; Kelton and Warkentin, 1995; Elalamy et al, 1996). Disadvantages of the PF4-heparin ELISA include its' lower specificity as compared to the SRA (Kelton and Warkentin, 1995; Arepally et al, 1995), its' inability to detect antibodies that may react with antigens other than heparin-PF4 (Amiral et al, 1996a), and its' inability to assess cross-reactivity of HIT plasma or sera for other anticoagulant glycosaminoglycans (Warkentin and Barkin, 1999). Despite its possible disadvantages, further research is warranted to evaluate the clinical utility of the PF4-heparin ELISA in different patient populations because the SRA is not widely available. 39 1.4.2.2.6 Diagnosis of HIT In certain clinical settings (e.g. critical care), patients may experience a dramatic decline in platelet count due to many potential causes. Unless the clinician is very aware of HIT, the diagnosis can easily be overlooked. HIT has been designated a clinicopathologic syndrome because to make a diagnosis, clinicians should consider both clinical and laboratory (pathologic) evidence (Warkentin et al,. 1998; Warkentin, 2001; Warkentin, 2001). Clinical evidence should include either an absolute or relative thrombocytopenia, occurring within an appropriate timeframe (i.e. the decrease in platelet count is observed after four days of heparin therapy or within hours or days of heparin exposure in patients who have received heparin therapy in the recent past (100 days)) (Warkentin and Kelton, 2001; Warkentin, 2001; Warkentin et al, 1998), with or without one or more known clinical sequelae of HIT (e.g. thrombosis, acute systemic reactions; skin reactions) (Warkentin, 2001; Warkentin et al, 1998; Chong, 1995). In addition, all potential causes for the thrombocytopenia, such as DIC, drug-induced thrombocytopenia, sepsis, hemodilution, and primary bone marrow disease should be ruled out. Laboratory evidence to "confirm" the diagnosis should include sensitive and specific diagnostic tests to detect the presence of HIT antibodies. The SRA is considered by many investigators to be the reference or gold standard (Warkentin, 2001; Warkentin et al, 1998; Morewood, 2000). However, the diagnosis of HIT must often be made prior to the availability of laboratory test results and hence, clinicians should be aware of the entire clinical picture (platelet counts, timing, and clinical events). In some patient populations, the diagnosis of HIT is not particularly difficult because thrombocytopenia may be infrequent and it is relatively easy to rule out other potential causes for it. However, in the critical care setting, thrombocytopenia occurs commonly, heparin is administered to most patients, and it is often difficult to rule out the numerous other potential causes for the thrombocytopenia. Thus, the diagnosis of HIT is a clinical challenge in critical 40 care patients. 1.4.2.2.7 Management of patients with HIT Immume-mediated HIT can be viewed as a syndrome of in vivo thrombin generation, therefore, it is recommended that therapy include: 1) discontinuation of heparin in order to interrupt the immune process; 2) rapid reduction of increased thrombin generation; and 3) treatment of HIT-associated thrombosis (Greinacher and Warkentin, 2001). Pharmacotherapy in most patients with HIT involves an agent that directly controls thrombin generation, and in some situations additional treatment approaches may be necessary (e.g. medical thrombolysis, surgical thromboembolectomy) (Greinacher and Warkentin, 2001; Warkentin et al., 1998). However, there is very little evidence for an optimal treatment approach for HIT, as no randomized trials without important methodological limitations have been published. Recommendations for treatment of HIT are based on observational studies that include prospective cohort treatment studies with historical controls (Greinacher et al., 1999; Greinacher et al., 1999a; Lewis et al., 2001), case-control series (Warkentin et al., 1997), and large case series (Magnani, 1993; Magnani, 1997; Warkentin and Kelton, 1996; Wallis etal, 1999). Discontinuation of heparin A number of case reports have described the occurrence of new, progressive, or recurrent thromboembolic events in patients with HIT during continued or repeated use of heparin (Greinacher and Warkentin, 2001). Thus, it is recommended that all sources of heparin (e.g. therapeutic intravenous or prophylaxis subcutaneous heparin, heparin flushes) be discontinued in patients meeting the clinical criteria for HIT and after all other possible causes for thrombocytopenia have been ruled out, until the results of diagnostic tests. However, discontinuation of heparin alone without initiating anticoagulation therapy has 41 been shown to be inadequate in patients diagnosed with HIT, as these patients are at high risk for thromboembolic complications (50%) in the week after its discontinuation (Greinacher and Warkentin, 2001; Wallis et al, 1999; Boon et al, 1994; Greinacher et al, 1999a; Warkentin and Kelton, 1996). Furthermore, for HIT patients with thrombosis in whom heparin administration is discontinued, the risk for successive thrombosis is also high (Greinacher and Warkentin, 2001; Greinacher et al, 1999a). In a meta-analysis of two historically controlled cohort studies in mixed patient populations, Greinacher et al (2000) reported that withholding treatment while awaiting laboratory results for heparin dependent antibodies (after stopping all heparin) was associated with an incidence of thrombotic events of 6.1% per patient day. Greinacher and Warkentin, (2001) recommend that alternative therapy should not be delayed while awaiting the results of diagnostic tests i f the patient is clinically suspected of having HIT. Anticoagulants for the management of HIT Certain anticoagulants should not be used in the management of HIT (Greinacher and Warkentin, 2001; Warkentin et al, 1998; Warkentin, 2000a). HIT has been reported to occur less often in post-operative patients treated with L M W H than in those treated with unfractionated heparin (Walenga et al, 2004; Warkentin et al, 2000; Warkentin et al, 1995). However L M W H is not recommended to manage HIT caused by heparin (Greinacher and Warkentin, 2001; Warkentin et al, 1998) because L M W H is virtually 100% cross-reactive with heparin-dependent antibodies in vitro (Greinacher et al, 1992; Warkentin, 2001a), and there have been reports of apparent failure (Warkentin, 1997a). Warfarin monotherapy is also not recommended during the first few days for acute management of HIT, as the full anticoagulation effect of this agent does not occur for approximately 4 - 5 days (Greinacher and Warkentin, 2001; Warkentin et al, 1998). Moreover, warfarin administration can potentially result in a syndrome known as venous limb gangrene (disturbances in procoagulant/anticoagulant balance), in which an 42 otherwise unremarkable DVT can develop to peripheral tissue necrosis requiring limb amputation (Greinacher and Warkentin, 2001; Warkentin et al, 1998; Warkentin et al., 1997; Warkentin et al., 1996a). Ancrod, a defibrinogenating snake venom, is another agent that has been used to treat HIT (Warkentin et al, 1998; Demers et al, 1991). Its thrombin-like action produces a rapid decrease in plasma fibrinogen concentrations. However, it is not recommended for the management of HIT for many reasons (Warkentin et al., 1998; Greinacher and Warkentin, 2001; Warkentin, 2000a; Warkentin et al., 1997), most important of which is that it does not block thrombin generation or microthrombus formation in HIT patients (Warkentin, 1998). It has been recommended that treatment of HIT focus on agents that will rapidly control thrombin generation (Warkentin and Greinacher, 2001; Warkentin et al., 1998; Warkentin, 2000a; Hirsh et al., 2001). Agents that are licensed and available include danaparoid (Orgaran®), which reduces thrombin generation by inhibiting factor Xa; and two direct acting thrombin inhibitors, lepirudin (Refludan®) and argatroban (Argatroban®, Novastan®) (Pravinkumar and Webster, 2003). Therapeutic-dose anticoagulation is recommended when treating HIT patients with any of these agents, and treatment should be initiated immediately after heparin cessation (Greinacher and Warkentin, 2001). While there is evidence that these three agents decrease new thrombotic events in patients with HIT (Greinacher et al., 1999; Greinacher et al., 1999a; Greinacher et al., 2000; Lewis et al, 2001; Magnani, 1993; Magnani, 1997), there are no prospective comparative studies of danaparoid, lepirudin, and argatroban, thus precluding specific recommendations about their relative effectiveness and safety. There are other anticoagulant agents that can reduce thrombin generation, bilvalirudin (Hirulog®) and desirudin (ReVase®) (Pravinkumar and Webster, 2003; Greinacher and Warkentin, 2001). However, there is only anecdotal evidence for managing HIT patients, and these two agents are currently not licensed for use in HIT. 43 Adjunctive Therapies The following therapies have been used and may benefit select patients with HIT (Pravinkumar and Webster, 2003; Greinacher and Warkentin, 2001; Warkentin et al., 1998): medical thrombolysis, surgical thromboembolectomy, intravenous gammaglobulin, plasmapheresis, and antiplatelet agents (dextran, ASA, dipyridamole, platelet glycoprotein (GP) Ilb/IIIa inhibitors) 1.4.3 Part B: Rationale and study goal No prospective studies on HIT incidence in the critical care setting had been performed prior to this investigation. HIT presents a diagnostic and therapeutic challenge to clinicians in this setting because heparin use and thrombocytopenia (i.e. non-HIT thrombocytopenia) are very common, and there are many other potential causative factors for thrombocytopenia. Thus, diagnosis of HIT is difficult because it is often hard to rule out other causes of thrombocytopenia in patients who meet the clinical criteria for HIT. Furthermore, diagnostic tests for HIT had not been evaluated in this patient population. The SRA is considered by many investigators to be the reference standard, however, this assay is not available at most institutions, and thus, cannot be used to make clinical decisions. The heparin-PF4 ELISA is relatively easy to perform and is commercially available, yet it had not been evaluated in the critical care setting and thus, its predictive performance was not known. The results of the HIT investigation reported herein have been published (Verma et al., 2003). Details of the findings are presented in the results section and considered in the discussion section of this thesis. The overall goal of Part B of the thesis was to estimate the incidence and evaluate the predictive performance the heparin-PF4 ELISA for the diagnosis of HIT in a cohort of ICU/CCU patients. 44 1.5 PART A and B: STUDY OBJECTIVES Part A: The specific objectives of the thesis were: 1) to estimate the incidence of thrombocytopenia in patients admitted to a community-based I C U / C C U , 2) to identify explanatory variables for the development of thrombocytopenia at admission and post-admission, and 3) to evaluate the performance of logistic regression models (i.e. by internal and external validation procedures) generated for thrombocytopenia. Part B: The specific objectives of the thesis were: 1) to estimate the incidence of HIT in critical care patients based on explicit clinical criteria and positive S R A (reference standard), and 2) to evaluate the predictive performance of a commercially available heparin-PF4 E L I S A for the diagnosis of HIT 45 METHODS 2.1 OVERVIEW OF RESEARCH This two year prospective, observational study consisted of two components. The first was to develop and validate models for thrombocytopenia in critical care patients using logistic regression analysis and internal and external validation methods. The second component was to estimate the incidence HIT and to evaluate the predictive performance of a commercially available diagnostic test, the heparin-PF4 ELISA, in critical care patients. The data for model development and internal validation were collected over two years at Lions Gate Hospital (LGH), and the data for external validation of the admission models were obtained from databases and chart reviews at St. Paul's Hospital (SPH). The data for the HIT component of this research were collected over two years from ICU/CCU patients admitted to L G H . The methods and statistical analyses related to model development for potential risk indicators associated with thrombocytopenia described herein are similar to those described in a preliminary study to the present investigation (Verma, 2000; Shalansky et al., 2002). Furthermore, the methods and statistical analysis pertaining to HIT in critical care patients have been cited in a recent publication (Verma et al., 2003). 46 PART A: DEVELOPMENT AND VALIDATION OF LOGISTIC REGRESSION MODELS FOR THROMBOCYTOPENIA 2.2 MODEL DEVELOPMENT FOR POTENTIAL RISK INDICATORS ASSOCIATED WITH THROMBOCYTOPENIA 2.2.1 Study design This was a prospective, observational, study. A database of patient characteristics relating to risk indicators for thrombocytopenia was maintained for patients meeting entry criteria. Risk indicators identified a priori, based on published information, were analyzed using multivariate logistic regression. 2.2.2 Study setting This study was conducted over two years, June 1997 to June 1999. Data were obtained from the L G H ICU/CCU, a non-teaching community-based ICU/CCU staffed with rotating intensivists and cardiologists. L G H is a 350 bed community-based hospital located in North Vancouver, British Columbia with an 11 bed ICU/CCU (6 ICU beds and 5 C C U beds) that admits all patients in the hospital who require mechanical ventilation, as well as any patients considered to have hemodynamic or respiratory instability to the extent that critical care monitoring is warranted. At the time of the study, approximately 975 patients were admitted to the ICU/CCU at L G H each year. 2.2.3 Patient selection The target population for this study included all patients over the age of 18 years admitted to the L G H ICU/CCU who had two or more platelet counts recorded, at least 12 hours 47 apart, following admission to the critical care unit. A l l patients were included unless they met any of the exclusion criteria: 1) a platelet count less than 100 x 109/L upon admission to the unit; 2) repeat admission to the unit; 3) concomitant participation in another study; 4) hereditary or congenital thrombocytopenia; 5) evidence of hypersplenism; 6) presence of mechanical heart valve; and 7) DIC, ITP or TTP at admission. The last three exclusion criteria refer to disease states that are known to be associated with thrombocytopenia. However, patients who developed DIC, ITP, or TTP during their stay in the ICU/CCU were to be included in the study. Patients discharged from the ICU/CCU to the ward who were then readmitted to the ICU/CCU within 24 hours of their discharge continued to have their data collected and recorded as though they had not left the ICU/CCU. A l l a priori documented risk indicators (laboratory values, medications, and procedures) that occurred on the ward were recorded. Patients discharged from the ICU/CCU to a ward and who were then readmitted to the ICU/CCU after 24 hours were considered as re-admissions and had only their first admission data included in the study. 2.2.4 Ethics approval The study protocol was approved by the Lions Gate Hospital Research Committee and the Clinical Research Ethics Board at U B C . The Certificates of Approval are attached (Appendices 1 and 2). The Research Ethics Board did not require patient consent because patient care was not influenced by the study data, data were collected from sources for study purposes only, and confidentiality was maintained. 2.2.5 Sample size for risk indicators associated with the development of thrombocytopenia Based on an expected incidence of thrombocytopenia (platelet count < 150 x 109/L) of 48 20%, as observed in previous studies involving critically i l l patients (Bonfiglio et al., 1995; Baughman et al., 1993), to obtain an exact 95% confidence interval (CI) of ± 3%, a sample of 670 (rounded to 700) patients was deemed to be sufficient (Table 3) (Number Cruncher Statistical Software (NCSS®, Kaysville, Utah)). 2.2.6 Data collection Clinical data were collected daily, prospectively for all study patients using a specifically prepared data collection form (Appendix 3). Specific definitions for each variable or group of variables were developed a priori for use in categorizing patient data. Data were collected from information routinely recorded during the course of patient care and from specimens drawn as part of usual therapeutic intervention or routine care. Eligible patients were followed prospectively during their stay in the ICU/CCU until discharge or death. Information unobtainable during daily data collection (e.g. total duration of hospital stay) was collected retrospectively, approximately six to eight weeks after discharge from the ICU/CCU. This was accomplished by reviewing the patient's chart in Medical Records. A l l medical charts were complete, so that there were no missing data. A l l data were recorded in a manner that ensured patient confidentiality. 2.2.6.1 Data management A l l completed data collection forms were coded and entered into a database in SPSS® 9.0 by the author to ensure quality and consistency of coding and data entry. Data were reviewed and verified in SPSS® 9.0. A l l entries that the author found ambiguous or problematic were queried and re-checked. 49 Table 3 Sample size estimation* Estimated Incidence 95% CI 95% CI 95% CI 95% CI 95% CI "'• . .of . • ± 1% ±2% ± 3 % ±5% ± 10% Thrombocytopenia 10% 3410 834 370 134 -15% 4863 1206 523 185 40 20% 6105 1505 670 232 57 25% 7154 1782 786 284 66 30% 8055 1982 870 315 70 * The table summarizes sample sizes required to estimate the incidence of thrombocytopenia in critically ill patients with specified confidence intervals. Calculations were based on a two-sided alpha of 0.05 (exact 95% CI). 50 2.2.7 Criteria for thrombocytopenia The primary outcome variable of the study was the development of thrombocytopenia at any time during a patient's stay in the ICU/CCU. In keeping with criteria recognized clinically, thrombocytopenia was defined in two ways. First, two or more consecutive platelet counts (more than 12 hours apart) below 150 x 109/L (the lower limit of normal recognized by the laboratory at LGH) (Warkentin and Kelton, 2000; Davis, 1998; Bessman, 1989; Handin, 2001; Lind, 1995; Sultan, 1985). This was also used as the criterion for HIT at the time the present investigation was designed (Warkentin et al., 1995). Second, one platelet count below 100 x 109/L. This criterion for thrombocytopenia had been used in other studies performed in critical care patients, and it has also been suggested to be the threshold at which the risk of bleeding increases in particular groups of patients (Williams et al, 1995a; Bithell, 1993a; Wittels et ah, 1990), especially those with recent surgery, impending surgery, a recent bleed, or anticoagulation therapy. The admission platelet count and time to occurrence of thrombocytopenia (from admission to the ICU/CCU) were recorded. 2.2.8 Determination of the platelet count For patients admitted to the ICU/CCU at L G H , admission (baseline) and daily platelet counts, when available, were recorded. Whole blood was collected in the presence of EDTA for platelet counts. Samples were typically analyzed within two hours of collection. At L G H , platelet counts were determined with an electronic (impedance) counter, the Coulter Counter S Plus Model STKR®. The intra-day coefficient of variation for normal platelet counts has been reported to range from 2 to 4 percent for automated counters, and to 11 percent or more for manual counters, such as phase microscopy (Hamilton, 1986; Williams, 1995). The reference range for a normal platelet count had been established for a number of years at L G H (Dr. Wolber M.D. , personal 51 communication, 1998). It was based on a clinically agreed normal range (reference laboratory mean ± 2 SD). Daily quality controls (internal control) were performed on the Coulter Counter. These consisted of a Coulter whole blood reagent control with established normal values for all cell components run once a day, as well as a patient normal whole blood measured four times per day. These should have a coefficient of variation within 10%. An external control provided by the College of American Pathologists was also measured at regular intervals. An internal control was measured in order to estimate the intra- and inter-day platelet count variability. This determination had to be done with a different sample each day due to the degradation of platelets after 8 to 12 hours. To estimate intra-day variability of the assay, blood samples from six patients were each split into four aliquots and platelet counts were measured using each aliquot. The intra-day coefficient of variation (CV) for each patient's sample was determined and used to calculate the mean C V for the group of six patients. To estimate the inter-day variability of the assay, daily blood samples for six days were obtained and analyzed from nine different non-thrombocytopenic patients. The inter-day C V was calculated for each of the nine patients and the mean of the nine patients' CVs was used to estimate the inter-day variability in platelet count. The mean intra-day CV was 2.9%. The range of platelet counts for the six patients over the four days was 98 - 3 86 (x 109/L). For the nine non-thrombocytopenic patients whose blood samples were assessed over six days, the mean inter-day CV of the platelet count was 10.5%. As this determination had to be done with a different sample each day the estimate of inter-day platelet count variability includes intra-patient variability and the variability inherent in the method. 2.2.9 Datasets and models developed at L G H Admission and exploratory post-admission models were developed for the entire cohort 52 of ICU/CCU patients and the subset of patients with an intensive care (ICU) admission diagnosis for the two criteria for thrombocytopenia described above (see 2.2.7). The admission models were generated using data obtained at admission, and exploratory post-admission models were generated using data obtained at admission and those collected during a patient's stay in the unit up to the development of thrombocytopenia or discharge/death. The post-admission models were considered exploratory because logistic regression analysis does not directly model time-dependent data, and the method described herein is therefore, not optimal for the type of data collected after admission (see Section 4.4). There were two datasets used to develop the models. The first dataset consisted of the subset of all ICU/CCU patients with admission platelet counts that were 150 x 109/L or greater (in this subset, the criterion for thrombocytopenia was two or more consecutive platelet counts (more than 12 hours apart) < 150 x 109/L). These patients comprised a subset of those referred to below. Admission and exploratory post-admission models were developed on data obtained from these patients, and on data from the ICU subset of these patients. The second dataset consisted of all ICU/CCU patients with admission platelet counts that were 100 x 109/L or greater (criterion for thrombocytopenia was one platelet count < 100 x 109/L). Admission and exploratory post-admission models were also developed on data obtained from these patients, and on data from the ICU subset of these patients. Thus, there were eight different models generated: M o d e l 1: Admission ICU/CCU for thrombocytopenia criterion < 150 x 109/L M o d e l 1 P A : Exploratory Post-Admission ICU/CCU for thrombocytopenia criterion < 150 x 109/L M o d e l 2: Admission ICU for thrombocytopenia criterion < 150 x 109/L 53 Model 2PA: Exploratory Post-Admission ICU for thrombocytopenia criterion < 150 x 109/L Model 3: Admission ICU/CCU for thrombocytopenia criterion < 100 x 109/L Model 3PA: Exploratory Post-Admission ICU/CCU for thrombocytopenia criterion < 100 x 109/L Model 4: Admission ICU for thrombocytopenia criterion < 100 x 109/L Model 4PA: Exploratory Post-Admission ICU for thrombocytopenia criterion < 100 x 109/L 2.2.10 Demographic and patient characteristics Initial admission evaluations included the recorded age, weight (absolute body weight), height, gender, race, and location patient was admitted from (i.e. emergency department, ward, or other hospital). 2.2.11 Risk indicators for thrombocytopenia A set of risk indicators (variables) previously identified as being associated with thrombocytopenia (Bonfiglio et al, 1995; Baughman et al, 1993; Hanes et al., 1997; Cawley et al., 1999; Stephan et al., 1999), along with other potential risk indicators identified from the literature were followed and recorded in the database. There were 8 and 90 potential risk indicators investigated for the admission and exploratory post-admission models, respectively. In this study, the term risk indicator was used instead of risk factor to identify certain characteristics that are associated with an increased risk of developing thrombocytopenia, 54 because it has not been clearly demonstrated that these characteristics are causally associated with this condition (Kraemer et al., 1997; Last, 1995). Potential risk indicators were documented up to and including the day before the development of thrombocytopenia, or for the entire duration of the ICU/CCU stay for each patient who did not develop thrombocytopenia. Some of the potential risk indicators have not been previously reported to be associated with the development of thrombocytopenia in particular patient groups. However, since these risk indicators are related to severity of illness, exposure to foreign surfaces, or invasive procedures, they were examined for their possible association with the development of thrombocytopenia. For all dichotomous variables, the presence of a dichotomous risk indicator was coded by a "1" and the absence of that risk indicator was coded as a "0". Potential risk indicators were categorized as indicated below. 2.2.11.1 Admission (baseline) risk indicators Patient demographics Age, gender, Acute Physiology and Chronic Health Evaluation (APACHE II) score (Cawley et al., 1999; Stephan et al., 1999), alcohol history, and year admitted to L G H (study performed over two years) were investigated as risk indicators for the development of thrombocytopenia and were documented for each study patient. Age and A P A C H E II score were classified as continuous variables, whereas gender, alcohol history, and year admitted to L G H were classified as dichotomous variables. Race was also coded as a dichotomous variable, based on whether or not patients were Caucasian. The Acute Physiology score (APS) and A P A C H E II are measures of severity of illness and are predictive of outcome (mortality) in critically i l l patients. The APS (Appendix 4) is a component of the A P A C H E II score. It is a weighting system, based on a scale of 0 to 4, used to assess the severity of acute disease (Knaus et al., 1985). The APS is comprised of the sum of the 55 weightings for 12 physiologic measures. It is determined from the worst physiologic value (i.e. greatest deviation from normal (highest or lowest)) during the initial 24 hours after ICU/CCU admission. If a patient was transferred from another hospital or ICU, the APS was determined upon admission to L G H ICU/CCU. If any physiologic measure used for calculating the APS was missing, it was assumed to be normal. The A P A C H E II (Appendix 4) is a clinician-evaluated instrument used to stratify acutely il l patients based on the severity of disease (Knaus et al., 1985). It is based on the premise that the severity of acute disease can be measured quantitatively by assessing the degree of abnormality of various physiologic measures. The A P A C H E II score represents the sum of weights assigned to 12 physiologic measures (APS), to age, and to a value for chronic health problems. Age and severe chronic health problems more or less reflect a patient's diminished physiologic reserve, and thus, they have been incorporated into the A P A C H E II score. Excessive alcohol use has been suggested to be a risk indicator for thrombocytopenia (Bogdonoff et al., 1990). History of moderate or excessive alcohol use was determined when there was evidence of a history of alcoholism or consumption of 3 or more alcohol drinks daily in view of previous reports suggesting that this level of alcohol intake may be associated with thrombocytopenia and platelet dysfunction (Rubin, 1999; Levine et al., 1986; Bogdonoff et al., 1990). In general, the necessary information regardinging previous alcohol and heparin use was obtained from the medical chart, discussions with the attending physician, nurses, or the patient or family (noted in the clinical record). Admission platelet count The admission platelet count was recorded for each patient eligible for the study. If the patient was transferred from a ward or from another hospital to the unit, the admission platelet 56 count was the first one obtained in the ICU/CCU. Medical procedures Surgery that occurred 24 hours prior to ICU/CCU admission was documented. Admission diagnoses Diagnostic categories were based on the most common International Classification of Diseases-9 (ICD-9) (Puckett, 1998) codes reported by the hospital's Medical Records Department for the ICU/CCU over the two-year period immediately prior to beginning data collection. The admission diagnoses for patients were taken from the ICU/CCU nurses' daily monitoring form or from the attending physician's notes in the medical chart, and categorized as outlined below: 1. Nervous System (neurologic) « Included surgical and non-surgical disorders, head trauma, and seizures. 2. Respiratory Surgery • Included all respiratory related surgeries, such as lobectomy, pneumoectomy, mediastinectomy. 3. Respiratory Non-Surgery • Included bronchitis/asthma, chronic obstructive pulmonary disease (COPD), pulmonary edema, respiratory failure, acute respiratory distress syndrome (ARDS), and other respiratory disorders (not including ones just mentioned). 4. Vascular Surgery • Included all vascular surgeries, such as aortic-femoral bypass surgery and aortic abdominal aneurysm surgery. 5. Cardiovascular Non-Surgery 57 • Included heart failure (HF), rhythm disturbances, or other cardiovascular disorders, but not acute myocardial infarction or unstable angina. 6. Acute Myocardial Infarction (AMI) • At the time this research was conducted, the criteria used for A M I were: physician documented A M I coinciding with the appearance of C K - M B (creatine kinase M B isoenzyme) in serum within 3 to 4 hours after A M I and electrocardiogram (ECG) changes such as ST-segment elevation and/or presence of Q waves. 7. Unstable Angina • Physician documented unstable angina coinciding with no C K - M B isoenzyme and E C G changes consistent with myocardial ischemia including non-specific ST-T changes, T-wave inversion, or ST-segment depression. 8. Gastrointestinal (Gl) • Included G l procedures and all G l disorders (including hepatobiliary and pancreatic disorders) except G l bleed. 9. Musculoskeletal and Connective Tissue • Included any trauma, injury or wound to the musculoskeletal system or any disease process that involved connective tissue as determined by the attending physician. 10. Endocrine • Included thyroid crisis or pheochromocytoma. 11. Diabetes Mellitus • Included diabetic ketoacidosis. 12. Kidney, Urinary Tract, and Reproductive Disorders • Included acute or chronic renal failure or any disease process that involved the urinary and reproductive systems as determined by the attending physician. 58 13. Infection (excluding sepsis) • Defined as patients with clinical signs (temp > 38.5 C, WBC > 11 x 10 /L), or those administered antibiotics for the infection. 14. Malignancy 15. Drug Overdose/Poisoning 16. Sepsis • The diagnosis of sepsis was noted when the physician recorded the diagnosis in the chart, there was a documented or suspected infection (a pathologic process induced by a microorganism) (Levy et al, 2003), and when the patient manifested 2 or more of the o o following conditions: temperature greater than 38 C or less than 36 C, respiratory rate greater than 20 breaths/minute, heart rate greater than 90 beats/minute, or partial pressure of carbon dioxide below 32 mm Hg, and white blood cell count greater than 12 x 109/L or less than 4 x 109/L (Bone et al., 1992). Also included in this definition of sepsis were patients with severe sepsis (sepsis associated with organ dysfunction, hypoperfusion, or hypotension) and septic shock (sepsis-induced hypotension along with perfusion abnormalities or organ dysfunction in spite of adequate fluid resuscitation). 17. Gastrointestinal Bleed • Physician documented G l bleed. While the unit does not explicitly distinguish between intensive care (ICU) and coronary care (CCU) admission diagnoses, patients' diagnoses at admission were classified as follows. Patients admitted for an acute myocardial infarction, unstable angina, or cardiovascular non-surgery were classified as C C U patients. Patients whose admission diagnosis was one of the other 14 diagnostic categories were classified as ICU patients. 59 2.2.11.2 Post-admission risk indicators Medications as risk indicators for thrombocytopenia Medications identified in previous studies (Hanes et al., 1997; Bonfiglio et al., 1995; Baughman et al, 1993), review articles (Bogdonoff et al., 1990; Wittels et al., 1990), and/or in hematology (Williams, 1995a) and internal medicine (Handin, 2001a) textbooks as risk factors for thrombocytopenia were selected as potential candidate variables for the study. However, only those medications that were on formulary at L G H were selected for the analysis. Previous heparin use, including any heparin-related products, which the patient was receiving or had received prior to admission to the ICU/CCU, was documented. This was determined by reviewing the patient's medical chart, hospital pharmacy records, the provincial prescription database (PHARMANET®), and interviewing the family physician, patient and/or the family, where possible. Medications previously identified as potential risk indicators for thrombocytopenia and on formulary at L G H included: heparin (unfractionated heparin), vasoactive agents (epinephrine, norepinephrine, dopamine (at dose rates > 2 ug/kg/min), isoproterenol, phenylephrine, and dobutamine), amrinone, auranofm (po), aurothianalate (iv), beta-lactam antibiotics (amoxicillin, ampicillin, cloxacillin, penicillin G, penicillin V , piperacillin, ticarcillin, cefaclor, cefamandole, cefotaxime, cefazolin, ceftazidime, ceftizoxime, ceftriaxone, cefuroxime, cephalexin), imipenem, vancomycin, amakacin, gentamicin, neomycin, tobramycin, antifungal agents (amphotericin B, flucytosine, fluconazole, ketoconazole), antineoplastic agents, ^-antagonists (cimetidine, ranitidine, and famotidine), thiazide (Aldacthiazide (hydrochlorothiazide and spironolactone), chlorthalidone, Dyazide® (hydrochlorothiazide and triamterene), hydrochlorothiazide, metolazone) and loop diuretics (ethacrynic acid, furosemide), phenytoin, salbutamol, ipratropium bromide, quinidine, quinine, sulfonamide derivatives (acetazolamide, trimethoprim-sulfamethoxazole, Azogantrisin® (sulfisoxazole and phenazopyridine), thiazides, furosemide, 60 sulfadiazine, sulfasalazine, sulfinpyrazone, sulfisoxazole, olsalazine, acetohexamide, chlorpropamide, tolbutamide, gliclazide, glyburide), digoxin, methyldopa, enoxaparin, tinzaparin, A S A , and NSAIDS (diclofenac, ibuprofen, indomethacin, ketoprofen, mefenamic acid, naproxen). Single doses of some medications have been reported to be associated with thrombocytopenia (Williams, 1995a). Such cases usually involve re-exposure to the medication, but the initial exposure may have been hours earlier or a gradual exposure over months or years (Bogdonoff et al., 1990; Williams, 1995a). In general, since many patients admitted to an ICU/CCU have had previous hospital admissions, and because many of these patients have previously received drug therapy, the minimum exposure for inclusion was one dose prior to thrombocytopenia. A l l medications, with the exception of heparin, identified a priori as possible risk indicators were recorded as a dichotomous variable. It was anticipated that there would be some medications that patients would not be exposed to, or some medications that few (< 5) patients would be exposed to. Therefore, in such cases, classes of medications were constructed and entered into logistic regression analysis based on similarities in chemical structure and pharmacological action. Each class of medication was analyzed as a dichotomous variable. Medications that one or more patients were exposed to were grouped into the following classes: 1. Penicillin antibiotics (amoxicillin, ampicillin, cloxacillin, penicillin G, pipercillin) 2. Cephalosporins (cefotaxime, cefazolin, ceftazidime, ceftizoxime, ceftriaxone, cefuroxime, cephalexin) 3. Histamine ^ -Antagonists (cimetidine, ranitidine) 4. Inotropes (dobutamine, dopamine, norepinephrine) 61 5. Sulfonamide derivatives (acetazolamide, trimethoprim-sulfamethoxazole, hydrochlorothiazide, dyazide, furosemide, gliclazide, glyburide, metolazone) Candidate medications not included in one of the 5 classes were left as individual medications. Doses, total duration of use, route of administration, medication frequency, and indication for use were only recorded for unfractionated heparin. The possible association of heparin with the development of thrombocytopenia was explored further by analyzing the daily dose of heparin administered to patients as a continuous and as a categorical variable. Daily dose of heparin exposure was categorized as follows: 1) no heparin administered to the patient; 2) doses to maintain IV line and pulmonary artery catheter patency (low dose) (< 1,000 units/day); 3) prophylactic doses (medium dose) (1,000-16,000 units/day); and 4) full anticoagulation (high dose) for thrombosis therapy (> 16,000 units/day). The reference group for interpretation was that in which no heparin was administered. Renal and hepatic dysfunction Changes in renal and hepatic function have been reported to be associated with thrombocytopenia (Bogdonoff et al., 1990; Baughman et al., 1993) and were investigated as risk factors in previous studies, and thus, were included in the analysis. Baseline (admission) and daily serum creatinine concentration, when available, were recorded and creatinine clearance was determined by the modified Cockcroft and Gault equation normalized for weight. (Cockcroft and Gault, 1976; Davis and Chandler, 1996). Estimated creatinine clearance < 30 ml/min/72kg or a 50% drop in creatinine clearance from baseline was considered abnormal. Baseline and daily liver function test results (Crawley et al, 2002; Bonfiglio et al, 1995), when available, were recorded. Aspartate aminotransaminase (AST), alanine aminotransaminase (ALT), and alkaline phosphatase (ALK) were considered elevated when their values exceeded 5 times the 62 upper limit of the normal range (Bonfiglio et al., 1995). Total and direct bilirubin were considered elevated when their values exceeded 3 times the upper limit of the normal range. In addition, INR was also documented, when available, for patients who were not receiving any anticoagulant therapy. Both renal and hepatic dysfunction were classified as dichotomous variables. Medical procedures The following medical procedures were documented either until the development of thrombocytopenia (i.e. one day before the onset of thrombocytopenia) or the end of the ICU/CCU stay when no thrombocytopenia occurred: units of packed red blood cells (PRBC) transfused (Hanes et al., 1997; Baughman et al., 1993), units of fresh frozen plasma (FFP) transfused, surgery (Hanes et al., 1997), pulmonary artery (Swan-Ganz) catheter insertion (Bonfiglio et al., 1995; Bogdonoff et al., 1990), or mechanical ventilation (Crawley et al., 2002). Each procedure was documented and included in the analysis, and was classified as a dichotomous variable. 2.3 DESCRIPTION OF DATA OBTAINED FROM ICU PATIENTS AT ST. PAUL'S HOSPITAL FOR EXTERNAL VALIDATION 2.3.1 Study setting Data for external validation were obtained from the ICU at St. Paul's Hospital (SPH), a tertiary referral hospital affiliated with the University of British Columbia staffed by rotating intensivists, critical care fellows, and medical residents. SPH is a 560 bed inner city hospital located in downtown Vancouver, British Columbia, and it has a 14 bed ICU. 63 2.3.2 Patient selection The inclusion and exclusion criteria were identical to those used for the L G H patients. Specifically, the target population for this study included all patients over the age of 18 years admitted to the SPH ICU during the period of November 1998 to August 2000 unless they met any of the exclusion criteria (see Section 2.2.3). 2.3.3 Ethics approval The study protocol for the external validation component of this research performed at SPH was approved by the UBC/Providence Health Care Research Ethics Board. The Certificate of Approval is attached (Appendix 5). 2.3.4 Data collection Clinical data for specific admission variables were collected retrospectively for all study patients admitted to the unit using the same data collection form as at L G H (Appendix 3). The data were obtained from information routinely recorded at admission and from specimens drawn at admission as part of usual therapeutic intervention or routine care. Since 1998, a computer database has been developed and continuously updated by study nurses to record the progress of patients in the ICU. This database was used to identify patients in this study and to obtain the demographic, diagnostic and procedural data, as well as clinical outcomes. Information was also acquired from the laboratory database, the medical records database, and chart reviews. Specific definitions for each admission variable or group of variables were developed a priori for use in categorizing patient data (Section 2.2.11.1). A l l medical charts were complete so that there were no missing data. A l l data were recorded in a manner that ensured patient confidentiality. 64 2.3.4.1 Data management Initially, all completed data collection forms were coded and entered into a database in SPSS® 9.0 by a pharmacy resident at SPH and verified by the author to ensure quality and consistency of coding and data entry. The pharmacy resident only collected and recorded data for patients who had an admission platelet count > 150 x 109/L and met the other inclusion/exclusion criteria. The author collected and recorded data for patients with an admission platelet count > 100 x 109/L and < 150 x 109/L. A l l entries that the author found ambiguous or problematic were queried and re-checked. 2.3.5 Criteria for thrombocytopenia As described in Section 2.2.7. 2.3.6 Determination of the platelet count For patients admitted to the ICU at SPH, admission (baseline) and daily platelet counts, when available, were obtained from the laboratory database. Whole blood was collected on EDTA for platelet counts. Samples were routinely analyzed within two hours of collection. At SPH, platelet counts were obtained with an electronic (impedance) counter, the Coulter Counter STKS®. No information was obtained regarding the intra- and inter-day CV of platelet counts of this instrument. 2.3.7 Demographic and patient characteristics Admission demographic and patient characteristics obtained for patients admitted to the SPH ICU included the age, gender, race, and location patient was admitted from (emergency department, ward, or other hospital). Variables were coded as described in Section 2.2.10. 65 2.3.8 Risk indicators for thrombocytopenia Data were collected for six admission risk indicators (variables) that were used in the external validation component of this research. Data for alcohol history was not obtained, as information for alcohol history was not recorded in any database, and the variable year admitted to L G H was not applicable to the SPH dataset. Specifically, data were collected for A P A C H E II score, surgery 24 hours prior to admission, age, admission platelet count, gender (male), and admission diagnoses. Information on A P A C H E II score, age, gender (male), and admission diagnoses were obtained from the ICU database. Data for surgery 24 hours prior to admission and admission platelet count were obtained from medical records and the laboratory databases, respectively. Missing data were acquired from chart reviews. The definitions and categories of admission diagnoses for patients admitted to the ICU at SPH were the same as those for patients admitted to the ICU/CCU at L G H (see Section 2.2.11.1). However, patients admitted to the ICU at SPH with an diagnosis of acute myocardial infarction or cardiovascular non-surgery were considered to be intensive care patients, as opposed to coronary care patients as at L G H , because they had an associated or underlying intensive care diagnosis, and most were mechanically ventilated. There is a separate C C U at SPH, however, prior to 2002, C C U patients requiring mechanical ventilation were transferred to the ICU. 2.3.9 Datasets used for external validation Two SPH patient datasets were used to externally validate the four L G H admission models. One SPH dataset consisted of ICU patients with admission platelet counts > 150 x 109/L, and the L G H < 150 x 109/L admission ICU/CCU (Model 1) and ICU (Model 2) models were applied to this dataset. The other SPH dataset consisted of ICU patients with admission platelet counts > 100 x 109/L, and the L G H < 100 x 109/L admission ICU/CCU (Model 3) and 66 ICU (Model 4) models were applied to this dataset. 2.4 STATISTICAL ANALYSIS 2.4.1 Data management A l l analyses were performed using SPSS® 9.0 Professional version. This included all descriptive statistics, univariate procedures, logistic regression analyses, area under ROC curve generation, and internal and external validation procedures. 2.4.2 Potential risk indicators for thrombocytopenia 2.4.2.1 Descriptive analysis Baseline demographic characteristics of the study sample were summarized in terms of the mean and SD for continuous variables and frequencies for dichotomous variables. Continuous data were analyzed using Student's t-test for independent samples. A l l statistical tests were two-sided, and p < 0.05 was considered statistically significant. Exact results for the 95% CIs were obtained from NCSS®, based on the binomial distribution. 2.4.2.2 Logistic regression Logistic regression is based on the principle of regressing a dichotomous dependent variable on a set of independent covariates (risk indicators) (Hosmer and Lemeshow, 2000; Hosmer and Lemeshow, 1989; Van Houwelingen and le Cessie, 1988). In the present study, logistic regression analysis was used to examine the individual and combined relations between the dichotomous (binary) outcome of thrombocytopenia and an a priori list of risk indicators during a patient's stay in the ICU/CCU. The goals of the logistic regression analysis were to 67 identify explanatory risk indicators for thrombocytopenia and obtain a model to predict the probability of developing thrombocytopenia based on the identified set of independent variables (Appendix 6). Coding of independent variables Independent variables were coded either as dichotomous or continuous, as indicated above. Continuous variables with a low frequency of occurrence were recoded as dichotomous variables. For example, units of packed red blood cell (PRBC) and fresh frozen plasma (FFP) transfusions were recoded as "1" i f a patient received a transfusion or "0" i f a patient did not receive a transfusion. In order to provide a clinically meaningful indication of how the odds (risk) for the development of thrombocytopenia changes with the specific continuous variables, A P A C H E II score, admission platelet count, heparin dose/day and age, were expressed as increments of (i.e. changes of) 5 units, 50 x 109/L, 1000 Units, and 5 years, respectively, (Hosmer and Lemeshow, 2000a). Admission diagnoses and heparin dose range were nominal scale variables, and were designated as categorical variables. Individual admission diagnoses were initially categorized as follows. For the models developed on the entire ICU/CCU cohort, a design (dummy) variable was created for each individual intensive care admission diagnosis, and the three coronary care admission diagnoses (acute myocardial infarction, unstable angina, and cardiovascular non-surgery) were grouped together and served as the reference group. Individual intensive care admission diagnoses with < 5% frequency or with no or only one patient who developed thrombocytopenia were combined together to create a new design variable called "other ICU diagnosis". However, admission diagnoses with < 5% frequency and > 1 case of thrombocytopenia that were clinically associated with the outcome, and that were noted to be associated with thrombocytopenia in previous studies (e.g. sepsis, musculoskeletal/connective 68 tissue, gastrointestinal, G l bleed, and vascular surgery admission diagnoses) were not included with the design variable "other ICU diagnosis". For models developed on the subset of ICU patients, the design variable, "other ICU diagnosis", was created in the same way as described above and served as the reference group. The design variables for heparin dosage range are described in Section 2.2.11.2. Univariate analysis For the exploratory post-admission models, univariate analysis was used to reduce the initial variable list by identifying those variables that might be individually associated with thrombocytopenia. For dichotomous variables, univariate analysis was performed with the chi-square test, whereas for continuous variables, analysis involved fitting a univariate logistic regression model to obtain the level of significance by the Wald statistic. Variables were selected as candidates for multivariate logistic regression when a p-value < 0.25 by the appropriate univariate test. Variables that exceeded this criterion, but were thought to be clinically important were also selected. Variables with small numbers in each group (exposure frequency < 5%) were not considered, except in cases where they were either clinically or statistically (i.e. by univariate analysis, p < 0.05, with > 1 case of thrombocytopenia) associated with thrombocytopenia, and were noted to be associated with thrombocytopenia in previous studies. Collinearity between risk indicators Independent variables were considered to demonstrate a high degree of association if they conveyed essentially the same information regarding the risk of thrombocytopenia. The strength of association was indicated by the size of the correlation coefficient. This was done by correlating each independent variable identified by univariate analysis with each of the other 69 variables using the bivariate correlation procedure in SPSS®. Variables were considered to demonstrate collinearity when the Pearson's correlation coefficient was greater than 0.7. The more clinically relevant of two collinear variables was chosen for inclusion in the multivariate analyses. Linearity of continuous variables Continuous variables were assumed to be linear in the logit (log-odds) at the variable selection stage. Once the variable was identified as important by univariate logistic regression (Wald's statistic), the correct scale or parametric relation was determined. This was done using the "design variables based on the quartiles of the distribution" method described by Hosmer and Lemeshow, (2000b). Continuous variables not linear in the logit were subjected to non-linear transformations (e.g. squared, square root, logarithmic, reciprocal). The non-linear transformation that resulted in either an increasing or decreasing linear trend in the estimated coefficients quartiles for the continuous variable was chosen (Hosmer and Lemeshow, 2000b). Method of independent variable entry for multivariate logistic regression Multivariate logistic regression was selected as an appropriate statistical technique, due to the dichotomous nature of the dependent variable (thrombocytopenia), to identify independent associations between each risk indicator and thrombocytopenia after adjusting for the other variables (Appendix 6). The backward stepwise method, based on the likelihood ratio statistic (SPSS® 9.0), was used (Hosmer and Lemeshow, 2000b). Variables were selected for stepwise removal and entry i f they met the following criteria: removal (pout = 0.05) and entry (pj„ = 0.04). Thus, all variables in the final model should have a p-value less than 0.05. The results from the multivariate logistic regression stepwise procedure were also used to further reduce the number of individual intensive care admission diagnostic design variables. 70 This was accomplished iteratively as follows. For the models developed using the ICU/CCU cohort, the individual admission diagnosis design variable with the largest p-value greater than a probability of 0.05 was included in the "other ICU diagnosis" design variable. If the change in the -2LL was not statistically significant (i.e. p > 0.05) (the likelihood ratio test statistic follows a chi-square distribution with 1 degree of freedom) (Appendix 6) and the beta coefficients of the remaining risk indicators did not change markedly, then the individual admission diagnosis remained in the "other ICU diagnosis" category. For models developed using the subset of ICU patients, the individual admission diagnosis design variable with a low odds (OR < 1.00) for the development of thrombocytopenia and the largest p-value greater than a probability of 0.05 was included in the "other ICU diagnosis" design variable. If the change in the -2LL was not statistically significant (i.e. p > 0.05) (the likelihood ratio test statistic follows a chi-square distribution with 1 degree of freedom) and the beta coefficients of the remaining risk indicators did not change markedly, then the individual admission diagnosis remained in the "other ICU diagnosis" category. The procedures just described were repeated for each individual admission diagnosis design variable until the no individual admission diagnosis design variable could be included in the "other ICU diagnosis" category or the individual admission diagnoses left all had a p-value < 0.05. Note, when removing an individual admission diagnosis design variable and including it with the "other ICU diagnosis" category, admission diagnoses selected were those with p > 0.05 and beta coefficients in the same direction (in the present investigation these risk indicators were associated with a lower odds (OR < 1.00) for thrombocytopenia). However, the individual admission diagnosis removed may have been a confounder. To assess this, the fitted models with and without the individual admission diagnosis design variable were examined to assess the changes in the estimated coefficients of the remaining risk indicators as described by Hosmer and Lemeshow, (2000a). 71 The estimated coefficients (maximum likelihood estimates), their standard errors, the Wald test statistic and associated p-value (based on a chi-square distribution), odds ratio [calculated as exp (B)], and 95% confidence intervals around the estimated odds ratio of thrombocytopenia were computed. Lastly, a mathematical expression relating the independent risk indicators to the logit (log odds of developing thrombocytopenia) was expressed. By solving the mathematical expression, the predicted probability of a patient developing thrombocytopenia can be determined (Appendix 6). Interaction terms in the model An interaction between two variables occurs when the effect of one of the variables is not constant over levels of the other. Following selection of the variables for the model, interactions between these variables were evaluated as described by Hosmer and Lemeshow, (2000b). This was accomplished by first creating an interaction term, which involved taking the product of the two variables involved. SPSS® performed this computation automatically. The interaction term was then assessed for its contribution to the model (significance) by using the likelihood ratio test. Decisions on inclusion of interaction terms were based on model statistics, as well as clinical considerations. Interactions with a p-value < 0.05 were deemed statistically significant and were entered into the model. Assessing the fit of the model The likelihood is the probability of the observed results given the parameter estimates. Logistic regression uses -2 log-likelihood (-2LL) as a measure of how well the estimated model fits the data. A reasonable model results in a high likelihood of the observed results and hence, a small - 2 L L (Appendix 6). A calibration plot of the percent (proportion) of patients with thrombocytopenia against 72 the deciles of predicted probability was used to illustrate how well the model described (fit) the observed data. In brief, SPSS® was used to generate deciles of predicted probability. The ascending values of the estimated predicted probabilities were divided into 10 groups, partitioned at the decile values. The mean predicted percent thrombocytopenia (i.e. mean expected proportion) of each decile was estimated. In addition, the observed proportion of patients within each decile was calculated. Observed and mean predicted percent were plotted to generate a calibration curve. The Hosmer-Lemeshow goodness of fit test was used to quantitatively evaluate how well the model described the observed data. This test is based on the null hypothesis that the model is a reasonable fit of the observed data. SPSS was used to generate the Hosmer-Lemeshow goodness of fit test statistic. This test statistic follows a chi-square distribution with eight degrees of freedom, and a p-value greater than 0.05 indicated that the model was a reasonable fit of the data. Sensitivity and specificity of the model The sensitivity, specificity, and overall classification of the model were also used to describe each model, and were obtained from the classification table in the SPSS® logistic regression printout. The sensitivity of a model was the proportion of patients observed to have developed thrombocytopenia that the model correctly predicted to have developed thrombocytopenia. A model's specificity was the proportion of patients who were not observed to have developed thrombocytopenia that the model correctly predicted not to have developed thrombocytopenia. The overall correct classification of a model was the proportion of patients correctly predicted to have and not have developed thrombocytopenia. The reported sensitivity, specificity, and overall classification ofthe model were based on a cutoff or decision threshold of 0.50 (default value in SPSS® 9.0). A plot of the cut-off probability (above which 73 thrombocytopenia is predicted) against the sensitivity and specificity was used to illustrate the trade-off between the sensitivity and specificity of the model at different cut-off probabilities. Regression diagnostics Regression diagnostics were used to examine how well the model described the observed data and the impact of individual covariate patterns in the model (i.e. identifying excessively influential observations, manifested by poorly fitting estimates) (Hosmer and Lemeshow, 2000c). A casewise listing of the values of the following variables was created in SPSS®: predicted probability, residual, studentized residual, leverage value, Cook's distance, and difference in beta coefficients. These measures were investigated in order to identify individual patients who did not fit the model well (outliers) and whose data may have had a strong influence on the coefficient estimates. This gave an indication of how well the model fit the observed data and how sensitive the model was to individual patients' data. The residual, studentized residual, leverage value, Cook's distance, and difference in beta coefficients were plotted against the predicted probability. A l l specific covariate patterns (i.e. cases) identified as outliers or with large values for either Cook's distance or difference in beta (i.e. delta beta) coefficients were examined for individual covariate values and changes in the individual coefficients (Cook's distance and delta betas), as well as changes in the individual coefficients, goodness-of-fit measures, and area under the receiver operating curve (ROC) when these influential covariate patterns were deleted and the model refit. Residual analysis The residual is the difference between the observed probability of thrombocytopenia (in this case "0" or "1") and the predicted probability of thrombocytopenia based on the model. The 74 studentized (deviance) residual was used to identify patients who appeared to be outliers according to their residuals (see below). The studentized (deviance) residual for a particular case is the change in the model deviance (-2 times the difference between log likelihoods of a reduced model and the saturated model (contains as many parameters as there are data points (Hosmer and Lemeshow, 2000; McCullagh and Nelder, 1989))) when that case is excluded (Hosmer and Lemeshow, 2000c; Vollmer, 1996). As a rule of thumb, 99% of the data should be within ± 3 standard deviations (SD) from the mean of the residuals (Draper and Smith, 1981). This rule is based on the approximate normality of the residuals. Leverage plots Leverage was used to identify patients who may have had a covariate pattern (a single set of values for the covariates in the model) that was unusual relative to the rest of the patients. The leverage value is defined as the relative influence of each observation on the model's fit, and the larger the value of this statistic, the more the observation influences that estimate of the regression coefficients. Influence of individual cases The effects of residual analysis and leverage are combined to generate a measure that expresses the influence of each patient on the estimated coefficients (Hosmer and Lemeshow, 1991). These cases have a large effect on the magnitude of the estimated coefficients in the model. Cook's distance and difference in beta coefficients were used to identify influential cases (i.e. to identify those covariate patterns that were poorly fit and those that had a great deal of influence on the values of the estimated coefficients). Cook's distance is defined as a measure of the influence of a case. It was used to approximate the overall change in the estimated coefficients due to the exclusion of the ith patient (Hosmer and Lemeshow, 1991). Difference in 75 beta coefficients was also used to measure influence. It measures the difference in the estimated coefficients for each independent variable when the case is omitted from the model. Hosmer and Lemeshow (2000c) suggest that the "influence diagnostic must be larger than 1.0" for an individual covariate pattern to have an effect on the estimated coefficients. Large values for either Cook's distance or difference in beta coefficients identified cases that were examined further. Evaluation of the proportion of explained variation The proportion of variation (R ) in the dependent variable, thrombocytopenia, explained by the variables in the model was calculated using SPSS®. Two R2 values were given: the Cox & Snell (Rc), which achieves a maximum value of 0.75, and the Nagelkerke (RN), which transforms the Rc so that it has a maximum value of 1 (Mittlbock and Schemper, 1996). The Pearson correlation (r) was also calculated and then squared to describe the association between observed and predicted outcomes in order to examine whether the occurrence of thrombocytopenia (observed outcome) was correlated with the predicted probability of an event occurring. Discriminating potential of the models using a receiver operating characteristic (ROC) curve The discriminating (predictive) ability of the model was assessed by the area under the receiver operating characteristic (ROC) curve, which is also referred to as the c-statistic (Steyerberg et al., 2001; Hosmer and Lemeshow, 2000c). The principles underlying development and interpretation of a ROC curve for this investigation have been described by others (Hanley et al, 1982; Hanley et al, 1983; Metz, 1978; McNeil and Hanley, 1984). The area under the ROC curve or c statistic, was determined from the Dorfman and Alf (1968) maximum likelihood estimation program. It represents the probability of correctly 76 ranking a randomly selected pair of thrombocytopenic and non-thrombocytopenic patients (Hanley and McNeil, 1982). In other words, it is the probability that a patient with thrombocytopenia has a higher predicted probability than a patient without thrombocytopenia. 2.4.3 Evaluation (Validation) of logistic regression models Admission and exploratory post-admission models developed at LGH were validated internally using a bootstrap re-sampling technique. In addition, to demonstrate that right censoring was not an issue when generating models by logistic regression, Cox proportional hazards modeling was performed and risk indicators identified and their associated impact of the hazard for thrombocytopenia were compared to those identified by logistic regression analysis. The 4 admission models generated at LGH were validated externally using data obtained from patients admitted to the ICU at SPH. 2.4.3.1 Internal validation Internal validity of logistic regression models developed on data obtained from ICU/CCU and ICU patients at LGH was determined by using the bootstrap procedure described by Steyerberg et al (2001) with slight modification, as described below. 2.4.3.1.1 Bootstrap procedure Internal validation of the eight admission and exploratory post-admission models generated from data obtained from LGH ICU/CCU and ICU patients was performed using the regular bootstrap technique as described by Steyerberg et al (2001). In brief, 200 random bootstrap samples were drawn with replacement from the original (derivation) LGH dataset consisting of all patients. The syntax used in SPSS® for generating bootstrap samples was a modification of procedure suggested by Rich Herrington and published on the internet 77 (http://wvw.mit.edu/UNT/departm (Appendix 7). Next, a multivariate logistic regression model was generated for each of the 200 bootstrap samples using the method described in Section 2.4.2.2, except the steps described earlier to reduce the number of individual intensive care admission diagnostic design variables by combining them into a single "other ICU diagnosis" category (Sections 2.4.2.2) were not performed when generating a model for each of the 200 bootstrap samples. Instead, the set of individual intensive care admission diagnostic design variables used in the final model in the original L G H dataset were used for generating a model for each of the 200 bootstrap samples. For the exploratory post-admission models, the set of candidate variables following univariate analysis (including the reduced set of individual intensive care admission diagnostic design variables) used in the final four L G H exploratory post-admission models were the same ones used to generate a model for each of the 200 bootstrap samples. Following the development of a model for each of the 200 bootstrap samples, the predictive performance (area under the ROC curve) for each was estimated and this estimate was designated the bootstrap performance. Next, each of the 200 models developed from the bootstrap samples was evaluated in the original (derivation) L G H sample and the predictive performance in the original sample was designated the test performance. The difference between the bootstrap performance and the test performance was an estimate of the optimism. This difference was estimated 200 times (i.e. once for each bootstrap sample). The 200 differences were averaged to obtain an estimate of the optimism (i.e. the extent to which the original model overestimates the area under the ROC). The optimism was subtracted from the predictive performance (area under the ROC) of the original model generated at L G H to provide an estimate of the internally validated performance: Internally validated performance = original model's performance - average optimism [bootstrap performance - test performance] The exact 95% CI around the optimism was calculated by obtaining the standard error of the 200 78 bootstrap differences and estimating the upper and lower 95% confidence limits (Number Cruncher Statistical Software (NCSS®, Kaysville, Utah)). A macro for the entire bootstrap procedure was written for SPSS, with help from Dr. John Spinelli (Appendix 8). 2.4.3.1.2 Cox proportional hazards modeling The steps used to generate a Cox proportional hazards model paralleled those of logistic regression modeling. The same list of risk indicators used for logistic regression were used for Cox regression, however the dependent variable was the logarithm of the relative hazard for the development of thrombocytopenia. SPSS® was used to perform Cox proportional hazards modeling. The method used to develop the model was similar to that for logistic regression (Section 2.4.2.2) except that it models the time to an event, and that the measure of effect is the Hazard Ratio, as opposed to the Odds Ratio in logistic regression. Patients were censored i f they did not develop thrombocytopenia before they were discharged or died in the ICU/CCU. 2.4.3.2 External validation External validation of the four admission logistic regression models developed using the data obtained from ICU/CCU and ICU patients at L G H was performed on retrospective data collected from patients admitted to the ICU at SPH. The L G H < 150 x 109/L admission ICU/CCU and ICU models were applied to the SPH dataset consisting of ICU patients with admission platelet counts > 150 x 109/L. The L G H < 100 x 109/L admission ICU/CCU and ICU models were applied to the SPH dataset consisting of ICU patients with admission platelet counts > 100 x 109/L. The predictive performance (area under the ROC area) and calibration (see Section 2.4.2.2) were assessed. 79 PART B: HEPARIN-INDUCED THROMBOCYTOPENIA 2.5 HEPARIN-INDUCED THROMBOCYTOPENIA IN CRITICAL CARE PATIENTS 2.5.1 Incidence of immune-mediated heparin-induced thrombocytopenia (HIT) 2.5.1.1 Study design This was a prospective, observational study conducted in conjunction with the development and validation of models for thrombocytopenia in critical care patients. 2.5.1.2 Study setting Data for this part of the study were also obtained from the L G H ICU/CCU as described in Section 2.2.2. 2.5.1.3 Patient selection To identify those patients who were potentially at risk for HIT, the study included all patients over the age of 18 years admitted to the L G H 11 bed ICU/CCU between June 1997 to June 1999 who: a) had a platelet count greater than 100 x 109/L upon admission to the unit; b) had received any dose of unfractionated heparin, by any route of administration; c) had at least 2 platelet counts performed, at least 12 hours apart, during their intensive or coronary care admission; d) were not participating in another study in which clinicians were blinded to the use of heparin; and e) were not diagnosed as having disease states that are known to be associated with thrombocytopenia, including hereditary or congenital thrombocytopenia, evidence of hypersplenism, presence of mechanical heart valve, DIC, ITP, or TTP. However, patients who 80 developed hypersplenism, DIC, ITP, or TTP during their stay in the ICU/CCU were to be included in the study. 2.5.1.4 Ethics approval As described in Section 2.2.4. 2.5.1.5 Sample size for the incidence of HIT The main objective for this part of the study was to obtain an estimate of the incidence of HIT in ICU/CCU patients. The sample size estimate was based on an expected incidence of HIT of 2.7% reported from a large study (designed with explicit clinical criteria for HIT, a well-defined patient population, and use of the SRA as the diagnostic test) of post-operative hip surgery patients who received unfractionated heparin for prevention of thrombosis (Warkentin et al., 1995). A sample size of 209 patients at risk for HIT (defined below) would be required to generate an incidence with an exact 95% CI of ± 2.1% of the estimated value (Number Cruncher Statistical Software (NCSS®, Kaysville, Utah)). To account for loss of patients post inclusion, a target sample size of 250 patients sufficiently exposed to heparin to be at risk for HIT was set. 2.5.1.6 Data collection Data collection for this part of the research was as described in Section 2.2.6. 2.5.1.6.1 Data management As described in Section 2.2.6.1. 81 2.5.1.7 Admission diagnostic categories As described in Section 2.2.11.1. 2.5.1.8 Definitions 2.5.1.8.1 Patients at risk for HIT Heparin-dependent antibodies that are responsible for HIT are generally not detectable before the fifth day of heparin administration (Warkentin et al., 1998; Warkentin and Kelton, 2001; Warkentin, 2001); however, patients previously exposed to heparin may present with heparin-dependent antibodies and some may develop HIT earlier than 5 days after starting heparin (the first day of heparin therapy was defined as day 0) (Warkentin and Kelton, 2001; Warkentin, 2001). Therefore, intensive and coronary care patients were defined to be at risk of developing HIT if they had received heparin for 5 or more days, or i f they received any heparin during the index admission after having been exposed to heparin within 8 weeks prior to admission to the unit. 2.5.1.8.2 Clinical criteria for HIT Patients at risk for HIT were considered to have met the clinical criteria for this adverse effect if they developed thrombocytopenia and the timing was sufficient for the patient to have developed heparin-dependent antibodies. The criteria for thrombocytopenia were two or more consecutive platelet counts < 150 x 109/L (lower limit of the normal range at LGH) or > 33% decrease in the platelet count (based on the fact that a relative decline of 30% - 50% was the suggested range at the time the study was initiated (Greinacher, 1995)). For those patients not previously exposed to heparin, the platelet count on the fourth day of heparin administration was 82 the baseline count used to identify those patients who subsequently developed thrombocytopenia. For patients who had been exposed to heparin in the previous 8 weeks, the baseline platelet count was that observed on the day heparin was started in the ICU/CCU. A l l heparin exposure was accounted for, including heparin in pressure bags to maintain patency of arterial lines, Swan-Ganz and central venous catheters. 2.5.1.8.3 Diagnosis of HIT among patients at risk Patients at risk for HIT were considered to have developed HIT if they met the clinical criteria and had a positive assay result using the reference standard, the ,4C-serotonin release assay (SRA). 2.5.1.9 Diagnostic testing for HIT 2.5.1.9.1 Sample selection for the SRA and the heparin-platelet factor 4 enzyme-linked immunosorbent assay (heparin-PF4 ELISA) This was an observational study and informed consent was not required. Diagnostic testing for HIT was performed on plasma or serum samples from suspected HIT patients using the SRA and heparin-PF4 ELISA. Assays were performed on excess plasma or serum collected for routine laboratory tests. One sample was collected at the time the patient met the clinical criteria for HIT (see Section 2.5.1.9.2). If a sample was not available, the laboratory was notified to save the next obtained sample. Only one sample was saved for a suspected HIT event. The samples that were saved contained serum or a minimum of 9 volumes of blood collected in 1 volume of 0.109 M trisodium citrate anticoagulant, whatever was most readily o available from other laboratory tests ordered. A l l samples were aliquoted and stored at - 20 C. No additional blood was drawn for this study and the results of these tests were not available to 83 anyone except the investigator. In addition, laboratory personnel, who performed all diagnostic testing, were unaware of patients' clinical status and platelet counts. 2.5.1.9.1.1 SRA assay The SRA assay was performed as described by Sheridan et al (1986), Warkentin et al (2000) and Warkentin and Greinacher (2001) in the Coagulation Laboratory, under supervision of Dr. Kelton, at McMaster University in Hamilton, Ontario. Serum samples were collected and saved during the study, and sent to Hamilton for SRA testing three times during the two-year study period. A positive test was defined as one in which heparin-dependent platelet activation (> than 20% release of 14C-serotonin) occurred at therapeutic concentrations of heparin (0.1-0.3 Units/mL), but was inhibited (< than 20% release of i4C-serotonin) at very high heparin concentrations (100 Units/mL) and in the presence of the platelet Fc-receptor-blocking monoclonal antibody (IV.3). 2.5.1.9.1.2 Heparin-PF4 ELISA For comparison with the SRA, a heparin-PF4 ELISA was also performed with the use of Asserachrom® Heparin-PF4 ELISA kits (Diagnostica Stago, Asserachrom® HIP A) four times during the two-year study by the same laboratory technician at L G H . The heparin-PF4 ELISA procedures were followed according to the monograph provided with the kits. A l l reagents were included in the kit with the exception of 3 M H2SO4 (Fisher Scientific), and all solutions were reconstituted by the technician at L G H . The heparin-PF4 ELISA results were interpreted as positive or negative by comparison with the reference standard provided in the kit. As indicated in the monograph, "all absorbance values greater than the stated percentage (%) of the absorbance value observed for the reference standard are considered positive". The absorbance value of the reference standard and the stated 84 value for a test to be positive were indicated on the Assay Value insert supplied with the kit. Similarly, all plasma or serum samples whose observed absorbance values were less than or equal to the stated percentage of the absorbance of the reference standard were considered negative (normal). For example, the reference standard provided with the kit for the third run had an observed absorbance (A492nm) of 1.90 (stated absorbance value of reference standard was o 2.20 + 0.42 (assayed by Diagnostica Stago at 20 C)). As stated in the kit insert, all plasma or serum absorbance values greater than 27% of the absorbance value of the reference standard (1.90) were to be considered positive. A l l positive results were repeated, as well as those serum or plasma samples with an A492nm ±0.100 of the criterion for a positive test (0.513 for the third run). Physicians were not made aware of the SRA or heparin-PF4 ELISA results. 2.5.1.9.2 Occurrence of false positives with the SRA and the heparin-PF4 ELISA 2.5.1.9.2.1 Sample selection for false positives In order to estimate the false positive rates of the SRA and heparin-PF4 ELISA, samples were collected from patients who received heparin for 5 or more days, but who did not develop thrombocytopenia (as this time frame has been reported to be long enough for the development of heparin-dependent antibodies) (Warkentin et al, 1998; Warkentin and Kelton, 2001; Warkentin, 2001). Samples from these patients were not selected according to any predefined criteria, but rather on the basis of availability after they had received heparin for 5 or more days. As stated in Section 2.5.1.9.1, both assays were performed on excess plasma or serum collected for routine laboratory tests. One sample was collected; i f a sample was not available, the laboratory was notified to save the next obtained sample. 85 2.6 STATISTICAL ANALYSIS 2.6.1 Incidence of HIT The incidence of HIT was estimated for patients at risk for HIT. This was based on the number of these patients who met the clinical criteria and had a positive assay result using the reference standard, the SRA. Exact results for the 95% CIs around the estimated incidences were obtained from NCSS®, based on the binomial distribution. 2.6.2 Predictive performance of the heparin-PF4 ELISA The positive (PPV) and negative (NPV) predictive values were calculated using Bayes' Theorem (Pagano and Gauvreau, 1993). To estimate the sensitivity of the heparin-PF4 ELISA, studies were identified from the literature meeting the following inclusion criteria: explicit clinical criteria for HIT (i.e. definition for thrombocytopenia and the timing of the thrombocytopenia) after other possible causes for thrombocytopenia were ruled out; patient samples tested positive by the SRA; and patient samples also tested by the heparin-PF4 ELISA (Warkentin et al., 2000; Pouplard et al., 1999; Arepally et al 1995). The sensitivity of this test was taken to be the weighted (by number of patients) average of the reported sensitivities, which was 95.3%. Since clinicians are encouraged to. use the diagnostic tests in patients who meet the clinical criteria for HIT, the prior probability used in calculating the PPV and N P V of the heparin-PF4 ELISA was the estimated incidence of HIT and 1 minus the incidence of HIT, respectively, among those patients who met the clinical criteria and tested positive by the SRA in the present study. The specificity of the heparin-PF4 ELISA was estimated using results from patients who met the clinical criteria for HIT and tested negative by the SRA. Exact results for the 95%) CIs were obtained from NCSS®, based on the binomial distribution. 86 The PPV (proportion of patients with the disease (HIT) who actually tested positive) of the heparin-PF4 ELISA in the present study was calculated as follows: PPV = Sensitivity x Incidence (Sensitivity x Incidence) + [(1 - Specificity) x (1 - Incidence)] The NPV (proportion of patients without the disease (no HIT) who actually tested negative) of the heparin-PF4 ELISA in the present study was calculated as follows: NPV = Specificity x (1 - Incidence) [(Specificity x (1 - Incidence)] + [(1 - Sensitivity) x (Incidence)] 2.6.3 False positive rates of the heparin-PF4 ELISA and SRA McNemar's chi-square test was used to compare the false positive rates of the SRA and heparin-PF4 ELISA among ICU/CCU patients at risk for HIT, but who had not developed thrombocytopenia. A l l quoted P-values were two-sided. 87 RESULTS PART A: DEVELOPMENT AND VALIDATION OF LOGISTIC REGRESSION MODELS FOR THROMBOCYTOPENIA 3.1 DEMOGRAPHIC CHARACTERISTICS AND CLINICAL COURSE IN THE ICU/CCU 3.1.1 Demographic characteristics of the LGH ICU/CCU patients included in the > 100 x 109/L and > 150 x 109/L datasets During the study period, 1813 patients were admitted to the ICU/CCU at L G H . Of these, 792 patients met the criteria for inclusion in the > 100 x 109/L dataset, and a subset of these, 707 patients, comprised the > 150 x 109/L dataset. There were 1021 and 1106 patients excluded from the > 100 x 109/L and > 150 x 109/L datasets, respectively, and the reasons for exclusion are listed in Table 4. Table 5 summarizes the admission demographic characteristics of the patients in the two datasets. The patients in this study sample were mainly Caucasian males, and the mean age was approximately 64 years. Patients admitted with a coronary care diagnosis were older than those with an intensive care admission diagnosis. The mean ages for intensive and coronary care patients were 62 and 67 years, respectively (p < 0.001). The median (25%, 75% percentiles) lengths of ICU/CCU and hospital stay were 3 (2, 6) and 10 (6, 20) days, respectively. 3.1.1.1 Severity of illness The admission mean A P A C H E II score for the patients in the > 150 x 109/L and > 100 x 109/L datasets were similar (Table 5). Patients admitted with an intensive care diagnosis were observed to have a higher mean A P A C H E II score than those with a coronary care admission diagnosis (19 as compared to 12 for both datasets). Based on the A P A C H E II scores, 88 Table 4 Reasons for excluding patients in the > 150 x 109/L and > 100 x 109/L datasets Exclusion Criteria > 150 x 109/L Dataset > 100 x 109/L Dataset < Two platelet counts 878 868 Admission Platelet Count < 150 x 109/L 126 N/A Admission Platelet Count < 100 x 109/L N/A 46 Repeat Admissions 45 47 Two Platelet Counts Within 12 Hours Only 36 39 Participation in Another Study 11 11 < 18 Years of Age 8 8 DIC/ITP/TTP At Admission 2 2 89 Table 5 Demographic characteristics for the ICU/CCU patients in the > 150 x 109/L and > 100 x 109/L datasets Study Sample Characteristics > 150 x 109/L Dataset > 100 x 109/L Dataset N =707 (%) N = 792 (%) Age (years) - Mean±SD 64.4 ± 15.3 64.5 ± 15.3 - Range 18-92 19-92 Gender Males 438 (62.0) 495 (62.5) Age - MeantSD 63.3 ± 14.8 63.3 ± 15.0 - Range 18-90 18-90 Females 269 (38.0) 297 (37.5) Age - MeantSD 66.3 ± 15.9 66.5 ± 15.7 - Range 19-92 18-92 Race Caucasian 632 (89.4) 703 (88.8) Non-Caucasian 75 (10.6) 89(11.2) APACHE II score - Mean±SD 15.6 ±8.7 15.5 ±8.6 - Range 1-46 1-46 Alcohol History 84(11.9) 92(11.6) Weight [actual body weight] (kg) - Mean±SD 76.2 ± 16.9 76.2 ± 16.9 - Range 32-135 32-135 Location patient admitted from: Emergency room 456 (64.5) 498 (62.9) Ward 201 (28.4) 237 (29.9) Other hospital 50(7.1) 57 (7.2) 90 the patients comprising the study sample would be considered mildly to moderately critically i l l (Knaus etal, 1985). 3.2 ADMISSION DIAGNOSES AND CLINICAL COURSE 3.2.1 Admission diagnoses The admission diagnoses of the study patients in both datasets are summarized in Table 6. Approximately one-half of all admissions were for coronary care diagnoses (acute myocardial infarction, unstable angina, cardiovascular non-surgery). For both datasets, the mean A P A C H E II score among patients admitted with an intensive care diagnosis (19 ± 9) was significantly different than the mean A P A C H E II score for those admitted with a coronary care diagnosis (12 ± 7) (p < 0.001). 3.2.2 Admission platelet count The mean admission platelet count (± SD) of the patients included in the > 150 x 109/L dataset was 242.2 ± 77.7 x 109/L (range 150 - 932 x 109/L). The mean admission platelet count for the 354 patients included in the > 150 x 109/L dataset who were admitted with an intensive care diagnosis (250.9 + 88.7 x 109/L) was significantly different than that for the 353 patients admitted with a coronary care diagnosis (233.4 ± 63.7 x 109/L) (p = 0.003). The mean admission platelet count (± SD) of the patients included in the > 100 x 109/L dataset was 229.0 ± 81.3 x 109/L (range 100 - 932 x 109/L). The mean admission platelet count for the 405 patients included in the > 100 x 109/L dataset who were admitted with an intensive care diagnosis (234.2 ± 92.0 x 109/L) was not significantly different than that for the 387 patients admitted with a coronary care diagnosis (223.6 ±68.1 x 109/L) (p = 0.064). 91 Table 6 Admission diagnoses for the ICU/CCU patients in the > 150 x 109/L and > 100 x 109/L datasets Admission Diagnoses > 150 x 109 Dataset Number of Patients (%) [N = 707] > 100 x 109 Dataset Number of Patients (%) [N = 792] Acute myocardial infarction 175 (24.8) 190 (24.0) Unstable angina 91 (12.9) 100 (12.6) Cardiovascular non-surgery 87(12.3) 97 (12.2) Respiratory non-surgery 110(15.6) 116(14.6) Nervous System 45 (6.4) 50 (6.3) Infection 40 (5.7) 45 (5.7) Gastrointestinal 33 (4.7) 38 (4.8) Musculoskeletal/Connective Tissue 25 (3.5) 28 (3.5) Vascular surgery 15 (2.1) 26 (3.3) Gastrointestinal bleed 17(2.4) 22 (2.8) Sepsis 16(2.3) 21 (2.7) Drug overdose 18(2.5) 20 (2.5) Respiratory surgery 17(2.4) 18(2.3) Diabetes mellitus 10(1.4) 10(1.3) Malignancy 7(1.0) 9(1.1) Kidney, Urinary tract, Reproductive 1 (0.1) 2 (0.3) Endocrine 0(0) 0(0) 92 3.2.3 Clinical course in the I C U / C C U 3.2.3.1 Incidence of thrombocytopenia As shown in Table 7, 122 (17.3%; 95% CI: 14.5% - 20.2%) patients had two consecutive platelet counts less than 150 x 109/L, and 84 (10.6%; 95% CI: 8.5% - 13.0%) had at least one platelet count less than 100 x 109/L. Of note, 67 (9.5%) additional patients in the > 150 x 109/L dataset had only one platelet count less than 150 x 109/L, after which the platelet count rose above this threshold. According to the definition, these patients were not considered to have developed thrombocytopenia. Among the 707 patients included in the > 150 x 109/L dataset, the mean admission platelet counts were 208.5 ± 56.8 and 249.2 ± 79.6 x 109/L for those who did and did not develop thrombocytopenia, respectively, and the mean (± SD) time to onset of thrombocytopenia was 2.8 ± 3.2 days (range 1-34 days). Similarly, for the patients included in the > 100 x 109/L dataset, the mean admission platelet counts were 173.7 ± 68.2 and 235.6 ± 80.3 x 109/L in those who did and did not develop thrombocytopenia, respectively, and the mean (± SD) onset of thrombocytopenia was 2.9 ± 3.9 days (range 1-34 days). For both criteria of thrombocytopenia, the incidence was lower among those with a coronary care admission diagnosis (Table 7). For a variety of reasons, investigators have used different thresholds for thrombocytopenia (see Section 1.2). The data in Table 8 summarize the frequencies of thrombocytopenia that would have been documented using different platelet count thresholds. The frequency of thrombocytopenia was consistently higher for those admitted with an intensive care diagnosis. Note, the incidence for platelet counts < 50 x 109/L and < 100 x 109/L may be an underestimate because patients with an admission platelet count below the threshold platelet count for each dataset were not included in this estimation. 93 Table 7 Incidence of thrombocytopenia among patients with intensive and coronary care admission diagnosis included in the two datasets Dataset Thrombocytopenia > 150 x 109/L Dataset Coronary Care Admission Diagnosis (N = 353) 28 (7.9%) [95% CI: 5.3%- 11.3%] Intensive Care Admission Diagnosis (N = 354) 94 (26.6%) [95% CI: 22.0%-31.5%] All Patients (N = 707) 122(17.3%) [95% CI: 14.5%-20.2%] > 100 x 109/L Dataset Coronary Care Admission Diagnosis (N = 387) 12(3.1%) [95% CI: 1.6%-5.4%] Intensive Care Admission Diagnosis (N = 405) 72(17.8%) [95% CI: 14.2%-21.9%] All Patients (N = 792) 84 (10.6%) [95% CI: 8.5%- 13.0%] 94 Table 8 Frequency of thrombocytopenia based on different criteria among patients included in the two datasets > 150 x 109/L Dataset Entire Study Sample (N = 707) ICU (N = 354) CCU (N = 353) ONE or more Platelet Counts: < 150 x 109/L < 100 x 109/L < 50 x 109/L 189 (26.5%) 47 (6.7%)* 15 (2.1%)* 131 (37.0%) 45 (12.7%)* 14 (4.0%)* 58 (16.4%) 2 (0.6%)* 1 (0.3%)* TWO or more Platelet Counts: < 150 x 109/L < 100 x 109/L < 50 x 109/L 122(17.3%) 37 (5.2%)* 13 (1.8%)* 94 (26.6%) 36 (10.2%)* 12(3.4%)* 28 (7.9%) 1 (0.3%)* 1 (0.3%)* > 100 x 109/L Dataset Entire Study Sample (N = 792) ICU (N = 405) CCU (N = 387) ONE or more Platelet Counts: < 100 x 109/L < 50 x 109/L 84(10.6%) 20 (2.5%)** 72(17.8%) 19 (4.7%)** 12(3.1%) 1 (0.3%)** TWO or more Platelet Counts: < 100 x 109/L < 50 x 109/L 61 (7.7%) 16(2.0%)** 55 (13.6%) 15 (3.7%)** 6 (1.6%) 1 (0.3%)** * missing all those admitted with a platelet count < 150 x 109/L as these patients were excluded from this dataset. ** missing all those admitted with a platelet count < 100 x 109/L as these patients were excluded from this dataset. 95 Figure 1 illustrates the time to the onset of thrombocytopenia for the patients in the two datasets, respectively. The time frame for the development of thrombocytopenia was short. Approximately 90% of these patients developed thrombocytopenia within 4 days. 3.2.3.2 Duration of ICU/CCU stay Figure 2A shows the length of ICU/CCU stay for the patients in the > 150 x 109/L dataset. Approximately 75% of the non-thrombocytopenic patients were discharged from or expired in the ICU/CCU within four days. About half the thrombocytopenic patients were discharged from, or expired in, the unit within six days, and there was a subgroup of 32 (26%) thrombocytopenic patients who had a length of stay on the unit of > 14 days. Figure 2B shows the length of ICU/CCU stay for the patients in the > 100 x 109/L dataset. Approximately 70% of the non-thrombocytopenic patients were discharged from or expired in the ICU/CCU within four days. About half the thrombocytopenic patients were discharged from or expired in the unit within six days, and there was a subgroup of 26 (31%) thrombocytopenic patients who had a length of stay on the unit of > 14 days. 3.3 LOGISTIC REGRESSION ANALYSES A l l eight multivariate logistic regression models were developed in a systematic way. Four admission and four exploratory post-admission models were developed (see Section 2.2.9), and the results for the ICU/CCU < 150 x 109/L admission model (Model 1) are provided in detail to illustrate the process used for all other models. Variables used in the development of the exploratory post-admission model included risk indicators present on admission to the ICU/CCU and risk indicators patients were exposed to in the unit up to the development of thrombocytopenia, or discharge or death i f thrombocytopenia did not develop. 96 Figure 1 Distribution of time to the onset of thrombocytopenia for patients included in the > 150 x 109/L (N = 122) and > 100 x 109/L (N = 84) datasets. ^mW, £150 X 10 9/L Dataset ^ 1 2>100 X 10 9/L Dataset 100 -90 -80 -70 h 11-12 13-14 >14 Time to Thrombocytopenia (Days) 97 Figure 2 A Distribution of length of stay in the ICU/CCU for thrombocytopenic (N = 122) and non-thrombocytopenic (N = 585) patients in the > 150 x 109/L study sample. T h r o m b o c y t o p e n i c s N o n T h r o m b o c y t o p e n i c s o c 0 cr 0) UL 1-2 3-4 5-6 7-8 9-10 11-12 13-14 T i m e in I C U / C C U (Days) Figure 2B Distribution of length of stay in the ICU/CCU for thrombocytopenic (N = 84) and non-thrombocytopenic (N = 708) patients in the > 100 x 109/L study sample. T h r o m b o c y t o p e n i c s N o n T h r o m b o c y t o p e n i c s 60 50 o c (D cr 0 40 1-2 3-4 5-6 7-8 9-10 11-12 13-14 >14 T i m e in I C U / C C U (Days ) 98 3.3.1 Multivariate ICU/CCU < 150 x 109/L admission (Model 1) and exploratory post-admission (Model 1 PA) models ICU/CCU < 150 x 109/L admission model (Model 1) Three variables (nine indicators), admission diagnosis, A P A C H E II score, and admission platelet count, were identified following multivariate logistic regression analysis and are shown in order of decreasing odds ratio for this admission model (Table 9A). Admission diagnosis was a categorical variable and six individual intensive care admission diagnoses in Model 1 were identified as being significantly associated with an increased risk for thrombocytopenia relative to coronary care admission diagnoses. Gastrointestinal and G l bleed admission diagnosis had the two highest odds ratios, respectively. Interestingly, 23 of the 33 (69.7%) patients with a gastrointestinal diagnosis had a surgical procedure performed 24 hours prior to admission to the Unit (note, surgery performed 24 hours prior to admission was assessed as a potential candidate variable, but was not independently associated with thrombocytopenia). The indicator "other ICU diagnoses", which included the individual intensive care admission diagnoses respiratory non-surgery, respiratory surgery, nervous system, drug OD/Poisoning, diabetes mellitus, kidney/urinary tract/reproductive, and malignancy, was not independently associated with thrombocytopenia (p = 0.081) relative to coronary care admission diagnoses. A higher A P A C H E II score was associated with the development of thrombocytopenia. Admission platelet count was associated with a decreased risk of thrombocytopenia for a 50 x 109/L increase in its value. For example, the estimated odds ratio for an increase of 50 x 109/L was 0.47, which indicates that for every increase of 50 x 109/L in admission platelet count, the predicted odds ratio of thrombocytopenia is reduced approximately by one-half. Table 9B shows some of the logistic regression model statistics for the ICU/CCU < 150 x 99 Table 9A MODEL 1; ICU/CCU < 150 x 10 9 /L admission model Variable Coefficient (P) SE* Odds Ratio 95% CI OR P-Value Admission Diagnosis Category** < 0.001 Gastrointestinal 2.22 0.45 9.19 3.77-22.37 < 0.001 Gl Bleed 2.19 0.59 8.94 2.82-28.39 < 0.001 Sepsis 2.05 0.65 7.76 2.18-27.59 0.002 Musculoskeletal/connective tissue 2.02 0.50 7.55 2.83-20.15 < 0.001 Vascular Surgery 1.46 0.65 4.32 1.22-15.31 0.023 Infection 1.15 0.46 3.15 1.27-7.84 0.013 Other ICU Diagnoses 0.53 0.30 1.69 0.94-3.06 0.081 APACHE II score3 0.46 0.071 1.58 1.37-1.81 < 0.001 Admission Platelet Count -0.76 0.12 0.47 0.37-0.60 < 0.001 Constant -0.45 0.54 0.41 * Standard Error (SE) ** Coronary care admission diagnoses was the reference category. a per 5 unit increase. A change in the log odds P for an increase of 5 units in the A P A C H E II score. b per 50 x 109/L increase. A change in the log odds P for an increase of 50 x 109/L in the admission platelet count. Table 9B MODEL 1: Logistic regression statistics Model Statistics -2 Log-Likelihood 493.20 Cox-Snell R2 0.20 Nagelkerke R2 0.33 Observed vs. Predicted Pearson's R2 0.25 Overall Correct Classification 84.6% Sensitivity 28.7% Specificity 96.2% Area Under ROC 0.834 100 109/L admission model (Model 1). The admission model resulted in a - 2 L L of 493.20. When the predicted probability of each case was correlated with the observed outcome of thrombocytopenia, the correlation coefficient, r, was 0.504, yielding an r 2 of 0.25. A classification table was used to assess how well the model fit the observed data at a cut-off probability of 0.5. Overall, 598 of 707 (84.6%) patients were correctly classified. The admission model had a sensitivity of 28.7%, meaning 35 of 122 patients who developed thrombocytopenia were correctly classified as developing thrombocytopenia, and a specificity of 96.2%, indicating that 563 of 585 patients who did not develop thrombocytopenia were correctly predicted by the model not to have developed thrombocytopenia at this cut-off. The regression equation associated with Model 1 (Table 9A) is: logit (thrombocytopenia) = -0.45 + 2.22 (gastrointestinal) + 2.19 (Gl Bleed) +1.15 (Infection) + 2.02 (musculoskeletal/connective tissue) + 2.05 (sepsis) + 1.46 (vascular surgery) + 0.53 (other ICU Diagnosis) + 0.46 (APACHE II score/5) - 0.76 (admission platelet count/50) To calculate the logit for a given patient, one would enter a "0" or a "1" for dichotomous variables and the actual value for continuous variables. Then from the logit, the predicted probability of thrombocytopenia for that patient can be calculated. Figure 3 illustrates the probability of developing thrombocytopenia for patients with just the mean admission values of the two continuous variables in the admission model (admission platelet count and A P A C H E II score) and no other risk indicators; and the probability of developing thrombocytopenia for patients with one of the three independent risk indicators(nine independent parameters) identified in the admission logistic regression model. Figure 4 illustrates the calibration curve for Model 1. The model did not appear to demonstrate any systematic bias, and was reasonably well calibrated, as indicated by the results 101 Figure 3 Effect of the individual risk indicators in the ICU/CCU < 150 x 109/L admission model (Model 1) on the predicted probability of developing thrombocytopenia. Dark coloured bars indicate the predicted probability of thrombocytopenia in patients whose platelet count and A P A C H E II score are equal to the respective sample admission mean values and who have no other risk indicators. Light coloured bars indicate the predicted probability of thrombocytopenia in patients with one of the individual dichotomous variables or whose value for the two continuous variables is one standard deviation above the mean value on admission ofthe study sample. 102 of the Hosmer-Lemeshow Goodness of Fit test (p = 0.276). As shown in Figure 5, a ROC curve was generated from the predicted probabilities and observed outcomes. The area under the ROC curve, or c statistic, was 0.834 (95% CI: 0.794 -0.873), and based on this value, excellent discrimination (Hosmer and Lemeshow, 2000c) was found between predicted and observed outcome. Figure 6 illustrates the trade-off between sensitivity and specificity of this model at increasing cut-off probabilities. The assumption of linearity in the logit was examined for all three continuous admission variables. For example, admission platelet count was a continuous variable identified as being associated with the development of thrombocytopenia by univariate analysis. Figure 7 shows the result of plotting the estimated coefficients of the three design variables and the mean of the quartiles of admission platelet count, and indicates that there was a linear trend between the logit and admission platelet count. The Pearson correlation coefficient, r, was 0.94 (rz = 0.89) (p = 0.058). Investigation of linearity in the logit was also performed for the other two continuous admission risk indicators, A P A C H E II score and age. The estimated coefficients and odds ratios indicated that these two risk indicators were linear in the logit, and thus, were not transformed. Possible interactions among the three risk indicators identified in Model 1 were investigated, and when each of the three interactions was individually entered into the multivariate model with the three risk indicators, none appeared to enhance the fit of the model. Therefore, no interaction term was added to the ICU/CCU < 150 x 109/L admission model. Regression diagnostic analyses were performed to identify patients whose observed outcome deviated from the expected or predicted outcome. Analyses of the residuals, studentized residuals, leverage, Cook's distance, and difference in beta coefficients for each of the three admission model variables versus the predicted probability were done in order to identify cases that were outliers. 103 Figure 4 Calibration curve of the ICU/CCU < 150 x 109/L admission model (Model 1): Mean predicted and observed percent of patients within the deciles of predicted probability. Figure 5 Area under the receiver operating characteristic (ROC) curve of the ICU/CCU < 150 x 109/L admission mode! (Model 1). 1.00 CO rz <D C O 0.00 .00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00 1 - Specificity 105 Figure 6 Trade-off between Sensitivity and Specificity of ICU/CCU < 150 x 109/L admission model (Model 1) with increasing cut-off probabilities. 106 Figure 7 Estimated beta coefficients and means of the quartiles of admission platelet count in assessing linearity in the logit. Pearson correlation coefficient, r was -0.94 (r2 = 0.89) (p = 0.058) The scatter plot for studentized residual and predicted probability is shown in Figure 8. This plot indicated that all, except one (patient # 389) of the studentized residuals were within ± 3 standard deviations. The data from this patient were determined to have been entered correctly into the SPSS® database. Patient # 389 was a 62 year-old female admitted to the ICU/CCU with a diagnosis of unstable angina after 11 days on the hospital ward. She had an admission platelet count of 464 * 109/L, an admission A P A C H E II score of 9 and had a Swan-Ganz catheter inserted upon admission to the unit. She experienced a hemorrhage on day 3 (nadir hemoglobin concentration on day 4 = 89 g/L; no transfusions given), and she developed thrombocytopenia on day 4 of her ICU/CCU stay. Therapeutic dose heparin therapy (i.e. > 16,000 Units/day), according to the unstable angina protocol, was started on admission and discontinued on day 3 (day of the hemorrhage). This patient was identified as an outlier because she developed thrombocytopenia, but had a predicted probability of developing thrombocytopenia of only 0.0013 (95% CI: 0.00064 - 0.0025). When patient # 389 was excluded from the dataset, the same three variables (nine indicators), admission diagnoses, A P A C H E II score, and admission platelet count, were identified with very similar coefficient values following multivariate logistic regression analysis as in the original model (Model 1) (Table 9A). Excluding patient # 389 from the dataset resulted in a slightly lower -2LL suggesting a slightly better fit of the observed data. Examination of this patient's covariate pattern demonstrated that she had a high admission platelet count, thus lowering her risk for developing thrombocytopenia. This was the most likely reason why she was identified as an outlier. Since this patient met the clinical criteria for admission to the unit (suspected unstable angina requiring intravenous nitroglycerin and hemodynamic monitoring), she was not excluded from the analysis even though her studentized residual was greater than 3 standard deviations from the mean. Leverage values were used for detecting observations that had a large impact on the 108 Figure 8 Scatter plot of Studentized Residuals and Predicted Probability for the ICU/CCU < 150 x 109/L admission model (Model 1). predicted probabilities. No patient had large leverage values, but data from patients identified as having higher leverage values were checked for correct entry into the database. Cook's distance and differences in betas (delta betas) were plotted against the predicted probabilities in order to examine the influence of each case. Figure 9 shows a scatter plot of Cook's distance and predicted probabilities of thrombocytopenia. Two cases (patients # 120 and 389) demonstrated somewhat extreme values, but < 1.0, and were checked for proper data entry into the database. Patient #120 was a 72 year-old female admitted to the ICU/CCU following femoral popliteal bypass graft surgery. On admission, she had a platelet count of 386 x 109/L, A P A C H E II score of 12, and she had received 2 Units PRBC prior to admission to the unit. She experienced a hemorrhage on day 2, and developed thrombocytopenia on day 3 of her ICU/CCU stay. This patient was identified as an influential case because she developed thrombocytopenia, but had a predicted probability of developing thrombocytopenia of only 0.023 (95% CI: 0.0053 - 0.094). Excluding her from the dataset resulted in the same three variable (nine indicators) being identified as in the model generated from the entire dataset (N - 707), and modest changes in the model's statistics. Therefore, she was not excluded from the analysis. ICU/CCU < 150 x 10 /L exploratory post-admission model (Model IP A) Of the 90 candidate variables evaluated for Model 1PA, 63 were not subjected to univariate analysis because < 5% of patients were exposed to them. However, candidate variables that were clinically and statistically (i.e. by univariate analysis, p < 0.05, with > 1 case of thrombocytopenia) associated with thrombocytopenia, and were noted to be associated with thrombocytopenia in previous studies (e.g. FFP transfusion (Brunner-Bolliger et al., 1997; Nijjar et al., 1987; Noe et al., 1982); imipenem (Cawley et al., 1999; Williams, 1995a); and hepatic dysfunction (Bonfiglio et al., 1995; Baughman et al., 1993)) were included in the multivariate analysis. Some of the medications were grouped into one of the five classes described in Section 110 Figure 9 Scatter plot of Cook's Distance and Predicted Probability for the ICU/CCU < 150 x 109/L admission model (Model 1). CD O Cd •+—» CO b CO ! k o o O 1.00 0.90 0.80 0.70 0.60 0.50 0.40 0.30 0.20 0.10 0.00 # 120 # 389 « o O 8 . a e — • 0.00 0.20 0.40 0.60 0.80 1.00 Predicted Probability i n 2.2.11.2 because of low patient exposure to the medications. Hence, 28 risk indicators were analyzed by univariate analysis, and the resulting measures of association are summarized in Appendix 8. Variables associated with thrombocytopenia with a value of p < 0.25 were considered for inclusion in subsequent multivariate analysis. Of the 28 variables subjected to univariate analysis, 18 were initially selected as candidates for multivariate logistic regression: A P A C H E II score, admission diagnosis, admission platelet count, surgery 24 hours prior to admission, history of alcohol use, ASA, gentamicin, ipratropium bromide, salbutamol, inotropes, cephalosporins, ^-antagonists, heparin dose/day, Swan-Ganz (pulmonary artery) catheter insertion, packed red blood cell transfusions, fresh frozen plasma transfusions, mechanical ventilation, and hepatic dysfunction. Collinearity was assessed among the 18 candidate variables, and was apparent in two pairs of variables. Salbutamol and ipratropium bromide demonstrated the highest association (r = 0.907). This meant that both variables conveyed essentially the same information for the development of thrombocytopenia. Salbutamol was selected for use in multivariate analysis over ipratropium bromide because of its greater frequency of use. Similarly, A P A C H E II score was selected over mechanical ventilation (r = 0.706) because the A P A C H E II score was performed on all patients and is a better measure of severity of illness than mechanical ventilation. Therefore, two variables were excluded because of collinearity with other candidate variables, leaving 16 variables as candidates for multivariate analysis. Results of multivariate logistic regression analysis are shown in Table 10A, in decreasing order of the odds ratio. Six variables (ten indicators) were identified as being independently associated with thrombocytopenia. The same three admission risk indicators were identified as noted in the ICU/CCU < 150 x 109/L admission model (Model 1): admission diagnosis, which included five individual admission diagnosis categories (all associated with an increase risk of thrombocytopenia relative to coronary care admission diagnoses), A P A C H E II score (associated 112 Table 10A MODEL 1PA: ICU/CCU < 150 x 109/L exploratory post-admission model Variable Coefficient (P) SE* Odds Ratio 95% CI OR P-Value Admission Diagnoses Category** < 0.001 Sepsis 2.15 0.65 8.56 2.38-30.83 0.001 Gastrointestinal 2.02 0.49 7.51 2.89-19.50 < 0.001 Musculoskeletal/connective tissue 1.91 0.51 6.75 2.46-18.50 < 0.001 Infection 1.36 0.49 3.91 1.49-10.27 0.006 Other ICU Diagnoses 0.58 0.31 1.79 0.99-3.24 0.056 FFP Transfusion 1.76 0.67 5.82 1.57-21.55 0.008 Swan-Ganz Catheter 1.50 0.31 4.50 2.44-8.28 < 0.001 PRBC Transfusion 1.29 0.33 3.63 1.91-6.88 < 0.001 A P A C H E II score3 0.24 0.079 1.27 1.09-1.49 0.002 Admission Platelet Countb -0.91 0.14 0.40 0.31-0.53 < 0.001 Constant 0.40 0.59 0.50 * Standard Error (SE) ** Coronary care admission diagnoses was the reference category. a per 5 unit increase. A change in the log odds P for an increase of 5 units in the A P A C H E II score. bper 50 x 109/L increase. A change in the log odds P for an increase of 50 x 109/L in the admission platelet count. Table 10B MODEL 1PA: Logistic regression statistics Model Statistics -2 Log-Likelihood 444.21 Cox-Snell R 2 0.25 Nagelkerke R 2 0.42 Observed vs. Predicted Pearson's R 0.33 Hosmer-Lemeshow Goodness-of-Fit p = 0.57 Overall Correct Classification 86.1% Sensitivity 43.4% Specificity 95.0% Area Under ROC 0.871 113 with an increased risk of thrombocytopenia for a 5 unit incremental increase in its value), and admission platelet count (associated with a decreased risk of thrombocytopenia for a 50 x 109/L incremental increase in its value). Three post-admission risk indicators, FFP transfusions, Swan-Ganz catheters, and PRBC transfusions, were associated with an increased risk for thrombocytopenia. Table 10B shows some of the logistic regression model statistics for the ICU/CCU < 150 x 109/L exploratory post-admission model. The area under the curve was 0.871 (95% CI: 0.836 - 0.906) suggesting excellent discrimination (Hosmer and Lemeshow, 2000c). The regression equation associated with Model 1PA (Table 10A) is: logit (thrombocytopenia) = 0.40 + 2.15 (sepsis) + 2.02 (gastrointestinal) + 1.91 (musculoskeletal/connective tissue) + 1.36 (infection) + 0.58 (other ICU Diagnosis) + 0.24 (APACHE II score) + 1.76 (FFP transfusion) + 1.50 (Swan-Ganz catheter) + 1.29 (PRBC transfusion) - 0.91 (admission platelet count) Figure 10 illustrates the probability of developing thrombocytopenia for patients with the mean admission values of the two continuous variables in Model 1PA (admission platelet count and A P A C H E II score) and no other risk indicators; and the probability of developing thrombocytopenia for patients with one of the five independent risk indicators (ten independent parameters) identified in Model 1PA (Table 10A). Figure 11 illustrates the calibration curve for the ICU/CCU < 150 x 109/L exploratory post-admission model. The model did not appear to demonstrate any systematic bias, and was well calibrated (Hosmer-Lemeshow Goodness of Fit test (p = 0.57)). The assumption of linearity in the logit was examined for all three continuous admission 114 Figure 10 Effect of the individual risk indicators in the ICU/CCU < 150 x 107L exploratory post-admission model (Model 1PA) on the predicted probability of developing thrombocytopenia. Dark coloured bars indicate the predicted probability of thrombocytopenia in patients whose platelet count and A P A C H E II score are equal to the respective sample admission mean values and who have no other risk indicators. Light coloured bars indicate the predicted probability of thrombocytopenia in patients with one of the individual dichotomous variables or whose value for the two continuous variables is one standard deviation above the mean value on admission of the study sample. co © Q_ O •*-> >> o o E o o co O 0 o -a 0 1.00 0.90 0.80 0.70 0.60 0.50 0.40 0.30 0.20 0.10 0.00 (5 c o o CD O CO a. C/3 X a D o CD LL N C C O O c CO o m cc CL LU X o < O L < o CL E < Post-Admission Risk Indicators 115 Figure 11: Calibration curve for the ICU/CCU < 150 x 109/L exploratory post-admission model (Model 1PA): Mean predicted and observed percent of patients within the deciles of predicted probability. variables (admission platelet count, A P A C H E II score, and age) as described in Section 3.3.1.4, and for the one continuous post-admission variable (heparin dose/day). A l l four risk indicators were linear in the logit and were not transformed. Among the six variables independently associated with thrombocytopenia in Model 1PA, a total of 15 possible 2-way interactions were screened. None of the interactions enhanced the fit of the model. Regression diagnostic analyses were performed, and as identified in the ICU/CCU < 150 x 109/L admission model (Model 1), the scatter plot of the studentized residual and the predicted probability indicated that all, except one (patient # 389) of the studentized residuals were within ± 3 standard deviations. This was the same outlier identified in the ICU/CCU < 150 x 109/L admission model (Model 1). When patient # 389 was excluded from the dataset, the same six variables (ten indicators) were identified as observed in Model 1PA (Table 10A) with very similar coefficient values, and the -2LL was marginally lower suggesting a slightly better fit of the observed data (data not shown). However, this patient was not excluded from the analysis because of the reason stated earlier. 3.3.2 Multivariate ICU < 150 x 109/L admission (Model 2) and exploratory post-admission (Model 2 PA) models ICU < 150 x 109/L admission model (Model 2) Three variables (six indicators), admission diagnoses, A P A C H E II score, and admission platelet count, were identified following multivariate logistic regression analysis and are shown in order of decreasing odds ratio (Table 11 A). Table 1 IB shows some of the logistic regression model statistics for the ICU < 150 x 109/L admission model (Model 2). Figure 12 illustrates the calibration curve for this model. The model did not appear to demonstrate any systematic bias, 117 Table 11A MODEL 2: ICU < 150 x 109/L admission model Variable Coefficient (P) SE* Odds Ratio 95% CI OR P-Value Admission Diagnoses Category** < 0.001 Gastrointestinal 1.56 0.45 4.76 1.98-11.45 0.001 G l Bleed 1.53 0.58 4.60 1.47-14.40 0.009 Musculoskeletal/connective tissue 1.38 0.50 3.96 1.47-10.62 0.009 Sepsis 1.33 0.62 3.77 1.11-12.77 0.033 A P A C H E II score3 0.49 0.086 1.61 1.36-1.91 < 0.001 Admission Platelet Countb -0.72 0.14 0.49 0.37-0.63 < 0.001 Constant -0.017 0.65 0.98 * Standard Error (SE) ** Other intensive care admission diagnoses (vascular surgery, diabetes mellitus, drug overdose/poisoning, endocrine, infection, kidney/urinary tract/reproductive, malignancy, nervous system, respiratory surgery, respiratory non-surgery) was the reference category. a per 5 unit increase. A change in the log odds (3 for an increase of 5 units in the A P A C H E II score. b per 50 x 109/L increase. A change in the log odds (3 for an increase of 50 x 109/L in the admission platelet count. Table 11B MODEL 2: Logistic regression statistics Model Statistics -2 Log-Likelihood 321.76 Hosmer-Lemeshow Goodness-of-Fit p= 1.00 Cox-Snell R 2 0.22 Nagelkerke R 2 0.32 2 Observed vs. Predicted Pearson's R 0.25 Overall Correct Classification 77.7% Sensitivity 38.3% Specificity 91.9% Area Under ROC 0.806 118 Figure 12: Calibration curve for the ICU < 150 x 109/L admission model (Model 2): Mean predicted and observed percent of patients within the deciles of predicted probability. and was reasonably well calibrated (Figure 12) (Hosmer-Lemeshow Goodness of Fit test (p -1.00)). This model demonstrated excellent discriminating ability (area under the curve was 0.806 (95% CI: 0.755 - 0.857)) (Hosmer and Lemeshow, 2000c). Possible interactions among the three risk indicators identified in this were investigated, and none appeared to enhance the fit of the model. Furthermore, none of the patients had studentized residuals greater than ± 3 standard deviations or large leverage, Cook's distance, or delta beta values. ICU < 150 x 109/L exploratory post-admission model (Model 2PA) The results of multivariate logistic regression analysis are shown in Table 12A, in decreasing order of the odds ratio. Six variables (eight indicators) were identified as being independently associated with thrombocytopenia, the same three admission risk indicators as noted in Model 2 (Table 11 A) (admission diagnosis, A P A C H E II score, and admission platelet count), and three post-admission risk indicators, FFP transfusions, Swan-Ganz catheters, and PRBC transfusions, which were all found to be associated with an increased risk for thrombocytopenia. Table 12B shows some of the logistic regression model statistics for Model 2PA, and Figure 13 illustrates the calibration curve for this model, which did not appear to demonstrate any systematic bias, and was reasonably well calibrated (Hosmer-Lemeshow Goodness of Fit test (p = 0.24)). The area under the curve, or c statistic, was 0.869 (95% CI: 0.827 - 0.910). None of the possible 2-way interactions appeared to enhance the fit of the model, and one of the patients had studentized residuals greater than ± 3 standard deviations or large leverage, Cook's distance, or delta betas values. 120 Table 12A MODEL 2PA: ICU < 150 x 109/L exploratory post-admission model Variable Coefficient (P) SE* Odds Ratio 95% CI OR P-Value Admission Diagnoses Category** < 0.001 Gastrointestinal 1.44 0.49 4.20 1.61-10.98 0.003 Sepsis 1.41 0.63 4.08 1.19-13.98 0.025 Musculoskeletal/connective tissue 1.34 0.53 3.82 1.36-10.70 0.011 FFP Transfusion 2.03 0.73 7.58 1.80-31.93 0.006 Swan-Ganz Catheter 1.49 0.36 4.42 2.17-9.02 < 0.001 PRBC Transfusion 1.43 0.37 4.17 2.02-8.61 < 0.001 A P A C H E II score3 0.33 0.097 1.39 1.15-1.68 0.001 Admission Platelet Countb -0.88 0.15 0.41 0.31-0.56 < 0.001 Constant 0.57 0.71 0.42 * Standard Error (SE) ** Other intensive care admission diagnoses (Gl Bleed, vascular surgery, diabetes mellitus, drug overdose/poisoning, endocrine, infection, kidney/urinary tract/reproductive, malignancy, nervous system, respiratory surgery, respiratory non-surgery) was the reference category. a per 5 unit increase. A change in the log odds P for an increase of 5 units in the A P A C H E II score. bper 50 x 109/L increase. A change in the log odds p for an increase of 50 x 109/L in the admission platelet count. Table 12B MODEL 2PA: Logistic regression statistics Model Statistics -2 Log-Likelihood 277.66 Cox-Snell R 2 0.31 Nagelkerke R 2 0.45 Observed vs. Predicted Pearson's R 0.36 Hosmer-Lemeshow Goodness-of-Fit p = 0.24 Overall Correct Classification 82.2% Sensitivity 57.5% Specificity 91.2% Area Under ROC 0.869 121 Figure 13 Calibration curve for the ICU < 150 x 109/L exploratory post-admission model (Model 2PA): Mean predicted and observed percent of patients within the deciles of predicted probability. 122 3.3.3 Multivariate ICU/CCU < 100 x 109/L admission (Model 3) and exploratory post-admission (Model 3 PA) models ICU/CCU < 100 x 109/L admission model (Model 3) There were five variables (nine indicators), admission diagnoses, A P A C H E II score, surgery 24 hours prior to admission, age and admission platelet count, identified following multivariate logistic regression analysis and they are shown in order of decreasing odds ratio (Table 13 A). Table 13B shows some of the logistic regression model statistics for this admission model (Model 3). Figure 14 illustrates the calibration curve for this model, which did not appear to demonstrate any systematic bias, and was reasonably well calibrated (Hosmer-Lemeshow Goodness of Fit test (p = 0.63)). This model demonstrated outstanding discriminating ability (Hosmer and Lemeshow, 2000c) (area under the curve was 0.925 (95% CI: 0.857 - 0.952)). Investigation of linearity in the logit indicated that there was not a linear relationship between the logit and admission platelet count. Four non-linear transformations were investigated for the admission platelet count (squared, square root, logarithmic, reciprocal) and only the reciprocal of the admission platelet count was linear in the logit. The reciprocal of the admission platelet count was multiplied by 1000 in order to interpret a 1 unit change for this non-linear transformation. Possible interactions among the five risk indicators identified in this model were investigated, and none appeared to enhance the fit of the model. The scatter plot of the studentized residual and the predicted probability indicated that all, except one (patient # 433) of the studentized residuals were within ± 3 standard deviations. Patient # 433 was an 85 year-old female admitted to the ICU/CCU with supra-ventricular tachycardia and hypotension following bowel resection surgery for colon cancer. On admission to the unit, she had a platelet count of 279 x 109/L, an A P A C H E II score of 15, had a Swan-Ganz catheter inserted, and was 123 Table 13A MODEL 3: ICU/CCU < 100 x 109/L admission model Variable Coefficient ( P ) SE* Odds Ratio 95% CI OR P-Value Admission Diagnoses Category** < 0.001 Musculoskeletal/connective tissue 3.13 0.67 22.88 6.13-85.41 < 0.001 Gl Bleed 2.85 0.71 17.22 4.32-68.59 < 0.001 Sepsis 2.62 0.71 13.78 3.44-55.12 . < 0.001 Gastrointestinal 2.47 0.61 11.81 3.59-38.82 < 0.001 Other ICU Diagnoses 0.67 0.44 1.96 0.83-4.62 0.13 (1/Admission Platelet Count) x 1000a 0.94 0.11 2.56 2.08-3.14 < 0.001 Surgery 24 hours prior to admission 0.69 0.34 2.00 1.02-3.91 0.044 APACHE II score13 0.69 0.097 1.99 1.65-2.41 .< 0.001 Agec -0.10 0.049 0.90 0.82-0.99 0.031 Constant -9.75 1.09 < 0.001 * Standard Error (SE) ** Coronary care admission diagnoses was the reference category. a per 1 unit increase. A change in the log odds P for an increase of 1 unit in the reciprocal admission platelet count (1 /admission platelet count) x 1000. b per 5 unit increase. A change in the log odds P for an increase of 5 units in the A P A C H E II score. c per 5 year increase. A change in the log odds p for an increase of 5 years in the patient's Age. Table 13B MODEL 3: Logistic regression statistics Model Statistics -2 Log-Likelihood 297.56 Cox-Snell R2 0.26 Nagelkerke R2 0.53 Hosmer-Lemeshow Goodness-of-Fit P = 0.63 Observed vs. Predicted Pearson's R" 0.42 Overall Correct Classification 92.9% Sensitivity 48.8% Specificity 98.2% Area Under ROC 0.925 124 Figure 14: Calibration curve for the ICU/CCU < 100 x 109/L admission model (Model 3): Mean predicted and observed percent of patients within the deciles of predicted probability. mechanically ventilated. She developed thrombocytopenia on day 5 of her ICU/CCU stay. In addition, she experienced a hemorrhage on day 5 (admission hemoglobin concentration 116 g/L; hemoglobin concentration on day 5 = 69 g/L; 1 unit PRBC transfusion given), just prior to the development of thrombocytopenia. Heparin prophylaxis was started on admission to the ICU/CCU (i.e. > 10,000 and < 16,000 Units/day), and discontinued on day 5 (day of the hemorrhage). This patient was identified as an outlier because she developed thrombocytopenia, but had a predicted probability of developing thrombocytopenia of only 0.0045 (95% CI: 0.0019 -0.0080). When patient # 433 was excluded from the analysis, a slightly different model resulted with four variables (eight indicators) identified following multivariate logistic regression analysis. Surgery 24 hours prior to admission was no longer an independent risk indicator. Excluding patient # 433 from the dataset resulted in a model slightly lower -2LL (294.83 as compared to 297.56) suggesting that this model was a slightly better fit of the observed data. In addition, there was a very modest increase in the area under the ROC curve (0.934 as compared to 0.925). Examination of this patient's covariate pattern showed that she had a cardiovascular non-surgical diagnosis, and thus, a lower risk for developing thrombocytopenia. This was the most likely reason why she was identified as an outlier. Since this patient met the clinical criteria for admission to the unit (arrhythmia and hypotension post-surgery requiring hemodynamic monitoring), she was not excluded from the analysis even though her studentized residual was greater than 3 standard deviations from the mean. ICU/CCU < 100 x 109/L exploratory post-admission model (Model 3PA) The results of multivariate logistic regression analysis are shown in Table 14A, in decreasing order of the odds ratio. Eight variables (eleven indicators) were identified as being independently associated with thrombocytopenia. Three of these were admission risk indicators, 126 Table 14A MODEL 3PA: ICU/CCU < 100 x 109/L exploratory post-admission models Variable Coefficient (P) SE* Odds Ratio 95% CI OR P-Value Admission Diagnoses Category** < 0.001 Musculoskeletal/connective tissue 2.94 0.70 18.89 4.82-74.04 < 0.001 Sepsis 2.61 0.72 13.61 3.31-55.94 < 0.001 Gastrointestinal 1.90 0.65 6.69 1.86-23.99 0.004 Other ICU Diagnoses 0.52 0.47 1.68 0.67-4.26 0.27 FFP Transfusion 1.93 0.66 6.91 1.91-25.00 0.003 Imipenem 1.24 0.59 3.47 1.10-10.95 0.034 Swan-Ganz Catheter 1.13 0.38 3.11 1.49-6.48 0.003 PRBC Transfusion 1.13 0.36 3.10 1.54-6.24 0.002 (1/Admission Platelet Count) x 1000a 1.01 0.12 2.75 2.18-3.46 < 0.001 A P A C H E II scoreb 0.43 0.101 1.53 1.26-1.87 < 0.001 Heparin dose/dayc -0.046 0.022 0.95 0.92-1.00 0.032 Constant -10.37 1.08 < 0.001 * Standard Error (SE) ** Coronary care admission diagnoses was the reference category. a per 1 unit increase. A change in the log odds p for an increase of 1 unit in the reciprocal admission platelet count (1/admission platelet count) x 1000. b per 5 unit increase. A change in the log odds P for an increase of 5 units in the A P A C H E 11 score. 0 per 1000 unit increase. A change in the log odds P for an increase of 1000 units in heparin dose/day Table 14B MODEL 3PA: Logistic regression statistics Model Statistics -2 Log-Likelihood 271.94 Cox-Snell R 2 0.28 Nagelkerke R 2 0.58 Observed vs. Predicted Pearson's R 0.47 Hosmer-Lemeshow Goodness-of-Fit p -0 .87 Overall Correct Classification 93.6% Sensitivity 53.6% Specificity 98.3% Area Under ROC 0.942 127 admission diagnosis, which included four individual admission diagnosis categories (all associated with an increased risk of thrombocytopenia relative to coronary care admission diagnoses), A P A C H E II score, and admission platelet count, while five were post-admission risk indicators. FFP transfusions, imipenem, Swan-Ganz catheters, and PRBC transfusions, were associated with an increased risk for thrombocytopenia, whereas heparin dose/day was associated with a slightly decreased risk for thrombocytopenia for each 1000 Unit increase in its value (i.e. estimated odds ratio for an increase of 1000 Units was approximately 0.95). Table 14B shows some of the logistic regression statistics for this model (Model 3PA), and Figure 15 illustrates the calibration curve for this model, which did not appear to demonstrate any systematic bias, and was reasonably well calibrated (Hosmer-Lemeshow Goodness of Fit test (p = 0.87)). The area under the curve was 0.942 (95% CI: 0.920 - 0.964) suggesting that this model had outstanding discriminating ability (Hosmer and Lemeshow, 2000c). Investigation of linearity in the logit for the continuous variables indicated that there was not a linear relationship between the logit and admission platelet count, it was transformed (reciprocal multiplied by 1000) in order to be linear in the logit. None of the possible interactions appeared to enhance the fit of the model, and none of the patients had studentized residuals greater than ± 3 standard deviations or large leverage, Cook's distance, or delta betas values. 3.3.4 Multivariate ICU < 100 x 109/L admission (Model 4) and exploratory post-admission (Model 4 PA) models ICU < 100 x 10g/L admission model (Model 4) Initially, there were five variables (eight indicators), admission diagnoses, A P A C H E II 128 Figure 15 Calibration curve for the ICU/CCU < 100 x 109/L exploratory post-admission model (Model 3PA): Mean predicted and observed percent of patients within the deciles of predicted probability. Mean Predicted Percent Observed Percent CO 100 90 80 70 60 50 40 30 20 10 tf • t f CM o o> c o CO CO CM c o tf W c o o CO m O o o o CM w c o c o o> O q o o q q q CO o> d q q q q d d d d d CM in i n c o T— CM c o d c n o CM CO tf m CO T- r-. CO o o o o o i — CM W c o c o q q q q q q q q c o d d d d d d d d d d Deciles Of Predicted Probability 129 score, surgery 24 hours prior to admission, age and admission platelet count, identified following multivariate logistic regression analysis (data not shown). Possible interactions among the five risk indicators identified in this admission model were investigated, and one interaction, surgery 24 hours prior to admission by A P A C H E II score, was found to be statistically significant and clinically plausible (p = 0.023). Thus, six variables (nine indicators), including the surgery 24 hours prior to admission by A P A C H E II score interaction were identified following multivariate logistic regression analysis and are shown in order of decreasing odds ratio (Table 15 A). Table 15B shows some of the logistic regression model statistics for the ICU < 100 x 109/L admission model (Model 4). The interaction noted in this model suggested that the impact of surgery 24 hours prior to admission was changed or modified by the value of A P A C H E II score. In this case, the impact of the two risk indicators together was greater than the additive effect of these two variables (beta coefficient for the interaction term was positive and statistically significant): Hence, the estimate of the odds ratio for surgery 24 hours prior to admission cannot be made without specifying the A P A C H E II score. The odds ratio for A P A C H E II score can be estimated as follows. For patients who did not have surgery 24 hours prior to admission, the estimated odds ratio of the A P A C H E II score for an increase of 5 units was 1.61. However, for patients who had surgery 24 hours prior to admission, the estimated odds ratio of the A P A C H E II score for an increase of 5 units was 2.77 (1.61 x 1.72). This indicates that for every increase of 5 units in the A P A C H E II score, the predicted odds ratio for thrombocytopenia was approximately 3 times. Figure 16 illustrates the calibration curve for Model 4, which did not appear to demonstrate any systematic bias, and was reasonably well calibrated (Hosmer-Lemeshow Goodness-of-Fit test (p = 0.23)). The area under the curve was 0.883 (95% CI: 0.841 - 0.925), suggesting that this model had excellent discriminating ability (Hosmer and Lemeshow, 2000c). Regression diagnostic were investigated and none of the patients had studentized residuals 130 Table 15A MODEL 4: ICU < 100 x 109/L admission model Variable Coefficient (P) SE* Odds Ratio 95% CI OR P-Value Admission Diagnoses Category** < 0.001 Musculoskeletal/connective tissue 2.14 0.61 8.46 2.58-27.72 < 0.001 Gl Bleed 2.08 0.64 7.97 2.26-28.06 0.001 Gastrointestinal 1.88 0.54 6.53 2.28-18.69 < 0.001 Sepsis 1.72 0.61 5.56 1.68-18.43 0.005 (1/Admission Platelet Count) x 1000a 0.83 0.11 2.29 1.83-2.86 < 0.001 APACHE II score15 By Surgery 24 hours prior to admission Interaction 0.48 0.22 1.61 • 1.06-2.45 0.026 APACHE II scoreb 0.54 0.12 1.72 1.35-2.19 < 0.001 Agec -0.14 0.052 0.87 0.79-0.96 0.008 Surgery 24 hours prior to admission -1.16 0.94 0.31 0.049-1.98 0.22 Constant -7.32 1.08 < 0.001 * Standard Error (SE) ** Other intensive care admission diagnoses (vascular surgery, diabetes mellitus, drug overdose/poisoning. endocrine, infection, kidney/urinary tract/reproductive, malignancy, nervous system, respiratory surgery, respiratory non-surgery) was the reference category. a per 1 unit increase. A change in the log odds (3 for an increase of 1 unit in the reciprocal admission platelet count (1/admission platelet count) x 1000. b per 5 unit increase. A change in the log odds p for an increase of 5 units in the A P A C H E II score. 0 per 5 year increase. A change in the log odds p for an increase of 5 years in the patient's Age. Table 15B MODEL 4: Logistic regression statistics Model Statistics -2 Log-Likelihood 239.89 Cox-Snell R2 0.29 Nagelkerke R2 0.48 Observed vs. Predicted Pearson's R 0.39 Hosmer-Lemeshow Goodness-of-Fit P = 0.23 Overall Correct Classification 88.4% Sensitivity 50.0% Specificity 96.7% Area Under ROC 0.883 131 Figure 16: Calibration curve for the ICU < 100 x 109/L admission model (Model 4): mean predicted and observed percent of patients within the deciles of predicted probability. Mean Predicted Percent Observed Percent CO — 100 90 80 70 60 50 +-» 40 30 20 10 c o CO o c o CM c o LO o 00 f - cn LO i — LO o o CM CD t o CM T - CO CO 0) c o o o q q q 1 — CM LO o> d d o d d o d d d d LO , 1 CO c6 c b CM c o IV c n LO i— LO c o CM o> o o CM i — ( O CO o> o o o o o o T— 1— CM LO d d d d d d d d d d Deciles Of Predicted Probability 132 greater than ± 3 standard deviations or large leverage, Cook's distance, or delta betas values. ICU < 100 x 109/L exploratory post-admission model (Model 4PA) The results of multivariate logistic regression analysis are shown in Table 16A, in decreasing order of the odds ratio. Eight variables (ten indicators) were identified as being independently associated with thrombocytopenia. Three of these were admission risk indicators, admission diagnosis, which included three individual admission diagnosis categories (all associated with an increased risk of thrombocytopenia relative to the other ICU diagnosis category), A P A C H E II score, and admission platelet count, and five were post-admission risk indicators. FFP transfusions, imipenem, Swan-Ganz catheters, and PRBC transfusions were associated with an increased risk, whereas heparin dose/day was associated with a slightly decreased risk for each 1000 Unit increase in its value. Table 16B shows some of the logistic regression statistics for this model, and Figure 17 illustrates the calibration curve for this model, which did not appear to demonstrate any systematic bias, and was reasonably well calibrated (Hosmer-Lemeshow Goodness of Fit test (p = 0.22)). This model demonstrated outstanding discriminating ability (area under the curve was 0.913 (95% CI: 0.879 - 0.947)) (Hosmer and Lemeshow, 2000c). None of the possible interactions among the risk indicators identified in Model 4PA appeared to enhance the fit of the models, and none of the patients had studentized residuals greater than ± 3 standard deviations or large leverage, Cook's distance, or delta beta values. 133 Table 16A MODEL 4PA: ICU < 100 x 109/L especially post-admission model Variable Coefficient (P ) SE* Odds Ratio 95% CI OR P-Value Admission Diagnoses Category** < 0.001 Musculoskeletal/connective tissue 2.32 0.59 10.13 3.21-31.99 < 0.001 Sepsis 2.07 0.62 7.89 2.36-26.39 0.001 Gastrointestinal 1.40 0.56 4.04 1.34-12.17 0.013 FFP Transfusion 1.93 0.68 6.86 1.80-26.07 0.005 Swan-Ganz Catheter 1.35 0.42 3.86 1.69-8.80 0.001 PRBC Transfusion 1.26 0.37 3.51 1.69-7.30 0.001 Imipenem 1.20 0.60 3.31 1.03-10.64 0.044 (1/Admission Platelet Count) x 1000a 0.90 0.13 2.46 1.90-3.18 < 0.001 A P A C H E II scoreb 0.43 0.11 1.53 1.23-1.90 < 0.001 Heparin dose/day0 -0.081 0.030 0.92 0.87-0.98 0.006 Constant -9.07 1.09 < 0.001 * Standard Error (SE) ** Other intensive care admission diagnoses (Gl Bleed, vascular surgery, diabetes mellitus, drug overdose/poisoning, endocrine, infection, kidney/urinary tract/reproductive, malignancy, nervous system, respiratory surgery, respiratory non-surgery) was the reference category. a per 1 unit increase. A change in the log odds (3 for an increase of 1 unit in the reciprocal admission platelet count (1 /admission platelet count) x 1000. b per 5 unit increase. A change in the log odds P for an increase of 5 units in the A P A C H E II score. c per 1000 unit increase. A change in the log odds P for an increase of 1000 units in heparin dose/day. Table 16B MODEL 4PA: Logistic regression statistics Model Statistics -2 Log-Likelihood 218.44 Cox-Snell R 2 0.33 Nagelkerke R 2 0.54 Observed vs. Predicted Pearson's R 2 0.44 Hosmer-Lemeshow Goodness-of-Fit p = 0.22 Overall Correct Classification 89.6% Sensitivity 55.6% Specificity 97.0% Area Under ROC 0.913 134 Figure 17: Calibration curve for the ICU < 100 x 109/L exploratory post-admission Model (Model 4 PA): mean predicted and observed percent of patients within the deciles of predicted probability. Mean Predicted Percent Observed Percent cd 100 0 Q . O +-* >* O o _Q E o CD "+-> a! o c 0 o 0 90 80 70 60 50 • i - * 40 30 20 10 tf O o co o o o CM o T— o d I CO •tf o o tf co o b I co o o 6 tf co CM o ao o d CO in o LO ao CM o CO co o> q d • i— LO O CO CO co o> o r--co CO d i tf CO 1^  CM in co m co co co d co tf co LO oo in Deciles Of Predicted Probability 135 3.4 MODEL EVALUATION 3.4.1 Internal validation of the admission and exploratory post-admission models generated at LGH Table 17 shows the results of internal validation by the bootstrap procedure for the models generated at L G H . For example, the estimated optimism of the ICU/CCU < 150 x 109/L admission model (Model 1) was + 0.023 (95% CI: 0.005 - 0.050). This is an indication of the extent to which the area under the ROC curve of this model overestimates the likely ability of the model to discriminate in a new sample of patients. Even after correction for the optimism, this model still had reasonably excellent discriminating ability. Correction for the optimism of the ICU/CCU < 100 x 109/L admission and exploratory post-admission models (Models 3 and 3PA) resulted in each model demonstrating "outstanding" discriminating ability. The ICU/CCU < 100 x 109/L admission model (Model 3) demonstrated the smallest optimism. The optimism was largest for the models generated from small sample sizes (i.e. the four ICU models (Models 2, 2PA, 4, and 4PA)). However, each of these models still demonstrated acceptable to excellent discriminating ability after correction for the optimism (Hosmer and Lemeshow, 2000c). For example, correction for the optimism resulted in the ICU < 150 x 109/L admission model (Model 2) retaining acceptable discriminating ability (reducing the area under the ROC curve from 0.806 to 0.773). 3.4.2 Comparison of logistic regression and Cox proportional-hazards regression for the admission and exploratory post-admission models generated at LGH Cox modeling was performed to explore whether right censoring might have had an impact on the explanatory variables identified for thrombocytopenia. Tables 18A to 18H summarize the variables that were retained by the Cox and logistic regression models with their 136 Table 17 Predictive performance and the estimated optimism of the 8 models developed at LGH Model Number Platelet Criterion Area Under ROC Curve Optimism Model 1 ICU/CCU < 150 x 109/L 0.834 (95% CI: 0.794 - 0.873) + 0.023 (95% CI: 0.005 - 0.050) Model 1PA ICU/CCU < 150 x 109/L 0.871 (95% CI: 0.836 - 0.906) + 0.023 (95% CI: 0.005-0.051) Model 2 ICU < 150 x 109/L 0.806 (95% CI: 0.755 - 0.857) + 0.033 (95% CI: 0.012 -0.068) Model 2PA ICU < 150 x 109/L 0.869 (95% CI: 0.827-0.910) + 0.029 (95% CI: 0.010 -0.062) Model 3 ICU/CCU < 100 x 109/L 0.925 (95% CI: 0.857 - 0.952) + 0.008 (95% CI: 0.001 - 0.032) Model 3PA ICU/CCU < 100 x 109/L 0.942 (95% CI: 0.920 - 0.964) + 0.021 (95% CI: 0.005 - 0.048) Model 4 ICU < 100 x 109/L 0.883 (95% CI: 0.841 - 0.925) + 0.021 (95% CI: 0.007 - 0.055) Model 4PA ICU < 100 x 109/L 0.913 (95% CI: 0.879 - 0.947) + 0.031 (95% CI: 0.013-0.067) 137 odds ratios and hazard rates. The risk indicators identified for ICU/CCU < 150 x 109/L admission model (Model 1) following Cox proportional-hazards analysis were very similar to those identified by logistic regression analysis (Table 18A), as were the magnitude and direction of the association between risk indicators and thrombocytopenia. However, there were a few differences: 1) there were four variables (nine indicators) independently associated with thrombocytopenia in the Cox proportional-hazards model, whereas three variables (nine indicators) were identified as independent risk indicators following logistic regression analysis; 2) age appeared as an independent risk indicator in the Cox proportional-hazards model; 3) infection was no longer identified as an individual admission diagnosis category independently associated with the development of thrombocytopenia; and 4) the admission diagnosis G l Bleed was associated with highest hazard (the instantaneous relative risk for thrombocytopenia, at any time, for patients admitted with a G l Bleed was 11.7-fold higher than a patient admitted with a coronary care diagnosis), whereas, gastrointestinal admission diagnosis was associated with the highest odds ratio for thrombocytopenia in the logistic regression model (Model 1). Very similar explanatory variables were identified for the ICU < 150 x 109/L, ICU/CCU < 100 x 109/L, and ICU < 100 x 109/L admission and post-admission models using logistic and Cox proportional-hazards regression (Tables 18B - 18H). 138 18A Comparison of logistic regression and Cox proportional-hazards for MODEL 1# Logistic Regression Model Cox-Proportional-Hazards Model Variable Odds Ratio Variable Hazard Ratio Admission Diagnosis Category* Admission Diagnosis Category* Gastrointestinal 9.19 Gastrointestinal 6.00 G l Bleed 8.94 G l Bleed 11.70 Sepsis 7.76 Sepsis 4.95 Musculoskeletal/connective tissue 7.55 Musculoskeletal/connective tissue 4.73 Vascular Surgery 4.32 Vascular Surgery 4.88 Infection 3.15 Other ICU Diagnoses 1.69 Other ICU Diagnoses 1.55 A P A C H E II score3 1.58 A P A C H E II score3 1.36 Agec 0.93 Admission Platelet Countb 0.47 Admission Platelet Countb 0.47 A blank space indicates a variable that was identified by one of the regression methods, but not the other. Bold type indicates a variable that was identified by the Cox proportional4iazards method and not by logistic regression. * Coronary care admission diagnosis was the reference category. a per 5 unit increase. A change in the log odds P for an increase of 5 units in the A P A C H E II score. bper 50 x 109/L increase. A change in the log odds P for an increase of 50 x 109/L in the admission platelet count. c per 5 year increase. A change in the log odds p for an increase of 5 years in the patient's Age. 139 18B Comparison of logistic regression and Cox proportional-hazards for MODEL 1PA# Logistic Regression Model Cox-Proportional-Hazards Model Variable Odds Ratio Variable Hazard Ratio Admission Diagnosis Category* Admission Diagnosis Category* Sepsis 8.56 Sepsis 10.60 Gastrointestinal 7.51 Gastrointestinal 4.74 Musculoskeletal/connective tissue 6.75 Musculoskeletal/connective tissue 8.21 Infection 3.91 Infection 2.55 Gl Bleed 7.32 Nervous System 2.59 Other ICU Diagnoses 1.79 Other ICU Diagnoses 1.55 FFP Transfusion 5.82 FFP Transfusion 5.12 Swan-Ganz Catheter 4.50 Swan-Ganz Catheter 2.53 PRBC Transfusion 3.63 PRBC Transfusion 1.68 A P A C H E II score3 1.27 A P A C H E II score3 1.16 Admission Platelet Countb 0.40 Admission Platelet Countb 0.42 Gentamicin 0.46 * A blank space indicates a variable that was identified by one of the regression methods, but not the other. Bold type indicates a variable that was identified by the Cox proportional-hazards method and not by logistic regression. * Coronary care admission diagnosis was the reference category. a per 5 unit increase. A change in the log odds P for an increase of 5 units in the A P A C H E II score. bper 50 x 109/L increase. A change in the log odds P for an increase of 50 x 109/L in the admission platelet count. 140 Table 18C Comparison of logistic regression and Cox proportional-hazards for MODEL 2' Logistic Regression Model Cox-Proportional-Hazards Model Variable Odds Ratio Variable Hazard Ratio Admission Diagnosis Category* Admission Diagnosis Category* Gastrointestinal 4.76 Gastrointestinal 3.09 G l Bleed 4.60 G l Bleed 5.83 Musculoskeletal/connective tissue 3.96 Musculoskeletal/connective tissue 3.51 Sepsis 3.77 Sepsis 3.20 A P A C H E II score3 1.61 A P A C H E II score3 1.30 Admission Platelet Countb 0.49 Admission Platelet Countb 0.51 * A blank space indicates a variable that was identified by one of the regression methods, but not the other. Bold type indicates a variable that was identified by the Cox proportional4iazards method and not by logistic regression. * Other intensive care admission diagnosis was the reference category. a per 5 unit increase. A change in the log odds (3 for an increase of 5 units in the A P A C H E II score. b per 50 x 109/L increase. A change in the log odds P for an increase of 50 x 109/L in the admission platelet count. 141 Table 18D Comparison of logistic regression and Cox proportional-hazards for MODEL 2PAn Logistic Regression Model Cox-Proportional-Hazards IV odel Variable Odds Ratio Variable Hazard Ratio Admission Diagnosis Category* Admission Diagnosis Category* Gastrointestinal 4.20 Gastrointestinal 2.68 Sepsis 4.08 Sepsis 5.18 Musculoskeletal/connective tissue 3.82 Musculoskeletal/connective tissue 4.71 Gl Bleed 3.44 FFP Transfusion 7.58 FFP Transfusion 5.10 Swan-Ganz Catheter 4.42 Swan-Ganz Catheter 2.46 PRBC Transfusion 4.17 PRBC Transfusion 1.71 APACHE II score3 1.39 APACHE II score3 1.16 Admission Platelet Count5 0.41 Admission Platelet Countb 0.45 A blank space indicates a variable that was identified by one of the regression methods, but not the other. Bold type indicates a variable that was identified by the Cox proportional-hazards method and not by logistic regression. * Other intensive care admission diagnosis was the reference category. a per 5 unit increase. A change in the log odds p for an increase of 5 units in the A P A C H E II score. bper 50 x 109/L increase. A change in the log odds P for an increase of 50 x 109/L in the admission platelet count. 142 Table 18E Comparison of logistic regression and Cox proportional-hazards for MODEL 3' Logistic Regression Model Cox-Proportional-Hazards Model Variable Odds Ratio Variable Hazard Ratio Admission Diagnosis Category* Admission Diagnosis Category* Musculoskeletal/connective tissue 22.88 Musculoskeletal/connective tissue 16.07 G l Bleed 17.22 G l Bleed 4.42 Sepsis 13.78 Sepsis 6.98 Gastrointestinal 11.81 Gastrointestinal 8.46 Other ICU Diagnoses 1.96 Other ICU Diagnoses 1.81 (1/Admission Platelet Count) x 1000a 2.56 (1/Admission Platelet Count) x 1000a 2.04 Surgery 24 hours prior to admission 2.00 A P A C H E II scoreb 1.99 A P A C H E II score" 1.41 Age c 0.90 A blank space indicates a variable that was identified by one of the regression methods, but not the other. Bold type indicates a variable that was identified by the Cox proportionaUhazards method and not by logistic regression. * Coronary care admission diagnosis was the reference category. a per 1 unit increase. A change in the log odds P for an increase of 1 unit in the reciprocal admission platelet count (1/admission platelet count) x 1000. b per 5 unit increase. A change in the log odds P for an increase of 5 units in the A P A C H E II score. 0 per 5 year increase. A change in the log odds p for an increase of 5 years in the patient's Age. 143 Table 18F Comparison of logistic regression and Cox proportional-hazards for MODEL 3PA Logistic Regression Model Cox-Proportional-Hazards Model Variable Odds Ratio Variable Hazard Ratio Admission Diagnosis Category* Admission Diagnosis Category* Musculoskeletal/connective tissue 18.89 Musculoskeletal/connective tissue 8.24 Sepsis 13.61 Sepsis 4.67 Gastrointestinal 6.69 Gastrointestinal 3.98 Other ICU Diagnoses 1.68 Other ICU Diagnoses 1.08 FFP Transfusion 6.91 FFP Transfusion 2.26 Imipenem 3.47 Swan-Ganz Catheter 3.11 PRBC Transfusion 3.10 PRBC Transfusion 1.92 (1/Admission Platelet Count) x 1000a 2.75 (1 /Admission Platelet Count) x 1000a 1.97 A P A C H E II score5 1.53 A P A C H E II scoreb 1.38 Heparin dose/dayc 0.95 Heparin dose/dayc 0.95 A blank space indicates a variable that was identified by one of the regression methods, but not the other. Bold type indicates a variable that was identified by the Cox proportional4iazards method and not by logistic regression. * Coronary care admission diagnosis was the reference category. a per 1 unit increase. A change in the log odds P for an increase of 1 unit in the reciprocal admission platelet count (1/admission platelet count) x 1000. b per 5 unit increase. A change in the log odds p for an increase of 5 units in the A P A C H E II score. 0 per 1000 unit increase. A change in the log odds P for an increase of 1000 units in heparin dose/day. 144 Table 18G Comparison of logistic regression and Cox proportional-hazards for MODEL 4' Logistic Regression Model Cox-Proportional-Hazards Model Variable Odds Ratio Variable Hazard Ratio Admission Diagnosis Category* Admission Diagnosis Category* Musculoskeletal/connective tissue 8.46 Musculoskeletal/connective tissue 5.38 G l Bleed 7.97 G l Bleed 5.83 Gastrointestinal 6.53 Gastrointestinal 4.81 Sepsis 5.56 Sepsis 3.73 (1/Admission Platelet Count) x 1000a 2.29 (1/Admission Platelet Count) x 1000a 1.80 A P A C H E II scoreb By Surgery 24 hours prior to admission Interaction 1.61 A P A C H E II score" 1.72 A P A C H E II score" 1.36 Age c 0.87 Age c 0.92 Surgery 24 hours prior to admission 0.31 A blank space indicates a variable that was identified by one of the regression methods, but not the other. Bold type indicates a variable that was identified by the Cox proportional-hazards method and not by logistic regression. * Other intensive care admission diagnosis was the reference category. a per 1 unit increase. A change in the log odds (3 for an increase of 1 unit in the reciprocal admission platelet count (1 /admission platelet count) x 1000. b per 5 unit increase. A change in the log odds P for an increase of 5 units in the A P A C H E II score. c per 5 year increase. A change in the log odds P for an increase of 5 years in the patient's Age. 145 Table 18H Comparison of logistic regression and Cox proportional-hazards for MODEL 4PA' Logistic Regression Model Cox-Proportional-Hazards Model Variable Odds Ratio Variable Hazard Ratio Admission Diagnosis Category* Admission Diagnosis Category* Musculoskeletal/connective tissue 10.13 Musculoskeletal/connective tissue 6.51 Sepsis 7.89 Sepsis 4.72 Gastrointestinal 4.04 Gastrointestinal 3.36 FFP Transfusion 6.86 FFP Transfusion 2.44 Swan-Ganz Catheter 3.86 PRBC Transfusion 3.51 PRBC Transfusion 2.28 Imipenem 3.31 (1/Admission Platelet Count) x 1000a 2.46 (1/Admission Platelet Count) x 1000a 1.74 A P A C H E II scoreb 1.53 A P A C H E II scoreb 1.34 Heparin dose/day0 0.92 Heparin dose/dayc 0.94 A blank space indicates a variable that was identified by one of the regression methods, but not the other. Bold type indicates a variable that was identified by the Cox proportional-hazards method and not by logistic regression. * Other intensive care admission diagnosis was the reference category. a per 1 unit increase. A change in the log odds (3 for an increase of 1 unit in the reciprocal admission platelet count (1/admission platelet count) x 1000. b per 5 unit increase. A change in the log odds P for an increase of 5 units in the A P A C H E II score. c per 1000 unit increase. A change in the log odds P for an increase of 1000 units in heparin dose/day. 146 3.4.3 External validation of the admission models generated at LGH A review of the ICU and laboratory databases yielded 927 admissions to the St. Paul's Hospital ICU between November 1998 and August 2000. Of these, 572 patients met the criteria for inclusion in the > 100 x 109/L dataset, and a subset of these, 445 patients, comprised the > 150 x 109/L dataset. There were 482 and 355 patients excluded from the > 150 x 109/L and > 100 x 109/L datasets, respectively. The reasons for exclusions from the > 150 x 109/L dataset included: 269 admitted with a platelet count < 150 x 109/L, 176 had insufficient platelet count data (< two platelet counts), 27 repeat admissions to the ICU, 15 admissions were less than 12 hours in duration, and one patient was less than 18 years of age. The reasons for exclusions from the > 100 x 109/L dataset included: 148 admitted with a platelet count < 100 x 109/L, 155 had insufficient platelet count data (< two platelet counts), 26 repeat admissions to the ICU, 24 admissions were less than 12 hours in duration, and 2 patients were less than 18 years of age. 3.4.3.1 Demographic characteristics of the SPH ICU datasets The case-mix of patients admitted to the SPH ICU was somewhat different than that of those included in the L G H datasets (Tables 19 and 20). The SPH ICU datasets were comprised of slightly younger patients, a lower proportion who had surgery 24 hours prior to admission, and a greater proportion of females, non-Caucasians, and those who expired in the unit than those included in the L G H ICU datasets. However, ICU patients admitted to both settings had similar mean admission A P A C H E II scores and admission platelet counts. When patients included in the SPH ICU datasets were compared to those in the entire L G H ICU/CCU datasets, the mean admission A P A C H E II score was slightly higher in SPH ICU patients, and an even greater proportion of SPH ICU patients expired in the unit compared to L G H ICU/CCU patients. But there was a smaller proportion of SPH ICU patients who had surgery 24 hours prior to admission compared to L G H ICU/CCU patients. 147 Table 19 Comparison of the demographic characteristics for the SPH ICU patients and LGH ICU/CCU and LGH ICU patients included in the > 150 x 109/L datasets Study Sample SPH ICU Patients LGH ICU Patients LGH ICU/CCU Characteristics N =445 (%) N = 354 (%) Patients N = 707 (%) Age (years) - Mean±SD 58.9 i 16.6 62.3 ±17.2 64.4 ± 15.3 - Range 18-94 18-89 18-92 Gender Males 228 (64.7) 193 (54.5) 438 (62.0) Age (years) - MeantSD 60.2 116.6 61.2 ± 17.3 63.3 ± 14.8 - Range 18-94 18-88 18-90 Females 157(35.3) 161 (45.5) 269 (38.0) Age (years) - MeantSD 56.7 i 17.2 63.6 ± 16.9 66.3 ± 15.9 - Range 18-87 19-89 19-92 Race Caucasian 345 (77.5) 319(90.1) 632 (89.4) Non-Caucasian 100 (22.7) 35 (9.9) 75 (10.6) APACHE II score - Mean±SD 22.3 ± 7.8 19.0 ±9.1 15.6 ±8.7 - Range 3-48 2-46 1-46 Admission Platelet Count (x 109/L) - MeantSD 255.5 ± 97.2 250.9 ±88.7 242.2 ± 77.7 - Range 151-885 151-932 151-932 Surgery 24 Hours Before 53 (11.9) 103 (29.1%) 128(18.1%) ICU/(CCU) Admission Location patient admitted from: Emergency room 125 (28.1) 180 (50.8) 456 (64.5) Ward 198 (44.5) 142 (40.1) 201 (28.4) Other hospital 122 (27.4) 32 (9.0) 50(7.1) Expired ICU 72(16.2) 46(13.0) 63 (8.9) Ward 58 (13.0) 26 (7.3) 35 (5.0) 148 Table 20 Comparison of the demographic characteristics for the SPH ICU patients and LGH ICU/CCU and LGH ICU patients included in the > 100 x 109/L datasets Study Sample SPH ICU Patients LGH ICU Patients LGH ICU/CCU Characteristics N =572 (%) N = 405 (%) Patients N = 792 (%) Age (years) - Mean±SD 59.8 i 16.6 62.3 i 17.1 64.5 ± 15.3 - Range 18-94 18-89 18-92 Gender Males 380 (66.4) 225 (55.6) 495 (62.5) Age - Mean±SD 60.7 i 16.2 61.0 i 17.5 63.3 ± 15.0 - Range 18-94 18-88 18-90 Females 192 (33.6) 180(44.4) 297 (37.5) Age - MeantSD 58.0 i 17.2 64.0 i 16.4 66.5 ± 15.7 - Range 18-87 19-89 18-92 Race Caucasian 440 (76.9) 362 (89.4) 703 (88.8) Non-Caucasian 132 (23.1) 43 (10.6) 89(11.2) APACHE II score - MeantSD 22.3 i 7.7 18.8 i 8.9 15.518.6 - Range 3-50 2-46 1-46 Admission Platelet Count (x 109/L) - MeantSD 226.6 i 101.6 234.2 ± 92.0 229.0 ± 89.3 - Range 101-885 101-932 101-932 Surgery 24 Hours Before 80 (14.0) 128 (31.6%) 154(19.4%) ICU/(CCU) Admission Location patient admitted from: Emergency room 167 (29.2) 196(48.4) 498 (62.9) Ward 265 (46.3) 172 (42.5) 237 (29.9) Other hospital 140 (24.5) 37(9.1) 57 (7.2) Expired ICU 102(17.8) 54(13.3) 73 (9.2) Ward 71 (12.4) 28 (6.9) 38 (4.8) 149 3.4.3.2 Admission diagnoses for SPH and LGH critical care patients The admission diagnoses of patients admitted to the SPH ICU was somewhat different than that of those included in the L G H datasets (Tables 21 and 22). There were a higher proportion of SPH ICU patients with an infection, sepsis, and kidney/urinary tract/reproductive admission diagnosis compared to L G H ICU patients. It was noted that SPH ICU patients with a kidney/urinary tract/reproductive admission diagnosis were comprised mainly of those with acute renal failure requiring dialysis, a procedure not performed at L G H . Nervous system and vascular surgery admission diagnoses were more prevalent among L G H ICU patients than SPH ICU patients. Some differences, such as for with an infection, sepsis, and respiratory non-surgery admission diagnoses, were accentuated for patients included in SPH ICU datasets compared to those in the entire L G H ICU/CCU datasets. There were no patients admitted to the SPH ICU with unstable angina, and there was a much lower proportion of these patients with an acute myocardial infarction admission diagnosis as compared to those included in the L G H ICU/CCU datasets. Conversely, similar proportions of SPH ICU and L G H ICU/CCU patients were admitted with a cardiovascular non-surgery diagnosis. However, as stated in Section 2.3.8, patients admitted to the SPH ICU with a acute myocardial infarction or cardiovascular non-surgery diagnosis were considered to be intensive care patients, as opposed to coronary care patients as at L G H , because they had an associated or underlying intensive care diagnosis, and most were mechanically ventilated. There is a separate C C U at SPH, however, prior to 2002, C C U patients requiring mechanical ventilation were transferred to the ICU 3.4.3.3 Incidence of thrombocytopenia among the SPH ICU patients Eighty-three ICU patients at SPH (18.7%; 95% CI: 15.1% - 22.6%) had two consecutive platelet counts less than 150 x 109/L, and 82 (14.3%; 95% CI: 11.6% - 17.5%) had at least one platelet count less than 100 x 109/L. These observed incidences were a slightly lower than those 150 Table 21 Admission diagnoses for the critical care patients included in the SPH and LGH > 150 x 109 datasets Admission Diagnoses SPH ICU Patients (N = 445) LGH ICU Patients (N = 354) LGH ICU/CCU Patients (N = 707) Respiratory non-surgery 111 (24.9) 110(31.1) 110(15.6) Infection 92 (20.7) 40(11.3) 40 (5.7) Cardiovascular non-surgery 64 (14.4) - 87(12.3) Sepsis 48 (10.8) 16(4.5) 16(2.3) Nervous System 28 (6.3) 45 (12.7) 45 (6.4) Kidney/Urinary tract/Reproductive 22 (4.9) 1 (0.3) 1 (0.1) Acute myocardial infarction 21 (4.7) - 175 (24.8) Gastrointestinal 20 (4.5) 33 (9.3) 33 (4.7) Drug overdose 16 (3.6) 18(5.1) 18(2.5) Musculoskeletal/Connective 8(1.8) 25 (7.1) 25 (3.5) Gastrointestinal bleed 8(1.8) 17(4.8) 17(2.4) Diabetes mellitus 4 (0.9) 10(2.8) 10(1.4) Vascular surgery 2 (0.4) 15 (4.2) 15(2.1) Malignancy 1 (0.2) 7 (2.0) 7(1.0) Respiratory surgery 0(0) 17(4.8) 17 (2.4) Unstable angina 0(0) - 91 (12.9) Endocrine 0(0) 0(0) 0(0) 151 Table 22 Admission diagnoses for the critical care patients included in the SPH and LGH > 100 x 109 datasets Admission Diagnoses SPH ICU Patients (N = 572) LGH ICU Patients (N = 405) LGH ICU/CCU Patients (N = 792) Respiratory non-surgery 141 (24.7) 116(28.6) 116(14.6) Infection 107(18.7) 45 (11.1) 45 (5.7) Cardiovascular non-surgery 94(16.4) - 97(12.2) Sepsis 61 (10.7) 21 (5.2) 21 (2.7) Nervous System 40 (7.0) 50(12.3) 50 (6.3) Kidney /Urinary tract/Reproductive 27 (4.7) 2 (0.5) 2 (0.3) Acute myocardial infarction 25 (4.4) - 190 (24.0) Gastrointestinal 25 (4.4) 38 (9.4) 38 (4.8) Drug overdose 20 (3.5) 20 (4.9) 20 (2.5) Musculoskeletal/Connective 13 (2.3) 28 (6.9) 28 (3.5) Gastrointestinal bleed 12(2.1) 22 (5.4) 22 (2.8) Diabetes mellitus 4(0.7) 10(2.5) 10(1.3) Vascular surgery 2 (0.3) 26 (6.4) 26 (3.3) Malignancy 1 (0.2) 9 (2.2) 9(1.1) Respiratory surgery 0(0) 18(4.4) 18(2.3) Unstable angina 0(0) - 100(12.6) Endocrine 0(0) 0(0) 0(0) 1 52 observed among ICU patients at L G H , but marginally higher than observed in the subsets of L G H ICU/CCU patients (Table 7). The mean time onset of thrombocytopenia among SPH ICU patients was very similar to that found in L G H ICU/CCU and ICU patients (see Section 3.2.3.1). 3.4.3.4 External validation: L G H I C U / C C U and I C U admission models applied to the two SPH I C U validation datasets When the L G H ICU/CCU < 150 x 109/L admission model (Model 1; Table 9A) was applied to the SPH > 150 x 109/L validation set, the area under the ROC curve decreased from 0.834 to 0.742 (95% CI: 0.684 - 0.800), indicating acceptable discriminating ability (Table 23). Figure 18A shows the calibration curve of Model 1 in this SPH validation dataset. Above a predicted probability of 8%, the model tended to systematically over-predict the development of thrombocytopenia. When the L G H ICU < 150 x 109/L admission model (Model 2; Table 11 A) was applied to this SPH validation set, the area under the ROC curve decreased from 0.806 to 0.750 (95% CI: 0.692 - 0.808), again indicating acceptable discriminating ability (Table 23). Figure 18B shows the calibration curve of Model 2 in this SPH validation dataset, which indicated that above a predicted probability of 14%, model 2 tended to systematically over-predict the development of thrombocytopenia. When the L G H ICU/CCU < 100 x 109/L (Model 3; Table 13A) and ICU < 100 x 109/L (Model 4; Table 15A) admission models were applied to the SPH > 100 x 109/L validation set, the models demonstrated excellent discriminating ability (Table 23). However, Models 3 and 4 tended to systematically over-predict the development of thrombocytopenia above predicted probabilities of 16.5% and 6%, respectively (Figures 19A and 19B). 153 Table 23 Predictive performance of the L G H I C U / C C U and ICU admission models when applied to the two SPH validation datasets Area Under the R O C Curve L G H Admission Model Original L G H Model Performance in the SPH Dataset I C U / C C U < 150 x 109/L Admission (Model 1) 0.834 (95% CI: 0.794 - 0.873) 0.742 (95% CI: 0.684 - 0.800) ICU < 150 x 109/L Admission (Model 2) 0.806 (95% CI: 0.755 - 0.857) 0.750 (95% CI: 0.692 - 0.808) I C U / C C U < 100 x 109/L Admission (Model 3) 0.925 (95% CI: 0.857 - 0.952) 0.808 (95% CI: 0.757 - 0.860) ICU < 100 x 109/L Admission (Model 4) 0.883 (95% CI: 0.841 -0.925) 0.806 (95% CI: 0.755 - 0.858) 154 Figure 18A: Calibration curve of the L G H I C U / C C U < 150 x 109/L admission model (Model 1) applied to the SPH > 150 x 109/L ICU dataset: Mean predicted and observed percent of patients within the deciles of predicted probability; Figure 18B: Calibration curve of the L G H ICU < 150 x 107L admission model (Model 2) applied to the SPH > 150 x 109/L ICU dataset: Mean predicted and observed percent of patients within the deciles of predicted probability. Mean Predicted Percent Observed Percent to ' c Q> CL O -t—» o o ja E o i _ £1 r -• — 50 5 4-» c (D «J c <D O CD 0. Dec i les of Pred icted Probabi l i ty 155 Figure 19A: Calibration curve ofthe LGH ICU/CCU < 100 x 109/L admission model (Model 3) applied to the SPH > 100 x 109/L ICU dataset: Mean predicted and observed percent of patients within the deciles of predicted probability. Figure 19B: Calibration curve of the LGH ICU < 100 x 109/L admission model (Model 4) applied to the SPH > 100 x 109/L ICU dataset: Mean predicted and observed percent of patients within the deciles of predicted probability. Mean Predicted Percent • Observed Percent .55 'c a> o. o >< o o .Q E o w_ sz \— sz *—1 i CO 4—' c •4—• co Q_ c a> o CL Dec i les of Pred icted Probabi l i ty 156 PART B: HEPARIN-INDUCED THROMBOCYTOPENIA 3.5 HEPARIN-INDUCED THROMBOCYTOPENIA IN CRITICAL CARE LGH PATIENTS 3.5.1 Description of the ICU/CCU patients in the HIT component of this study Table 24 indicates that among all patients admitted to the L G H ICU/CCU and meeting the inclusion criteria, 267 were deemed to be at risk for HIT. Among these, 40 (15.0%; 95% CI, 10.7% - 19.3%) met the a priori clinical criteria for this adverse effect, illustrating how common it is for clinicians in the critical care setting to be treating patients who may potentially have HIT. 3.5.2 Demographic characteristics of patients at risk and who met the clinical criteria for HIT Table 25 summarizes the demographic and clinical characteristics of the patients at risk and those who ultimately met the clinical criteria for HIT. These patients were predominantly Caucasian and males, and the mean age was approximately 70 years. The duration of ICU/CCU and hospital stay for the patients who met the clinical criteria was longer than that of all patients at risk for HIT. 3.5.3 Admission diagnoses Table 26 shows the eight most common admission diagnoses of the patients at risk and those who met the clinical criteria for HIT. Collectively, 167 of 267 (62.5%) patients at risk were considered to have an intensive care admission. There were nine intensive care admission diagnoses not shown, which together accounted for < 9% of admissions. Among the 40 patients 157 Table 24 Description of the patients in the heparin-induced thrombocytopenia component of this study Patient Description Number of Patients Patients admitted to the ICU/CCU 1813 Patients receiving heparin meeting the inclusion criteria3 748 Patients at risk for HITb 267 Patients meeting the clinical criteria for HIT0 40 a See inclusion criteria Section 2.5.1.3. b Intensive and coronary care patients were defined to be at risk of developing HIT if they had received heparin for 5 or more days, or if they received any heparin during the index admission after having been exposed to heparin within 8 weeks prior to admission to the unit (Section 2.5.1.8.1). c Two or more consecutive platelet counts < 150 x 109/L or > 33% decrease in platelet count 5 or more days after initiating heparin, or any time after starting heparin for patients exposed to heparin within the previous 8 weeks (Section 2.5.1.8.2). 158 Table 25 Demographic characteristics of the 267 patients at risk and the 40 patients who met the clinical criteria for HIT Study Sample Characteristics At Risk for HITC Clinical Criteria for HITC N = 267 N = 40 Age (years) - Mean±SD 67.6+13.1 70.4+11.3 - Range 18-89 44-87 Gender Males 155 (58.1) 24 (60.0) Age - Mean±SD 66.8 ± 13.3 69.0 ± 9.3 - Range 20-89 48-85 Females 112(41.9) 16 (40.0) Age - Mean + SD 68.6 ± 12.9 72.4 ± 13.8 - Range 18-88 44-87 Race Caucasian 238 (89.1) 37(92.5) Non-Caucasian 29 (10.9) 3 (7.5) APACHE II score - Mean±SD 19.6 ±9.3 22.5 ± 8.8 - Range 4-45 7-39 Admission Platelet Count (x 109/L) - Mean + SD 240.8 ± 94.8 228.1 ±87.3 - Range 102-932 104-510 Weight [actual body weight] (kg) - Mean±SD 75.3 ± 16.7 74.0 ± 15.0 - Range 41-133 44-106 Duration ICU/CCU stay (days) - MeantSD 11.5 ± 11.6 19.7 ± 18.2 - Range 1-82 1-82 Duration hospital stay (days) - Mean±SD 36.3 ±49.1 53.4 ±55.6 - Range 2-369 3-225 a Patients at risk for HIT were: see Section 2.5.1.9.1. b Patients at risk who met the clinical criteria for HIT: see Section 2.5.1.9.2. c Mean ± SD for continuous variables; number (%) for categorical variables. 159 Table 26 Admission diagnoses ofthe 267 patients at risk and the 40 patients who met the clinical criteria for HIT Admission Diagnoses At Risk for HIT N = 267 Clinical Criteria for HIT N = 40 Acute myocardial infarction 37(13.6) 4(10.0) Unstable angina 34(12.7) 1 (2.5) Cardiovascular non-surgery 29(10.9) 2 (5.0) Respiratory non-surgery 81 (31.3) 17(42.5) Nervous System 19(7.1) 4(10.0) Infection 15 (5.6) 1 (2.5) Gastrointestinal 18(6.7) 6(15.0) Sepsis 8 (3.0) 2(5.0) 160 who met the clinical criteria for HIT, 33 (82.5%) were considered to have an intensive care admission. Overall, most admissions were due to respiratory non-surgery, gastrointestinal, and acute myocardial infarction diagnoses. 3.5.4 Heparin administration Table 27 shows that the duration and dosages of heparin administered to patients at risk were similar to those who met the clinical criteria for HIT. Most of those at risk and who met clinical criteria for HIT received prophylactic dose of heparin (medium dose) (1,000-16,000 units/day) each day. Lee and Warkentin, 2001 suggested that low dose heparin administration is associated with a lower risk of HIT, and of those at risk and who met the clinical criteria for HIT in the present investigation, only 8% and 10% received low dose heparin, respectively, and thus had a relatively low perceived risk of developing HIT. 3.5.5 Timing and change in the platelet count decline among patients meeting the clinical criteria for HIT The timing and degree of platelet decline among the patients who met the clinical criteria for HIT are outlined in Table 28. Most of these patients (70%) had not been previously exposed to heparin therapy, and of these patients without previous heparin exposure, most experienced a relative (> 33% decrease in the platelet count) as opposed to an absolute (two platelet counts < 150 x 109/L) thrombocytopenia. 3.5.6 Laboratory analysis of clinically suspected HIT samples Serum or plasma samples for diagnostic testing were available for 32 of these 40 patients. The 32 samples were tested with both the heparin-PF4 ELISA and SRA assays. 161 Table 27 Heparin administration for the 267 patients at risk and the 40 patients who met the clinical criteria for HIT Heparin Administration At Risk for HIT3 Clinical Criteria for HIT3 N = 267 N = 40 Duration of heparin therapy (days) - Mean + SD 9.5 ±8.3 10.3 ± 10.1 - Range 1-47 1-47 Heparin dosage (Units/day) - MeanlSD 15,034.0 ± 10,314.7 12,988.5 ± 9,402.4 - Range 244-47,150 311-45,030 Low dose heparin 20 (7.5%) 4(10.0%) Medium dose heparin 154 (57.5%) 25 (62.5%) High dose heparin 93 (34.8%) 11 (27.5%) Mean + SD for continuous variables; number (%) for categorical variables. 162 Table 28 Change in platelet counts among patients meeting the clinical criteria for HIT Platelet Count Change Days of Heparin Therapy During Index Admission Number of Patients N = 40(%) Patients exposed to heparin within the previous 8 weeks Two Platelet Counts < 150 x 109/L < 5 days 3 (7.5%) > 33% Decrease in the Platelet Count < 5 days 9 (22.5%) Patients NOT previously exposed to heparin Two Platelet Counts < 150 x 109/L > 5 days 11 (27.5%) > 33% Decrease in the Platelet Count > 5 days 17(42.5%) 163 3.5.7 Incidence of HIT Table 29 shows the incidence of HIT among the patients at risk and who met the clinical criteria for HIT. Of all the patients tested, only one was observed to test positive by the SRA, and the incidence in ICU/CCU patients at risk was estimated to be 0.39% (95% CI, 0.01% -2.1%). Critical care physicians are frequently confronted with patients who receive heparin and develop thrombocytopenia. Among the 32 patients who met the clinical criteria for HIT, the incidence was estimated to be 3.1% (95% CI, 0.08% - 16.2%). In order to estimate the upper bound of the 95% CI, the estimated incidences of HIT among the ICU and C C U patients at risk are shown in Table 29. Details of the ICU patient diagnosed with HIT are described in Section 3.5.9. The findings of this investigation highlight the fact that the incidence of HIT in these ICU/CCU patients who present as though they have HIT is very low, and underscores the difficulty critical care physicians face in diagnosing this limb- and life-threatening adverse drug reaction. 3.5.8 Diagnostic testing with the heparin-PF4 ELISA In order to estimate the specificity of the heparin-PF4 ELISA, samples for the 31 patients who met the clinical criteria for HIT and tested negative by the SRA were evaluated. Twenty-two of these samples tested negative by the heparin-PF4 ELISA and thus, the apparent specificity was 71.0% (95% CI, 52.0 - 85.8%). As 3.1% of the patients who met the clinical criteria had HIT, this can be considered an estimate of the prior probability (see Section 3.5.7). This value along with the observed specificity of the heparin-PF4 ELISA (71%), and the literature derived sensitivity of 95% (see Section 2.6.2) provided the basis for evaluating the predictive performance (PPV and NPV) of the heparin-PF4 ELISA. The calculated PPV and NPV were 9.5% (95% CI, 0.25% - 44.5%) and 100% (95% CI, 84.6% - 100%), respectively. 164 Table 29 Estimated Incidence of HIT Heparin Administration Incidence (%) 95% CI (%) ICU/CCU patients at risk for HIT 0.39 0.01-2.1 (N = 259)* ICU patients at risk for HIT 0.62 0.02-3.4 (N= 162) C C U patients at risk for HIT 0 0 -3 .7 (N = 97) ICU/CCU patients meeting clinical criteria 3.1 0.08-16.2 (N = 32) Samples were not available for 8 patients who met the clinical criteria thus, 259 patients at risk were evaluated for HIT. 165 Both the heparin-PF4 ELISA and SRA have been reported to provide false positive results among patients without thrombocytopenia (Warkentin et al, 2000; Chong et al., 1993; Amiral et al., 1996; Bachelot-Loza et al., 1998). Of 91 samples that were available from patients exposed to heparin for > 5 days, but who did not develop thrombocytopenia (56 from patients considered to have an intensive care admission), 14 (15.4%; 95% CI, 8.7% - 24.5%) tested positive by the heparin-PF4 ELISA. Ninety of these samples were tested using the SRA (55 from patients considered to have an intensive care admission) and 3 (3.3%; 95% CI, 0.69% -9.4%>) tested positive. This difference in false positive rates was significant (p < 0.003). 3.5.9 Clinical course of the HIT patient The patient who was deemed to have HIT was a 77 year-old Caucasian male admitted to hospital for central abdominal pain. He was diagnosed with small bowel obstruction and underwent a laparotomy for small bowel resection. Five days post surgery he experienced respiratory distress and was admitted to the ICU/CCU (considered a respiratory non-surgical cause for admission). His concurrent medical history consisted of type II diabetes mellitus and cellulitis of the left foot. A Swan-Ganz catheter and an arterial line were inserted to guide therapy, and heparin was initiated at 5,000 units subcutaneously twice daily for thromboembolic prophylaxis upon admission to the ICU/CCU (there was no documented heparin administration prior to his admission to the unit). Capped ports on the Swan-Ganz catheter were flushed a total of 4 times with 300 Units of heparin per flush. In addition, heparin was infused through pressure bags on a daily basis to keep the Swan-Ganz catheter and the arterial line patent. The patient was intubated 2 days after admission, and on day 10 he met the clinical criteria for HIT. Figure 20 summarizes his platelet count profile in the ICU/CCU. While his platelet count never fell below 150 x 109/L, it had decreased by 34% from baseline on day 10, and a serum sample obtained on the morning of his death tested positive by both the SRA and heparin-PF4 ELISA. 166 Figure 20 Platelet count profile of HIT patient. This patient developed thrombocytopenia having experienced a 34% decline from baseline on the 10th day of heparin therapy. The sample tested positive by both the S R A and heparin-PF4 E L I S A . 400 350 300 250 200 150 100 50 Baseline Sample Heparin Started Thrombocytopenia 0 1 2 3 4 5 6 7 8 9 10 DAYS 167 He died of respiratory complications later on day 10, however, it was not possible to determine whether he experienced a thrombotic event, as no notation was made in the medical chart and no autopsy was performed. Note, prior to developing thrombocytopenia, he received several medications that have been suggested to be associated with the development of thrombocytopenia: ampicillin, ceftizoxime, cefuroxime, dobutamine, dopamine, furosemide, heparin, ipratropium bromide, norepinephrine, and salbutamol (Bogdonoff et al, 1990; Williams, 1995a). Based on admission and post-admission characteristics, his predicted probability for developing thrombocytopenia with the ICU < 150 x 109/L exploratory post-admission model (Model 2PA; Tablel2A) was 0.271. 3.5.10 Clinical course of a thrombocytopenic patient who did not develop HIT To illustrate the difficulty in diagnosing HIT in critical care patients using clinical criteria alone, Figure 21 outlines the platelet count profile of a patient who demonstrated the classical clinical presentation of HIT, but tested negative by both SRA and heparin-PF4 ELISA. The patient was a 71 year-old Caucasian male admitted to the ICU/CCU from a general medical ward for ventilatory support, which was required because of acute respiratory distress following anterior resection of the rectosigmoid (considered a respiratory non-surgical admission diagnosis). Heparin was initiated at 5,000 units subcutaneously twice daily for thromboembolic prophylaxis following admission (there was no documented heparin administration prior to his admission to the unit). Heparin was also infused through two pressure bags to keep arterial and central venous pressure (CVP) lines patent. Seven days after admission, the patient received 2 units of packed red blood cells. He met the clinical criteria for HIT on day 9 (two consecutive platelet counts < 150 x 109/L), and a serum sample obtained on day 9 tested negative by both the 168 SRA and heparin-PF4 ELISA. Clotting was observed in his bladder and left leg on day 11. In addition, he experienced bleeding in his mouth on day 12, and heparin therapy was stopped on day 14. He expired of respiratory failure on day 15. Prior to developing thrombocytopenia, this patient was administered the following medications that have been associated with thrombocytopenia: cefotaxime, ceftazidime, furosemide, heparin, ipratropium bromide, ranitidine, and salbutamol. Based on admission and post-admission characteristics, his predicted probability for developing thrombocytopenia with the ICU < 150 x 109/L exploratory post-admission model (Model 2PA; Table 12A) was 0.409. 169 Figure 21 Platelet count profile of patient meeting the clinical criteria for HIT and having a negative ELISA and SRA test. This patient developed thrombocytopenia having experienced two consecutive platelet counts < 150 x 109/L from baseline on the 9* day of heparin therapy. O X o o r -LU < _ J 300 250 200 150 100 50 0 Baseline Thrombocytopenia Heparin Started Sample I I I I L J L 0 1 2 3 4 5 6 7 8 9 10 1112 13 14 1516 DAYS 170 DISCUSSION PART A: DEVELOPMENT AND VALIDATION OF LOGISTIC REGRESSION MODELS FOR THROMBOCYTOPENIA 4.1 INCIDENCE OF THROMBOCYTOPENIA This was the first large prospective study to investigate the incidence of thrombocytopenia in critically i l l patients. Thrombocytopenia was common in this cohort of patients, and the incidence was consistent with that reported in a preliminary study involving a smaller subset of these patients (Shalansky et al., 2002; Verma, 2000). The observed incidence of thrombocytopenia in the present investigation was higher for intensive care than for cardiac care patients (Table 7), and the observed incidences in these groups are consistent with previous studies of intensive care (Baughman et al, 1993; Bonfiglio et al, 1995; Hanes et al, 1997; Cawley et al, 1999; Stephen et al, 1999; Strauss et al, 2002) (Table 1) and cardiac care patients (McClure et al, 1999; Eikelboom et al, 2001; Bovill et al, 1991; Harrington et al, 1994; Berkowitz et al, 1998; Sane et al, 2000) (Table 2). In the present investigation, patients admitted with an intensive care diagnosis were observed to have a higher mean A P A C H E II score than those admitted with a coronary care diagnosis and hence, were more severely i l l . This may in part explain the difference in incidence of thrombocytopenia between these two subgroups of ICU/CCU patients. In this investigation, two different criteria were used to define thrombocytopenia. In the only other study that used a criterion of < 150 x 109/L (Strauss et al, 2002), the reported incidence of thrombocytopenia among their cohort of medical ICU patients (44%) was higher than observed in the present investigation (27%), and this difference may in part be explained by different inclusion criteria. Strauss et al (2002) excluded patients with an ICU stay of < 48 hours 171 and thus, may have selected more severely i l l patients. Other investigators have used a criterion of one platelet count < 100 x 109/L (Baughman et al., 1993; Bonfiglio et al., 1995; Cawley et al., 1999; Stephen et al., 1999; Hanes et ah, 1997), and the reported incidence of thrombocytopenia ranged from 13% (Cawley et al, 1999) to 41% (Hanes et al., 1997). The incidence observed among the intensive care patients in this investigation (10.6%; 95% CI: 8.5% - 13.0%) was at the lower end of the reported range, and thus may reflect differences in the critical care settings, as previous studies were performed in tertiary care/academic hospitals, whereas this study was performed in a community hospital. Previous reports of the occurrence and incidence of thrombocytopenia in cardiovascular patients were post-hoc analyses of data from clinical trials of specific drug therapies (Sane et al., 2000; McClure et al, 1999; Berkowitz et al, 1998; Harrington et al, 1994; Eikelboom et al, 2001; Bovill et al, 1991) (Table 2). The reported incidence of thrombocytopenia (a platelet count < 100 x 109/L) ranged from 1.0% to 7.3%, which is consistent with that observed among C C U patients in the present investigation (3.1%). 4.2 MULTIVARIATE LOGISTIC REGRESSION MODELING 4.2.1 Admission models This was the first investigation to develop admission models for thrombocytopenia using data obtained from critically i l l patients, and there were a few notable findings. First, the ICU/CCU and ICU admission models demonstrated excellent discriminating ability (area under the ROC curve) (Hosmer and Lemeshow, 2000c) (Tables 9B, 11B, 13B, 15B). Admission models developed using data from ICU/CCU patients had larger areas under the ROC curves (discriminating ability) than models developed on data from the subset of ICU patients, due to the larger sample size of the combined cohort. Second, similar risk indicators were identified in 172 the admission models using both criteria for thrombocytopenia (< 150 x 109/L and < 100 x 109/L). Three risk indicators, admission diagnoses, A P A C H E II score, and admission platelet count, were consistently identified in all four admission models. Two additional risk indicators (surgery 24 hours prior to admission and age) were identified in the two < 100 x 109/L admission models (Models 3 and 4; Tables 13A and 15A). And third, the risk indicators identified in the ICU/CCU admission models (Models 1 and 3; Tables 9A and 13A) were very similar to those identified in ICU models (Models 2 and 4; Tables 11A and 15A). This is likely because patients with an intensive care admission diagnosis had a markedly higher incidence of thrombocytopenia thus, the risk indicators identified in all models were driven primarily by this subset patients. These findings suggest that thrombocytopenia was related to admission characteristics and/or admission diagnosis, reflected by the fact that most patients developed thrombocytopenia within the first few days of admission (60% by day 2 and 90% by day 4). Other investigators also noted that most critically ill patients who developed thrombocytopenia did so within the first 2 to 5 days (Stephen et al, 1999; Stephen et al, 1999a; Strauss et al, 2002; Hanes et al, 1997). Previous investigators have included admission variables in their analyses (Baughman et al, 1993; Bonfiglio et al, 1995; Stephen et al, 1999; Strauss et al, 2002), but none generated separate admission models in their studies. The ICU/CCU < 150 x 10 9/L admission model (Model 1; Table 9A) was similar to the baseline (admission) model developed in a preliminary study performed in a smaller subset of these ICU/CCU patients (Verma, 2000; Shalansky et al, 2002). A P A C H E II score, admission platelet count, and 5 individual admission diagnoses were consistent with those identified in the present investigation. However, the individual admission diagnoses were not coded as a categorical variable. While the results were similar, including admission diagnoses as a categorical variable is less likely to produce spurious associations. 173 Consideration of admission risk indicators for thrombocytopenia Consideration of risk indicators identified by the multivariate logistic regression admission models can provide some information about the underlying biological or clinical mechanisms responsible for the development of thrombocytopenia. It is difficult to compare the results of the present investigation with those reported by others because admission models were not developed in previous studies. Comparisons between the present and previous investigations will be done when the exploratory post-admission models are discussed (see Section 4.2.2). A higher A P A C H E II score was identified as an independent risk indicator, indicating that the risk of thrombocytopenia increased with disease severity. A lower admission platelet count was associated with an increased risk for the development of thrombocytopenia. This is as expected since patients with an admission platelet count close to the thresholds require a relatively small decline in their platelet count to meet the respective criterion for thrombocytopenia. Similarly, patients admitted with low admission platelet counts who are exposed to other risk indicators that may decrease the platelet count, are more likely to develop thrombocytopenia. Also, a patient's platelet count may already be falling when admitted to the critical care unit due to an underlying disease process, and thus the admission platelet count may be a marker for a process that has been ongoing prior to admission. Admission diagnosis was independently associated with thrombocytopenia, and four individual intensive care admission diagnoses were observed to be associated with an increased risk for thrombocytopenia in all four admission models: sepsis, musculoskeletal/connective tissue, gastrointestinal, and G l bleed. An association between sepsis (which includes patients with SIRS, sepsis, severe sepsis, or septic shock) and thrombocytopenia has been suggested for many years (Lee et al., 1993; Brun-Buisson et al., 1995; Oppenheimer et al., 1976; Wilson et al., 1982; Milligan et al., 1974), and review articles of thrombocytopenia in critically i l l (intensive care) patients have listed and 174 described sepsis as cause of thrombocytopenia (Bogdonoff et al, 1990; DeLoughery, 2002). In a "handbook of evidence-based critical care", Marik (2001) stated that sepsis is the most common cause of thrombocytopenia in the ICU, though evidence for this was not provided. In the present investigation, patients admitted with sepsis had a higher A P A C H E II score than non-septic patients (24 as compared to 15 for both datasets) and thus, were more severely i l l . However, sepsis was independently associated with thrombocytopenia, even after adjusting for A P A C H E II score. Thrombocytopenia has been reported to occur early in the course of sepsis (Neame et al, 1980), and it is believed that a series of complex events leads to thrombocytopenia in septic patients. Possible mechanisms suggested for thrombocytopenia in these patients include: platelet interaction with bacteria and endotoxin resulting in platelet activation and degranulation (Hinshaw et al., 1982), DIC leading to consumption of platelets (Bogdonoff et al., 1990; Neame et al., 1980; Kelton et al., 1979), an immune-mediated response that results in non-specific platelet-antibody binding (platelet-associated IgG) (Stephan et al, 2000; Kelton et al, 1979) and specific platelet antibodies directed against glycoprotein Ilb/IIIa or Ib/IX (Stephan et al., 2000), and hematophagocytic syndrome (Baker and Levin, 1998; Stephan et al., 1997; Francois et al, 1997). The increased formation of platelet aggregates and enhanced platelet clearance that occur in septic patients indicate that activated platelets do not remain in the circulation, but rather are cleared (Gawaz et al., 1995). As defined by Bone et al (1992), sepsis is currently classified as a "clinical syndrome defined by the presence of both infection and a systemic inflammatory response" (Levy et al., 2003). However, it is important to note that, frequently, an infection is strongly suspected, but there is no evidence of pathogenic or potentially pathogenic microorganisms (Levy et al, 2003). Hence, investigators strongly suspect sepsis in some patients, without microbiological confirmation. In the present investigation, the definition of sepsis included: documentation of 175 sepsis at admission by the physician in the patient's chart, documented or suspected infection, and clinical criteria related to signs and symptoms of a systemic response. Sixteen and 21 patients, who were included in the ICU/CCU > 150 x 109/L and > 100 x 109/L datasets, respectively, were classified as having an admission diagnosis of sepsis, however only 3 and 4 of these patients had documented laboratory evidence of bacterial infections, respectively. Patients with a musculoskeletal/connective tissue admission diagnosis were mainly comprised of those who had suffered traumatic injuries (> 75%), usually as a result of motor vehicle accidents, skiing accidents, or falls. Although many of these patients were noted to have suffered blood loss, no correlation was observed between musculoskeletal/connective tissue admission diagnosis and PRBC transfusion. Platelets are lost during hemorrhage and consumed at the site of the injury, which could have resulted in a decrease in the platelet count in patients admitted with a musculoskeletal/connective tissue diagnosis in this investigation. Yeaman (1997) suggested that platelets respond rapidly and are consumed at sites of endothelial damage due to trauma, and Akca et al. (2002) argued that critically i l l trauma patients may experience a loss of circulating platelets, which can contribute to the development of thrombocytopenia. Thus, it is possible that vascular injury and the loss of blood experienced by these patients contributed to the development of thrombocytopenia. Gastrointestinal disorders, such as small bowel diverticula, inflammatory bowel disease, and ulcerative colitis, have been reported to be associated with thrombocytopenia (Klee et al., 1997; Mones, 1983; Kocoshis et al., 1979). In the present investigation, gastrointestinal diagnosis included both surgical and non-surgical patients. Surgery and other invasive procedures that lead to blood loss can result in the development of thrombocytopenia (Vincent et al, 2002; Chang, 1996; Hanes et al, 1997; Stephan et al, 1999). In the present investigation, approximately 70% of the patients admitted with a gastrointestinal diagnosis had a prior surgical procedure, and 90% of these had the surgery within 24 hours prior to admission to the unit. 176 However, surgery 24 hours prior to admission was also identified as an independent risk indicator for thrombocytopenia in Models 3 and 4 (Tables 13A and 15A), suggesting that gastrointestinal admission diagnosis and surgery 24 hours prior to admission were both providing independent information about thrombocytopenia. Gastrointestinal (Gl) bleed was also independently associated with an increased risk for thrombocytopenia. It is likely that this association was due to blood loss and platelet consumption, as circulating platelets are utilized in the normal hemostatic system to limit blood loss (Handin, 2001a). Low admission platelet count was likely not the sole reason for the thrombocytopenia because the mean admission platelet count for the 22 patients admitted with a G l bleed was 232 x 109/L (range 111 - 606 x 109/L) and G l bleed was independently associated with thrombocytopenia after adjusting for the admission platelet count. It has also been suggested that thrombocytopenia is a common cause of bleeding in specific patient populations (Levine, 1999; Arrowsmith et al, 1999; Bick et al, 1996), and Cook et al (1994) observed that coagulopathy (defined as a platelet count < 50,000 x 109/L, an INR > 1.5 [i.e., prothrombin time >1.5 times the control value], or a partial-thromboplastin time >2.0 times the control value) was an independent risk factor for G l bleeding. However, there is little prior evidence suggesting that G l bleeds are associated with the development of thrombocytopenia. Two additional individual intensive care admission diagnoses, infection and vascular surgery, were independently associated with thrombocytopenia in Model 1 (Table 9A). In the present investigation, an admission diagnosis of infection included documentation of this diagnosis by the physician in the patient's chart, and clinical signs of infection or administration of antibiotics. None of these patients met the criteria for sepsis (Section 2.2.11.1). However, it is possible that some of these patients had underlying pathology similar to patients with a diagnosis of sepsis. In a review article of thrombocytopenia in the critically i l l patients, Bogdonoff et al (1990) suggested that increased platelet destruction by immune and non-immune 177 mechanisms and suppression of bone marrow thrombopoiesis are possible explanations for the thrombocytopenia seen with bacterial and viral infections. Bacteria have been observed to directly interact with platelets (Herzberg et al, 1983; Usui et al., 1987), resulting in aggregation and clearance from the circulation (Clawson and White 1971; Clawson and White 1971a). Other postulated mechanisms of non-immune mediated platelet destruction include: adhesion and aggregation of platelets to endothelium damaged by infectious organisms or their products (e.g. endotoxin) (Bogdonoff et al, 1990; Issekutz and Ripley, 1986), consumption of platelets during a bacterial infection, which usually occurs during DIC (Wilson et al., 1982), direct damage to platelets by exotoxin from gram-positive bacteria (George and El-Harake, 1995), bone marrow suppression (Pappas et al., 2004), and reactive hematophagocytosis (i.e. infection hematophagocytic syndrome) (Fisman, 2000; Risdall et al, 1984; Lortholary et al, 1990). Viruses (e.g. viral hepatitis, varicella, cytomegalovirus) have been suggested to directly affect megakaryocytes or platelets (Yeaman, 1997; Zucker-Franklin 1994). An immune mechanism has been suggested for the thrombocytopenia in some patients with viral infections and gram-positive or gram-negative bacterial (Bogdonoff et al, 1990; Zucker-Franklin 1994; Pappas et al, 2004). Wilson et al (1982) suggested possible mechanisms for platelet destruction, which included non-specific binding of IgG to platelet bound bacterial endotoxins or bacterial or viral fragments forming immune complexes that subsequently bind to platelet F e receptors. Arnold et al (2004) observed autoantodies directed against platelet glycoproteins Ilb/IIIa and Ib/IX in a patient with infective endocarditis and suggested that autoantibody-mediated platelet destruction was an important contributing factor in the development of thrombocytopenia. Vascular surgery was also identified as an independent risk indicator in admission Model 1 (Table 9A). Invasive procedures like vascular surgery lead to blood loss that can result in the development of thrombocytopenia (Vincent et al, 2002; Chang, 1996; Hanes et al, 1997; Stephan et al, 1999). Furthermore, it is possible that exposure of the vasculature to foreign 178 surfaces during surgery may result in aggregation of platelets to injured endothelium (Packham, 1988). Two other admission risk indicators were independently associated with thrombocytopenia in Models 3 and 4 (Tables 13A and 15A), age and surgery 24 hours prior to admission. Increasing age was associated with a decreased risk for thrombocytopenia. Approximately half the patients admitted were relatively young ICU patients and the other half tended to be older cardiac patients with a markedly lower risk for development of thrombocytopenia (Table 7). C C U admission diagnosis, which was the reference, group that individual ICU admission diagnoses were compared to, was also independently associated with a decreased risk for thrombocytopenia. This suggests that age was not simply a marker for C C U admission diagnosis, but that there were other factors involved in the apparent "protective" effect of age. Surgery 24 hours prior to admission was associated with an increased risk for the development of thrombocytopenia. In the present investigation, patients who had undergone a surgical procedure 24 hours prior to admission were mainly comprised of those admitted to the unit for post-operative management (75% for both datasets). Although these patients had a lower mean admission platelet count compared to those who did not undergo surgery (data not shown), after adjusting for admission platelet count, surgery 24 hours prior to admission was identified as an independent risk indicator and appeared to be providing additional information with regard to the development of thrombocytopenia in this cohort of patients. In a prospective observational study, Akca et al (2002) collected data from 40 ICUs and analyzed the time course of the serial daily platelet counts in relation to thrombocytopenia and mortality for 1,449 critically i l l patients (513 (35%) of whom were admitted following elective or emergency surgery). The investigators observed that an initial reduction in platelet count within the first days of ICU admission was followed by an increase in their cohort of patients. They argued that surgical patients may experience a loss of circulating platelets, which can contribute to the 179 development of thrombocytopenia i f the platelet count on admission or just after admission was close to the threshold for thrombocytopenia. Others have observed a decrease in the platelet count in the first 2 - 4 days after surgery as compared to pre-operative levels (O'Brien et al., 1984; Economopoulos et al, 1977), and hence surgical patients who are exposed to other risk indicators that may decrease the platelet count, are more likely to develop thrombocytopenia. In summary, the admission models generated in the present investigation fit the observed data well and demonstrated excellent discriminating ability. The explanatory variables identified in these models are either directly causative for thrombocytopenia or are indicators for other causal factor(s). In either case, the development of thrombocytopenia is evidently a multi-factorial outcome, as a combination of risk indicators is involved in decreasing the platelet count below the threshold. 4.2.2 Exploratory post-admission models Exploratory post-admission models were developed to compare the results of this investigation with others and to investigate whether post-admission risk indicators provided additional information (i.e. modified the risk) for the development of thrombocytopenia in this patient cohort. The exploratory post-admission models demonstrated excellent (Models 1PA and 2 PA) and outstanding (Hosmer and Lemeshow, 2000c) (Models 3PA and 4 PA) discriminating ability, respectively. There was a modest increase in the discriminating ability (area under the ROC curve) of each exploratory post-admission model compared to its admission model indicating that the post-admission risk indicators were providing additional information about thrombocytopenia. Furthermore, there was a slight improvement in the fit of the four exploratory post-admission models in describing the observed data, as indicated by a slightly lower -2LL and higher Nagelkerke R compared to the admission models. Similar risk indicators were identified in the four exploratory post-admission models. In 180 addition to the three admission variables, disease severity (APACHE II score), admission diagnoses, and admission platelet count, three post-admission risk indicators were consistently associated with thrombocytopenia in the four exploratory post-admission models and these were: FFP transfusion, Swan-Ganz (pulmonary artery) catheter insertion, and PRBC transfusion. Two additional risk indicators (imipenem and heparin dose/day (associated with a decreased risk for thrombocytopenia)) were identified in the two < 100 x 109/L exploratory post-admission models (Models 3PA and 4PA; Tables 14A and 16A). The risk indicators identified in the two ICU/CCU exploratory post-admission models (Models 1PA and 3PA; Tables 10A and 14A) were very similar to those identified in the 2 ICU models (Models 2PA and 4PA; Tables 12A and 16A), largely because patients with an intensive care admission diagnosis had a markedly higher incidence of thrombocytopenia (Table 7) and the risk indicators identified reflected the subset of ICU patients. The risk indicators independently associated with thrombocytopenia in Model 1PA (Table 10A) were very similar to those noted in a preliminary study performed in a smaller subset of these same patients (Shalansky et al., 2002; Verma, 2000), FFP transfusion, pulmonary artery catheter insertion, PRBC transfusion, admission platelet count, and four individual "most responsible" diagnoses. Using data from patients admitted to a medical ICU, Strauss et al (2002) also developed a post-admission model for thrombocytopenia defined as < 150 x 109/L using logistic regression. They identified four independent risk factors, of which a higher initial (admission) platelet count (associated with a reduced risk of thrombocytopenia) and a higher initial SOFA score (a measure of disease severity like the A P A C H E II score) were similar to those found in the ICU/CCU and ICU < 150 x 109/L exploratory post-admission models (Models 1PA and 2PA; Tables 10A and 12A). 181 Consideration of the post-admission risk indicators The risk indicators identified in the exploratory post-admission models can be compared to those reported by other investigators who also developed post-admission models in their cohorts of ICU patients (Baughman et al, 1993; Bonfiglio et al, 1995; Hanes et al, 1997; Cawley et al, 1999; Stephen et al, 1999; Strauss et al, 2002). However, it should be noted that all the previous studies included smaller samples of ICU patients. As well, the three studies that evaluated admission diagnosis as a risk factor did not code admission diagnoses as a categorical variable (Bonfiglio et al, 1995; Cawley et al, 1999; Strauss et al, 2002). Furthermore, three groups of investigators used stepwise multiple linear regression rather than logistic regression to develop their post-admission models (Baughman et al, 1993; Bonfiglio et al, 1995; Cawley et al, 1999). The three admission risk indicators that were consistently identified in all four exploratory post-admission models (Models 1PA - 4PA), A P A C H E II score, admission platelet count, and admission diagnosis, have been described above. Others have observed these or similar risk factors in their post-admission models. The A P A C H E II score was also identified by another investigator (Stephan et al, 1999), who noted that an A P A C H E II score > 15 was associated with the development of thrombocytopenia in surgical ICU patients. Strauss et al (2002) reported that a higher A P A C H E II score was associated with thrombocytopenia following univariate analysis, but not after multivariate logistic regression analysis. However, these researchers found that a higher initial SOFA score, which is a measure of severity of illness that quantitatively and objectively describes the degree of organ dysfunction/failure (Ferreira et al, 2001; Vincent et al, 1996), was independently associated with thrombocytopenia in their cohort of patients. Hanes et al (1997) used the trauma score (Champion et al, 1981) to assess severity of the injury. However, it is difficult to compare the A P A C H E II score and the trauma or SOFA scores as measures of disease severity, as different parameters are used in their calculation. 182 Other investigators (Stephan et al, 1999; Strauss et al, 2002) have also reported that higher admission platelet count was associated with a decreased risk for thrombocytopenia, and Bonfiglio et al (1995) noted that the baseline (admission) platelet count accounted for the largest proportion of the variance for the development of thrombocytopenia following stepwise linear regression analysis. Three intensive care admission diagnoses were consistently associated with thrombocytopenia in all four exploratory post-admission models (Models 1PA - 4PA), sepsis, gastrointestinal, and musculoskeletal/connective tissue admission diagnoses. Previous investigators who included admission diagnoses in their regression analyzes did not classify diagnosis as a categorical variable (Bonfiglio et al, 1995; Cawley et al, 1999; Strauss et al, 2002) or clearly define each diagnosis. Patients in the present investigation who had a documented admission diagnosis of sepsis included those with SIRS and sepsis, and this appears to be similar to previous investigations (Baughman et al, 1993; Hanes et al, 1997; Stephan et al, 1999; Cawley et al, 1999). However, only Stephan et al (1999) and Baughman et al (1993) found sepsis to be independently associated with thrombocytopenia following multivariate analysis. In another investigation, Bonfiglio et al (1995) did not define SIRS or sepsis, but they found that the combined diagnostic categories of respiratory failure and sepsis syndrome/septic shock were associated with thrombocytopenia following univariate analysis. Recently, Strauss et al (2002) reported that septic shock and bacteremic sepsis were associated with thrombocytopenia following univariate analysis, but not following multivariate logistic regression analysis. These investigators did not define sepsis, although they did provide a definition for shock. None of the previous investigators reported infection as a risk indicator for the development of thrombocytopenia, and it is unclear whether they included it as a candidate variable. However, Stephan et al (1999) suggested that physicians should by aware that the onset of thrombocytopenia may be indicative of an acute infection. 183 Bonfiglio et al (1995) included gastrointestinal admission diagnosis as a candidate variable, but found that it was not associated with thrombocytopenia following univariate analysis. Interestingly, a greater proportion of their patients (11%) had this diagnosis on admission compared to those in the present investigation (5%). It does not appear that previous investigators included musculoskeletal/connective tissue admission diagnosis as a candidate variable in their analyses. However, Hanes et al (1997) noted that non-head injury was independently associated with the development of thrombocytopenia in a small sample of trauma ICU patients, while Stephan et al (1999) reported that trauma was not associated with thrombocytopenia in their small cohort of surgical ICU patients. Similarly, previous investigators have not included G l bleeds as a potential risk indicator for thrombocytopenia in critically i l l patients; however, Stephan et al (1999) reported that a variable for episodes of bleeding or transfusions was independently associated with the development of thrombocytopenia. In addition, Baughman et al (1997) noted that the most common underlying admission diagnosis in their patient cohort was G l bleeds (20%), but it is unclear whether they analyzed the association between this or other underlying admission diagnoses and thrombocytopenia. Surgery at admission was investigated as a potential risk factor for thrombocytopenia by Stephan et al (1999) and Hanes et al (1997) in 63 trauma ICU and 147 surgical ICU patients, respectively, but neither found that this risk factor was independently associated with thrombocytopenia. However, Stephen et al (1999) reported that surgery as an admission category was associated with thrombocytopenia in their post-admission model following univariate analysis. It should be noted that these were small studies with relatively few cases of thrombocytopenia. In the present investigation, three post-admission risk indicators, FFP transfusion, pulmonary artery catheter insertion, and PRBC transfusion, were consistently associated with 184 thrombocytopenia in all four exploratory post-admission models (Models 1PA - 4 PA; Tables 10A, 12A, 14Aand 16A). Previous authors reported that FFP transfusion was associated with a reduction in platelet count in mixed patients groups (Solenthaler et al, 1999; Brunner-Bolliger et al., 1997; Noe et al., 1982); however, none of the studies analyzing risk factors for thrombocytopenia in critically il l patients identified FFP transfusion as an independent predictor. Thrombocytopenia associated with FFP transfusions has been suggested to result from the dilutional effect of PRBC administered with FFP (Bogdonoff et al, 1990). However, it is not clear how many units of blood products transfused within 24 hours is large enough to cause a dilutional effect. Investigators have suggested that more than 20 units of whole blood or red blood cells transfused within 24 hours may result in a decrease in the platelet count and lead to thrombocytopenia (Leslie and Toy, 1991; Riska et al., 1988), whereas Marik (2001) suggested that more the 10 units of whole blood transfused within 24 hours may result in thrombocytopenia. In the present investigation, among 16 patients who received both FFP and PRBC transfusions, 4 units of PRBC (and 2 units of FFP) in 24 hours was the largest quantity transfused. Thus, it is unlikely that the thrombocytopenia resulted from the dilutional effect of PRBC administered with FFP. Some investigators have suggested that alloantibodies, present in transfused FFP, directed against the PI A I antigen located on platelet glycoprotein (GP) Ilia receptor results in increased platelet destruction by macrophages in the reticuloendothelial system (Solenthaler et al., 1999; Nugent, 1992; Brunner-Bolliger et al, 1997; Nijjar et al, 1987). FFP transfusions are indicated for patients who are actively bleeding or facing a hemostatic challenge with a deficiency of multiple coagulation factors (Gernsheimer, 2003; Drews, 2003; American Red Cross, 2002; Marik, 2001; Bucur et al, 2000). This includes patients with severe liver disease who may have low concentrations of vitamin K-dependent clotting factors (i.e. II, VII, IX X); those who receive massive transfusions of PRBC, which lack 185 coagulation factors; patients who need correction of coagulopathies before invasive or surgical procedures (Drews, 2003); patients with DIC, usually due to a specific underlying cause; and patients who need a rapid reversal of the anticoagulant effect of warfarin because of active bleeding, emergency surgery, or serious trauma. Investigators have reported that following retrospective audits of patients administered FFP transfusions, 60% (Kakker et al, 2004) and 37% (Schofield et al, 2003) of transfusions were given for inappropriate indications, such as hypoproteinemic states, anemia, volume depletion, and bleeding without coagulation deficiency (Kakker et al., 2004). FFP transfusions should not to be used for the following indications: volume expansion (unless the patient also has a major coagulopathy and is bleeding) (Gernsheimer, 2003; Bucur et al., 2000), to treat polycythemia (unless there is a co-existant coagulopathy), as a source of nutrients (Bucur et al., 2000), or to improve immune function in septic patients (Gibson et ah, 2004). In the present investigation, only 5 of the 21 patients who received FFP transfusions had a documented active bleeding episode prior to or during administration of the transfusion. The administration of FFP transfusions is not free of risks. Risks include allergic reactions, circulatory (fluid) overload, infectious complications, transfusion-related acute lung injury, and antibodies present in FFP that may react with the recipient's red blood cells, causing a positive direct antiglobulin test, and possibly hemolysis (American Red Cross, 2002; Kakker et al., 2004; Marik, 2001). In addition, if FFP transfusions do increase the risk of thrombocytopenia, then it could increase patients' subsequent risk of bleeding. Pulmonary artery catheter insertion was also associated with an increased risk for thrombocytopenia in the four exploratory post-admission models (Models 1PA - 4 PA; Tables 10A, 12A, 14A and 16A). Pulmonary artery catheters have previously been implicated in the development of thrombocytopenia (Kim et al., 1980; Miller et ah, 1984; Layon, 1999; McNulty et al., 1998; Vicente Rull et al, 1984). However, two groups of investigators (Kim et al., 1980; 186 Vicente Rull et al, 1984) never observed a decline in the platelet count < 150 x 109/L in critically i l l ICU patients with pulmonary artery catheters, but they did note a "statistically significant" drop in the platelet count from baseline. Baughman et al (1993) found that pulmonary artery catheter use was associated with thrombocytopenia following univariate analysis, but not after multivariate linear regression analysis. Bonfiglio et al (1995) reported that hemodynamic instability (pulmonary artery catheter and administration of vasoactive adrenergic compounds) was independently associated with thrombocytopenia after multivariate linear regression analysis, but it is unclear whether this association was a result of one of these interventions, both interventions together, or whether these interventions were indicators of another causal factor. In a prospective study, Cawley et al (1999) observed that insertion of invasive central or arterial lines was independently associated with thrombocytopenia in critically i l l mixed surgical-trauma patients. In the present investigation, data for central lines were not collected; however, such lines may be comparable to pulmonary artery catheters in that they introduce a foreign surface and can lead to non-immune platelet destruction similar to pulmonary artery catheters (Bogdonoff et al., 1990). As expected, patients who had a pulmonary artery catheter inserted were noted to have a higher mean A P A C H E II score (25 ± 9) than patients without a pulmonary artery catheter (14 ± 7) in both datasets, indicating that catheters were used in patients who were more severely i l l . The interaction between pulmonary artery catheter insertion and A P A C H E II score was explored, but was not statistically significant in this patient cohort. Therefore, pulmonary artery catheters have a direct effect on platelet count or are indicators for some other unidentified process associated with thrombocytopenia, or both. There is no established explanation for the decline in the platelet count experienced by some patients following insertion of these catheters; however, it has been suggested that these catheters have a local or systemic anti-platelet effect resulting in non-immune platelet destruction (Bogdonoff et al., 1990). Furthermore, heparin is bonded to the surface of pulmonary artery 187 catheters and low doses of heparin are continuously infused to keep them patent. However, it is unlikely thrombocytopenia was due to heparin because the anticoagulant was not associated with an increased risk for thrombocytopenia, but was found to be associated with a decreased risk in Models 3PA and 4PA (Tables 14A and 16A). The benefit of pulmonary artery catheters has been debated for the past two decades (Sandham et al, 2003; Dalen, 2001; Connors et al, 1996; Brandstetter et al, 1998; Bender, 1999; Dalen and Bone, 1996; Robin, 1985). In general, these catheters are used for monitoring unstable critically i l l patients in order to: assess left or right ventricular function, or both; monitor changes in hemodynamic status; guide therapy with pharmacologic and non-pharmacologic agents; and provide prognostic information (Voyce and McCaffree. 2003). These catheters may be useful to evaluate and optimize intravascular status in refractory shock, ARDS and acute lung injury, acute myocardial infarction complicated by acute cardiac failure, and high risk surgical patients (perioperative management) (Marik, 2001a). However, prospective randomized controlled trials have never been performed to clearly define indications for its clinical use (Bender, 1999; Parsons, 2003), and interpretation of the data generated by these catheters and the goals of therapy directed by them remain to be clarified (Parsons, 2003). Evidence from randomized and non-randomized studies in critical and non-critical care settings suggests that overall, patients managed with pulmonary artery catheters have either worse outcomes (Connors et al, 1996; Polanczyk et al., 2001) or no difference in outcomes (Richard et al, 2003; Y u et al, 2003; Barone et al, 2001; Sandham et al, 2003; Chittock et al, 2004) when compared to those managed without pulmonary artery catheters. However, none of these studies, especially those performed in critically i l l patients, evaluated the impact these catheters had on a patients' platelet count (i.e. the development of thrombocytopenia). Recently, however, Rhodes et al (2002) performed a prospective, randomized, controlled pilot study designed to assess the effect of pulmonary artery catheter monitoring on survival and other clinical outcomes in 188 critically i l l patients. These investigators reported that there was no significant difference in mortality between the two groups, but observed that those randomized to the pulmonary artery catheter arm developed thrombocytopenia (not defined) more frequently (p < 0.03) and experienced a greater degree of renal dysfunction. Although the authors of this pilot study did not define thrombocytopenia, it is the only randomized study performed that has provided some evidence of a causal association between pulmonary artery catheter insertion and thrombocytopenia. This is consistent with the rinding in the exploratory post-admission models (Models 1 P A - 4 P A ; Tables 10A, 12A, 14A, 16A) in the present investigation. The ongoing debate about the appropriateness of pulmonary artery catheter use in critically i l l patients and the increased risk of thrombocytopenia and other outcomes (mortality) warrant further large prospective randomized studies that include investigation of the association of thrombocytopenia with pulmonary artery catheter use in these patients. PRBC transfusion was also associated with an increased risk for thrombocytopenia in all four exploratory post-admission models (Models 1PA - 4 PA; Tables 10A, 12A, 14A and 16A). In previous studies, PRBC transfusions (Baughman et al, 1993) and the number of PRBC units transfusions administered (Hanes et al., 1997) were associated with thrombocytopenia following univariate analysis. Stephan et al (1999) reported that a variable "episodes of bleeding or transfusions" was associated with thrombocytopenia following multivariate analysis. However, their variable "transfusions" consisted of R B C and platelet transfusions, and it is unclear whether "transfusions", or more specifically RBC transfusions, were independently associated with the development of thrombocytopenia. Neither Strauss et al (2002) nor Cawley et al (1999) included PRBC transfusions as a candidate variable in their logistic regression model. However, Cawley et al (1999) noted that a limitation of their study was the lack of monitoring of blood products administered, while Strauss et al (2002) and Stephan et al (1999a) reported that most PRBC transfusions were administered after thrombocytopenia developed. 189 It has been suggested that the dilution in the platelet count observed in patients transfused with red blood cells or whole blood is related to the number of transfused units, and occurs relatively early after the transfusion (Hardy et al., 2004; Hanes et al, 1997; Counts et al., 1979; Wittels et al., 1990; Reed et al., 1986; Noe et al., 1982). However, it is unclear how many units comprise a large transfusion. The post-transfusional decline in platelet count can be ascribed to the following: dilution of platelets in the circulation by blood products containing low concentrations of viable platelets (Bogdonoff et al., 1990; Wittels et al., 1990) or that do not contain platelets (Hardy et al., 2004), or sequestration of platelets by the spleen following blood transfusions (Bareford et al., 1987). In the present investigation, the maximum number of units of PRBC transfused in 24 hours was 6, and the maximum number transfused during a patient's stay in the ICU was 15 (over 8 days). There has been much debate regarding the risks and benefits of red blood cell transfusion in the critically i l l . In the last several years, studies have been published to evaluate and re-define the threshold hemoglobin concentration that warrants transfusion in critically i l l patients (Chohan et al., 2003; Hebert et al., 1999; Hebert et al., 1998). In addition, several studies have documented that administration of red blood cell transfusions for treatment of anemia in hemodynamically stable critically i l l patients resulted in worse clinical outcomes (Corwin et al., 2004; Vincent et al, 2002a; Napolitano, 2004). In the critical care setting, indications for PRBC transfusion include: a hemoglobin concentration < 90 g/L in patients with evidence of cardiac decompensation and/or history of coronary artery disease, acute G l bleed with hematocrit < 0.30, and a hemoglobin concentration < 75 g/L in all other intensive care patients (Marik, 2001). However, clinicians should evaluate each patient individually, paying particular attention to the patient's age, volume status, bleeding risk, and presence of ischemic heart disease (Drews, 2003; Grensheimer, 2003). The risks of PRBC transfusions are very similar to those stated for FFP transfusions. In the present investigation, physicians ordered PRBC transfusions when patients' 190 hemoglobin concentration was declining. In most cases it was between 90 - 100 g/L at the time PRBCs were administered (data not shown). Based on the results of the present and previous investigations, it is thus unknown whether PRBC transfusion is a marker for other potential causes or is a direct contributor to the thrombocytopenia in critically i l l patients, or both. There were two additional risk indicators identified in models 3PA and 4PA (Tables 14A and 16A), imipenem administration and heparin dose/day. Imipenem, a broad spectrum carbapenem antibiotic, was the only medication associated with an increased risk for thrombocytopenia. Cawley et al (1999) found imipenem to be associated with thrombocytopenia, but only following univariate analysis. In another investigation, Strauss et al (2002) reported that carbapenems were not associated with thrombocytopenia following univariate analysis. There is little published evidence of how imipenem may cause thrombocytopenia; however, in a review article on the hematologic side effects of drugs, Lubran (1989) suggested that thrombocytopenia may result from an immune mechanism that results in peripheral platelet destruction. In the present investigation, imipenem was generally administered to patients as treatment for an infection. Patients given imipenem had a higher mean A P A C H E II score (26 ± 10) than patients who were not administered imipenem (15 ± 8); however, the interaction between imipenem and A P A C H E II score was not statistically significant. Thus, imipenem either has a direct effect on platelet count or was an indicator for some other unidentified process associated with thrombocytopenia. Heparin was widely used in the ICU/CCU patients (86%), and increasing heparin dose/day was found to be independently associated with a decreased risk for thrombocytopenia in Models 3PA and 4PA (Tables 14A and 16A). In previous studies involving critically ill patients, heparin was not found to be associated with thrombocytopenia (Baughman et al, 1993; Bonfiglio et al, 1995; Hanes et al, 1997; Cawley et al, 1999; Stephan et al, 1999; Strauss et 191 al, 2002), and it does not appear there are any reports of heparin associated with a decreased risk for thrombocytopenia in particular patient populations. It is possible that the present cohort of critically i l l ICU/CCU patients represented a different subgroup of patients who had a lower risk of developing thrombocytopenia with heparin administration. A possible explanation for a decreased risk has been postulated by Kitchens (2004), who reported that occasionally, patients with acute venous thomboembolism will develop thrombocytopenia, referred to as acute thrombosis-associated thrombocytopenia. Platelets are consumed at the surface of these large clots, resulting in thrombocytopenia, especially i f the platelet count was close to the absolute threshold for thrombocytopenia. Heparin is active in preventing or treating thromboembolic disease and thus, may decrease the consumption of platelets at the site(s) of the thromboembolism, which may result in an increase in the platelet count. In the present investigation, 19 of 792 (2.4%) patients included in the > 100 x 10 9/L dataset had a documented thromboembolism, and of these only 3 were given full anticoagulation (high dose) for thrombosis therapy (> 16,000 units/day). Thus, it is unlikely that heparin reduced the risk of thrombocytopenia by the above mechanism in this study. Despite the fact that heparin has been suggested to cause a non-immune related thrombocytopenia referred to as HAT (Chong, 1992; Greinacher 1995; Chong 1995; Chong and Castaldi, 1986), it was not independently associated with an increased risk for thrombocytopenia in the present investigation. There do not appear to be any well-designed randomized controlled studies evaluating the impact of heparin therapy on the platelet count during the first few days of administration, as previous studies have been small and have produced conflicting (inconclusive) results (Gollub and Ulin, 1962; Davey and Lander, 1968; Saffle et al., 1980; Schwartz et al., 1985). The results of the present investigation suggest that HAT was not a strong contributor to thrombocytopenia in this sample of critically i l l patients, and thus poses the question whether HAT is a real phenomenon. 192 Heparin-induced thrombocytopenia (HIT) appears to be relatively uncommon (< 5%) immune-mediated adverse effect that can result in serious limb- and life-threatening outcomes. However, it is highly unlikely that HIT would emerge as an independent risk indicator following multivariate logistic regression analysis based on the results of the present investigation because the observed incidence of HIT was so low (one patient) that it would not be observed to contribute to an overall increased risk for thrombocytopenia in logistic regression models. It would require many cases of HIT for it to be identified as an independent risk indicator. In summary, the exploratory post-admission models provided additional information for thrombocytopenia in this patient cohort. There was a modest increase in the discriminating ability and slight improvement in the fit of the four exploratory post-admission models as compared to its admission model. The post-admission explanatory variables identified in these models are either directly causative for thrombocytopenia or are indicators for some other factor(s). The results of the present and other observational investigations, as well as those from a recent randomized controlled pilot study provide some evidence of a causal association between pulmonary artery catheter insertion and thrombocytopenia. In general, thrombocytopenia in the critical care setting is evidently a multi-factorial outcome, as a combination of risk indicators is involved in decreasing the platelet count below the threshold for most patients. Risk indicators not identified Critically i l l patients are exposed to many potential risk indicators previously associated with thrombocytopenia, but most of these were not identified as independent risk indicators in the present investigation. In review articles of thrombocytopenia in the critical care setting, DIC, ITP, and TTP are commonly reported to be potential causes for thrombocytopenia (DeLoughery, 2002; Bogdonoff et al, 1990; Wittels et al, 1990), and Strauss et al (2002) reported that 193 evolution of DIC was independently associated with the development of thrombocytopenia in their cohort of patients. In the present investigation, these three risk indicators were not identified as risk indicators largely due to the exclusion criteria, and the fact that no cases of ITP or TTP, and only two cases of DIC were observed post-admission. Certain medications, such as vancomycin (Bonfiglio et al, 1995; Cawley et al., 1999; Wazny and Ariano, 2000), quinine, and quinidine (Bogdonoff et al., 1990; Wittels et al., 1990; Wazny and Ariano, 2000), have been suggested to be associated with thrombocytopenia; however, due to the small number of patients exposed to these variables (< 2%), they were not included in the multivariate analysis. In a review article of H2-antagonist-induced thrombocytopenia Wade et al (2002) listed published reports implicating this drug class as a cause of thrombocytopenia. ^-antagonists are commonly administered to critically i l l patients for stress ulcer prophylaxis (Lam et al., 1999; Cook et al., 1998), and in the present investigation, 30% of the ICU/CCU patients were administered these agents. In previous studies involving critically i l l patients (Hanes et al., 1997; Cawley et al, 1999; Stephen et al, 1999; Bonfiglio et al, 1995; Strauss et al, 2002), H 2 -antagonists were not reported to be associated with thrombocytopenia. Bonfiglio et al (1995), however, found that duration of ^-antagonist therapy accounted for 1% of the overall variance in platelet count following stepwise linear regression analysis. Wade et al (2002) argued that Wi-antagonist-induced thrombocytopenia does in fact occur, but that the incidence has been over-reported, and Marik (2001) suggested that the true incidence of this drug-induced adverse effect is probably < 0.1%. 4.3 MODEL EVALUATION Regression analyses are used to develop predictive models, and the apparent performance of these models on the derivation set is generally better than the performance of these same models on new data (Justice et al, 1999; Harrell et al, 1996; Bleeker et al, 2003; Steyerberg et 194 al, 2001). In general, few investigators have validated or tested their predictive logistic regression models on a new dataset. None of the investigators who developed multivariate models to describe thrombocytopenia (Baughman et al, 1993; Bonfiglio et al, 1995; Hanes et al, 1997; Cawley et al, 1999; Stephen et al, 1999; Strauss et al, 2002; Shalansky et al, 2002) performed validation studies to assess the predictive performance of their models on a new data set. In the present investigation, internal and external validation of the logistic regression models generated at L G H were performed by the regular bootstrap technique and on data obtained from a tertiary care ICU (SPH), respectively. 4.3.1 Internal validation Internal validation of the L G H admission and exploratory post-admission models was performed by the regular bootstrap technique (Steyerberg et al, 2001). The models demonstrated low bias, which suggested that they would not be overly optimistic (indication of the extent to which the A U C of the original model overestimates the likely ability of the model to discriminate in a new sample of patients) when applied to a similar group of critically i l l patients. The values for optimism of the models developed in this investigation (0.008 - 0.033) were lower than that observed by Bleeker et al (2003) (0.070) who developed and validated a logistic regression model for the presence of serious bacterial infections in 376 children with fever without source, but higher than those reported by Lee et al (1995) (0.002) and Steyerberg et al (2000) (0.001) who validated logistic regression models for 30-day mortality on data obtained from acute myocardial infarction patients included in the GUSTO-I trial (GUSTO-I investigators, 1993). However, there were considerably more patients (> 40,000) included in the latter two studies, which likely explains the low values for optimism obtained by those investigators. It has been reported that the regular bootstrap technique results in stable and nearly 195 unbiased estimates of performance when compared to other internal validation techniques, such as data-splitting and cross-validation (Steyerberg et al., 2001). In a recent study, Steyerberg et al (2003) recommended the bootstrap procedure for internal validation because it provided reasonably valid estimates of the expected optimism in predictive performance. In the present investigation, 200 random bootstrap samples were drawn with replacement from the original (derivation) L G H datasets. Each random bootstrap sample represented a simulation of a future set of patients at the same or a very similar institution. Evaluation of the model's performance in these random bootstrap samples provided an estimate of how it might perform if it was used to predict thrombocytopenia at L G H or a very similar critical care setting. The results obtained from the bootstrap procedure suggest that these models should perform well in similar settings. Cox Proportional-Hazards Regression In the present investigation, explanatory variables identified by logistic and Cox proportional-hazards regression were consistent (Tables 18A - 18H). This indicates that right censoring was not an issue in the logistic regression analyses for the development of thrombocytopenia in this cohort of critically i l l patients. 4.3.2 External validation Models often have high predictive potential on the data from which they were derived, but typically their performance on new sets of data will decrease. Most researchers who have published reports evaluating predictive models developed by logistic regression have focused on internal validity. Very few have assessed the external validity of their models. Internal validation is performed on data from the original patient sample and does not provide a good indication of bias when these models are applied in a new group of patients at a different setting (Bleeker et al., 2003). In order to investigate the generalizability of these models, the 196 performance should be evaluated on new data collected from a different, but plausibly related population. External validity is considered a stronger test of a model's performance than internal validity because it accounts for the differences in the case-mix between clinical settings (Steyerberg et al, 2001). In the present investigation, external validation of the four L G H admission models was performed on data from ICU patients admitted to SPH. The L G H admission models demonstrated acceptable (Models 1 and 2; Tables 9A and 11 A) and excellent (Models 3 and 4; Tables 13A and 15A) discriminating ability (Hosmer and Lemeshow, 2000c) (Table 23), respectively. Not surprisingly, there was a slight decrease in predictive performance when each L G H admission model was applied to the SPH data. The two L G H ICU admission models (Models 2 and 4) appeared to perform slightly better than the two L G H ICU/CCU admission models (Models 1 and 3) (Table 23) as illustrated by a smaller relative decrease in predictive performance for model 2 (0.806 to 0.750 (7%)) as compared to model 1 (0.834 to 0.742 (11%)). A smaller relative decrease in predictive performance was also noted for model 4 (9%) as compared to model 3 (13%). These results are not unexpected as the SPH ICU case-mix is more comparable to the L G H ICU than the L G H ICU/CCU case-mix. One notable finding following external validation of the four L G H admission models on SPH ICU data was that they tended to systematically over-predict the development of thrombocytopenia in SPH patients, especially at higher predicted probabilities (Figures 18A, 18B, 19A, and 19B). Some loss of calibration was not surprising given the differences in the patients admitted to the two different critical care settings. For example, SPH is an inner-city, tertiary care, teaching hospital whereas, L G H is a community-based, non-teaching hospital. Patients admitted to the critical care unit at SPH are ICU patients and not a mix of ICU and C C U patients as seen at L G H . There was a notable difference in patient demographics and admission diagnoses among the ICU patients admitted to L G H as compared to those at SPH (Tables 19 -197 22). There was a slightly higher proportion of males, but lower proportion of Caucasians admitted to the ICU at SPH as compared to L G H . In addition, SPH ICU patients had a slightly higher A P A C H E II score and more expired in the ICU, suggesting a somewhat higher severity of illness. The SPH ICU was comprised of a higher proportion of patients with an infection and sepsis admission diagnosis, but lower proportion of those who had surgery 24 hours prior to admission as compared to the L G H ICU. Despite these differences in case-mix, the four L G H admission models still demonstrated acceptable and excellent discriminating ability when applied to the SPH ICU data. Another possible reason why the L G H admission models over-predict the development of thrombocytopenia in SPH patients might be the interventions or procedures being performed in the SPH ICU. Physicians at SPH may be intervening by giving platelet transfusions at higher platelet count thresholds or performing some other procedure, so that some ICU patients will not develop thrombocytopenia, even though they were predicted to by the L G H admission model. The results of the internal and external validation analyses obtained in the present investigation were much better than those reported in a recent study by Bleeker et al (2003). They developed a prediction model using logistic regression for the presence of serious bacterial infections in a sample of 376 pediatric patients with fever without source, and validated it internally using the regular bootstrap technique, and externally on data obtained from a combined group of 179 pediatric patients admitted to the same hospital and a different hospital during a different time period. Nine predictors were identified in the derivation dataset, and the area under the ROC curve of this predictive model was 0.83. Results of internal validation suggested substantial optimism (0.070; area under the ROC curve following bootstrap correction was 0.76), and when the model was applied to data from 179 patient external validation set, there was a large decrease in predictive perfonnance (area under the ROC curve was 0.57). The authors suggested that the poor performance of their model in the external validation set might 198 have been due to overfitting of the data (i.e. large number of candidate variables to the number of events (infections)); slightly different case-mix in the external dataset; and general aspects such as the selection of patients (e.g. referral pattern) between those included in the derivation and external validation datasets. In the present investigation, the range of the optimism for the L G H admission and exploratory post-admission models (0.008 - 0.033) was lower and the discriminating ability with the external sample was greater than observed by Bleeker et al 2003. 4.3.3 Model evaluation summary and conclusions In the present investigation, internal validation of the L G H admission and exploratory post-admission models suggested excellent discriminating ability and low bias, especially for the ICU/CCU and ICU < 100 x 109/L models, which suggests that these models should perform reasonably well in similar settings External validation of the four L G H admission models on data from SPH ICU patients demonstrated acceptable to excellent discriminating ability, but these models tended to over-predict the development of thrombocytopenia (i.e. poor calibration), possibly due to differences in case-mix between L G H and SPH or interventions performed at SPH resulting in fewer patients developing thrombocytopenia. The relatively high discriminating ability suggests that the explanatory variables identified in L G H admission models provide important information with regard to the development of thrombocytopenia. On the other hand, the calibration of the L G H admission models on the SPH data was not very good, suggesting that applying these models to individual patients would result in systematic inaccuracy in the prediction of thrombocytopenia in future patients. Thus, i f the intention is to predict thrombocytopenia in individual ICU patients at SPH or another setting, for research or clinical purposes, the optimum approach would likely be to develop logistic regression models at the specific site, using methodology similar to that described herein. 199 4.4 LIMITATIONS OF THE PRESENT INVESTIGATION Right censoring of data A common consideration of outcome studies is right censoring of the data due to different lengths of follow-up. In principle, logistic regression analysis is performed by modeling a group of explanatory variables, measured at a single point in time (e.g. baseline). Patients are then followed for a defined period of time, and the outcome of interest is noted for each patient. To date, many investigators have designed studies that do not readily allow for a defined follow-up period and thus, have not accounted for right censoring of the data. For example, studies conducted in a variety of clinical settings (Brown et al., 1988; Higgins et al., 2003; Engoren et al., 1999; Tostivint et al., 2002; Isaacman et al., 2000), as well as those conducted to investigate thrombocytopenia in critical care patients (Hanes et al., 1997; Stephen et al., 1999; Strauss et al., 2002; McClure et al., 1999; Shalansky et al., 2002) have not accounted for right censoring of the data. Several authors (Ingram and Kleinman, 1989; Annesi et al., 1989; Green and Symons, 1983; Hauck, 1985; Brenn and Arnesen, 1985; Steenland et al, 1986) have concluded that when the outcome is relatively infrequent and the follow-up period short, logistic regression modeling produces similar results to Cox proportional hazards, which inherently accounts for censoring of the data. Green and Symons, (1983) and Ingram and Kleinman, (1989) noted that models generated using Cox proportional hazards and logistic regression analyses resulted in similar regression estimates and standard errors when the incidence of the disease is between 10% and 19%. This was demonstrated in a mortality study involving 671 patients who had coronary angiography and met explicit clinical criteria for coronary revascularization (Kravitz et al., 1995). Seventy (10.4%) of these patients died, and following logistic regression and Cox proportional hazards analyses, the regression estimates for death at one year among the patients 200 who received coronary revascularization were very similar (OR 0.49; 95% CI: 0.28 - 0.86; relative hazard 0.59; 95% CI: 0.36 - 0.97). In the present investigation the incidence of thrombocytopenia among patients in the ICU/CCU > 150 x 109/L and > 100 x 109/L datasets was 17.3% and 10.6%, respectively (Table 7), and most patients developed thrombocytopenia within the first few days of ICU/CCU stay (60% by day 2 and 90% by day 4) (Figure 1). The explanatory variables identified by logistic and Cox proportional-hazards regression analyzes were similar, and the regression estimates for these variables were comparable, especially ICU Models 2, 2A, 4 and 4A (Tables 18C, 18D, 18G, and 18H). Thus, the exploratory variables identified were not greatly affected by right censoring of the data (Tables 18A - 18H). Time-dependent variables A limitation specifically related to the development of the exploratory post-admission models (Models 1PA - 4PA; Tables 10A, 12A, 14A 16A) was the inclusion of time-dependent post-admission variables (e.g. FFP transfusion, pulmonary artery catheter insertion). These are independent variables that change value at a particular point in time over the course of a longitudinal study (Katz, 2003; Wolfe and Strawderman, 1996). Other investigators who used logistic regression analysis to examine risk indicators for thrombocytopenia in critical care patients also included time-dependent variables in their analysis without adjusting for their time-varying nature (Hanes et al., 1997; Stephen et al., 1999; Strauss et al., 2002; McClure et ah, 1999). In addition, many investigators have used logistic regression to model patient data obtained from a variety of clinical settings and have not adjusted for the time-varying nature of the candidate variables (Brown et al., 1988; Higgins et al., 2003; Engoren et al., 1999; Tostivint et al, 2002; Sanders et al, 2001; Slotman et al, 2000; Rue et al, 2001). Logistic regression analysis does not directly model time-dependent data. Investigators 201 often include time-dependent variables in their studies when they believe that changes in these variables over time may be more predictive of the outcome than their measurement at baseline (Cupples et al., 1988). For studies in the critical care setting, some investigators have accounted for time-dependent variables by developing a new logistic regression model for each day of ICU stay (Wagner et al, 1994; Lemeshow et al, 1994) or by pooling repeated observations of risk factors and then performing logistic regression (referred to as person-time interval logistic regression, as each unit of analysis is a person-time interval) (Cupples et al., 1988; D'Agostino et al., 1990). In a strict sense, researchers who develop models that incorporate time-dependent variables should adjust for the time-dependent nature of these variables. If not, they might potentially miss identifying causal factors or identify spurious factors associated with the outcome of interest. In the present investigation, the admission models (Models 1 -4 ; Tables 9A, 1 IA, 13A, 15A) demonstrated excellent discriminating ability, and the post-admission risk indicators identified in the exploratory post-admission models (Models 1PA - 4PA; Tables 10A, 12A, 14A, 16A) provided additional information with respect to the development of thrombocytopenia, as demonstrated by the small increase in the area under the ROC curve and improved fit of these models. Most of the admission risk indicators were retained in the exploratory post-admission models (admission diagnosis, A P A C H E II score, admission platelet count). This indicates that the admission risk indicators were still associated with thrombocytopenia, even after inclusion of the post-admission risk indicators, and it appears the effect of the admission risk indicators was not manifested through their effect on the post-admission (time-dependent) variables. In this context, not accounting for the time varying nature of the post-admission variables appears to have provided new and potentially useful information. Bull and Spiegelhalter (1997) noted that, in contrast to randomized studies, the basis for selection and timing of interventions in observational studies has not been specifically 202 determined. In general, randomized comparative trials are conducted to control for confounders and should provide evidence whether time-dependent variables are associated with the outcome of interest. However, it is unclear what impact time-dependent variables have in observational studies that use multivariate regression analyses. In the present investigation, pulmonary artery catheter insertion was observed to be independently associated with thrombocytopenia and this is supported by evidence from a recent randomized controlled trial (Rhodes et al., 2002) of a possible causal association between pulmonary artery catheter insertion after admission and the occurrence thrombocytopenia. Future prospective studies of thrombocytopenia and other outcomes in critical care patients should be conducted to compare logistic regression analysis with other methods over a relatively short follow-up period, in which adjustments are made for post-admission time-dependent variables. Limitations associated with case-mix The identification of explanatory variables by multivariate modeling is limited to the case-mix. For example, even if similar definitions for thrombocytopenia are used and similar risk indicators evaluated, results cannot necessarily be transferred to other critical care settings, because the case-mix and treatment protocols may vary from one setting to another. Models should be developed at a specific site, especially if they are to be used to predict an outcome (e.g. thrombocytopenia). That is, for a model to perform well, it should be generated in the population of interest in which it is going to be used, or i f the intent is to develop a more universally applicable model to predict thrombocytopenia, the model should be generated from a larger dataset with a diverse population mix. The latter would be difficult to do as the study would need to be performed in a large number of critically i l l patients from many institutions using uniform definitions for medical/treatment interventions. 203 Other limitations There were other shortcomings associated with the design of this investigation. First, even though this was a prospective study, there was a reliance on physician and nursing charting, as data were obtained from ICU/CCU charts. Second, every attempt was made to verify pre-admission exposure to certain risk indicators (e.g. medications (heparin), procedures (surgery 24 hours prior to admission)), however, it is possible that information regarding these variables was missing, incorrectly charted, or not recalled by the patient/family. Third, it is possible that changes in clinical practice during the 2-year data collection period may have had an effect on the risk indicators identified. However, there were very few changes to the medical and nursing staff, and there did not appear to be major changes in medical interventions during this time. It is possible that changes in clinical practice after data collection might have an effect on future use of these models in this (LGH) as well as other critical care settings. Lastly, not all patients had their platelet counts measured daily; hence, it was possible that some patients may have developed thrombocytopenia earlier or that cases of thrombocytopenia were missed because platelet counts were measured every second or third day. Platelet counts were measured from specimens drawn as part of usual therapeutic intervention or routine care. Since informed consent was not required, blood samples were not requested to measure patients' platelet counts. 4.5 POTENTIAL CLINICAL APPLICATIONS OF LOGISTIC REGRESSION MODELING FOR THROMBOCYTOPENIA Thrombocytopenia and the risk of bleeding Bleeding is the major complication of thrombocytopenia, and it is especially problematic in critically i l l patients, who may already be prone to bleeding from stress ulcers (Cook et al., 1994), and who may be exposed to invasive diagnostic or therapeutic procedures such as catheterization or surgery. In these patients, the propensity to bleed has been reported to increase 204 as the platelet count declines below the normal range (Strauss et al., 2002; Vanderschueren et al., 2000; Chakraverty et al., 1996). Recently, Strauss et al (2002) reported that, compared to patients without thrombocytopenia, the risk of bleeding was more than three times higher in those with platelet counts of < 150 x 109/L and six times higher in those with platelet counts of < 50.0 x 109/L. In addition, bleeding, including G l bleeds, has been reported to increase the risk of mortality in critically i l l patients (Cook et al., 2001; Stephan et al., 1999a; Anonymous, 1996). Vanderschueren et al (2000) noted that bleeding secondary to thrombocytopenia may result in adverse outcomes, and Chakraverty et al (1996) reported that bleeding was the major contributor to death of ICU patients with clinical coagulopathy (bleeding unexplained by local or surgical factors). Critically i l l patients who develop thrombocytopenia have an increased risk of bleeding due to a variety of contributing factors, including exposure to certain medications (heparin, ASA), associated coagulopathy (e.g. vitamin K deficiency), and invasive procedures (e.g. pulmonary artery catheter insertion) (Wade et al, 2002; Chang, 1996; Wittels et al., 1990). In the present investigation there were relatively few risk indicators associated with thrombocytopenia identified by logistic regression modeling, and the admission models contained most of the information for discriminating between thrombocytopenic and non-thrombocytopenic patients. Models such as those developed herein could be used to provide knowledge of the risk indicators associated with the development of thrombocytopenia to help clinicians make informed decisions (i.e. weigh the risk-to-benefit ratio) of therapies or procedures that may contribute to a decrease in the platelet count and increase patients' risk for bleeding episodes. Typically, critical care patients are transfused with platelets when the count is < 20 x 109/L or lower in the absence of trauma, surgery, or active bleeding (Tinmouth and Freedman, 2003; British Committee for Standards in Haematology, Blood Transfusion Task Force). For 205 those who are actively bleeding, undergoing major surgery or invasive procedures, or who have undergone major surgery, it is generally recommended that platelet transfusions be administered to keep the platelet count above 50 x 109/L (Wittels et al., 1990; Drews, 2003; Anonymous, 1996; Nacht, 1992). However, these guidelines for platelet transfusions are based on evidence from other patient groups (oncology, hematology, surgical). There appears to be little evidence to guide therapeutic decisions regarding platelet transfusions for critical care patients. Future studies should be done to help establish guidelines for platelet transfusion that are specific to critical care patients. Then, models for thrombocytopenia may be useful in guiding clinical decision making. Bleeding episodes as a risk factor for thrombocytopenia The results of the present investigation are intriguing as it appears that episodes of bleeding may be associated with the development of thrombocytopenia. First, a number of the risk indicators identified in the admission and exploratory post-admission models appear to be indicators for episodes of bleeding. For example, a number of patients admitted with gastrointestinal, G l bleed, musculoskeletal/connective tissue, or vascular surgery diagnosis, had a bleeding episode prior to the development of thrombocytopenia, as did a number of patients who received FFP and PRBC transfusions. Thus, bleeding episodes may be a causative factor for thrombocytopenia. Second, the 2 patients identified as outliers in the present investigation (patient # 389 in the Models 1 and 1PA; patient # 433 in Model 3) had very low predicted probabilities of developing thrombocytopenia, yet they both developed thrombocytopenia that was preceded by a bleeding episode(s). Third, it was observed (data not shown) that patients with the most extreme values for Cook's distance (but still < 1.0) developed thrombocytopenia after a bleeding episode. In a prospective study involving surgical ICU patients, Stephen et al (1999) identified a variable "episodes of bleeding or transfusions" as an independent risk factor 206 for thrombocytopenia, and in a review article Vincent et al (2002) suggested that blood loss is a common cause of thrombocytopenia in critically i l l patients. A variable for episodes of bleeding was not defined a priori in the present investigation and thus, was not included in the analysis. Future research investigating risk indicators for thrombocytopenia in critical care patients should include and examine the association of episodes of bleeding (defined a priori) for the development of thrombocytopenia. Risk indicators and bleeding associated with severe thrombocytopenia Several investigators have defined severe thrombocytopenia as a platelet count < 50 x 109/L (Strauss et al., 2002; Baughman et al, 1993), and Strauss et al (2002) reported that the risk of bleeding was six times higher in patients with severe thrombocytopenia. No studies have been done to examine risk indicators independently associated with severe thrombocytopenia, and in the present investigation models were not developed because there were too few patients who met this criterion (Table 8). Stephan et al (1999a) performed a case-control study to investigate the effect of severe thrombocytopenia on mortality and blood transfusions in surgical ICU patients. They reported that severe thrombocytopenia was associated with increased mortality (relative risk 2.7; 95% CI: 1.02 - 7.10) and excess blood product (RBC, FFP, or platelet transfusions) consumption (relative risk 1.52; 95% CI: 1.05 - 2.20). They suggested that a platelet count < 50 x 109/L may be a marker for severe illness and increased risk of mortality, and that most of the RBC, FFP, and platelet transfusions were administered after the development of severe thrombocytopenia. This reflects current practice and suggests that patients may benefit from initiation of platelet transfusions at higher platelet counts. Thus, a study designed using similar methodology to the present investigation may provide evidence of risk indicators independently associated with severe thrombocytopenia. This might have an impact on physicians' decision making regarding medical and/or therapeutic interventions, 207 which may decrease a patient's platelet count and increase the bleeding risk. Furthermore, knowledge of risk indicators for severe thrombocytopenia may influence physicians' decisions regarding platelet count thresholds for platelet transfusions as described above. Potential application of logistic regression models for ruling out HIT (i.e. identification of other causes of thrombocytopenia in patients who meet the clinical criteria for HIT) In the present investigation, thrombocytopenia was common in ICU/CCU patients, and those admitted with an intensive care diagnosis had a higher incidence than those admitted with a coronary care diagnosis. Most ICU patients in this study cohort (84%) received heparin, and 43%) of these patients were at risk for developing HIT. Intensive care physicians often find it difficult to diagnose HIT because there are many other more common causes for thrombocytopenia in this patient population. One of the steps in diagnosing HIT involves ruling out other potential causes for the thrombocytopenia (i.e. non-HIT thrombocytopenia). Since there are many such causes in critically ill patients, clinicians could use the risk indicators from these models to judge whether causes other than HIT can explain the thrombocytopenia in any given case. Thus, these risk indicators could be used qualitatively to rule out HIT in a subset of those patients who meet the clinical criteria for this syndrome. Second, ICU models, such as those derived in this study, could be used to quantify the predicted probability of thrombocytopenia for each patient a priori, and thus identify those who were at high and low risk for thrombocytopenia prior to its onset. Intensive care physicians could use this information to rule out HIT in those with a high predicted probability for thrombocytopenia (non-HIT thrombocytopenia), or make decisions regarding diagnostic testing for HIT and continuing heparin therapy in those with a low predicted probability for non-HIT thrombocytopenia. However, future studies would be warranted to quantify the optimum model 208 and cut-point in order to identify patients with high and low predicted probabilities for thrombocytopenia (non-HIT thrombocytopenia) and to evaluate the utility of this approach. PART B: HEPARIN-INDUCED THROMBOCYTOPENIA 4.6 HEPARIN-INDUCED THROMBOCYTOPENIA IN CRITICAL CARE PATIENTS Heparin-induced thrombocytopenia (HIT) is a serious complication of heparin therapy that reportedly has a high rate of morbidity (thrombosis and amputation) and mortality (Chong, 1995; King and Kelton, 1984). Thrombocytopenia occurs commonly in the critical care setting, and most critically i l l patients receive heparin. This prospective investigation (Verma et al., 2003) is the first to provide an estimate of the incidence of HIT in critical care patients. Based on the clinical criteria and a positive SRA (the reference standard assay), the observed incidence of HIT among 259 community-based ICU/CCU patients at risk for this syndrome was 0.39% (95% CI, 0.01% - 2.1%o). As this estimate was based on a single intensive care patient who met the clinical criteria for HIT, it was not possible to determine whether the incidence of HIT in patients admitted with an intensive care diagnosis differed from that in patients with a coronary care diagnosis. Eight other patients met the clinical criteria for HIT, but the diagnosis could not be determined, as samples were not available for testing. It is possible, though not highly likely, that one or more of these patients also had HIT. However, had one additional case tested positive by the SRA, the estimate ofthe incidence of HIT would still be relatively low (< 1%). Fifty-four patients admitted to the ICU/CCU were excluded because they had admission platelet counts < 100 x 109/L, and thus had absolute thrombocytopenia on admission. Such patients are presumably at risk, however there are no published guidelines for diagnosing HIT in patients with pre-existing thrombocytopenia (platelet count < 100 x 109/L). 209 The incidence of HIT using the SRA has been investigated by other researchers in different groups of patients; however, the results have varied. Warkentin et al (1995) conducted a prospective randomized trial comparing unfractionated heparin and a low molecular weight heparin (enoxaparin) in 665 postoperative hip surgery patients. Among the 332 patients who were randomized to receive unfractionated heparin, the reported incidence of HIT was 2.7% (95% CI, 1.3% - 5.1%). Patients were diagnosed with HIT if they experienced at least one platelet count below 150 x 109/L after 5 days of treatment with heparin and had a positive SRA result. Eleven patients met the clinical criteria and 9 of these (81.8%) tested positive by the SRA. Subsequently, Warkentin et al (2000) re-analyzed data from a subset of these 332 patients using revised clinical criteria for HIT: > 50% decline in the platelet count from the postoperative peak that occurred between days 5 and 14 after surgery, and where no other cause for thrombocytopenia was apparent. With these revised criteria, they found that the HIT incidence in these postoperative orthopedic patients was 4.9% (95% CI, 2.4% - 8.8%). However, it should be noted that the authors performed a post-hoc determination of a new criterion for HIT, and this report illustrates the degree to which modification of the clinical criteria for HIT can influence the incidence of HIT being reported. Lower incidences of HIT have been reported by investigators who also used the SRA (Rao et al., 1989; Warkentin et al., 2000). In a multicenter study, Rao et al (1989) investigated the incidence of HIT in 193 medical patients who had received heparin for at least 5 days. These investigators reported that none of the patients developed thrombocytopenia, however their definition of thrombocytopenia was more stringent than other investigators' (i.e. a platelet count below lOOx 109/L on two occasions or below 75 x 109/L on one occasion). However, as none of their patients developed thrombocytopenia, the SRA was not utilized and the reported incidence of HIT was 0%. In the other study, Warkentin et al (2000) investigated the incidence of HIT in 100 cardiac surgery patients administered heparin. The clinical criteria they used were > 50% 210 decline in the platelet count from the postoperative peak that occurred between 5 and 14 days after surgery, and where no other cause for thrombocytopenia was apparent. One patient met the clinical criteria and tested positive by the SRA, yielding a HIT incidence of 1% (95% CI, 0.03% - 5.5%). Taken together, the latter two studies and the present investigation suggest that the incidence of HIT is < 1% in patients at risk, at least in medical, cardiac, and critically i l l ICU/CCU patients. Other than the study by Warkentin et al (2000), there do not appear to be any prospective studies that have reported an estimate of the HIT incidence using well-defined a priori clinical criteria (which would include specifying the timing of the thrombocytopenia in relation to heparin administration and excluding other potential causes of thrombocytopenia) and the SRA as the reference standard. There are other studies that have reported the incidence of HIT, but interpretation of these is complicated by inconsistencies in the clinical criteria used to define patients at risk (Pouplard et al., 1999; Kappers-Klunne et ah, 1997; Ansell et al, 1980; Powers et al., 1984), different assays to "confirm" the diagnosis of HIT (Girolami et al., 2003; Harbrecht et al, 2003; van Eps et al, 2001; Pouplard et al, 1999; Kappers-Klunne et al, 1997), and a lack of a clear indication of samples tested by the SRA (Ansell et al, 1980; Powers et al, 1984). Based on the studies that used well-defined a priori clinical criteria and the SRA as the reference standard (Verma et al, 2003; Warkentin et al, 2000; Rao et al, 1989; Warkentin et al, 1995; Lee and Warkentin, 2001), the incidence of HIT appears to differ among different patient populations, and the incidence in ICU/CCU patients may be lower than in other patient groups. Alternatively, it may be the case that the incidence is similar across patient groups (as confidence intervals from most of these studies overlap), but the studies performed to date have been too small to sufficiently obtain precise estimates of HIT incidence. To address the possibility that the HIT incidence differs among different patient populations, further investigations involving large numbers of patients in different clinical settings would need to be 211 conducted. These studies should be prospective and multi-centered, using uniform clinical criteria, including both absolute and relative thrombocytopenia and a well-defined baseline for identifying thrombocytopenia (Warkentin and Kelton, 2001; Warkentin, 2001), and they should use the SRA as the diagnostic test. The high apparent sensitivity and low false positive rate of the SRA support the use of this test as the reference standard (Leo and Winteroll, 2003; Warkentin et al., 1998; Warkentin and Greinacher, 2001; Morewood, 2000). However, the SRA is costly and technically difficult to perform (Chong et al., 1993; Warkentin and Greinacher, 2001; Greinacher et ah, 1991; Stewart et al., 1995; Arepally et al., 1995), and therefore is not widely used in clinical laboratories. ELISA methods have been developed to overcome these limitations, and at least two different ELISA tests have become commercially available (Warkentin and Greinacher, 2001), but their utility for HIT has not been systematically evaluated across patient populations. In the present investigation, the estimated positive and negative predictive values for the heparin-PF4 ELISA were 9.5% (95% CI, 0.25% - 44.5%) and 100% (95% CI, 84.6% - 100%), respectively. This result suggests that, in ICU/CCU patients, the heparin-PF4 ELISA is not useful for confirming a diagnosis of HIT. Interestingly, had the upper limit of the 95% CI for the prior probability of HIT (16.2%) and the estimated specificity of the heparin-PF4 ELISA (85.8%) been used, the estimated PPV would be only 56%, suggesting that the heparin-PF4 ELISA still would not be useful for confirming a diagnosis of HIT. However, due to its high estimated NPV, this test could be used to help rule out HIT as a possible cause for the thrombocytopenia. This may be useful in critical care settings where most patients receive heparin (86% in the present investigation) and have a high incidence of thrombocytopenia due to causes other than HIT (Strauss et al, 2002; Shalansky et al., 2002; Verma, 2000). Others have also suggested that a negative test result by the ELISA essentially rules out HIT (high NPV) (Warkentin and Heddle, 2003; Warkentin and Greinacher, 2003; Spinier and Dager, 2003). In 212 the present investigation, a negative heparin-PF4 ELISA result would have ruled out HIT in 22 of 31 (71%) non-HIT patients who met the clinical criteria and had a serum sample available, thus obviating a decision regarding a change in heparin therapy in the majority of such cases. However, as the estimated PPV and NPV were based on only one case of HIT, future investigations regarding the clinical utility of this test are warranted. Few investigators have reported the predictive values of the ELISA. Warkentin and Greinacher (2001) estimated the post-test probability for HIT if the ELISA result was positive or negative in post-operative cardiac and orthopedic patients. They calculated the post-test probability in the two patient populations based on what they considered to be a moderate (0.50) or high (0.90) pre-test (i.e. prior) probability for HIT, which they defined as the "physician's estimate of the likelihood of HIT in a given clinical setting". By their calculation, in post-operative cardiac patients with moderate or high pre-test probabilities for HIT, the positive post-test probabilities (i.e. PPV) would be 66% and 94%, respectively, for the ELISA based on a sensitivity and specificity of 95% and 50%, respectively. In post-operative orthopedic patients with moderate or high pre-test probabilities for HIT, the PPVs would be 92% and 99%, respectively, for the ELISA based on a sensitivity and specificity of 95% and 92%, respectively. The PPVs noted by Warkentin and Greinacher (2001) are higher than estimated in the present investigation, and this is largely due to the high pre-test probabilities they used in their estimation. They suggested that "a high pre-test probability (0.9) could be a patient whose timing of onset of thrombocytopenia suggests HIT, and there is a thrombotic event". It appears that they obtained this estimate in post-operative orthopedic patients (Warkentin et al, 2000; Warkentin et al., 1995), who, generally, have fewer potential risk factors for thrombocytopenia than critically i l l patients. This estimate was used to calculate the PPVs and NPVs in both post-operative orthopedic and cardiac patients. In post-operative orthopedic patients in whom HIT was not certain, they suggested a moderate pre-test probability of 0.5, and calculated the 213 predictive values based on this estimate. However, the prior probability of HIT is either unknown or appears to be lower than this in other patient populations. For example, the results of the present study indicate that, in ICU/CCU patients who meet the clinical criteria for HIT, the prior probability is only 3.1%. Warkentin and Greinacher (2001) also calculated negative post-test probabilities (i.e. NPV) for the ELISA based on the same sensitivities and specificities as stated above. In post-operative cardiac patients with moderate (0.5) or high (0.9) pre-test probabilities for HIT (i.e. associated pre-test probabilities for not having HIT would be 0.5 and 0.1, respectively), they reported the NPVs would be 9% and 47%, respectively, and in post-operative orthopedic patients with the same pre-test probabilities, the NPVs were reported to be 5% and 33%, respectively. However, their calculations appear to be incorrect. Using their data, the NPVs for post-operative cardiac patients with the same pre-test probabilities (i.e. 0.5 and 0.1 for not having HIT) should be 91% and 53%, respectively, and for post-operative orthopedic patients with these same pre-test probabilities, the NPVs should be 95% and 67%, respectively. These NPVs are lower than estimated herein, and this largely reflects the very high prior probability of not having HIT (96.9%) among the ICU/CCU patients who met the clinical criteria in the present investigation. Pouplard et al (1999a) evaluated the sensitivity, specificity, and the PPV and NPV of the heparin-PF4 ELISA in a subgroup of highly selected patients referred for hematology consult because of clinically suspected HIT and HIT-related thrombosis. These authors reported the PPV and NPV of the ELISA to be 93% and 95%, respectively, based on a sensitivity and specificity of 97% and 86%, respectively, which they estimated from results obtained in patients with and without clinical HIT. Their calculated PPV was higher than estimated in the present investigation in large part due to the high prior probability (0.67) they used in their estimation, and thus it would be incorrect to apply their estimates when diagnosing HIT in other unselected populations, such as critically i l l patients. 214 It has been suggested by other investigators that the false positive rate of the ELISA is greater than that of the SRA in non-thrombocytopenic patients who receive heparin for 5 or more days (Chong et al., 1993; Warkentin and Greinacher, 2001; Harbrecht et al., 2003; Pouplard et al., 1999; Amiral et al., 1996; Bachelot-Loza et al., 1998). In addition, the formation of antibodies to heparin-PF4 complexes may be greater in certain patient populations (e.g. cardiac surgical patients) (Warkentin et al., 2000; Bauer et al., 1997; Visentin et al., 1996). In the present investigation, the false positive rate ofthe heparin-PF4 ELISA (15%) was greater than that of the SRA (3%) (p < 0.003), which is consistent with findings of others. Thus, the predictive performance of diagnostic tests that directly detect heparin-PF4 antibodies may be dependent on the patient population in which it is used (Warkentin et al., 2000; Warkentin and Greinacher, 2001; Bauer et al, 1997; Pouplard et al, 1999). Had the heparin-PF4 ELISA been used as the diagnostic test in the present study, the estimated incidence of HIT would have been 3.9%, ten-fold higher than that obtained with the SRA. Laboratory confirmation of HIT has been problematic, because no single test has optimal sensitivity and specificity. The antigen assays (e.g. heparin-PF4 ELISA) have a high sensitivity, but may frequently be positive in patients without HIT. Functional assays (e.g. SRA, heparin-induced platelet activation (HIPA) test) also have high sensitivity and have very good specificity. Investigators have suggested that by performing a combination of antigen and functional assays, the sensitivity and specificity may be improved (Pouplard et al, 1999a; Warkentin et al, 1998; Eichler et al, 1999; Harenberg et al, 2001). Pouplard et al (1999a) found that performing both the heparin-PF4 ELISA and SRA simultaneously improved diagnostic accuracy. They noted that when both the heparin-PF4 ELISA and SRA were negative, the NPV approached 100%), and when both the heparin-PF4 ELISA (based on a sensitivity and specificity of 97% and 86%) and SRA (based on a sensitivity and specificity of 88%) and 100%) assays were positive, the specificity and PPV approached 100%). It is very important to note that these estimates were 215 based on a prior probability of 0.67, which is 20-fold higher than observed in the present investigation. Very few clinical institutions can perform both assays, and in particular functional assays (e.g. SRA), which are technically difficult to perform and time-consuming. Thus, the present investigation provides evidence of the possible use of a commercially available heparin-PF4 ELISA in critical care patients, which can be performed in a relatively short period of time (results were available in approximately 4 hours) in order to provide clinicians information with which to rule out HIT in a subset of patients at risk. It is important that the clinical criteria for HIT include both a relative and absolute thrombocytopenia (Warkentin and Kelton, 2001; Warkentin, 2001). This was illustrated in the HIT case presented in Figure 20. This patient experienced a decline > 33% in his platelet count from baseline, but his platelet count remained above 150 x 109/L. Had an absolute thrombocytopenia been used as the only criterion to identify potential HIT patients, this case would have been missed. In addition, this case illustrates the need to evaluate the magnitude of relative platelet count decline that should arouse clinical suspicion of HIT. A relative platelet count decline of > 50% has been suggested, but this was determined from a post-hoc analysis in post-operative orthopedic patients (Warkentin et al., 2003), and thus the optimum relative decline in the platelet count needs to be determined (Warkentin, 2001). Had a > 50% decline in the platelet count been used to define thrombocytopenia, this patient would not have met the clinical criteria and the diagnosis would have been missed, resulting in a lower incidence of HIT than observed. It is also possible that this patient's diagnosis of HIT was a false positive result, as the SRA was observed to have a 3% false positive rate in the present study, and false positive results have been noted by other investigators (Warkentin et al., 2000; Bauer et al., 1997; Chong etal, 1993). Heparin use and non-HIT thrombocytopenia are common in critical care patients (Shalansky et al., 2002; Verma, 2000; Strauss et al., 2002), and many of these patients meet the 216 clinical criteria for HIT (15% in the present investigation), but do not have this syndrome. The non-HIT case summarized in Figure 21 illustrates some common issues related to HIT diagnosis in these patients. This patient's platelet profile and timing of the thrombocytopenia were typical of published HIT cases, however his serum tested negative by both the heparin-PF4 ELISA and SRA. The more accessible heparin-PF4 ELISA could be useful for ruling out HIT in such cases because of its high apparent NPV. It is possible that this patient had a false negative result, but it is unlikely given the apparent high sensitivities of both the SRA and ELISA. 4.7 CONCLUSIONS AND IMPLICATIONS FOR FUTURE RESEARCH The incidence of HIT among community-based intensive and coronary care patients appears to be low (< 1%), and the diagnosis of this adverse effect is complicated by the fact that critical care patients often develop non-HIT thrombocytopenia, but also meet the clinical criteria for HIT. The observed incidence of HIT was lower than, but was not inconsistent with other studies. The low PPV ofthe heparin-PF4 ELISA precludes it from being used to replace the SRA for diagnosing HIT in this patient population. However, the apparent high NPV and accessibility of the heparin-PF4 ELISA make it a possible alternative for ruling out HIT in these patients. It may also be possible to use logistic regression models for non-HIT thrombocytopenia in conjunction with the heparin-PF4 ELISA to assist clinicians in ruling out HIT in patients who meet the clinical criteria for this syndrome. In principle, a stepwise approach could be considered whereby clinicians would first use the information provided by the logistic regression model to identify those patients who very likely have thrombocytopenia due to some cause(s) other than heparin. This should reduce the number of patients that physicians would then need to test for heparin-dependent antibodies. Next, the heparin-PF4 ELISA would be used to test the serum of the patients in whom HIT is still suspected. Utilizing the apparent high NPV of this assay, clinicians could rule out HIT in those who test negative, thus further reducing the number 217 of patients in whom a decision regarding heparin therapy (i.e. whether to discontinue heparin) would need to be made. However, as the incidence of HIT and clinical utility of the heparin-PF4 ELISA were based on one case of HIT in the present study, further investigation of the HIT incidence and clinical utility of this assay in critically i l l patients is warranted. 218 REFERENCES AbuRahrna A F , Boland JP, Witsberger T. Diagnostic and therapeutic strategies of white clot syndrome. A m J Surg 1991; 162:175-179. Akca S, Haji-Micheal P, de Mendonca A , Suter P, Levi M , Vincent JL. Time course of platelet counts in critically i l l patients. Crit Care Med 2002; 30:753-756. A m Surg. 2001 Jul; 67(7):674-9. American Red Cross. Practice guidelines for blood transfusion: a compilation from recent peer-reviewed literature. 2002. 20-25. (http://w\v\v.redcross.org/services/biomed/profess/pgbtscreen.pdf) Amiral J, Bridey F, Dreyfus M , Vissoc A M , Fressinaud E, Wolf M , et al. Platelet factor 4 complexed to heparin is the target for antibodies generated in heparin-induced thrombocytopenia [letter]. Thromb Haemost 1992; 68:95-96. Amiral J, Bridey F, Wolf M . Antibodies to macromolecular platelet factor 4-heparin complexes in heparin-induced thrombocytopenia: a study of 44 cases. Thromb Haemost 1995; 73:21-28. Amiral J, Marfaing-Koka A , Wolf M , Alessi M C , Tardy B, Boyer-Neumann C, et al. Presence of auto-antibodies to interleukin-8 or neutrophil-activating peptide-2 in patients with heparin-associated-thrombocytopenia. Blood 1996a; 88:410-416. Amiral J, Wolf M , Fischer A M , BoyerNeumann C, Vissac A M , Meyer D. Pathogenicity of IgA and/or IgM antibodies to heparin-PF4 complexes in patients with heparin-induced thrombocytopenia. Br J Haematol 1996; 92:954-959. Annesi I, Moreau T, Lellouch J. Efficiency of the logistic regression and Cox proportional hazards models in longitudinal studies. Stat Med 1989; 8:1515-1521. Anonymous. Practice Guidelines for blood component therapy: A report by the American Society of Anesthesiologists Task Force on Blood Component Therapy. Anesthesiology 1996; 84:732-747. Ansell J, Slepchuk N , Kumar R, Southard L, Deykin D. Heparin induced thrombocytopenia: a prospective study. Thromb Haemost 1980; 43:61-65. Ansell JE, Price JM, Shah S, Beckner RR. Heparin-induced thrombocytopenia: What is its real frequency? Chest 1985; 88:878-882. Arepally G, Reynolds C, Tomaski A , Amiral J, Jawad A, Poncz M , Cines DB. Comparison of PF4/Heparin ELISA assay with the 14C-serotonin release assay in the diagnosis of heparin-induced thrombocytopenia. Am J Clin Pathol 1995; 104:648-54. Arepally G M , Mayer IM. Antibodies from patients with heparin-induced thrombocytopenia stimulate monocytic cells to express tissue factor and secrete interleukin-8. Blood 2001;98:1252-1254. 219 Arnold D M , Smaill F, Warkentin TE, Christjanson L, Walker I. Cardiobacterium hominis endocarditis associated with very severe thrombocytopenia and Platelet Autoantibodies. Am J Hematol 2004; 76:373-377. Arrowsmith ER, Greer JP, Macon WR. Complications of hemostatic neoplasms. In: Lee GR, Foerster J, Lukens J, Paraskevas F Greer JP, Rodgers G M , eds. Wintrobe's clinical hematology. 10 ed. Baltimore M D : Williams and Wilkens, 1999:2033-2075. Aster R. Heparin-induced thrombocytopenia and thrombosis. N Engl J Med 1995; 323:1374-1376. Babcock RB, Dumper CW, Scharfman WB. Heparin-induced thrombocytopenia. N Engl J Med 1976; 295:237-241. Bachelot-Loza C, Saffroy R, Lasne D, Chatellier G, Aiach M , Rendu F. Importance of the cgammaRIIa-Arg/His-131 polymorphism in heparin-induced thrombocytopenia diagnosis. Thromb Haemost 1998; 79:523-528. Baker GR, Levin J. Transient Thrombocytopenia Produced by Administration of Macrophage Colony-Stimulating Factor: Investigations of the Mechanism. Blood 1998; 91:89-99. Bareford D, Chandler ST, Hawker RJ, Jackson N , Smith M , Boughton BJ. Splenic platelet sequestration following routine blood transfusion is reduced by filtered/washed product. Br J Haematol 1987; 67:177-180. Barone JE, Tucker JB, Rassias D, Corvo PR. Routine perioperative pulmonary artery catheterization has no effect on rate of complications in vascular surgery: a meta-analysis. Am Surg 2001;67:674-679. Bauer TL, Arepally G, Konkle B A , Mestichelli B , Shapiro SS, Cines DB, et al. Prevalence of heparin-associated antibodies without thrombosis in patients undergoing cardiopulmonary bypass surgery. Circulation 1997; 95:1242-1246. Baughman RP, Lower EE, Flessa HC, Tollerud DJ. Thrombocytopenia in the intensive care unit. Chest 1993; 104:1243-1247. Bell WR, Tomasulo PA, Alving F M , Duffy TP. Thrombocytopenia occurring during administration of heparin. A prospective study in 52 patients. Ann Intern Med 1976; 85:155-160. Bender JS. When is the pulmonary artery catheter needed in the care of the surgical patient? Adv Surg 1999;32:365-384. Berkowitz SD, Sane DC, Sigmon K N , Shavender JH, Harrington RA, Tcheng JE, et al. Occurrence and clinical significance of thrombocytopenia in a population undergoing high-risk percutaneous coronary revascularization. Evaluation of c7E3 for the Prevention of Ischemic Complications (EPIC) Study Group. J Am Coll Cardiol 1998; 32:311-319. 220 Bessman D. Determination of platelet volume and number. Methods in Enzymol 1989; 169:164-172. Bick RL, Strauss JF, Frenkel EP. Thrombosis and hemorrhage in oncology patients. Hematol Oncol Clin North Am 1996; 10:875-907. Bithell TC. Platelets and Megakaryocytes. In: Lee GR, Bithell TC, Foerster J, Athens JW, Lukens JN, eds. Wintrobe's Clinical Hematology. Philadelphia, PA. Lea & Febiger; 1993:511-539. Bithell TC. Thrombocytopenia. Pathophysiology and Classification. In: Lee GR, Bithell TC, Foerster J, Athens JW, Lukens JN, eds. Wintrobe's Clinical Hematology. Philadelphia, PA. Lea & Febiger; 1993a:1325-1328. Bleeker SE, Moll HA, Steyerberg EW, Donders AR, Derksen-Lubsen G, Grobbee DE, et al. External validation is necessary in prediction research: a clinical example. J Clin Epidemiol 2003; 56:826-832. Bogdonoff DL, Williams M E , Stone DJ. Thrombocytopenia in the critically i l l patient. J Crit Care 1990; 5:186-205. Bone R, Francis P, Pierce A. Intravascular coagulation associated with the adult respiratory distress syndrome. A m J Med 1076; 61:585-589. Bone RC, Balk RA, Cerra FB, Dellinger RP, Fein A M , Knaus W A , et al. Definitions for sepsis and organ failure and guidelines for the use of innovative therapies in sepsis. Chest 1992; 101:1644-1655. Bonfiglio M F , Traeger S M , Kier K L , Martin BR, Hulisz DT, Verbeck SR. Thrombocytopenia in intensive care patients: A comprehensive analysis of risk factors in 314 patients. Ann Pharmacother 1995; 29:835-842. Boon D M , Michiels JJ, Stibbe J, van Vliet HH, Kappers-Klurine M C . Heparin-induced thrombocytopenia and antithrombotic therapy. Lancet 1994; 344:1296. Boshkov L K , Warkentin TE, Hayward C P M , Andrew M , Kelton JG. Heparin-induced thrombocytopenia and thrombosis: clinical and laboratory studies. Br J Haematol 1993; 84:322-328. Bovill EG, Terrin M L , Stump DC, Berke A D , Frederick M , Collen D, et al for the TIMI Investigators. Hemorrhagic events during therapy with recombinant tissue-type plasminogen activator, heparin, and aspirin for acute myocardial infarction: results of the Thrombolysis in Myocardial Infarction (TIMI) Phase II trial. Ann Intern Med 1991; 115:256-265. Brandstetter RD, Grant GR, Estilo M , Rahim F, Singh K, Gitler B. Swan-Ganz catheter: misconceptions, pitfalls, and incomplete user knowledge—an identified trilogy in need of correction. Heart Lung 1998; 27:218-222. Brass LF. Thrombin and platelet activation. Chest 2003; 124:18S-25S. 221 Brenn T, Amesen E. Selecting risk factors: a comparison of discriminant analysis, logistic regression and Cox's regression model using data from the Tromso Heart Study. Stat Med 1985; 4:413-423. Brieger DB, Mak K H , Kottke-Marchant K, Topol EJ. Heparin-induced thrombocytopenia. J A m Coll Cardiol 1998; 31:1449-1459. British Committee for Standards in Haematology, Blood Transfusion Task Force. Guidelines for the use of platelet transfusions. Br J Haematol 2003; 122:10-23. Brown RB, Klar J, Teres D, Lemeshow S, Sands M . Prospective study of clinical bleeding in intensive care unit patients. Crit Care Med 1988; 16:1171-1176. Browner WS, Newman TB, Cummings SR. Designing a new study: III diagnostic tests. In: Cummings SR, Hulley SB, eds. Designing clinical research. Baltimore, M D ; Williams and Wilkins; 1988:87-97. Brun-Buisson C, Doyon F, Carlet J, Dellamonica P, Gouin F, Lepoutre A et al. Incidence, risk factors, and outcome of severe sepsis and septic shock in adults: a multicenter prospective study in intensive care units. J A M A 1995; 274:968-974. Brunner-Bolliger S, Kiefel V , Horber FF, Nydegger UE, Berchtold P. Antibody studies in a patient with acute thrombocytopenia following infusion of plasma containing anti-PIA 1. Am J Hematol 1997; 56:119-121. Bucur SZ, Berkman E M , Hillyer CD. Transfusion of plasma derivatives: fresh-frozen plasma, cryoprecipitate, albumin, and immunoglobulins. In: Hoffman R, Benz EJ Jr, Shattil SJ, Furie B, Cohen HJ, Silberstein L E , McGlave P eds. Hematology: Basic principles and practice. 3 r d ed. Philadelphia, Pa. J.B. Churchill-Livingstone; 2000:2263-2272. Bull K , Spiegelhalter DJ. Tutorial in biostatistics. Survival analysis in observational studies. Stat Med 1997; 16:1041-1074. Buller HR, Agnelli G, Hull RD, Hyers T M , Prins M H , Raskob GE. Antithrombotic Therapy for Venous Thromboembolic Disease: The Seventh A C C P Conference on Antithrombotic and Thrombolytic Therapy. Chest 2004; 126(3 Suppl):401S-428S. Burstein SA. Thrombocytopenia due to decreased platelet production. In: Hoffman R, Benz EJ Jr, Shattil SJ, Furie B, Cohen HJ, Silberstein L E , McGlave P eds. Hematology: Basic principles and practice. 3 r d ed. Philadelphia, Pa. J.B. Churchill-Livingstone; 2000:2115-2125. Cawley M J , Wittbrodt ET, Boyce E G Skaar DJ. Potential risk factors associated with thrombocytopenia in a surgical intensive care unit. Pharmacotherapy 1999; 19:108-113. Chakraverty R, Davidson S, Peggs K , Stross P, Garrard C, Littlewood TJ. The incidence and cause of coagulopathies in an intensive care population. Br J Haematol 1996; 93:460-463. Champion HR, Sacco WJ, Carnazzo AJ Copes W, Fouty WJ. Trauma Score. Crit Care Med 1981;9:672-676. 222 Chang JC. Postoperative thrombocytopenia: with etiologic, diagnostic, and therapeutic consideration. Am J Med Sci 1996; 311:96-105. Chittock DR, Dhingra V K , Ronco JJ, Russell JA, Forrest D M , Tweeddale M , Fenwick JC. Severity of illness and risk of death associated with pulmonary artery catheter use. Crit Care Med 2004; 32:911-915. Chohan SS, McArdle F, McClelland DBL, Mackenzie J, Walsh TS. Red cell transfusion practice following the transfusion requirements in critical care (TRICC) study: prospective observational cohort study in a large U K intensive care unit. Vox Sanguinis 3003; 84:211-218. Chong B H , Berndt M C . Heparin-induced thrombocytopenia. Blut 1989; 58:53-57. Chong B H , Burgess J, Ismail F. The clinical usefulness of the platelet aggregation test for the diagnosis of heparin-induced thrombocytopenia. Thromb Haemost 1993; 69:344-350. Chong B H , Castaldi PA. Platelet proaggregating effect of heparin: possible mechanism for non-immune heparin-associated thrombocytopenia. Aust N Z J Med 1986; 16:715-716. Chong B H , Eisbacher M . Pathophysiology and laboratory testing of heparin-induced thrombocytopenia. Semin Hematol 1998; 35(4 suppl 5):3-8. Chong B H , Grace CS, Rozenberg M C . Heparin-induced thrombocytopenia: effect of heparin platelet antibody on platelets. Br J Haematol 1981; 49:531-540. Chong B H . Heparin-induced thrombocytopenia. AustNZ J Med 1992; 22:145-152. Chong B H . Heparin-induced thrombocytopenia. Br J Haematol 1995;89:431-9. Chong B H . Heparin-induced thrombocytopenia. J Thromb Haemost 2003; 1:1471-1478. Cines DB, Kaywin P, Bina M , Tomaski A , Schreiber A D . Heparin-associated thrombocytopenia. N Engl J Med 1980; 303:788-795. Clawson CC, White JG. Platelet interaction with bacteria. I. Reaction phases and effects of inhibitors. A m J Pathol 1971; 65:367-380. Clawson CC, White JG. Platelet interaction with bacteria. II. Fate of the bacteria. Am J Pathol 1971a; 65:367-380. Cockgroft DW, Gault M H . Prediction of creatinine clearance from serum creatinine. Nephron 1976; 16:31-41. Coller BS. Platelets and thrombolytic therapy. N Engl J Med 1990; 322:33-42. Connors AF Jr, Speroff T, Dawson N V , Thomas C, Harrell FE Jr, Wagner D, et al. The effectiveness of right heart catheterization in the initial care of critically i l l patients. J A M A 1996; 276:889-897. 223 Contreras, M . The appropriate use of platelets: An update from the Edinburgh Consensus Conference. Brit J Haematol 1998; 101 (Suppl 1):10-12. Cook D, Guyatt G, Marshall J, Leasa D, Fuller H , Hall R, et al. A comparison of sucralfate and ranitidine for the prevention of upper gastrointestinal bleeding in patients requiring mechanical ventilation. New Engl J Med 1998; 338:791-797. Cook DJ, Fuller HD, Guyatt GH, Marshall JC, Leasa D, Hall R, et al. Risk factors for gastrointestinal bleeding in critically i l l patients. Canadian Critical Care Trials Group. -N Engl J Med 1994;330:377-381. Cook DJ, Griffith LE , Walter SD, Guyatt GH, O'Meade M , Heyland DK, et al. The attributable mortality and length of intensive care unit stay of clinically important gastrointestinal bleeding in critically i l l patients. Crit Care 2001; 5;368-375. Corwin HL, Gettinger A , Levy M M , Fink MP, Pearl RG, Abraham E, et al. Anemia and blood transfusion in the critically i l l : current clinical practice in the United States-—The CRIT Study. Crit Care Med 2004; 32:39-52. Counts RB, Haisch C, Simon TL, Maxwell N G , Heimbach D M , Carrico CJ. Hemostasis in massively transfused trauma patients. Ann Surg 1979; 190:91-99. Cox DR. The Analysis of Binary Data. Methuen, London. 1 - 142. 1970 Cummins D, Halil O, Amin S. Which patients undergoing cardiopulmonary bypass should be assessed for development of heparin-induced thrombocytopenia? Thromb Haemost 1995; 73: 890. Cupples L A , D'Agostino RB, Anderson K, Kannel B. Comparison of baseline and repeatecd measure covariate techniques in the Framingham heart study. Stat Med 1988; 7:1205-218. D'Agostino RB, Lee M L , Belanger AJ . Relation of pooled logistic regression to time dependent Cox regression analysis: the Framingham heart study. Stat Med 1990; 9:1501-1515. Dalen JE, Bone RC. Is it time to pull the pulmonary artery catheter? J A M A 1996; 276:916-918. Dalen JE. The pulmonary artery catheter-friend, foe, or accomplice? J A M A 2001; 286:348-350. Davey M G , Lander H . Effect of infected heparin on platelet levels in man. J Clin Path 1968; 21 55-59. Davis GA, Chandler M H . Comparison of creatinine clearance estimation methods in patients with trauma. Am J Health Syst Pharm. 1996; 53:1028-1032. Davis G l . Quantitative and qualitative disorders of platelets. In: Steine-Martin E H , Lotspeich-Steininger C A , Koepke JA, eds. Clinical hematology; principles, procedures, and correlations. New York, N Y : Lippincott; 1998:717-734. 224 DeLoughery TG. Thrombocytopenia and platelet dysfunction in critical care patients. In: Irwin RS, Rippe J M eds. Intensive care medicine. 5 t h ed. Philadelphia, Pa. Lippincott Williams & Wilkins; 2003:1240-1250. DeLoughery TG. Thrombocytopenia in critical care patients. J Intensive Care Med 2002; 17:267-282. Demers C, Ginsberg JS, Brill-Edwards P, Panju A , Warkentin TE, Anderson DR, et al. Rapid anticoagulation using ancrod for heparin-induced thrombocytopenia. Blood 1991; 78:2194-2197. Denomme GA. The platelet Fc receptor in heparin-induced thrombocytopenia. In: Warkentin TE, Greinacher A , eds. Heparin-induced Thrombocytopenia. 2 n d edition. New York, N Y : Marcel Dekker, Inc.; 2001:189-214. DeSancho MT, Rand JH. Bleeding and thrombotic complications in critically i l l patients with cancer. Crit Care Clin 2001; 17:599-622. Dorfman DD, A l f E Jr. Maximum likelihood estimation of parameters of signal detection theory--a direct solution. Psychometrika 1968; 33:117-124. Draper NR, Smith H. Applied regression analysis. 2 n d ed. New York, N Y . John Wiley and Sons, 1981:204. Drews RE. Critical issues in hematology: anemia, thrombocytopenia, coagulopathy, and blood product transfusions in critically i l l patients. Clin Chest Med 2003; 24:607-622. Economopoulos TC, Papayannis A G , Stathakis NE, Arapakis G, Gardikas C. Postoperative platelet function in patients on small subcutaneous doses of heparin. Acta Haematol 1977; 57:266-271. Eichler P, Budde U , Haas S, Kroll H , Loreth R M , Meyer O, et al. First workshop for detection of heparin-induced antibodies: validation of the heparin-induced platelet-activation test (HIPA) in comparison with a PF4/heparin ELISA. Thromb Haemost 1999; 81:625-629. Eikelboom JW, Anand SS, Mehta SR, Weitz JI, Y i C, Yusuf S. Prognostic significance of thrombocytopenia during hirudin and heparin therapy in acute coronary syndrome without ST elevation: Organization to Assess Strategies for Ischemic Syndromes (OASIS-2) study. Circulation 2001; 103:643-650. Elalamy I, Potevin F, Lecrubier C, Bara L, Marie JP, Samama M M . A fatal low-molecular-weight heparin-associated thrombocytopenia after hip surgery: possible usefulness of PF4-heparin ELISA test. Blood Coagul Fibrinolysis 1996; 7:665-671. Engoren M , Buderer NF, Zacharias A , Habib RH. Variables predicting reintubation after cardiac surgical procedures. Ann Thorac Surg 1999; 67:661-665. Ernest D, Fisher M M . Heparin-induced thrombocytopenia complicated by bilateral adrenal hemorrhage. Intensive Care Med 1991; 17:238-240. 225 Favaloro EJ, Bernal-Hoyos E, Exner T, Koutts J. Heparin-induced thrombocytopenia: laboratory investigation and confirmation of diagnosis. Pathology 1992; 24:177-183. Ferreira FL, Bota DP, Bross A , Melot C, Vincent JL. Serial evaluation of the SOFA score to predict outcome in critically ill patients. J A M A 2001; 286:1754-1758. Fisman D N . Hemophagocytic syndromes and infection. Emerg Infect Dis 2000; 6:601-608. Francois B, Trimoreau F, Vignon P, Fixe P, Praloran V , Gastinne H. Thrombocytopenia in the Sepsis Syndrome: Role of Hemophagocytosis and Macrophage Colony-stimulating Factor. Am J Med 1997; 103:114-120. Freedman M D . Pharmacodynamics, clinical indications, and adverse effects of heparin. J Clin Pharmacol 1992; 32:584-596. Gawaz M , Fateh-Moghadam S, Pilz G, Gurland HJ, Werdan K. Platelet activation and interaction with leucocytes in patients with sepsis or multiple organ failure. Eur J Clin Invest 1995;25:843-851. Geerts W H , Pineo GF, Heit JA, Bergqvist D, Lassen MR, Colwell CW, et al. Prevention of Venous Thromboembolism: The Seventh A C C P Conference on Antithrombotic and Thrombolytic Therapy. Chest 2004; 126(3 Suppl):338S-400S. George JN, El-Harake M , Aster RH. Thrombocytopenia due to enhanced platelet destruction by immunologic mechanisms. In: Beutler E, Lichtricn M A , Coller BS, Kipps TJ, eds. Williams hematology. Fifth edition. Toronto, Ont. McGraw-Hill, Inc; 1995:1315-1355. George JN, El-Harake M . Thrombocytopenia due to enhanced platelet destruction by non-immunologic mechanisms. In: Beutler E, Lichtricn M A , Coller BS, Kipps TJ, eds. Williams hematology. Fifth edition. Toronto, Ont. McGraw-Hill, Inc; 1995:1290-1315. Gernsheimer T. Transfusion therapy: blood components and transfusion complications. In: Irwin RS, Rippe J M eds. Intensive care medicine. 5 t h ed. Philadelphia, Pa. Lippincott Williams & Wilkins; 2003:1278-1283. Gibson BE, Todd A, Roberts I, Pamphilon D, Rodeck C, Bolton-Maggs P, et al. Transfusion guidelines for neonates and older children. Br J Haematol 2004; 124:433-453. Girolami B, Prandoni P, Stefani P M , Tanduo C, Sabbion P, Eichler P, et al. The incidence of heparin-induced thrombocytopenia in hospitalized medical patients treated with subcutaneous unfractionated heparin: a prospective cohort study. Blood 2003; 101:2955-2959. Gollub S, Ulin A W . Heparin-induced thrombocytopenia in man. J Lab Clin Med 1962; 59:430-435. Goyette RE, ed. Hematology. A comprehensive guide to the diagnosis and treatment of blood disorders. Los Angeles, C A : Practice Management Information Corporation, 1997:756,765. 226 Granger CB, Hirsch J, Califf R M , Col J, White HD, Betriu A et al. Activated partial thromboplastin time and outcome after thrombolytic therapy for acute myocardial infarction: results from the GUSTO-I trial. Circulation 1996; 93:870-878. Green D, Martin GJ, Shoichet SH, DeBacker N , Bomalaski JS, Lind RN. Thrombocytopenia in a prospective, randomized, double-blind trail of bovine and porcine heparin. A m J Med Sci 1984;288:60-64. Green MS, Symons MJ . A comparison of the logistic risk function and the proportional hazards model in prospective epidemiologic studies. J Chron Dis 1983; 36:715-724. Greinacher A , Amiral J, Dummel V , Vissac A , Kiefel B, Mueller-Eckhardt C. Laboratory diagnosis of heparin-associated thrombocytopenia and comparison of platelet aggregation test, heparin-induced platelet activation test, and platelet factor4/heparin enzyme-linked immunosorbent assay. Transfusion 1994a; 34:381-385. Greinacher A , Eichler P, Lubenow N , Kwasny H , Luz M . Heparin-induced thrombocytopenia with thromboembolic complications: meta-analysis of 2 prospective trials to assess the value of parenteral treatment with lepirudin and its therapeutic aPTT range. Blood 2000; 96:846-851. Greinacher A , Janssens U , Berg G, Bock M , Kwasny H , Kemkes-Matthes B , et al. Lepirudin (recombinant hirudin) for parenteral anticoagulation in patients with heparin-induced thrombocytopenia. Heparin-Associated Thrombocytopenia Study (HAT) investigators. Circulation 1999; 100:587-593. Greinacher A , Michels I, Kiefel B, Mueller-Eckhardt C. A rapid and sensitive test for diagnosing heparin-associated thrombocytopenia. Thromb Haemost 1991; 66:734-736. Greinacher A , Michels I, Mueller-Eckhardt C. Heparin-associated thrombocytopenia: the antibody is not heparin specific. Thromb Haemost 1992; 67:545-549. Greinacher A , Potzsch B, Amiral J, Dummel V , Eichner A , Mueller-Eckhardt C. Heparin-associated thrombocytopenia: isolation of the antibody and characterization of a multimolecular PF4-heparin complex as the major antigen. Thromb Haemost 1994; 71:247-251. Greinacher A , Volpel H , Janssens U , Hach-Wunderle V , Kemkes-Matthes B, Eichler P, et al. Recombinant hirudin (lepirudin) provides safe and effective anticoagulation in patients with heparin-induced thrombocytopenia: a prospective study. Circulation 1999a; 99:73-80. Greinacher A , Warkentin TE. Treatment of heparin-induced thrombocytopenia: an overview. In: Warkentin TE, Greinacher A , eds. Heparin-induced Thrombocytopenia. 2 n d edition. New York, N Y : Marcel Dekker, Inc.; 2001:291-322. Greinacher A . Antigen generation in heparin-associated thrombocytopenia: The non-immunologic type and the immunologic type are closely linked in their pathogenesis. Semin Thromb Hemost 1995;21:106-116. GUSTO-I Investigators. An international randomized trial comparing four thrombolytic strategies for acute myocardial infarction. N Engl J Med 1993; 329:673-682. 227 Hach-Wunderle V , Kainer K , Krug B, Muller-Berghaus G, Potzsch B. Heparin-associated thrombosis despite normal platelet counts. Lancet 1994; 344:469-470. Hach-Wunderle V , Kainer K, Salzmann G, Muiler-Berghaus G, Potzsch B. Heparin-related thrombosis despite normal platelet counts in vascular surgery. Am J Surg 1997; 173:117-119. Hamilton GC. Platelet Count. Emerg Med Clin Nor Am 1986; 4:75-85. Handin RI. Bleeding and thrombosis. In: Braunwald E, Fauci AS, Kasper DL, Hauser SL, Longo DL, Jameson JL, eds. Harrison's: Principle of internal medicine. Toronto, Ont. McGraw-Hill, Inc; 2001:354-360. Handin RI. Disorders of platelets and vessel wall. In: Isselbacher K J , Braunwald E, Wilson JD, Martin JB, Fauci AS , Kasper DL, eds. Harrison's: Principle of internal medicine. Toronto, Ont. McGraw-Hill, Inc; 2001a:745-750. Hanes SD, Quarles DA, Boucher B A . Incidence and risk factors of thrombocytopenia in critically i l l trauma patients. Ann Pharmacother 1997; 31:285-289. Hanley JA, McNeil, BJ. A method of comparing the areas under receiver operating characteristic curves derived from the same cases. Radiology 1983; 148:839-843. Hanley JA, McNeil, BJ. The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology 1982; 143:29-36. Harbrecht U , Bastians B, Kredteck A , Hanfland P, Klockgether T, Pohl C. Heparin-induced thrombocytopenia in neurologic disease treated with unfractionated heparin. Neurology 2004; 62:657-659. Hardy JF, de Moerloose P, Samama M . Massive transfusion and coagulopathy: pathophysiology and implications for clinical management. Can J Anesth 2004; 51:293-310. Harenberg J, Wang L, Hoffmann U , Huhle G, Feuring M . Improved laboratory confirmation of heparin-induced thrombocytopenia type II. Time course of antibodies and combination of antigen and biologic assays. Am J Clin Pathol 2001; 115:432-438. Harrell FE Jr, Lee K L , Mark DB. Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med 1996; 15:361-87. Harrington R A , Becker RC, Ezekowitz M , Meade TW, O'Connor C M , Vorchheimer DA, et al. Antithrombotic Therapy for Coronary Artery Disease: The Seventh A C C P Conference on Antithrombotic and Thrombolytic Therapy. Chest 2004; 126(3 Suppl): 513S-548S. Harrington R A , Sane DC, Califf R M , Sigmon K N , Abbottsmith CW, Candela RJ, et al. Clinical importance of thrombocytopenia occurring in the hospital phase after administration of thrombolytic therapy for acute myocardial infarction. J A m Coll Cardiol 1994; 23:891-898. 228 Hauck WW. A comparison of the logistic risk function and the proportional hazards model in prospective epidemiologic studies. J Chron Dis 1985; 38:125-126. Hebert PC, Wells G, Blajchman M A , Marshall J, Martin C, Pagliarello P, et al. A multicenter, randomized, controlled clinical trial of transfusion requirements in critical care. N Engl J Med 1999; 340:409-417. Hebert PC, Wells G, Martin C, Tweeddale M , Marshall J, Blajchman M , et al. A Canadian survey of transfusion practices in critically i l l patients. Transfusion Requirements in Critical Care Investigators and the Canadian Critical Care Trials Group. Crit Care Med 1998; 26:482-487. Hechtman HB, Lonergan EA, Shepro D. Platelet and leukocyte lung interactions in patients with respiratory failure. Surgery 1978; 83:155-163. Heffner JE, Sahn SA, Repine JE. The role of platelets in the adult respiratory distress syndrome: culprit or bystanders? Am Rev Respir Dis 1987; 135:482-492. Herzberg M C , Brintzenhofe K L , Clawson CC. Aggregation of human platelets and adhesion of Streptococcus sanguis. Infect Immun 1983; 39:1457-1469. Higgins TL, McGee WT, Steingrub JS, Rapoport J, Lemeshow S, Teres D. Early indicators of prolonged intensive care unit stay: impact of illness severity, physician staffing, and pre-intensive care unit length of stay. Crit Care Med 2003; 31:45-51. Hinshaw L B , Archer LT, Beller-Todd B K . Hematologic disturbances during sepsis: platelets and leukocytes. Advan Shock Res 1982; 7:1-6. Hirch J, Fuster V . Guide to anticoagulant therapy. Part 1: Heparin. Circulation 1994; 89:1449-1468. Hirsh J, Raschke R. Heparin and low-molecular-weight heparin: the Seventh A C C P Conference on Antithrombotic and Thrombolytic Therapy. Chest 2004; 126(3 Suppl): 188S-203S. Hirsh J, Warkentin TE, Raschke R, Granger C, Ohman E M , Dalen JE. Heparin and low-molecular-weight heparin: mechanism of action, pharmacokinetics, dosing considerations, monitoring, efficacy, and safety. Chest 1998; 114 (Suppl):489S-510S. Hirsh J, Warkentin TE, Shaughnessy SG, Anand SS, Halperin JL, Raschke R, et al. Heparin and low-molecular-weight heparin: mechanisms of action, pharmacokinetics, dosing, monitoring, efficacy, and safety. Chest 2001; 119(1 Suppl):64S-94S. Hirsh J. Heparin. N Engl J Med 1991; 324:1565-1574. Home M K III. Nonimmune heparin-platelet interactions: implications for the pathogenesis of heparin-induced thrombocytopenia. In: Warkentin TE, Greinacher A , eds. Heparin-induced Thrombocytopenia. 2 n d edition. New York, N Y : Marcel Dekker, Inc.; 2001:123-136. Hosmer DW, Lemeshow S. Applied logistic regression. 1st ed. New York, N Y . John Wiley and Sons, 1989. 1-24. 229 Hosmer DW, Lemeshow S. Applied logistic regression. 2 n d ed. New York, NY. John Wiley and Sons, 2000. 1-30. Hosmer DW, Lemeshow S. Applied logistic regression. 2 n d ed. New York, NY. John Wiley and Sons, 2000a. 47-90. Hosmer DW, Lemeshow S. Applied logistic regression. 2 n d ed. New York, NY. John Wiley and Sons, 2000b. 91-142. Hosmer DW, Lemeshow S. Applied logistic regression. 2 n d ed. New York, NY. John Wiley and Sons, 2000c. 143-202. Hosmer DW, Lemeshow S. The importance of assessing the fit of logistic regression models: a case study. Am J Public Health 1991: 81:1630-1635. Ingram DD, Kleinman JC. Empirical comparisons of proportional hazards and logistic regression models. Stat Med 1989; 8:525-538. Isaacman DJ, Shults J, Gross TK, Davis PH, Harper M . Predictors of bacteremia in febrile children 3 to 36 months of age. Pediatrics 2000; 106:977-982. Issekutz A C , Ripley M . The effect of intravascular neutrophil chemotactic factors on blood neutrophil and platelet kinetics. Am J Hematol 1986; 21:157-171. Justice A C , Covinsky K E , Berlin JA. Assessing the generalizability of prognostic information. Ann Intern Med 1999; 130:515-524. Kakkar N, Kaur R, Dhanoa J. Improvement in fresh frozen plasma transfusion practice: results of an outcome audit. Transfus Med 2004; 14:231-235. Kappers-Klunne M C , Boon DMS, Hop WCJ, Michiels JJ, Stibbe J, van der Zwaan C, et al. Heparin-induced thrombocytopenia and thrombosis: a prospective analysis of the incidence in patients with heart and cerebrovascular diseases. Br J Haematol 1997; 96:442-446. Katz, M H . Multivariable analysis: a primer for readers of medical research. Ann Intern Med 2003; 138:644-650. Kelton JG, Neame PB, Gauldie J, Hirsh J. Elevated platelet-associated IgG in the thrombocytopenia of septicemia. N Engl J Med 1979; 300:760-764. Kelton JG, Smith JW, Warkentin T E , Hayward CPM, Denomme GA, Horsewood P. Immunoglobulin G from patients with heparin-induced thrombocytopenia binds to a complex of heparin and platelet factor 4. Blood 1994; 11:3232-3239. Kelton JG, Warkentin TE. Diagnosis of heparin-induced thrombocytopenia. Still a journey, not yet a destination. Am J Clin Path 1995; 104:611-613. 230 Kickler TS. Principles of platelet transfusion therapy. In: Hoffman R, Benz EJ Jr, Shattil SJ, Furie B, Cohen HJ, Silberstein LE , McGlave P eds. Hematology: Basic principles and practice. 3 r d ed. Philadelphia, Pa. J.B. Churchill-Livingstone; 2000:2248-2257. Kim Y L , Richman K A , Marshall BE. Thrombocytopenia associated with Swan-Ganz catheterization in patients. Anesthesiology 1980; 53:261-262. King DJ, Kelton JG. Heparin-associated thrombocytopenia. Ann Intern Med 1984; 100:535-540. Kitchens CS. Thrombocytopenia due to acute venous thromboembolism and its role in expanding the differential diagnosis of heparin-induced thrombocytopenia. A m J Hematol 2004; 76:69-73. Klee FE, Osswald BR, Wysocki S. Severe abdominal pain and thrombocytopenia—typical symptoms of occult jejunal diverticulum perforation? J Gastroenterol 1997; 32:246-250. Klein H G , Bell WR. Disseminated intravascular coagulation during heparin therapy. Ann Intern Med 1974; 80:477-481. Knaus W A , Draper EA, Wagner DP, Zimmerman JE. A P A C H E II: A severity of disease classification system. Crit Care Med 1985; 13: 818-827. Kocoshis SA, Gartner JC Jr, Gaffney PC, Gyrboski JD. Thrombocytopenia in ulcerative coloitis. JPediat 1979; 95:83-84. Kraemer, HC, Kazdin A E , Offord DR, Kessler RC, Jensen PS, Kupfer DJ. Coming to terms with the terms of risk. Arch Gen Psychiatry 1997; 54:337-343. Kravitz RL, Laouri M , Kahan JP, Guzy P, Sherman T, Hilborne L , Brook R H . Validity of criteria used for detecting underuse of coronary revascularization. J A M A 1995; 274:632-638. Lam NP, Le PD, Crawford SY, Patel S. National survey of stress ulcer prophylaxis. Crit Care Med 1999; 27:98-103. Last J M . A dictionary of epidemiology. 3 r d ed. Toronto, Ont.: Oxford University Press, 1995. 148-149. Laster J, Silver D. Heparin-coated catheters and heparin-induced thrombocytopenia. J Vase Surg 1988; 7:667-672. Layon AJ . The pulmonary artery catheter: non-existential entity of occasionally useful tool? Chest 1999; 115:859-862. Lee D H , Warkentin TE, Denomme G A , Hayward C P M , Kelton JG. A diagnostic test for heparin-induced thrombocytopenia: detection of platelet microparticles using flow cytometry. Brit J Haematol 1996; 95:724-731. 231 Lee D H , Warkentin TE. Frequency of heparin-induced thrombocytopenia. In: Warkentin TE, Greinacher A , eds. Heparin-induced Thrombocytopenia. 2 n d edition. New York, N Y : Marcel Dekker, Inc.; 2001:87-121. Lee KJT, Jui KP , Tan WC. Thrombocytopenia in sepsis: a predictor of mortality in the intensive care unit. Singapore Med J 1993; 34:245-246. Lee K L , Woodlief L H , Topol EJ, Weaver WD, Betriu A , Col J, et al. Predictors of 30-day mortality in the era of reperfusion for acute myocardial infarction. Results from an international trial of 41,021 patients. Circulation 1995; 91:1659-1668. Lemeshow S, Klar J, Teres D, Avrunin JS, Gehlbach SH, Rapoport J, et al. Mortality probability models for patients in the intensive care unit for 48 or 72 hours: a prospective, multicenter study. Crit Care Med 1994; 22:1351-1358. Leo A , Winteroll S. Laboratory diagnosis of heparin-induced thrombocytopenia and monitoring of alternative anticoagulants. Clin Diagn Lab Immunol 2003; 10:731-740. Leslie SD, Toy PT. Laboratory hemostatic abnormalities in massively transfused patients given red blood cells and crystalloid. A m J Clin Pathol 1991; 96:770-3. Levi M , ten Cate H. Disseminated intravascular coagulation. N Engl J Med 1999; 8:586-592. Levine RF, Spivak JL, Meagher R, Sieber F. Effect of ethanol on thrombopoiesis. Br J Haematol 1986; 62:345-354. Levine SP. Thrombocytopenia caused by immunologic platelet destruction. In: Lee GR, Foerster J, Lukens J, Paraskevas F Greer JP, Rodgers G M , eds. Wintrobe's clinical hematology. lOed. Baltimore M D : Williams and Wilkens, 1999:1583-1611. Levy M M , Fink MP, Marshall JC, Abraham E, Angus D, Cook D, et al. 2001 SCCM/ESICM/ACCP/ATS/SIS international sepsis definitions conference. Crit Care Med 2003;31:1250-1256. Lewis BE, Wallis DE, Berkowitz SD, Matthai WH, Fareed J, Walenga J M , et al. Argatroban anticoagulant therapy in patients with heparin-induced thrombocytopenia. Circulation 2001; 103:1838-1843. Lewis BE , Wallis DE, Berkowitz SD, Matthai WH, Fareed J, Walenga JM, et al; ARG-911 Study Investigators. Argatroban anticoagulant therapy in patients with heparin-induced thrombocytopenia. Circulation 2001; 103:1838-1843. Lind SE. The hemostatic system. In: Handin R l , Lux SE, Stossel TP, eds. Blood. Principles and practice of hematology. Philadelphia, PA; JB Lippincott Company; 1995:949-973. Lortholary A , Raffi F, Aubertin P, Barrier JH, Boibieux A, Peyramond D. HIV-associated haemophagocytic syndrome. Lancet 1990; 336:1128. Lubran M M . Hematologic side effects of drugs. Ann Clin Lab Sci 1989; 19:114-121. 232 Magnani H N . Heparin-induced thrombocytopenia (HIT): an overview of 230 patients treated with Orgaran (Org 10172). Thromb Haemost 1993; 70:554-561. Magnani H N . Orgaran (danaparoid sodium) use in the syndrome of heparin-induced thrombocytopenia. Platelets 1997; 8:74-81. Makhoul RG, Greenberg CS, McCann R L . Heparin-associated thrombocytopenia and thrombosis: a serious clinical problem and potential solution. J Vase Surg 1986; 4:522-528. Marik PE. Handbook of evidence-based critical care. Hematology in the intensive care unit. New York, N Y . Springer; 2001:421-432. Marik PE. Handbook of evidence-based critical care. Hemodynamic monitoring: pulmonary artery and arterial catheterization. New York, N Y . Springer; 2001a:123-139. McClure M W , Berkowitz SD, Sparapani R, Tuttle R, Kleiman NS, Berdan L G , et al. Clinical significance of thrombocytopenia during a non-ST-elevation acute coronary syndrome. The platelet glycoprotein Ilb/IIIa in unstable angina: receptor suppression using integrilin therapy (PURSUIT) trial experience. Circulation 1999; 99:2892-2900. McNeil, BJ, Hanley JA. Statistical approaches to the analysis of receiver operating characteristic (ROC) curves. Med Decis Making 1984; 4:137-150. McNulty SE, Maguire DP, Thomas RE. Effect of heparin-bonded pulmonary artery catheters on the activated coagulation time. J Cardiothorac Vase Anesth 1998; 12:533-535. Menon V , Harrington RA, Hochman JS, Cannon CP, Goodman SD, Wilcox RG, et al. Thrombolysis and Adjunctive Therapy in Acute Myocardial Infarction: The Seventh ACCP Conference on Antithrombotic and Thrombolytic Therapy. Chest 2004; 126(3 Suppl): 549S-575S. Messmore H L , Jeske WP, Wehrmacher W H , Walenga JM. Benefit-risk assessment of treatments for heparin-induced thrombocytopenia. Drug Saf 2003; 26:625-641. Metz CE. Basic principles of ROC analysis. Semin Nuc Med 1978; 8:283-298. Miller JJ, Venus B, Mathru M . Comparison of the sterility of long-term central venous catheterization using single lumen, triple lumen, and pulmonary artery catheters. Crit Care Med 1984; 12:634-637. Miller M L . Heparin-induced thrombocytopenia. Cleve Clinical J Med 1989; 56:483-490. Milligan GF, MacDonald JA, Mellon A , Ledingham IM. Pulmonary and hematologic disturbances during septic shock. Surg Gynecol Obstet 1974; 138:43-49. Mittlbock M , Schemper M . Explained variation for logistic regression. Stat Med 1996; 15:1987-1997. 233 Mones RL. Thrombocytopenia and hypofibrinogenemia in association with inflammatory bowel disease. J Pediat Gastroenterol Nutrit 1983; 2:175-177. Morewood G. Uncomplicated heparinization for cardiopulmonary bypass in a patient with anti-heparin antibodies. J Cardiothorac Vase Anesth 2000; 14:309-311. Murphy S and Gardner FH. Platelet preservation: effects of storage temperature on maintenance of platelet viability—deleterious effect on refrigerated blood. N Engl J Med 1969; 280:1094-1098. Nacht A . The use of blood products in shock. Crit Care Clin 1992; 8:255-291. Nand S, Wong W, Yuen B, Yetter A , Schmulbach E, Gross Fisher S. Heparin-induced thrombocytopenia with thrombosis: Incidence, analysis of risk factors, and clinical outcomes in 108 consecutive patients treated at a single institution. Am J Hematol 1997; 56:12-16. Napolitano L M . Current status of blood component therapy in surgical critical care. Curr Opin Crit Care 2004; 10:311-317. Nathan S, Schaer GL. Acute coronary syndromes and acute myocardial infarction. In: Parrillo JE, Dellinger RP eds. Critical care medicine: principles of diagnosis and management in the adult. 2 n d ed. Philadelphia, Pa. Mosby; 2001:569-610. Neame PB, Kelton JG, Walker IR, Stewart IO, Nossel HL, Hirsh J. Thrombocytopenia in septicemia: the role of disseminated intravascular coagulation. Blood 1980; 56:88-92. Newman P M , Chong B H . Heparin-induced thrombocytopenia: new evidence for the dynamic binding of purified anti-PF4-heparin antibodies to platelets and the resultant platelet activation. Blood 2000;96:182-187. Nguyen P, Droulle C, Potron G. Comparison between platelet factor 4/heparin complexes ELISA and platelet aggregation test in heparin-induced thrombocytopenia. Thromb Haemost 1995; 74:804-805. N i H , Freedman J. Platelets in hemostasis and thrombosis: role of integrins and their ligands. Transfus and Apheresis Sci 2003; 28:257-264. Nijjar TS, Bonacosa IA, Israels L G . Severe acute thrombocytopenia following infusion of plasma containing anti-PIA 1. Am J Hematol 1987; 25:219-221. Noe D A , Graham S M , Luff R, Sohmer P. Platelet counts during rapid massive transfusions. Transfusion 1982; 22:392-395. Nugent DJ. Alloimmunization to platelet antigens. Semin Hematol 1992;29:83-88. O'Brien JR, Etherington M D , Shuttleworth RD, Davison S. Platelet and other tests followed sequentially for 14 days after operation. Clin Lab Haematol 1984; 6:239-245. 234 Oppenheimer L , Hryniuk W M , Bishop AJ . Thrombocytopenia in severe bacterial infections. J Surg Res 1976;20:211-214. Ortel TL. Heparin-induced thrombocytopenia. SeminHematol 1998; 35(4 suppl 5): 1-2. Packham M A . The behavior of platelets at foreign surfaces. Proc Soc Exp Biol Med 1988; 189:261-74. Pagano M , Gauvreau K. Principles of biostatistics. Belmont, C A : Wadsworth Publishing Company; 1993. Pappas G, Kitsanou M , Christou L , Tsianos E. Immune thrombocytopenia attributed to brucellosis and other mechanisms of Brucella-induced thrombocytopenia. A m J Hematol 2004; 75:139-141. Parsons PE. Progress in research on pulmonary-artery catheters. N Engl J Med 2003; 348:66-68. Patnode N M , Gandhi PJ. Drug-induced thrombocytopenia in the coronary care unit. J Thromb Thrombolysis 2000; 10:155-167. Personal communication, Dr. Wobler, Pathologist, Lions Gate Hospital Hematology Laboratory, 1998. Phelan B K . Heparin-associated thrombosis without thrombocytopenia. Ann Intern Med 1983; 99:637-638. Platell CFE, Tan EGC. Hypersensitivity reactions to heparin: delayed onset thrombocytopenia and necrotizing skin lesions. Aust NZ J Surg 1986; 56:621-623. Polanczyk CA, Rohde L E , Goldman L , Cook EF, Thomas EJ, Marcantonio ER, et al. Right heart catheterization and cardiac complications in patients undergoing noncardiac surgery: an observational study. J A M A 2001; 286:309-314. Poon M C . Management of thrombocytopenic bleeding: is there a role for recombinant coagulation factor Vila? Curr Hematol Rep 2003; 2:139-147. Popma JJ, Berger P, Ohman E M , Harrington RA, Grines C, Weitz JL Antithrombotic Therapy During Percutaneous Coronary Intervention: The Seventh A C C P Conference on Antithrombotic and Thrombolytic Therapy. Chest 2004; 126(3 Suppl): 576S-599S. Pouplard C, Amiral J, Borg JY, Laporte-Simitsidis S, Delahousse B, Gruel Y . Decision analysis for use of platelet aggregation test, carbon 14 serotonin release assay, and heparin-platelet factor 4 enzyme-linked immunosorbent assay for diagnosis of heparin-induced thrombocytopenia. Am J Clin Pathol 1999a; 111:700-706. Pouplard C, Iochmann S, Renard B, Herault O, Colombat P, Amiral J, et al. Induction of monocyte tissue factor expression by antibodies to heparin-platelet factor 4 complexes developed in heparin-induced thrombocytopenia. Blood 2001; 97:3300-3302. 235 Pouplard C, May M A , Iochmann S, Amiral J, Vissac A M , Marchand M , et al. Antibodies to platelet factor-4 after cardiopulmonary bypass in patients anticoagulated with unfractionated heparin or a low-molecular-weight-heparin: clinical implications for heparin-induced thrombocytopenia. Circulation 1999; 99:2530-2536. Powers PJ, Kelton JG, Carter CJ. Studies on the frequency of heparin-associated thrombocytopenia. Thromb Res 1984: 33:439-443. Pravinkumar E, Webster NR. HIT/HITT and alternative anticoagulation: current concepts. Br J Anaesth. 2003; 90:676-685. PuckettCD. The educational annotation of ICD-9-CM. Reno: Channel Publishing, 1998. Ramirez-Lassepas M , Cipolle RJ, Rodvold K A , Seifert RD, Strand L , Taddeini L, et al. Heparin-induced thrombocytopenia in patients with cerebrovascular ischemic disease. Neurology 1984; 34:736-740. Rao A K , White GC, Sherman L, Colman R, Lan G, Ball AP. Low incidence of thrombocytopenia with porcine mucosal heparin. A prospective multicenter study. Arch Intern Med 1989; 149:1285-1288. Reed R L 2nd, Ciavarella D, Heimbach D M , Baron L, Pavlin E, Counts RB, et al. Prophylactic platelet administration during massive transfusion: a prospective, randomized, double-blind clinical study. Ann Surg 1986; 203:40-48. Rhodes A , Cusack RJ, Newman PJ, Grounds R M , Bennett ED. A randomised, controlled trial of the pulmonary artery catheter in critically i l l patients. Intensive Care Med 2002; 28:256-264. Rhodes GR, Dixon RH, Silver D. Heparin induced thrombocytopenia with thrombotic and hemorrhagic manifestations. Surg Gynecol Obstet 1973; 136:409-416. Rice L, Attisha WK, Drexler A , Francis JL. Delayed-onset heparin-induced thrombocytopenia. Ann Intern Med 2002; 136:210-215. Richard C, Warszawski J, Anguel N , Deye N , Combes A , Barnoud D, et al; French Pulmonary Artery Catheter Study Group. Early use of the pulmonary artery catheter and outcomes in patients with shock and acute respiratory distress syndrome: a randomized controlled trial. J A M A 2003; 290:2713-2720. Risdall RJ, Brunning RD, Hernandez JI, Grodon DH. Bacteria-associated hemophagocytic syndrome. Cancer. 1984;54:2968-2972. Riska EB, Bostman O, von Bonsdorff H , Hakkinen S, Jaroma H, Kiviluoto O et al. Outcome of closed head injuries exceeding 20-unit blood transfusion needed. Injury 1988; 19:273-276. Roberts B, Rosato FE, Rosato EF. Heparin—a cause of arterial emboli? Surgery 1964; 55:803-808. Robin ED. The cult of the Swan-Ganz catheter: overuse and abuse of pulmonary flow catheters. Ann Intern Med 1985; 103:445-449. 236 Rubin R. Effect of ethanol on platelet function. Alcohol Clin Exp Res 1999;23:1114-1118. Rue M , Quintana S, Alvarez M , Artigas A . Daily assessment of severity of illness and mortality prediction for individual patients. Crit Care Med 2001; 29:45-50. Ryan T J , Anderson J L , Antman E M , Braniff B A , Brooks N H , Califf R M , et al. A A C / A H A guidelines for the management of patients with acute myocardial infarction. A report of the American College of Cardiology/American Heart Association task force on practice guidelines (committee on management of acute myocardial infarction). J Am Coll Cardio 1996; 28:1328-1428. Saffle J R , Russo J , Dukes GE, Warden GD. The effect of low-dose heparin therapy on serum platelet and transaminase level. J Surg Res 1980; 28:297-305. Saint S Matthay M A . Risk reduction in the intensive care unit. Am J Med 1998; 105:515-523. Sanders SA, Graham C A , Bass J L , Bancroft J . A prospective study of the effects of oral contraceptives on sexuality and well-being and their relationship to discontinuation. Contraception 2001; 64:51-58. Sandham J D , Hull RD, Brant RF, Knox L , Pineo GF, Doig C J , et al; Canadian Critical Care Clinical Trials Group. A randomized, controlled trial of the use of pulmonary-artery catheters in high-risk surgical patients. N Engl J Med 2003; 348:5-14. Sane DC, Damaraju L V , Topol E J , Cabot CF, Mascelli M A , Harrington R A , et al. Occurrence and clinical significance of pseudothrombocytopenia during abciximab therapy. J Am Coll Cardiol 2000; 36:75-83. Scalon J D , FreireSM, Cunha TA. Validation of models for predicting the use of health technologies. Med Decis Making 1996; 18:311-319. Schmitt BP, Adelman B. Heparin-associated thrombocytopenia: a critical review and pooled analysis. A m J Clin Sci 1993; 305:208-215. Schneider RC Zapol W M , Carvalho A. Platelet consumption and sequestration in severe acute respiratoryfailure. Am Rev Respir Dis 1980; 122:445-451. Schofield W N , Rubin GL, Dean M G . Appropriateness of platelet, fresh frozen plasma and cryoprecipitate transfusion in New South Wales public hospitals. Med J Aust 2003; 178:117-121. Schwartz K A , Royer G, Kaufman DB, Permer J A . Complications of heparin administration in normal individuals. A m J Hematol 1985; 19:355-363. Shalansky S J , Verma A K , Levine M , Spinelli, J J , Dodek P M . Prospective analysis of risk markers for thrombocytopenia in critical care patients. Pharmacotherapy 2002; 22:803-813. Sheridan D, Carter C, Kelton J G . A diagnostic test for heparin-induced thrombocytopenia. Blood 1986;67:27-30. 237 Singer RL , Mannion JD, Bauer TL, Armenti FR, Edie RN. Complication from heparin-induced thrombocytopenia in patients undergoing cardiopulmonary bypass. Chest 1993; 104:1436-1440. Slotman GJ. Prospectively validated predictions of shock and organ failure in individual septic surgical patients: the Systemic Mediator Associated Response Test. Crit Care 2000; 4:319-326. Solenthaler M , Krauss JK, Boehlen F, Koller R, Hug M , Lammle B. Fatal fresh frozen plasma infusion containing H P A - l a alloantibodies. Br J Haematol 1999; 106:258-259. Spinier SA, Dager W. Overview of heparin-induced thrombocytopenia. A m J Health-Syst Pharm 2003; 60(Suppl 5):S5-S11. Steenland K , Beaumont J, Hornung R. The use of regression analyses in a cohort mortality study of welders. J Chron Dis 1986; 39:287-294. Stephan F, Cheffi M A , Kaplan C, Maillet J, Novara A , Fagon J, et al. Autoantibodies against platelet glycoproteins in critically i l l patients with thrombocytopenia. Am J Med 2000; 108:554-560. Stephan F, de Montblanc J, Cheffi A , Bonnet F. Thrombocytopenia in critically i l l surgical patients: a case-control study evaluating attributable mortality and transfusion requirements. Crit Care 1999a; 3:151-158. Stephan F, Hollande J, Richard O, Cheffi A , Maier-Redelsperger M , Flahault A . Thrombocytopenia in a surgical ICU. Chest 1999; 115:1363-1370. Stephan F, Thioliere B, Verdy E, Tulliez M . Role of hemophagocytic histiocytosis in the etiology of thrombocytopenia in patients with sepsis syndrome or septic shock. Clin Infect Dis 1997; 25:1159-1164. Stewart M E , Etches M W , Boshkov L K , Gordon PA. Heparin-induced thrombocytopenia: an improved method of detection based on lumi-aggregometry. Br J Haematol 1995; 91:173-177. Steyerberg EW, Bleeker SE, Moll HA, Grobbee DE, Moons K G M . Internal and external validation of predictive models: a simulation study of bias and precision in small samples. J Clin Epidemiol 2003; 56:441^47. Steyerberg EW, Harrell FE, Borsboom GJ, Eijkemans MJ, Vergouwe Y , Habbema JD. Internal validation of predictive models. Efficiency of some procedures for logistic regression analysis. J Clin Epidemiol 2001; 54:774-781. Strauss R, Wehler M , Mehler, K Kreutzer D, Koebnick C, Hahn EG. Thrombocytopenia in patients in the medical intensive care unit: Bleeding prevalence, transfusion requirements, and outcome. Crit Care Med 2002; 30:1765-1771. Suh JS, Aster RH, Visentin GP. Antibodies from patients with heparin-induced thrombocytopenia/thrombosis recognize different epitopes on heparimplatelet factor 4 complexes. Blood 1998; 91:916-922. 238 Sultan C, Gouault-Heilman M , Imbert M , eds. Manuel of hematology. John Wiley and Sons, 1985. 129. Susla G M , Takanishi G, Dellinger RP. Drugs used in critical care medicine. In: Parrillo JE, Dellinger RP eds. Critical care medicine: principles of diagnosis and management in the adult. 2 n d ed. Philadelphia, Pa. Mosby; 2001:312-368. The RISC group. Risk of myocardial infarction and death during treatment with low dose aspirin and intravenous heparin in men with unstable coronary artery disease. Lancet 1990; 336:827-830. Theroux P, Ouimet H , McCans J, Latour JG, Joly P, Levy G, et al. Aspirin, heparin, or both to treat acute unstable angina. N Engl J Med 1988; 319:1105-1111. Tinmouth AT, Freedman J. Prophylactic platelet transfusions: which dose is the best dose? A review ofthe literature. Transfus Med Rev 2003; 17:181-193. Tostivint I, Mougenot B, Flahault A , Vigneau C, Costa M A , Haymann JP, et al. Adult haemolytic and uraemic syndrome: causes and prognostic factors in the last decade. Nephrol Dial Transplant 2002; 17:1228-1234. Towne JB, Bernhard V M , Hussey C, Garancis JC. White clot syndrome. Peripheral vascular complications of heparin therapy. Arch Surg 1979; 114:372-377. Trowbridge A A , Caraveo J, Green JB III, Amaral B, Stone MJ . Heparin-related immune thrombocytopenia. Studies of antibody-heparin specificity. Am J Med 1978; 65:277-283. Trujillo TC, Nolan PE. Unfractionated heparin in acute coronary syndromes: has its time come and gone? A m J Health Syst Pharm 1998; 55:2402-2409. Usui Y , Ohshima Y, Yoshida K. Platelet aggregation by group B streptococci. J Gen Microbiol 1987; 133(Pt 6): 1593-1600. Van Eps RS, Wester JPJ, de Ruiter FE et al. The incidence of heparin-induced thrombocytopenia in critically i l l patients with the multiple organ dysfunction syndrome, (abstract) Thromb Haemost 2001; 86(Suppl July):P2719. Van Houwelingen JC, le Cessie S. Logistic regression, a review. Stat Neerl 1988; 42:215-232. Vanderschueren S, De Weerdt A , Malbrain M , Vankersschaever D, Frans E, Wilmer A , et al: Thrombocytopenia and prognosis in intensive care. Crit Care Med 2000; 28:1871-1876. Verma A K , Levine M , Shalansky S, Carter CJ, Kelton JG. Frequency of heparin-induced thrombocytopenia in critical care patients. Pharmacotherapy 2003; 23:745-753. Verma A K . A prospective observational study of risk indicators associated with the development of thrombocytopenia in a community-based ICU/CCU. M.Sc. Thesis. Vancouver B.C., Canada: Faculty of Pharmaceutical Sciences, University of British Columbia; 2000. 239 Vicente Rull JR, Loza Aguirre J, de la Puerta E, Moreno Millan E, Penas Maldonado L , Perez Clausell C. Thrombocytopenia induced by pulmonary artery flotation catheters: a prospective study. Intensive Care Med 1984; 10:29-31. Vincent JL, Baron JF, Reinhart K , Gattinoni L, Thijs L, Webb A , et al. A B C (Anemia and Blood Transfusion in Critical Care) Investigators. Anemia and blood transfusion in critically ill patients. J A M A 2002a; 288:1499-1507. Vincent JL, Moreno R, Takala J, Willatts S, De Mendonca A , Bruining H et al. The SOFA (Sepsis-related Organ Failure Assessment) score to describe organ dysfunction/failure. On behalf of the Working Group on Sepsis-Related Problems of the European Society of Intensive Care Medicine. Intensive Care Med 1996; 22:707-710. Vincent JL,Yagushi A , Pradier O. Platelet function in sepsis. Crit Care Med 2002; 30 (Suppl):S313-S317. Visentin GP, Ford SE, Scott PJ, Aster RH. Antibodies from patients with heparin-induced thrombocytopenia/thrombosis are specific for platelet factor 4 complexed with heparin or bound to endothelial cells. J Clin Invest 1994; 93:81-88. Vollmer RT. Multivariate statistical analysis for pathologists: Part 1, the logistic model. Am J Clin Pathol 1996; 105:115-126. Voyce SJ, McCaffree DR. Pulmonary artery catheters. In: Irwin RS, Rippe J M eds. Intensive care medicine. 5 t h ed. Philadelphia, Pa. Lippincott Williams & Wilkins; 2003:45-66. Wade EE, Rebuck JA, Healey M A , Rogers FB. H2 Antagonist-induced thrombocytopenia: is this a real phenomenon? Intensive Care Med 2002; 28:459-465. Wagner DP, Knaus W A , Harrell FE, Zimmerman JE, Watts C. Daily prognostic estimates for critically i l l adults in intensive care units: results from a prospective, multicenter, inception cohort analysis. Crit Care Med 1994; 22:1359-1372. Walenga JM, Jeske WP, Fasanella AR, Wood JJ, Ahmad S, Bakhos M . Laboratory diagnosis of heparin-induced thrombocytopenia. Clin Appl Thrombosis/Hemostasis 1999; 5(suppl 1): S21-S27. Walenga JM, Jeske WP, Prechel M M , Bacher P, Bakhos M . Decreased prevalence of heparin-induced thrombocytopenia with low-molecular-weight heparin and related drugs. Semin Thromb Hemost 2004; 30(suppl l):69-80. Wallis DE, Workman K L , Lewis BE, Steen L , Pifarre R, Moran JF. Failure of early heparin cessation as treatment for heparin-induced thrombocytopenia. A m J Med 1999; 106:629-635. Walls JT, Curtis JJ, Silver D, Boley T M , Schmaltz RA, Nawarawong W. Heparin-induced thrombocytopenia in open heart surgical patients: sequelae of late recognition. Ann Thorac Surg 1992; 53:787-791. 240 Ware JA, Coller BS. Platelet morphology, biochemistry, and function. In: Beutler E, Lichtricn M A , Coller BS, Kipps TJ, eds. Williams hematology. Toronto, Ont. McGraw-Hill, Inc; 1995:1161-1202. Warkentin T, Hirte H , Anderson DR, Wilson WEC, 0,Connell GJ, Lo RC. Transient global amnesia associated with acute heparin-induced thrombocytopenia. Am J Med 1994a; 97:489-492. Warkentin TE, Barkin RL. Newer strategies for the treatment of heparin-induced thrombocytopenia. Pharmacotherapy 1999; 19:181-195. Warkentin TE, Chong B H , Greinacher A . Heparin-induced thrombocytopenia: Towards consensus. Thromb Haemost 1998; 79:1-7. Warkentin TE, Elavathil LJ , Hayward CP, Johnston M A , Russett JI, Kelton JG. The pathogenesis of venous limb gangrene associated with heparin-induced thrombocytopenia. Ann Intern Med 1997; 127:804-812. Warkentin TE, Greinacher A . Heparin-induced thrombocytopenia and cardiac surgery. Ann Thorac Surg 2003; 76:2121 -2131. Warkentin TE, Greinacher A . Laboratory testing for heparin-induced thrombocytopenia. In: Warkentin TE, Greinacher A , eds. Heparin-induced Thrombocytopenia. 2 n d edition. New York, N Y : Marcel Dekker, Inc.; 2001:231-270. Warkentin TE, Hayward C P M , Boshkov L K , Santos A V , Sheppard JA, Bode AP , et al. Sera from patients with heparin-induced thrombocytopenia generate platelet-derived microparticles with procoagulant activity: an explanation for the thrombotic complications of heparin-induced thrombocytopenia. Blood 1994; 84:3691-3699. Warkentin TE, Hayward C P M , Smith C A , Kelly P M , Kelton JG. Determinants of platelet variability when testing for heparin-induced thrombocytopenia. J Lab Clin Med 1992; 120:371-379. Warkentin TE, Heddle N M . Laboratory diagnosis of immune heparin-induced thrombocytopenia. Curr Hematol Rep 2003; 2:148-157. Warkentin TE, Kelton JG. A 14-year study of heparin-induced thrombocytopenia. Am J Med 1996; 101:502-507. Warkentin TE, Kelton JG. Delayed-onset heparin-induced thrombocytopenia and thrombosis. Ann Intern Med 2001a; 135:502-506. Warkentin TE, Kelton JG. Heparin-induced thrombocytopenia. Progr Hemost Thromb 1991; 10:1-34. Warkentin TE, Kelton JG. Management of thrombocytopenia. In: Colman W, Hirsh J, Marder V J , Salaman, eds. Hemostasis and Thrombosis: Basic principles and clinical practice. 3 r d ed. Philadelphia, Pa. J.B. Lippincott Company; 1994:469-486. 241 Warkentin TE, Kelton JG. Temporal aspects of heparin-induced thrombocytopenia. New Engl J Med 2001; 344:1286-1292. Warkentin TE, Kelton JG. Thrombocytopenia due to platelet destruction and hypersplenism. In: Hoffman R, Benz EJ Jr, Shattil SJ, Furie B, Cohen HJ, Silberstein L E , McGlave P eds. Hematology: Basic principles and practice. 3 r d ed. Philadelphia, Pa. J.B. Churchill-Livingstone; 2000:2138-2153. Warkentin TE, Levine M N , Hirsh J, Horsewood P, Roberts RS, Gent M , et al. Heparin-induced thrombocytopenia in patients treated with low-molecular-weight heparin or unfractionated heparin. New Engl J Med 1995; 332:1330-1335. Warkentin TE, Levine M N , Hirsh J, Klama L N , Kelton JG. Formation of heparin-induced thrombocytopenia IgG (HIT-IgG) without thrombocytopenia: analysis of a clinical trial (abstract). Blood 1995a; 86 (Suppl l):537a. Warkentin TE, Roberts RS, Hirsh, J Kelton JG. An improved definition of immune heparin-induced thrombocytopenia in postoperative orthopedic patients. Arch Intern Med 2003; 163:2518-2524. Warkentin TE, Sheppard JI, Horsewood P, Simpson PJ, Moore JC, Kelton JG. Impact of the patient population on the risk for heparin-induced thrombocytopenia. Blood 2000; 96:1703-1708. Warkentin TE, Souter RL, Panju A , Ginsberg JS. Acute systemic reactions to intravenous bolus heparin therapy: Characterization and relationship to heparin-induced thrombocytopenia (abstract). Blood 1992; 80(suppl l):160a. Warkentin TE. Clinical picture of heparin-induced thrombocytopenia. In: Warkentin TE, Greinacher A , eds. Heparin-induced Thrombocytopenia. 2 n d edition. New York, N Y : Marcel Dekker, Inc.; 2001:43-86. Warkentin TE. Heparin-induced skin lesions. Br J Haematol 1996; 92:494-497. Warkentin TE. Heparin-induced thrombocytopenia and its treatment. J Thromb Thrombolysis 2000a; 10:S29-S35. Warkentin TE. Heparin-induced thrombocytopenia, heparin-induced skin lesions, and arterial thrombosis. Thromb Haemost 1997; 77(suppl 1):562. Warkentin TE. Heparin-induced thrombocytopenia. Curr Hematol Rep 2002; 1:63-72. Warkentin TE. Heparin-induced thrombocytopenia. Pathogenesis, frequency, avoidance and management. Drug Saf 1997a; 17:325-41. Warkentin TE. Heparin-induced thrombocytopenia: IgG-mediated platelet activation, platelet microparticle generation, and altered procoagulant/anticoagulant balance in the pathogenesis of thrombosis and venous limb gangrene complicating heparin-induced thrombocytopenia. Trans Med Rev 1996a; 10:249-258. 242 Warkentin TE. Heparin-induced thrombocytopenia: yet another treatment paradox? Thromb Haemost 2001a; 85:947-949. Warkentin TE. Heparin-induced thrombocytopenia: a clinicopathologic syndrome. Thromb Haemost 1999; 82 (Suppl):439-447. Warkentin TE. Heparin-induced thrombocytopenia: pathogenesis and management. Br J Haematol 2003; 121:535-555. Warkentin TE. Laboratory testing for heparin-induced thrombocytopenia. J Thromb Thrombolysis 2000; 10:S34-S45. Warkentin TE. Platelet count monitoring and laboratory testing for heparin-induced thrombocytopenia. Arch Pathol Lab Med 2002a; 126:1415-1423. Warkentin TE. Limitations of conventional treatment options for heparin-induced thrombocytopenia. Semin Hematol 1998; 35(suppl 5): 17-25. Wazny L D , Ariano RE. Evaluation and management of drug-induced thrombocytopenia in the acutely i l l patient. Pharmacotherapy 2000; 20:292-307. Williams RT, Damaraju L V , Mascelli M A , Barnathan ES, Califf R M , Simoons M L et al. Anti-PF4/Heparin antibodies: an independent predictor of 30-day myocardial infarction after acute coronary ischemic syndromes. Circulation 2003; 107:2307-2312. Williams WJ, Morris M W , Nelson DA. Examination of the blood. In: Beutler E, Lichtricn M A , Coller BS, Kipps TJ, eds. Williams hematology. Toronto, Ont. McGraw-Hill, Inc; 1995:8-25. Williams WJ. Classification and clinical manifestations of disorders of hemostasis. In: Beutler E, Lichtricn M A , Coller BS, Kipps TJ, eds. Williams hematology. Fifth edition. Toronto, Ont. McGraw-Hill, Inc; 1995a: 1276-1281. Wilson JJ, Neame PB, Kelton JG. Infection-induced thrombocytopenia. Semin Thromb Hemost 1982; 8:217-233. Wittels EG, Siegel RD, Mazur E M . Thrombocytopenia in the intensive care setting. J Intensive Care Med 1990; 5:224-240. Wolfe R A , Strawderman RL. Logical and statistical fallacies in the use of Cox regression models. Am J Kidney Dis 1996; 27:124-129. Yeaman MR. The role of platelets in antimicrobial host defense. Clin Infect Dis 1997; 25:951— 970. Y u DT, Piatt R, Lanken PN, Black E, Sands K E , Schwartz JS, et al. Relationship of pulmonary artery catheter use to mortality and resource utilization in patients with severe sepsis. Crit Care Med 2003;31:2734-2741. Ziporen L , L i ZQ, Park KS , Sabnekar P, Liu W Y , Arepally G, et al. Defining an antigenic epitope on platelet factor 4 associated with heparin. Blood 1998; 92:3250-3259. 243 Zucker-Franklin D. The effect of viral infections on platelets and megakaryocytes. Semin Hematol. 1994;31:329-37. 244 APPENDIX 3 Database worksheets PATIENT DEMOGRAPHICS Patient Name: MRN: Study Number: ICU/CCU Bed Number: PHN: Age: Date of Birth: Gender: M / F Race: A P A C H E II Score: Alcohol History: Date Admitted to ICU/CCU: Where did patient come from: ER, WARD, OTHER HOSP Duration of ICU/CCU Stay Before Thrombocytopenia: Total Duration of Stay in the ICU/CCU: /Total Stay in Hospital Weight: Height: P N L P LABORATORY VALUES: DATE"; TIME HEMOGLOBIN PLATELETS ' RENAL • ; ' DYSFUNCTION • : .- HEPATIC C fSFUNC HON S Cre CreCl : AST , ALT Alk Totbil Dirbil INR Admission platelet count: Did the patient develop thrombocytopenia: If yes, what was the count: Date: Platelet count: minimum: mean: threshold: Hemoglobin: minimum: mean: 247 MEDICATIONS PRIOR AND DURING ICU/CCU STAY PRIOR TO THROMBOCYTOPENIA: N.B. RECORD ALL MEDICATIONS IF NO THROMBOCYTOPENIA N.B. ONLY RECORD MEDICATIONS TAKEN FOR THE PAST 3 MONTHS. MEDICATIONS PAST USE : START/STOP . DATE MEDICATIONS PAST USE START/STOP DATE ACETAZOLAMIDE FUROSEMIDE ACETOHEXAMIDE GENTAMICIN ALDACTHIAZIDE GLICLAZIDE AMIKACIN GLYBURIDE AMOXICILLIN HYDROCHLOROTHIAZIDE AMPHOTERICIN B IBUPROFEN AMPICILLIN IMIPENEM AMRINONE INDOMETHIC1N ANTINEOPLASTIC AGENT IPRATROPIUM BR ASA ISOPROTERENOL AURANOFIN (PO) KETOCONAZOLE AUROTH1ANALATE (IV) KETOPROFEN AZOGANTRIZIN MEFENAMIC ACID CEFACLOR METHYLDOPA CEFAMANDOLE METOLAZONE CEFAZOLIN NAPROXEN CEFOTAXIME NOREPINEPHRINE CEFTAZIDIME OLSALAZINE CEFTIZOXIME PENICILLIN G CEFTRIAXONE PENICILLIN V CEFUROXIME PHENYLEPHRINE CEPHALEXIN PHENYTOIN CHLORPROPAMIDE PIPERACILLIN CHLORTHALIDONE QUINIDINE CIMETIDINE QUININE CLOXACILLIN RANITIDINE COTRIMOXAZOLE SALBUTAMOL DICLOFENAC SULFADIAZINE DIG0X1N SULFASALAZINE DOBUTAMINE SULFINPYRAZONE DOPAMINE SULFISOXAZOLE DYAZIDE TICARCILLIN EPINEPHRINE TINZAPARIN ETHACRYNIC ACID TOBRAMYCIN FLUCONAZOLE TOLBUTAMIDE FLUCYTOSINE VANCOMYCIN Total number of medications patient is exposed to prior to thrombocytopenia during ICU/CCU stay: Total number of medications if patient does not develop thrombocytopenia during ICU/CCU stay:_ 248 HEPARIN Date heparin first started: Platelet count when heparin first started: D A T E : ' M E A N D O S E / R O U T E . D U R A T I O N I N D I C A T I O N Did the patient ever receive heparin in the past: If yes, when: Was heparin discontinued as a result of thrombocytopenia: Days patient on heparin: D O S E R A N G E : T O T A L H E P A R I N D O S E : FULL ANTICOAGULATION FOR THROMBOSIS H E R A P Y : 1 5 , 0 0 0 U / D A Y PROPHYLACTIC DOSES: 1 0 , 0 0 0 - 1 5 , 0 0 0 U / D A Y DOSES TO MAINTAIN IV LINE PATENCY AND PULMONARY ARTERY CATHETERS : < 1 0 , 0 0 0 U /DAY Dose heparin per day: C A T H E T E R T Y P E S T A R T / S T O P D A T E > U N I T S O F H E P A R I N S W A N G A N Z ( P A ) A R T E R I A L L I N E C V P L I N E Infusion Rate: 6U/hr (144U/d) Include IO0U/24hrs for flushes ELISA TEST: Is the test performed on this patient: If yes, is the test for early, late, or no thrombocytopenia, but patient on heparin: A492nm value: Result: 249 DIAGNOSES: Diagnosis at thrombocytopenia or discharge: D I A G N O S E S Y E S / N O . D I A G N O S E S Y E S / N O A C U T E M Y O C A R D I A L I N F A R C T I O N K I D N E Y , U R I N A R Y T R A C T , R E P R O D U C T I V E C A R D I O V A S C U L A R S U R G E R I E S M A L I G N A N C Y C A R D I O V A S C U L A R N O N S U R G E R I E S M U S C U L O S K E L E T A L & C O N N E C T I V E TISSUE D I A B E T E S M E L L I T U S N E R V O U S S Y S T E M D R U G O V E R D O S E / P O I S O N I N G S R E S P I R A T O R Y N O N S U R G I C A L E N D O C R I N E A N D N U T R I T I O N R E S P I R A T O R Y S U R G I C A L G A S T R O I N T E S T I N A L SEPSIS G l B L E E D U N S T A B L E A N G I N A I N F E C T I O N Admitting Diagnosis: PROCEDURES: P R O C E D U R E S :' PRIOR T O I C U / C C U D U R I N G I C U / C C U S T A R T D A T E P U L M O N A R Y A R T E R Y C A T H E T E R P L A C E M E N T T R A N S F U S I O N S (TOT. N U M B E R A N D V O L U M E ) PACKED RED BLOOD CELLS FRESH FROZEN PLASMA PLATELETS C A R D I A C V A L V E S O R PROSTHESIS S U R G I C A L P R O C E D U R E S ( A L L ) M E C H A N I C A L V E N T I L A T I O N C A R D I O P U L M O N A R Y B Y P A S S S U R G E R Y Surgical Procedure(s)/Date: FINAL CLINICAL OUTCOME: Transfer from ICU/CCU: Discharge from ICU/CCU: Expired: Did patient expire on ward after leaving the ICU/CCU: E V E N T Y E S / N O D A T E T H R O M B O E M B O L I S M H E M O R R H A G E SKIN E R U P T I O N 250 APPENDIX 4 APACHE II score THE APACHE II SEVERITY OF DISEASE CLASSIFICATION SYSTEM V A R I A B L E +4 +3 +2 +1 0 +1 +2 +3 +4 Temperature-rectal (°C) 0 >41° 0 39°-40.9° O 38.5°-38.9° O 36°-38.4° O 34°-35.9° O 32°-33.9° O 30°-31.9° O <29.9° Mean Arterial Pressure (mm Hg) 0 > 160 0 130-159 O 110-129 O 70-109 O 50-69 O <49 Heart Rate (ventricular response) 0 > 180 O 140-179 O 110-139 O 70-109 O 55-69 O 40-54 O <39 Respiratory Rate (non-ventilated or ventilated) 0 >50 O 35-49 O 25-34 O 12-24 O 10-11 O 6-9 O <5 Oxygenation: A-a D 0 2 or Pa0 2 (mm Hg) 0 >500 O 350-499 O 200-349 O <200 b) F i 0 2 < 0.5 record only Pa0 2 O P02 > 70 O PO251-70 P02 55-60 PO, <55 Arterial pH 0 >7.7 O 7.6-7.69 O 7.5-7.59 O 7.33-7.49 O 7.25-7.32 O 7.15-7.24 O < 7.15 Serum Sodium (mMol/L) 0 > 180 O 160-179 O 155-159 O 150-154 O 130-149 O 120-129 O 111-119 O < 110 Serum Potassium (mMol/L) 0 >7 O 6-6.9 O 5.5-5.9 O 3.5-5.4 O 3-3.4 O 2.5-2.9 o <2.5 Serum Creatinine (mg/lOOmL) (Double point score for acute renal failure) O >3.5 O 2-3.4 O 1.5-1.9 O 0.6-1.4 O <0.6 Hematocrit (%) 0 >60 O 50-59.9 O 46-49.9 O 30-45.9 O 20-29.9 O <20 White Blood Count (total/mmJ) (in 1000s) 0 >40 O 20-39.9 O 15-19.9 O 3-14.9 O 1-2.9 O < 1 Glasgow Coma Scale (GCS) Score = 15 minus actual GCS A) Total Acute Physiology Score (APS)—sum of the 12 individual variable points Serum H C 0 3 (venous—m<ol/L) |Not preferred; use if no ABGs) 0 >52 O 41-51.9 O 32-40.9 O 22-31.9 O 18-21.9 O 15-17.9 O < 15 B) AGE POINTS: Assign points to age as follows: A G E (yrs) Points <44 0 44-54 2 55-64 3 65-74 4 >75 5 C) CHRONIC HEALTH POINTS If the patient has a history of severe organ system insufficiency or is immuno-compromised assign points as follows: a. ) for non-operative or emergency postoperative patients—5 points or b. ) for elective postoperative patients—2 points DEFINITIONS Organ insufficiency or immuiio-compromiscd state must have been evident prior to this hospital admission and conform to the following criteria: LIVER: Biopsy proven cirrhosis and documented portal hypertension; episodes of past upper CI bleeding attributed to portal hypertension; or prior episodes of hepatic failure/encephalopathycoma Calculate the following < 24 hrs after ICU admission: 1. MAP = 2 (DBP) + SBP 3 CARDIOVASCULAR: New York Heart Association Class IV RESPIRATORY: Chronic restrictive, obstructive, or vascular disease resulting in severe exercise restriction, i.e. unable to climb stairs or perform household duties; or documented chronic hypoxia, hypercapnia, secondary polycythemia, severe pulmonary hypertension (> 40 mm Hg), or respirator dependency RENAL: Receiving chronic dialysis IMMUNOCOMPROMISED: The patient has received therapy that suppresses resistance to infection, e.g. immuno-suppression, chemotherapy, radiation, long-term or recent steroids, or has a disease that is sufficiently advanced to suppress resistance to infection, e.g. leukemia, lymphoma, AIDS APACHE II SCORE Sum of A + B + C 2. A-aD0 2 = 7 1 3 ( F i 0 2 ) - P a C 0 2 - P a 0 2 0.8 3. SC.r (mg/1 OOmLi = SCr (umol/L) 88.40 A) APS Points^ B) Age Points_ C) Chronic Health Points_ Total APACHE II 251 APPENDIX 6 Logistic Regression Regression models are used to describe the relation between a dependent (response) variable and one or more independent (explanatory) variables. The goal of any regression analysis is to build the most reasonable and parsimonious model that remains logical from a clinical point of view (Hosmer and Lemeshow, 1989). The most common example of regression analysis is linear regression modeling, in which the dependent variable is assumed to be continuous. However, when researchers are concerned with a dichotomous dependent variable, logistic regression modeling becomes the standard method of analysis. Logistic regression is based on the principle of regressing a dichotomous dependent variable on a set of independent covariates (risk indicators). The theory of logistic regression was developed by Cox (1970), and several textbooks have presented the theory and applications of the logistic regression model (e.g. Hosmer and Lemeshow, 2000). The goal of logistic regression analysis in the present investigation is to achieve a predictive mathematical model that describes the relation between thrombocytopenia and a set of independent variables. For a set of k independent variables denoted by the vector x = (xi, X2,....,X|<), the logistic regression model that the outcome is present is p ( y = l | x ) = e ( P o + p i x l +P2x2+....+Pkxk) | + g ( P o + pixl+p2x2+....+pkxk) The model can be used to predict the probability for each patient (case) to develop the dichotomous outcome. If the probability of the outcome is p, then p/(l - p) is the odds of the outcome occurring and log e [p/(l - p)] is referred to as the log odds of the outcome occurring. This can be expressed in a linear model as follows: 253 loge (p/1 - p) = p o + PlX, + (32X2 +....+ PkXk The log odds is also called the logit of p. It is the natural logarithm of the ratio of two probabilities. The logit acts as a "link" function linking the predicted probabilities to a linear combination of the independent variables where: po is the intercept coefficient. It is the log odds of the dichotomous outcome variable in the reference group when all the independent variables are 0. Since 0 may be outside the clinically meaningful range for some independent variables, for example platelet count, this coefficient may have no clinical interpretation, but is required in the model. Pi to Pk are the coefficients of the independent variables. They represent the increase in the log odds that the dichotomous outcome will develop per unit increase in Pk, given that all other independent variables remain constant. In other words, it is the effect of Pk on the outcome, adjusted for all other independent variables, xi to Xk are the independent variables. Logistic regression is used to identify the effect of individual variables and give an estimate of the odds ratio (i.e. estimate of the relative risk) for the individual effect. The odds ratio is defined as the odds in favour of an outcome among individuals exposed to a risk indicator divided by the odds of the event among individuals unexposed to the risk indicator. Interestingly, the odds ratio is a close approximation of the relative risk for rare diseases or outcomes. The coefficients in the logistic regression model represent the change in the logit of one unit in the independent variable. Interpretation of the coefficient depends on being able to place meaning on the difference between two logits. For example, when the independent 254 variable is dichotomous, and has the value 0 and 1, the odds coefficient P is the logarithm of the odds ratio of those exposed relative to those unexposed. When a logistic regression model contains a continuous independent variable, interpretation of the coefficient depends on how the variable is entered into the model and the specific units of the variable. Assuming that the logit is linear in the continuous variable (x), the slope coefficient, p, represents the change in the log odds for an increase of 1 unit in X . In many cases, an increase of 1 unit will not be biologically or clinically meaningful. Therefore, a change of 10, 25, or 50 units might be considered more useful. The likelihood of the model, which is used in estimation and in significance testing, expresses the probability of the observed data as a function of the unknown parameters. The regression coefficients are estimated by the method of maximum likelihood. The method of maximum likelihood chooses the parameters which give the highest probability of the observed data. Standard errors of the estimates and 95% confidence intervals are estimated using large sample theory. The statistical significance of the estimated regression coefficients are computed using the likelihood ratio test. The likelihood ratio test statistic is -2 times the difference between the log likelihoods of two fitted models, one of which is a subset of the other. A fitted model is the logistic model with the estimated parameters inserted. The distribution of the likelihood ratio statistic is closely approximated by the chi-square distribution for large sample sizes. The degrees of freedom of the approximating chi-square distribution is equal to the difference in the number of regression coefficients in the two models. The test is named as a ratio rather than a difference since the difference between two log likelihoods is equal to the log of the ratio of the two likelihoods. The value, -2, adjusts LR so the chi-square distribution can be used to approximate its distribution. Note that as the likelihood (L) increases, -2LL decreases, and hence decreasing values of -2LL are associated with improved models. 255 APPENDIX 7 Syntax for Generating Bootstrap Samples with Replacement i nput p r o g r a m , l o op c a se= 1 to 7 0 7 . + l oop #i=1 to 1. + c o m p u t e ident if=trunc(un iform(707) )+1 . + e n d c a s e . + e n d l oop . + l e a v e c a s e . e n d l oop . e n d f i l e . e n d input p r o g r a m , sor t c a s e s by ident if . e x e c u t e . 256 APPENDIX 8 Macro for the Bootstrap Procedure define !path() "c:\My Documents\Bootstrap ModelsV lenddefine . define !path2(name=!tokens(1) /extn=!tokens(1) !default(".sav")) !quote(!concat(!unquote(!eval(!path)),!unquote(!name),!unquote(!extn))) lenddefine . define bstrplr(). set mprint on. Ido !sample=1 !to200. !LET !var=!concat(var,!sample). !LET !file=!uhquote("ICUCCU_Admission_150_Data"). !LET !file1=!concat(bpred, Isample). !LET !file2=!concat(boots,Isample). ILET !file3=!concat(bootf,Isample). get file=!path2 name=!file. input program. loop case=1 to 707. + loop #i=1 to 1. + compute identif=trunc(uniform(707))+1. + end case. + end loop. + leave case. end loop. end file. end input program, sort cases by identif. compute bsamp=1. execute. save outfile=!path2 name=!file2. MATCH FILES /FILE=* /FILE=!path2 name=!file / R E N A M E (addxcatl addxcat4 admdxall admiss50 admissio age age5 alcoholh apacheii didtcpde gastrin2 gibled2 infecti2 male musculo2 otheric4 sepsis2 studynum surgbfic vasclsu2 yearlghl = dO d1 d2 d3 d4 d5 d6 d7 d8 d9 d10 d11 d12 d13 d14 d15 d16 d17 d18 d19 d20) /BY identif /DROP= dO d1 d2 d3 d4 d5 d6 d7 d8 d9 d10 d11 d12 d13 d14 d15 d16 d17 d18 d19 d20. E X E C U T E . match files /file=* /table=!path2 name=!file /by identif. execute. if (sysmis(bsamp)) didtcpde=$sysmis. LOGISTIC REGRESSION VAR=didtcpde /METHOD=BSTEP(LR) apacheii male alcoholh addxcat4 surgbfic age5 admiss50 yearlghl 257 / CONTRAST (addxcat4)=lndicator(1) / S A V E P R E D /PRINT=CI(95) /CRITERIA PIN(.04) POUT(.05) ITERATE(20) CUT(.5). save outfile=!path2 name=!file3 /keep identif didtcpde pre_1. A G G R E G A T E /OUTFILE=!path2 name=!file1 /BREAK=identif /!var= FIRST(preJ) . execute. Idoend lenddefine. define bstrpmf(). set mprint on. !I_ET !file=!unquote("ICUCCU_Admission_150_Data"). get file=!path2 name=!file /keep=identif didtcpde. save outfile-bpred.sav'. !do !sample=1 !to 200. !LET !file1=!concat(bpred,!sample). MATCH FILES /FILE=* /FILE=!path2 name=!file1 /BY identif. EXECUTE . save outfile=!path2 name=bpred. Idoend lenddefine. define bauca(). set mprint on. Ido !sample=1 Ito 200. ILET !file3=!concat(bootf,!sample) get file=!path2 name=!file3. ROC pre_1 BY didtcpde (1) /PLOT = NONE /CRITERIA = CUTOFF(INCLUDE) TESTPOS(LARGE) DISTRIBUTION(FREE) Cl(95) /MISSING = EXCLUDE . Idoend lenddefine. define bauct(). set mprint on. get file=!path2 name='bpred'. Ido !sample=1 Ito 200. ILET !var=!concat(var,!sample). ROC Ivar BY didtcpde (1) /PLOT = NONE /CRITERIA = CUTOFF(INCLUDE) TESTPOS(LARGE) DISTRIBUTION(FREE) Cl(95) /MISSING = EXCLUDE . 258 Idoend lenddefine. *edit the file 'bootstrap.sps'. *change the directory in the first line to the directory you are working in. *(the data file 'ICUCCU_Admission_150_Data.sav' should also be in that directory). * change the file name of the data file in line five of the macro bstrplr. *** NOTE: no spaces in the file name. **************************************** * bstrplr is a macro which generates the bootstrap samples, does 200 log reg, * and saves the predicted values in 200 files. * bstrpmf is a macro which merges the 200 predicted values into one file. * bstrauc is a macro which computes auc's for the 200 predicted values. * run the following commands. * open and run the file 'bootstrap.sps'. bstrplr. bstrpmf. bauca. * export output file to text file, bauct. * export output file to text file. 2 5 9 APPENDIX 9 Univariate Analysis for Model 1PA (ICU/CCU < 150 x 109/L Exploratory Post-Admission Model) Admission Variables Thrombocytopenia (N = 122) No Thrombocytopenia (N = 585) P-value Males 75 (61.5%) 363 (62.1%) 0.905 Alcohol history 21 (17.2%) 63 (10.8%) 0.045 Admission Diagnoses < 0.001 Acute Myocardial Infarction 12 (9.8%) 163 (27.9%) < 0.001 Cardiovascular Non-Surgery 13 (10.7%) 74 (12.6%) 0.542 Unstable Angina 3 (2.5%) 88(15.0%) < 0.001 Vascular surgery 4 (3.3%) 11 (1.9%) 0.330 Diabetes mellitus 1 (0.8%) 9(1.5%) 0.541 Drug overdose 3 (2.5%) 15(2.6%) 0.947 Endocrine N/A N/A N/A Gastrointestinal 15 (12.3%) 18(3.1%) < 0.001 Gastrointestinal bleed 7 (5.7%) 10(1.7%) 0.008 Infection 11 (9.0%) 29 (5.0%) 0.078 Kidney, Urinary tract Reproductive 0 (0%) 1 (0.2%) 0.648 Malignancy 0 (0%) 7(1.2%) 0.225 Musculoskeletal/connective tissue 10(8.2%) 15 (2.6%) 0.002 Nervous system 13 (10.7%) 32 (5.5%) 0.033 Respiratory non-surgery 22(18.0%) 88(15.0%) 0.407 Respiratory surgery 1 (0.8%) 16 (2.7%) 0.209 Sepsis 7 (5.7%) 9(1.5%) 0.005 Surgery within 24 hours of admission 38(31.1%) 90(15.4%) < 0.001 Age 64.4 ± 16.9 64.5 ± 15.0 0.941 Admission Platelet Count 208.5 ±56.8 249.2 ± 79.6 < 0.001 APACHE II score 21.0 ± 10.0 14.4 ±8.0 < 0.001 Year at LGH 68 (55.7%) 294 (50.3%) 0.271 260 Univariate Analysis Continued Admission Variables Thrombocytopenia (N = 122) No Thrombocytopenia (N = 585) P-value ASA 24 (19.7%) 295 (50.4%) < 0.001 Digoxin 23 (18.9%) 101 (17.3%) 0.675 Gentimicin 11 (9.0%) 36 (6.2%) 0.248 Heparin 102 (83.6%) 506 (86.5%) 0.403 Heparin dose/day 10548.7 ±9811.8 16107.2 ±2353.8 < 0.001 Ipratropium Br 60 (49.2%) 194 (33.2%) 0.001 Phenytoin 10(8.2%) 34 (5.8%) 0.321 Salbutamol 63 (51.6%) 214(36.6%) 0.002 Medication class inotropes 53 (43.4%) 100(17.1%) < 0.001 Medication class cephalosporins 60 (49.2%) 182 (31.1%) < 0.001 Medication class H2-antagonists 50 (41.0%) 151 (25.8%) 0.001 Medication class penicillins 11 (9.0%) 45 (7.7%) 0.622 Medication class sulfa-drugs 54 (44.3%) 282 (48.2%) 0.428 Swan Ganz catheter 49 (40.2%) 63 (10.8%) < 0.001 PRBC Transfusions 37 (30.3%) 45 (7.7%) < 0.001 FFP Transfusions 9 (7.4%) 6(1.0%) < 0.001 Surgical Procedures 15(12.3%) 76(13.0%) 0.835 Mechanical ventilation 71 (58.2%) 131 (22.4%) < 0.001 Renal dysfunction 18(14.8%) 68(11.6%) 0.336 Hepatic dysfunction 11 (9.0%) 11 (1.9%) < 0.001 261 

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.831.1-0092279/manifest

Comment

Related Items