UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Integration of genomics into clinical care : methods for economic evaluation Najafzadeh, Mehdi 2012

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

Item Metadata

Download

Media
24-ubc_2012_spring_najafzadeh_mehdi.pdf [ 3.64MB ]
Metadata
JSON: 24-1.0072612.json
JSON-LD: 24-1.0072612-ld.json
RDF/XML (Pretty): 24-1.0072612-rdf.xml
RDF/JSON: 24-1.0072612-rdf.json
Turtle: 24-1.0072612-turtle.txt
N-Triples: 24-1.0072612-rdf-ntriples.txt
Original Record: 24-1.0072612-source.json
Full Text
24-1.0072612-fulltext.txt
Citation
24-1.0072612.ris

Full Text

INTEGRATION OF GENOMICS INTO CLINICAL CARE: METHODS FOR ECONOMIC EVALUATION by MEHDI NAJAFZADEH B.Sc., Sharif University of Technology, 1997 M.Sc., Institute for Research on Planning & Development, 2001 M.A., The University of British Columbia, 2005  A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY in THE FACULTY OF GRADUATE STUDIES (Pharmaceutical Sciences) THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver)  March 2012 © Mehdi Najafzadeh, 2012  Abstract Background: As genomic technologies become more affordable, the demand for having these data will increase. Decision-makers must anticipate the increasing influence of genomics on heath care systems and take into account the expectations of patients, the public, health care providers, and industry in this regard. This thesis demonstrates applications of several methods for evaluation of genomic technologies in medicine. Using four case studies, I have highlighted the advantages that each method can offer given the nature and scope of the research question in each case study. Objectives: My specific objectives in the case studies were: 1) To elicit the preferences of cancer patients as well as the public for a hypothetical, genetically-guided treatment for cancer (a discrete choice experiment ); 2) To estimate the relative importance of attributes which influence physicians’ decisions for using personalized medicine in their practice (a Best Worst Scaling choice experiment); 3) To evaluate the impact of three potential genomic/proteomic tests on the long term burden of COPD in Canada (a system dynamics model); 4) To measure the cost-effectiveness of adding a new molecular diagnostic test (DX) to the current diagnostic strategy for thyroid cancer (using a discrete event simulation). Methods: Through these case studies, I have demonstrated the particular advantages of using discrete choice experiment (DCE), best-worst scaling (BWS) experiment, system dynamics simulation, and discrete event simulation (DES) for evaluations of genomic technologies. Results: Using four case studies I exemplified the questions that emerge in the process of integrating genomics into clinical care. In addition to bridging the methodological gaps by incorporating several novel methods (BWS, dynamic systems, and DES), the selected case studies illustrated the practical issues regarding the integration of genomics into clinical care from the perspective of patients, the public, health care providers, and decisionmakers. Conclusion: Although the methods previously developed for health technology assessment can be applied to the evaluation of genomic technologies as well, methodological challenges in the evaluation of genomic applications entail utilizing more diverse and more sophisticated analytical tools.  ii  Preface This doctoral thesis summarizes the research that I conducted related to the integration of genomics into clinical care. All four topics in my thesis were initially identified by my PhD supervisor, Dr Carlo Marra. In all of four studies, I was responsible for the statistical design of experiments, data collection, statistical analysis of data, and simulations. I also wrote all four manuscripts. I initiated using three of these methods (best worst scaling, dynamic systems simulation, and discrete event simulation) in our research center. My supervisor provided me with comments and advice throughout, and provided comments on the thesis chapters and associated manuscripts. The co-authors further commented on later versions of the manuscripts. I received the following certificate of approvals for conducting the choice experiments: 1 Barriers to integrating personalized medicine into clinical practice. Behavioural Research Ethics Board, the University of British Columbia (H09- 01797). 2 Preferences of lymphoma patients and the public regarding genetic testing. British Columbia Cancer Agency Research Ethics Board (H09- 00236). I have submitted four co-authored manuscripts to peer reviewed journals (Chapters 2, 3, 4 and 5). Chapter 1: Introduction. Mehdi Najafzadeh was responsible for literature search, summarizing, and writing Chapter 1. Dr. Carlo Marra revised and approved contents of this chapter. Drs. Larry Lynd and Stirling Bryan also revised this chapter.  iii  Chapter 2: Genetic Testing to Determine Drug Response: Measuring Preferences of Patients and the Public Using Discrete Choice Experiment (DCE). Mehdi Najafzadeh was responsible for the design, data collection, statistical analysis, and writing of the manuscript. Dr. Carlo Marra was responsible for conception of the study and also contributed to interpretation of the results and revision of manuscript. Dr. Joseph Connors provided access to the patients who participated in the study. Drs. Larry Lynd, Karissa Johnston, Stuart Peacock, and Marco Marra helped interpret the results and contributed to the revision of the manuscript. Chapter 3: Barriers for Integrating Personalized Medicine into Clinical Practice: A BestWorst Scaling Choice Experiment. Mehdi Najafzadeh was responsible for the design, data collection, statistical analysis, and writing the manuscript. Dr. Carlo Marra was responsible for conception of the project and also contributed to the interpretation of the results and revision of the manuscript. Drs. Larry Lynd, Stirling Bryan, Aslam Anis, Jennifer Davis and Marco Marra helped interpret the results and revise the manuscript. A version of this chapter has been accepted for publication. Najafzadeh M, Lynd LD, Davis JC, Bryan S, Anis A, Marra MA, Marra CA. Barriers for integrating personalized medicine into clinical practice: a best worst scaling choice experiment. Forthcoming in Genetics in Medicine (GIM-D-11-00182 R1). Chapter 4: Future Impact of Genomic and Proteomic Tests on the Burden of COPD: A Dynamic Population Model. Mehdi Najafzadeh was responsible for the design, modeling, and writing of the manuscript. Dr. Carlo Marra was responsible for conception of the idea and also contributed to the interpretation of the results and revision of manuscript. Drs.  iv  Larry Lynd, Don Sin, Mohsen Sadatsafavi, Mark FitzGerald, and Bruce MacManus contributed to the interpretation of the results and helped revise the manuscript. Chapter 5: Cost-Effectiveness of Using a Novel Genomic Test (GenomeDx) for Diagnosis of Thyroid Cancer. Mehdi Najafzadeh was responsible for design, modeling, and writing the manuscript. Drs. Carlo Marra and Larry Lynd were responsible for the conception of the idea and also contributed to the interpretation of the results and revision of manuscript. Dr. Sam Wiseman provided his expertise to ensure the clinical validity of the model, and contributed to the interpretation of the results and the revision of the manuscript. Chapter 6: Integrated Discussion. Mehdi Najafzadeh was responsible for summarizing the results of the thesis and for writing an integrated discussion. Dr. Carlo Marra revised and approved contents of this chapter. Dr. Stirling Bryan also provided his comments for improving this chapter.  v  Table of Contents Abstract .................................................................................................................................................... ii Preface .................................................................................................................................................... iii Table of Contents..................................................................................................................................... vi List of Tables ............................................................................................................................................ ix List of Figures ............................................................................................................................................ x Glossary .................................................................................................................................................. xii Acknowledgements ............................................................................................................................... xiii Dedication ............................................................................................................................................. xiv Chapter 1: Introduction............................................................................................................................. 1 1.1 Overview of Thesis Objectives and Themes .........................................................................................1 1.1.1 Thesis Objectives ...........................................................................................................................1 1.1.2 Thesis Themes ................................................................................................................................1 1.2 Literature Review .................................................................................................................................4 1.2.1 Genomics and Future of Health Care .............................................................................................4 1.2.1.1 Personalized Medicine, the Ultimate Destination of Genomics .................................................4 1.2.1.2 Technology, the Deriving Engine for Genomics ..........................................................................5 1.2.1.3 Patients and Their Hope for Early Arrival of Personalized Medicine ..........................................8 1.2.1.4 The Public’s Expectations and Concerns about Personalized Medicine .....................................9 1.2.1.5 Physicians and Their Role in the Provision of Genomic Guided Care .......................................12 1.2.1.6 Decision Makers and the Challenges of Assessing Genomic Technologies ..............................14 1.2.2 Methods for Evaluation of Genomic Technologies ..................................................................16 1.2.2.1 New Technologies, New Evaluation Tools ................................................................................16 1.2.2.2 Preference Elicitation Techniques ............................................................................................21 1.2.2.3 Modeling and Simulation ..........................................................................................................31 1.3 Overview of Thesis Chapters ................................................................................................................39 Chapter 2: Genetic Testing to Determine Drug Response: A Discrete Choice Experiment (DCE)................ 40 2.1 Background ..........................................................................................................................................40 2.2 Methods ...............................................................................................................................................42 2.2.1 Questionnaire Design...................................................................................................................42 2.2.2 Procedure.....................................................................................................................................44 2.2.3 Study Sample ...............................................................................................................................46 2.2.4 Statistical Analysis ........................................................................................................................49 2.3 Results..................................................................................................................................................49 2.3.1 Sample Characteristics .................................................................................................................49 2.3.2 Estimation Results........................................................................................................................50  vi  2.3.3 Latent Class Analyses ...................................................................................................................53 2.4 Discussion ............................................................................................................................................56 Chapter 3: Barriers to Integrating Personalized Medicine into Clinical Practice: A Best-Worst Scaling (BWS) Experiment. ............................................................................................................................................ 59 3.1 Background ..........................................................................................................................................59 3.2 Methods ...............................................................................................................................................61 3.2.1 Study Design ................................................................................................................................61 3.2.2 Questionnaire Design...................................................................................................................62 3.2.3 Study Sample ...............................................................................................................................65 3.2.4 Statistical Analysis ........................................................................................................................65 3.3 Results..................................................................................................................................................66 3.3.1 Sample Characteristics .................................................................................................................66 3.3.2 Model Estimation .........................................................................................................................68 3.3.3 Latent Class Analysis ....................................................................................................................70 3.4 Discussion ............................................................................................................................................72 Chapter 4: Future Impact of Genomic and Proteomic Tests on the Burden of COPD: A System Dynamics Model. .................................................................................................................................................... 76 4.1 Background ..........................................................................................................................................76 4.2 Methods ...............................................................................................................................................77 4.2.1 Structure of the Model ................................................................................................................77 4.2.2 Epidemiology ...............................................................................................................................79 4.2.3 Mortality ......................................................................................................................................82 4.2.3 Quality of Life ...............................................................................................................................82 4.2.4 Costs.............................................................................................................................................83 4.2.5 Model Assumptions .....................................................................................................................84 4.2.6 Modeling of Interventions ...........................................................................................................84 4.2.7 Model Validation and Sensitivity Analysis ...................................................................................86 4.3 Results..................................................................................................................................................86 4.3.1 Base Case Projections ..................................................................................................................86 4.3.2 Cost-Effectiveness of Interventions .............................................................................................86 4.4 Discussion ............................................................................................................................................98 Chapter 5: Cost-Effectiveness of Using A Molecular Diagnostic Test to Improve Pre-Operative Diagnosis of Thyroid Cancer: A Discrete Event Simulation (DES). ............................................................................... 101 5.1 Background ........................................................................................................................................101 5.2 Methods .............................................................................................................................................102 5.2.1 Model Deign...............................................................................................................................102 5.2.2 Data Sources and Assumptions..................................................................................................103 5.2.3 Quality of Life .............................................................................................................................107 5.2.4 Costs...........................................................................................................................................108 5.2.5 Model Outcomes .......................................................................................................................109 5.2.6 Univariate Sensitivity Analysis ...................................................................................................110  vii  5.2.7 Probabilistic Sensitivity Analysis (PSA) .......................................................................................110 5.3 Results................................................................................................................................................111 5.3.1 Results of Base Case Analysis .....................................................................................................111 5.3.2 Results of the Sensitivity Analyses .............................................................................................117 5.4 Discussion ..........................................................................................................................................120 Chapter 6: Integrated Discussions ......................................................................................................... 124 6.1 Important Findings and Implications .................................................................................................124 6.2 Limitations of My Research ...............................................................................................................130 6.3 Future Direction of Research .............................................................................................................132 6.4 Knowledge Transfer ...........................................................................................................................133 References ............................................................................................................................................ 135 Appendices ........................................................................................................................................... 147 Appendix A: Supplementary Material for Chapter 4 .................................................................................147 A.1 Interpolation of Rates in Subgroups .......................................................................................147 A.2 Background Mortality Rates ...................................................................................................150 A.3 Calculation of Direct and Indirect Costs .................................................................................150 A.4 Progression Rates Between Different Stages of COPD ...........................................................151 A.5 Model Structure in Vensim.....................................................................................................153 Appendix B: Supplementary Material for Chapter 5..................................................................................192  viii  List of Tables Table 1.1 Examples of attributes and levels for a DCE ...................................................................27 Table 1.2 Sample DCE questionnaire .............................................................................................27 Table 1.3 A sample BWS choice task ..............................................................................................28 Table 1.4 Comparison of various modeling techniques .................................................................33 Table 2.1 Attribute and levels included in the DCE questionnaire .................................................43 Table 2.2 A Sample choice task ......................................................................................................44 Table 2.3 Scenarios for DCE............................................................................................................45 Table 2.4 Characteristics of participants ........................................................................................47 Table 2.5 Estimated utilities and Marginal Willingness to Pay (MWTP) in 3 samples ....................51 Table 2.6 Estimated utilities, latent class analyses.........................................................................54 Table 3.2 A sample best worst choice task ....................................................................................64 Table 3.3 Characteristics of participants (N=197) ..........................................................................67 Table 3.4 Estimation results from latent class analysis – conditional logit model .........................69 Table 4.1 Mortality rates and prevalence of COPD by age, smoking status, and severity .............80 Table 4.2 Utilities, unit costs, and exacerbation rates by disease severity ....................................81 Table 5.1 Model parameters ........................................................................................................106 Table 5.2 Model parameters ........................................................................................................107 Table 5.3 Utility weights and unit costs .......................................................................................108 Table 5.4 Base case results, indeterminate cases ........................................................................113 Table 5.5 Results for all patients with palpable nodules ..............................................................114 Table A.1 Calculation of rates in subgroups .................................................................................148  ix  List of Figures Figure 1.1 The overall structure of my thesis ...................................................................................3 Figure 3.1 Utility weight estimates for attribute-levels .................................................................71 Figure 3.2 Average importance of attributes .................................................................................72 Figure 4.1 Model structure.............................................................................................................78 Figure 4.2 Total societal cost due to COPD ....................................................................................87 Figure 4.3 Total QALYs lost due to COPD .......................................................................................88 Figure 4.4 Population trend in Canada ...........................................................................................89 Figure 4.5 Number of smokers .......................................................................................................89 Figure 4.6 Number of previous smokers ........................................................................................90 Figure 4.7 Number of never smokers .............................................................................................90 Figure 4.8 Prevalence of COPD, all stages ......................................................................................91 Figure 4.9 Prevalence of mild COPD ...............................................................................................91 Figure 4.10 Prevalence of moderate COPD ....................................................................................92 Figure 4.11 Prevalence of severe COPD .........................................................................................92 Figure 4.12 Total deaths due to COPD ...........................................................................................93 Figure 4.13 Number of smokers with COPD ...................................................................................93 Figure 4.14 Number of previous smokers with COPD ....................................................................94 Figure 4.15 Number of never smokers with COPD .........................................................................94 Figure 4.16 Proportion of smokers among all COPD cases ............................................................95 Figure 4.17 Proportion of previous smokers among all COPD cases ..............................................95 Figure 4.18 Proportion of never smokers among all COPD cases ..................................................96 Figure 4.19 Total societal cost due to COPD, including testing costs .............................................96 Figure 4.20 Total societal cost due to COPD, discounted ...............................................................97 Figure 4.21 Total QALYs lost due to COPD, discounted..................................................................97 Figure 5.1 Model structure...........................................................................................................104  x  Figure 5.2 Effect of sensitivity and specificity of DX test on the outcomes..................................116 Figure 5.3 Net monetary benefit of DX test for different test costs ............................................117 Figure 5.4 Results of univariate sensitivity analysis .....................................................................119 Figure 5.5 Results of probabilistic sensitivity analysis ..................................................................120 Figure A.1 Overall structure of the model ....................................................................................153 Figure A.2 Detailed structure of the Vensim model .....................................................................154 Figure B.1 Overall structure of discrete event simulation model in Arena ..................................192 Figure B.2 Detailed structure of discrete event simulation model in Arena ................................193  xi  Glossary Abbreviation  Definition  BIBD  Balanced Incomplete Block Design  BWS  Best- Worst Scaling (choice experiment)  DCE  Discrete Choice Experiment  DES  Discrete Event Simulation  DTC  Direct to Consumer  ELSI  Ethical, Legal, and Social Issues (related to genomics)  ENCODE  Encyclopedia of DNA Elements  EVPI  Expected Value of Perfect Information  EVPPI  Expected Value of Partially Perfect Information  GE3Ls  Genomic Ethical, Environmental, Economic, Legal, and Social (issues related to genomics)  GEI  Gene Environment Interaction  GLM  Generalized Linear Models  GWAS  Genome Wide Association Studies  ICER  Incremental Cost Effectiveness Ratio  LCA  Latent Class Analysis  MRS  Marginal Rates of Substitution  MWTP  Marginal Willingness to Pay  NPV  Negative Predictive Value  PPV  Positive Predictive Value EuroQol-5D  QALYs  Quality Adjusted Life Years  SNP  Single Nucleotide Polymorphism  VOI  Value of information  xii  Acknowledgements I am grateful to my supervisor, Dr. Carlo Marra, for his excellent supervision and guidance throughout the Ph.D. His style of mentorship is based on encouragement and the provision of everything that one needs to gain the capacity for independent research. I am truly thankful for all of the support that he gave me. Dr. Larry Lynd also has been a great mentor to me since I started my work at CORE. I have always enjoyed working with him and I am grateful for his support over the years. I thank Drs. Aslam Anis, Stirling Bryan, and Marco Marra who guided me throughout the completion of my thesis, and offered their valuable time whenever I needed their advice. I also would like to thank Mohsen Sadatsafavi, my classmate and colleague. Talking with him is always inspiring and any simple conversation with him turns into an exciting learning experience. I thank the Canadian Institutes for Health Research (CIHR) for providing financial support during my PhD program through a Frederick Banting and Charles Best Canada Graduate Scholarship and UBC for their support with a Four-Year Fellowship award. The research in chapter 2 was funded by Genome Canada/Genome BC and I thank all lymphoma patients in British Columbia who provided their opinions in the experiment and voluntarily accepted to be part of the research. I also thank British Columbia Clinical Genomic Network (BCCGN) for supporting the research in chapter 3.  xiii  Dedication This thesis is a product of all the education that I have received over many years. I dedicate this work to my parents, Davood Najafzadeh and Nadereh Safarloo, who taught me the value of patience, hard work, and caring for others. I also dedicate this thesis to Naghmeh, my beloved wife. Without her support, completion of this work would not have been possible.  xiv  Chapter 1: Introduction 1.1 Overview of Thesis Objectives and Themes 1.1.1 Thesis Objectives In this thesis, I have applied several evaluation methods to examine genomic technologies. Using four case studies, I have highlighted the advantages of each method given the nature and scope of the research question. I utilized the following methods which have been developed in other areas and have only occasionally been applied to answer health-related questions: a discrete choice experiment, a best-worst scaling experiment, a system dynamics simulation, and a discrete event simulation. Through these case studies, I have demonstrated the particular advantages of these methods in the assessment of genomic technologies.  1.1.2 Thesis Themes Throughout my thesis, the case studies exemplify the questions that emerge in the process of integrating genomics into clinical care. Although my primary motivation was bridging the methodological gaps by incorporating several novel methods, the case studies were also selected to illustrate the practical issues regarding the integration of genomics into clinical care from the perspective of patients, the public, health care providers, and decision-makers. In this chapter (chapter 1), I have reviewed the literature about how advances in genomics will affect different stakeholders including patients, the public, physicians, and 1  health care policy makers. Secondly, I have described the theories and methods for preference elicitation and simulation modeling in the context of comparative effectiveness analysis. In chapter 2, I measured the preferences of cancer patients as well as the public for a hypothetical, genetically-guided treatment for cancer. Using a discrete choice experiment (DCE), I demonstrated how individuals make trade-offs between different attributes of a genetic test when deciding about cancer treatment options. I also showed how preferences of cancer patients differ from those of the public, and how the type and prognosis of cancer can affect the preferences for a genetically-guided treatment. In chapter 3, using a Best Worst Scaling (BWS) choice experiment, I estimated the relative importance of attributes which influence physicians’ decisions for using personalized medicine in their practice. Specifically, I determined their relative preferences for reimbursement, privacy, costs, education and integration into guidelines. In chapter 4, I developed a system dynamics model to evaluate the impact of three potential genomic/proteomic tests on the long term burden of COPD in Canada. Specifically, I examined the impact of biomarkers that could be used to predict the risk of developing COPD in smokers, to control progression of COPD, and to reduce the risk of COPD exacerbations on economic, quality of life, and mortality outcomes related to COPD.  2  In chapter 5, I developed a discrete event simulation (DES) to measure the costeffectiveness of adding a new molecular diagnostic test (DX) to the current diagnostic strategy for thyroid cancer. Finally, in chapter 6, I provide a conclusion that summarizes the benefits of using these methods in comparative effectiveness analyses of genomic technologies and I discuss the future directions of research in this area. Chapter 3: Barriers to Integrating Personalized Medicine into Clinical Practice: A Best-Worst Scaling Experiment Chapter 4: Future Impact of Genomic and Proteomic Tests on the Burden of COPD: A System Dynamics Model Chapter 5: CostEffectiveness of Using a Molecular Diagnostic Test to Improve Diagnosis of Thyroid Cancer: A Discrete Event Simulation Model  Health Care Providers  Decision Makers  Genomics  Patients  Chapter 2: Genetic Testing to Determine Drug Response: Measuring Preferences of Patients and the Public Using Discrete Choice Experiment (DCE)  The Public  Figure 1.1 The overall structure of my thesis This figure highlights the impact of genomics on heath care providers, patients, the public, and decision makers. Furthermore, this figure illustrates the case studies in each chapter of this thesis to demonstrate the potential research questions in relation with heath care providers, patients, the public, and decision maker and shows the methods that have been used for answering those questions.  3  1.2 Literature Review 1.2.1 Genomics and Future of Health Care 1.2.1.1 Personalized Medicine, the Ultimate Destination of Genomics The practice of medicine always has been personalized to the extent that relevant individualized information was available and integrated into care. Physicians usually consider factors such as familial history and environmental factors in their diagnosis and treatment of their patients. However, genomics has generated individualized information to the extent that has never been possible before resulting in a significant leap forward in personalized medicine. As new technologies are developed, genomic information will become part of individuals’ health records in the future. We are currently in the era that we will witness development of the necessary frameworks to ensure appropriate and beneficial integration of genomic information into healthcare [1,2]. Personalized medicine – i.e. the tailoring of interventions based on an individual’s genetic information- is one of the most important applications of genomics and represents a paradigm shift that will likely change routine clinical practice in the near future [3-7]. As a diagnostic tool, genomic/proteomic tests will potentially result in an increase in the accuracy of diagnoses or the ability to more accurately subtype diseases. Genomic information also has the potential to improve the efficacy and safety of treatments by helping us to identify therapeutics with a higher likelihood of being effective, or alternatively, to rule out the drugs with a higher probability of side effects for specific patients. In diseases like cancer, these advantages can result in a substantial reduction in 4  side effects and costs of treatment[4,8-10] and improve the therapeutic harm-benefit ratio by targeting individuals most likely to respond without side effects. Genomic information also can be used to predict the risk of developing common diseases and as such, can expand the paradigm of preventive medicine.[3,11] There are, however, several major obstacles for making personalized medicine an integral part of clinical practice. Translation of genomic data into useful clinical information is a complex endeavor. Our understanding about gene-gene and gene- environment interactions (GEI) will only be expanded by initiating well-designed experimental and observational studies. For example, in reality only a fraction of phenotypes can be explained by genetic factors, and environmental factors play a significant role in gene expression. Research on GEIs is a developing field that requires substantial work [12]. Ethical, legal, and social issues (ELSI) associated with the increasing access to genomic information are also causing significant concerns and need to be addressed in the coming years. Scheuner and colleagues have provided a useful review of the challenges for the delivery of personalized medicine and have indicated the knowledge gaps that need to be filled [13].  1.2.1.2 Technology, the Deriving Engine for Genomics Only 20 years ago, a typical project for finding a gene required years of investigation by a large team of scientists at a cost of millions of dollars. Nowadays, a similar project can be completed by a single graduate student in matter of days [7]. Owing to advances in genomic technologies, the cost of DNA sequencing has dropped 14,000 fold between 5  1999 and 2009 [7], and now a whole genome can be sequenced for less than $10,000 [14]. Using massively parallel sequencing technologies, scientists can sequence the human genome with unprecedented speed and accuracy [15]. Using these new technologies, the genetic data that is needed for clinical research can be generated by means of a few experiments with a minimal cost. The benefit of genomic data relies on its successful translation into clinical knowledge. Following the completion of the Human Genome Project in 2003 [16], several projects have improved our knowledge about genomics. The HapMap Project (2002-2005) had a significant role in improving our knowledge about common variants in human genome. The HapMap was a multi-country project that aimed at finding genetic similarities and differences in human beings by comparing genetic sequences of individuals at specific chromosomal regions (Single Nucleotide Polymorphisms (SNP)). The results of HapMap project provided information about the common genes that might impact diseases and response to medications. A project which started in 2003, the Encyclopedia of DNA Elements (ENCODE), provided a map of genes both in terms of their physical location on the DNA sequence and their functional role. Currently, genomes of 13 individuals have been sequenced in their entirety and are publicly accessible [7,17]. An international initiative to sequence the genomes of thousands individuals with European, Asian, and African ancestors- The 1000 Genomes Project- is also underway [18]. In the last decade, genome-wide association studies (GWAS) that are based on finding common variants have substantially increased our knowledge about the risk of common diseases. Furthermore, owing to advances in sequencing technologies, we are now able to 6  sequence protein coding areas on the genome (the “exome”) for a few thousand dollars that can play significant roles in elucidating mechanisms and treatments in uncommon diseases [7]. The impact of genomics on the pharmaceutical industry is also imminent. Allegedly, the failure of the blockbuster model for drug development is largely responsible for the exorbitant costs of drug development [19]. In addition, it is estimated that most drugs only work for less than half of patients they are prescribed for [20]. Pharmacogenetics, a science which improves the prediction of drug response and toxicity based on genomic variation could be an effective strategy to containing increasing costs of drug development and increasing the safety and effectiveness pharmacotherapeutics [19]. Pharmacogenetics aims at developing more cost-effective drug strategies by lowering drug research and development costs and limiting treatment failures due to inefficacy and toxicity. Vectibix (panitumumab), a drug for the treatment of colon cancer, is an illustrative example. Vectibix was rejected by European regulators as it was effective in only 10% of cases, but when patients with a KRAS mutation 1 were excluded, the response rate was much higher such that the drug was conditionally approved by the panel [19]. Similarly, the US Food and Drug Administration (FDA) changed the labeling for the anticoagulant warfarin to encourage physicians to order a genetic test prior to prescribing this agent. Also, several large clinical trials are studying the effect of certain genes on warfarin metabolism in order to facilitate appropriate initial dose selection to decrease  1  KRAS is a gene that encodes KRAS protein in human. Mutation in KRAS gene is responsible for several types of cancer.  7  the risk of hemorrhagic complications [4]. In fact, a number of large companies have become pioneers in pharmacogenomically guided, personalized medicine (e.g. Roche and Genzyme Genetics) and smaller companies (e.g. Celera, Genomic Health AviaraDX, Decode, DiaDexus, Exagen) have also been actively pursuing discoveries in pharmacogenomics [19]. Increasing capacity for genome sequencing and advances in bioinformatics have also facilitated the development of targeted or whole-genome sequencing methods in the search for potential treatment modalities based on genomic data. When the methods of data analysis can match our capacity for genomic data generation, we can expect significant improvements in diagnosis and treatments based on genomic information [21,22].  1.2.1.3 Patients and Their Hope for Early Arrival of Personalized Medicine Genomic information can potentially be used to select therapies with a higher probability of treatment success and a better safety profile for specific patients [23,24]. For example, genetic testing can determine if Selzentry (maraviroc) is effective for treatment for a specific HIV patient by identifying genetically which HIV strain the patient is infected with [23]. A systematic review [25] has shown that pharmacogenomics also has a significant role in reducing the number of adverse drug reactions. Advantages of genomic information are not confined to improved treatment. We also expect more accurate diagnostic tools that can detect the disease prior to onset of clinical symptoms by using genomic and proteomic markers and may enhance the role of prevention [26]. 8  Genomics has already started to alter cancer treatment. By comparing the genome of a patient’s tumor with his DNA, genetic alterations responsible for the development of cancer can be identified. This information can then in turn facilitate the development of targeted treatments that are more effective and cause less adverse effects. Our further understanding of this may lead to early detection or prediction of such alterations in individuals [27]. For example, the presence of a KRAS mutation in the tumor is a strong predictor of increased response of a patient with colorectal cancer to Cetuximab [23]. Whether having genomic information always leads to beneficial outcomes in patients is not completely clear. Access to genomic information may be useless or even can result in complex issues. For example, a patient’s knowledge about his higher than average risk of developing Alzheimer disease in the absence of effective interventions is associated with little benefit. Also consider the situation when a patient’s genome is examined for presence of autism and we also find out about her high risk for developing breast cancer. In this case, whether this incidental finding should be disclosed to patient and how it may affect other family members result in perplexing questions [28].  1.2.1.4 The Public’s Expectations and Concerns about Personalized Medicine The implications of genomics are not limited to the patients who are seeking better care. Healthy individuals have already started to use the results of genomic tests to try to improve their wellbeing. Currently, there are several commercial enterprises that, for a fee, will assess a person’s genetic profile to estimate their risk for developing more than 9  80 different diseases or to provide information such as ancestry background, nicotine dependence, or even athletic capacity [11,29-31]. The benefits and harms associated with access to direct to consumer (DTC) genetic testing is controversial [32,33]. Some believe that individuals should be allowed to learn about their genetic information, and in theory, this testing is not fundamentally different from measuring blood pressure or cholesterol levels. The argument is based on the very idea that people should not be denied access to information. However, the counter-argument is based around the quality of information, whether the appropriate use of this information should be regulated, and if the benefits for society need to be evaluated similar to any other health intervention [1,34,35]. As such, some experts have more skeptical viewpoints about widespread access to genomic information, particularly without the involvement of a health professional [32,36]. Current direct-to-consumer genetic tests have significant technical drawbacks. These genetic tests mainly rely on single–nucleotide polymorphisms (SNPs), which have significant limitations in accurately predicting risk and outcomes. Studies using SNPs only look at approximately 500,000 known variants per assay, which is less than 0.1% of the entire DNA sequence. Using SNPs as markers for disease risk have at least two serious drawbacks: the method does not search for uncommon gene variants that are responsible for many phenotypes, and it ignores copy number variation regions that are responsible for many phenotypic consequences that are especially important for diseases like cancer. Nevertheless, current limitations will be mostly overcome by replacing this methodology with complete sequencing of the genome [29]. Genomics will continue to shift the focus of medicine from an interventional to preventive paradigm. This shift will 10  likely be an influential strategy to improve health outcomes and control health expenditure at the same time [1]. Besides the technological issues, there are other concerns about increasing access to genomic information. Concerns about the privacy of genomic information including discrimination based on genetics (for example, in employment or health insurance), interpretation of genomic information, and the effect of genomic information on families (for example, genetic test results for one of family members may have implications for others) have raised a wide range of questions about the ethical, legal, social, and economic impacts of genomics[37]. The effect of genomic testing on behavioral and life style changes is an area that should be explored further. Research in this area must be concentrated on observational studies or clinical trials that can provide hard evidence on the utility of genetic testing in changing individuals’ harmful behaviors like smoking, and drinking while increasing beneficial behaviors like losing weight and exercising [34,38]. The literature documenting how information from genetic testing may affect individuals’ personal health related decisions either positively or negatively is still premature. Some studies have explored or speculated about the possible positive and negative effects of genomic risk information on decision making [38-44]. For example, Palmer and colleagues have shown that individuals who perceive themselves at higher risk for colon cancer more appropriately follow up with screening programs [39]. There are also a number of studies that have explored the psychological effects of receiving the results of genetic tests [45-49], how 11  people comprehend the overall concept of genetic risk [45,50-56], and the factors that influence participation in genetic carrier testing [57,58]. Also a few studies have looked at the effect of the results of genetic testing on families [40,41]. There is some evidence suggesting that individuals may adjust their economic and financial decisions after knowing their risk of developing diseases that has been determined based on their genetic information [59-66]. Also, there are ongoing debates about the privacy of genomic information and the potential for discrimination against individuals by employers or insurance companies based on their genetic risks [31,35,42,59,60,62,67-71]. Equity is an important issue that may be directly affected by genomics. The differential utilization of genetic testing in different populations has the potential to increase health disparities. Differences in insurance coverage, education, ethnicity, cultural beliefs, or socio economic status across populations can affect access to genomic testing. These challenges are important part of ongoing Ethical, Legal, and Social Issues (ELSI) related research [57,72] and as genomic testing becomes more available, the need for having solid evidence and methodologies to address these challenges is mounting.  1.2.1.5 Physicians and Their Role in the Provision of Genomic Guided Care We can imagine that in the not-too-distant future when patients visit their physicians, they may come with a request for genome sequencing or provide such data from other sources with the hope that this information will lead to improved, personalized treatments or help to predict disease risks and outcomes [4,6,34]. Many scientists believe 12  that as advances in technology make genomic testing more affordable, public demand for having genomic data with the hope for using that information in general clinical practice will increase. This wave will hit front-line physicians who may be faced with a plethora of patients’ expectations of integrating their genomic data into their clinical care [1,4]. Physicians’ and other health care professionals’ readiness to adapt their practices to incorporate this new information is a critical factor to materialize the promise of personalized medicine. There are ongoing discussions about the challenges that genomics will impose on physicians [73-77]. Furthermore, there are few studies that have investigated the role of primary care physicians in the development and application of personalized medicine to routine clinical care and the challenges that these new genetic tools may bring to this context [78-90]. Kumar et al. [86] interviewed 44 physicians and found that physicians were concerned about their lack of training, ethical dilemmas due to therapeutic gaps, and the need for support from geneticists. Carroll et al. [81] conducted four focus groups in Ontario to explore family physicians’ experience in dealing with genetic susceptibility to cancer. They concluded that developing educational tools and guidelines for genetic testing and improving genetic information communications with patients are crucial for physicians to integrate genetic testing into their practice. A study done on 112 physicians by Hindorff and colleagues [85] suggests that physicians who feel more confident about their interpretation of genetic test results more frequently used the factor V Leiden (FLV) genetic tests.  13  1.2.1.6 Decision Makers and the Challenges of Assessing Genomic Technologies Health care decision-makers must anticipate the increasing influence of genomics on heath care systems and take into account the expectations of patients, the public, health care providers, and industry in this regard. They also must plan to create the capacity to evaluate clinical and economic impacts of the application of genomics in health care. This knowledge will be necessary to integrate the utilization of genomic applications with positive impact on public health and address the concerns and issues that may arise [91]. Currently, decisions about the reimbursement of genomic tests are generally based on favorable evidence in terms of clinical utility of the tests without considering the associated costs [92]. Given the rapidly increasing number of genomic tests and the limited resources for health care, obviously this approach is not sustainable. Comparative effectiveness and, in particular, economic evaluation of these technologies must be considered as a part of reimbursement decision-making in order to optimize the value delivered to society by balancing between the clinical benefits and health care costs [2,92,93]. To achieve this goal, the methods for comparative effectiveness and evaluation of interventions should shift from population level approaches to consider “patient centered outcomes”(http://www.valuebasedcancer.com/article/cer-and-personalizedmedicine-face-hurdles). This change in evaluation methods can improve patients’ health outcomes and potentially lead to substantial savings simultaneously [92]. For example, expensive treatments for cancer that cause clinically significant side effects with little or no clinical benefits in some patients might be avoided. At the same time, new targeted 14  treatments with a higher chance of benefit in certain patients or subpopulations will have a greater potential to emerge. Without a doubt, health policy makers are responsible for establishing the legal and ethical framework to promote appropriate use of genomics. Relevant institutions or working groups have already been established in some jurisdictions in order to undertake this role. The Human Genetic Commission in the UK (http://www.hgc.gov.uk/) and Evaluation of Genomic Applications in Practice and Prevention (EGAPP) in the US are two examples. Initiatives such as Genomic Ethical, Environmental, Economic, Legal, and Social (GE3Ls) motivated by Genome Canada (http://www.genomecanada.ca/en/ge3ls) and the Ethical, Legal, and Social Issues (ELSI) project, part of Human Genome Project, in the United Sates (http://www.ornl.gov/sci/techresources/Human_Genome/elsi/elsi.shtml ) have been established to tackle the challenges in these areas. The design of policies to improve quality of genomic information, to prevent discrimination, and to preclude health disparities are of particular importance for the successful integration of genomics into health care systems. Health decision makers also have an important role in the development of novel genomic technologies through the appropriate allocation of public investment in biotechnology research. The ability to determine the projects with high potential value to society is crucial to achieving a satisfactory return on investment. Evaluation methods based on simulation and value of information analysis have significant roles for success in this area and can be used for prioritizing future investments [94].  15  1.2.2 Methods for Evaluation of Genomic Technologies 1.2.2.1 New Technologies, New Evaluation Tools Although the methods previously developed for health technology assessment can be applied to the evaluation of genomic technologies as well, methodological challenges in the evaluation of genomic applications entail utilizing more diverse and more sophisticated analytical tools [8,94-103]. In this section, I have briefly reviewed the new methodological challenges that are associated with the evaluation of genomic technologies. Significance of Preference Elicitation Methods for Assessment of Genomic Technologies In some cases, benefits or harms of genetic testing cannot be captured by the commonly used outcomes such as quality adjusted life years (QALYs). For example, imagine a diagnostic genetic test that can be used to improve a patient’s health outcomes but also reveals additional information about their family members which impacts both the individual’s and the family’s quality of life. In this case, conventional outcome measures such as the QALYs cannot measure the spill-over effect [102] of genomic information. Additionally, diagnostic or predictive tests of disease susceptibility can result in the labeling of individuals with potentially harmful effects for their entire life. Generally, this effect cannot be captured in a traditional cost-effectiveness analysis. Cutting edge methods for preference measurement can overcome some of these methodological limitations and therefore, are of particular interest in this context.  16  There is little data about the demand for genomic tests or genomically guided treatments. We have little information about the revealed preferences of patients, the public, or physicians since genomic applications are not widely available in the market at this point. Therefore, experiments that measure stated preferences as a surrogate for revealed preferences will yield useful information in this regard. Challenges of Modeling Techniques for Assessment of Genomic Technologies The size of the economic models for the evaluation of genomically guided treatments or diagnostic tests are often much larger compared to the models designed for the assessment of drugs. For example, when we model the effect of a genomic test to guide the treatment of cancer, the possible treatments and disease pathways are generally conditional on the genetic test results. Therefore, in many cases, gene characteristics, test characteristics, disease characteristics, and treatment characteristics should be included in the model [95]. However, a decision tree or Markov model that can capture all of these aspects can become very large and unwieldy. As such, we can use individual level simulation which can potentially provide more flexibility for incorporating those characteristics in the model. In cost effectiveness of preventive genetic tests, the model should be able to properly capture the long term effects at the population level, especially given that there is a large lag time between the time that the intervention (i.e. genetic test) takes place and the time that the outcomes will be materialized. Therefore, system dynamics modeling can be very helpful to analyze long term effects of interventions. 17  Modeling of an individualized treatment requires individual level simulation. The ability of conventional cohort methods (e.g. Markov models) for modeling individual level diagnosis, treatments, prognosis, and outcomes is limited. For example, the outcomes of a decision analytic model usually represent the average values over a cohort and patient level results are rarely available. Also, incorporating any interactions between individual level characteristics, disease characteristics, treatment properties, and the outcomes in cohort models is often not possible. There is significant uncertainty about the characteristics of genomic based interventions that need to be modeled. The majority of the genomic-based diagnostic tools or tailored treatments are currently in the development phase and have not been fully validated or implemented yet. Therefore, assessment of these tests often involves the evaluation of hypothetical scenarios based on the range of possible values about the various characteristics of an intervention such as efficacy, effectiveness, safety, cost, or diagnostic accuracy. As such, extensive sensitivity analysis, appropriate assessment of uncertainty, and value of information analysis are necessary parts of any robust evaluation in this area. Value of information analysis is particularly useful for determining the areas of research with a higher expected return on investment and can be useful for prioritizing future investment in the absence of sufficient data [94]. Molecular diagnostic tests involve additional complexities compared to usual diagnostic tests. A certain gene variant may affect other health conditions besides the disease being diagnosed or treated. Also the effect of a gene variant on health outcomes can depend on 18  individual characteristics, environmental factors, the presence of other diseases, or disease stage [94]. Any modeling effort using usual methods (such as Decision Trees and Markov Models) that aims to include those parameters might become prohibitively complex. Challenges in Design and Analysis of Observational Studies for Assessment of Genomic Technologies There are ongoing observational studies which are examining the use of genetic tests and their impact on health outcomes and health resource utilization. These observational studies can also be used to study gene-environment interactions. The expansion of electronic heath records and the increasing possibility for having larger data sources will create more opportunities for this type of research in the near future. However, analyses of this observational data create several methodological issues that require the application of advanced statistical methods. Joint causality, missing variables, and selection bias that are inherently part of observational data are a few examples in this regard that are widely present in studies that involve genetic information [104]. Although there is a general belief that clinical trials provide the highest quality of evidence, the trade-off between the benefits of using observation studies versus waiting for the results of clinical trials that usually take years to appear must be carefully considered [94]. In comparison with clinical trials for measuring efficacy of drugs or treatments, designing clinical trials for genetically guided treatments involves additional complexities. The traditional paradigm for the design of clinical trials suggests that inclusion criteria should  19  be set as broad as possible in order to increase the generalizability of the results. Furthermore, clinical trials generally are not designed to support the post-hoc subgroup analyses. In contrast, in the context of personalized medicine, we are particularly interested in finding subgroups of patients who may or may not respond to a certain genetically guided treatment and to find the genetic variations that can predict response. The clinical trials for measuring the efficacy of Iressa® (gefitinib) - a drug for lung cancerprovide clear examples of this. Iressa was initially shown to be no better than placebo in two large clinical trials. However, studies conducted later indicated that Iressa results in significant tumor regression in 10% of patients who have certain genetic mutations in their tumors, a finding that was masked in the larger clinical trials. This example highlights the need for appropriately designed clinical trials that facilitate appropriate subgroup analyses. There are also other aspects that should be considered in the design of clinical trials for genetic related studies. Unlike usual interventions, in the case of genetically guided treatments we are interested in prediction rather than inference. For example, we are interested to find the genes that can predict a positive response to treatments; whether contribution of some of those genes in this prediction is non-significant is not our primary concern. New approaches in designing clinical trials, such as adaptive designs and prediction based designs are currently being developed that can potentially provide individualized information needed for genetically guided treatments [103,105]. In the following sub-sections (1.2.2.2 and 1.2.2.3), I will explain several methods that can be used to overcome some of aforementioned methodological challenges. Preference  20  elicitation techniques and modeling and simulation methods will be discussed and their particular importance for evaluation of genomic technologies will be highlighted.  1.2.2.2 Preference Elicitation Techniques Theories of Utility The concept of utility was initially proposed to describe the level of pleasure or pain that human beings experience under different circumstances. This definition, however, was dismissed by positivist psychologists at the beginning of the twentieth century on the premise that pleasure or pain is a personal experience and cannot be measured [106]. Since then, utility theories have been based on choices or decisions that are made by individuals and therefore utility is defined as the relative preference or desire of an individual over a set of goods. These choices, generally known as revealed preferences, are directly observable and provide evidence of individuals’ rankings over a given set of goods 2. Although under mild conditions, an individual’s utility function can be defined to map ordinal preferences into cardinal utility values, economists emphasize an ordinal interpretation of utility factions [107,108]. For example, if 𝑈 is the utility function of an  individual and 𝑈(𝑥1 ) > 𝑈(𝑥2 ), we can only conclude that the individual prefers 𝑥1 to 𝑥2 , but the magnitude of difference 𝑈(𝑥1 ) − 𝑈(𝑥2 ) does not convey any information about  the intensity of the preference. The concept of ordinal utility based on reveled  preferences has been widely developed in economics and constitutes the foundation of modern consumer demand theories. At the aggregate societal level, the same notion of 2  “Good” refers to an abstract construct that is used in economic literature, therefore can refer to desirable or undesirable commodities, services, or outcomes.  21  utility has also been used by welfare economists who consider the maximization of the aggregate utility of individuals as the moral criteria that creates the most ‘happiness’ at the societal level. Expected Utility Theory Von Neumann and Morgenstern expanded the notion of utility to incorporate preferences of individuals over sets of lotteries. Von Neumann- Morgenstern utilities indicate preferences of individuals over sets of gambles instead of certain outcomes and therefore, are able to capture risk preferences of individuals and also can be interpreted as cardinal values. Von Neumann- Morgenstern utilities are the foundation of both game theory and expected utility theory and serve as the cornerstone of modern decision science. The assumptions of the Von Neumann- Morgenstern utility theorem (completeness, transitivity, continuity, and independence) are the necessary and sufficient conditions to have a theoretically rational individual and a Von NeumannMorgenstern utility function [107,108]. Expected utility theory particularly has been the central theory for health related decision making and measuring preference-based health related quality of life (e.g. standard gamble utilities) as it provides a normative framework for decision analysis and “rationality”. This theory can be formally described as the following (Kahnemann 1973): U ( x1 , p1 ,..., xn , pn ) = p1U ( x1 ) + ... + pnU ( xn )  (Eq. 1.1)  U ( w + x1 , p1 ,..., w + x n , p n ) > U ( w) ⇔ U ( x1 , p1 ,..., x n , p n ) > 0  (Eq. 1.2)  22  U ′′(.) > 0  (Eq. 1.3)  where 𝑥𝑖 is the outcome of event 𝑖, 𝑝𝑖 is the probability of event 𝑖 happening, and 𝑤 is individual’s asset at the beginning of the lottery. Equation 1.1 is called the expectation rule, which states that utility of a lottery is equivalent to expected value of its components. Equation 1.2, or asset integration principle, says that a rational person will participate in the lottery if, and only if, the utility of the lottery outcomes in combination with his current asset 𝑤 is higher than utility of his current asset. Finally, the last equation suggests concavity of the utility function, meaning that a rational individual is risk averse  and therefore prefers certain outcomes to equivalent risky options. Random Utility Theory Random utility theory provides a conceptual framework to describe how individuals make their choices based on the properties of these choices. The departure points of random utility theory from expected utility theory are the following assumptions: 1) each option (good) can be described by a finite number of properties or attributes; and 2) the probability of each option being selected is a function of its attributes rather than the option per se. Formally speaking, according to the general framework of random utility theory[109], the utility of individuals can be described by the following functional form: 𝑈𝑖𝑗 = 𝑉𝑖𝑗 + 𝜀𝑖𝑗  (Eq. 1.4)  where 𝑈𝑖𝑗 is the utility of individual 𝑖 as a result of choosing alternative 𝑗. 𝑈𝑖𝑗 , in turn,  consists of a systematic part (𝑉𝑖𝑗 ) and a random part (𝜀𝑖𝑗 ) where we assume that 𝜀𝑖𝑗 are 23  independently, identically distributed extreme value type I (Gumble distribution). Assuming the indirect utility function 𝑉𝑖𝑗 as a linear function of attributes, following the conditional logit model holds, i.e.: exp (Vij )  Pij = ∑L  (Eq. 1.5)  𝑉𝑖𝑗 = ∑𝐾 𝑘=1 𝛽𝑘 𝑥𝑖𝑗𝑘  (Eq. 1.6)  l=1 exp (Vik)  where 𝑃𝑖𝑗 is the probability of individual 𝑖 choosing alternative 𝑗, 𝑥𝑖𝑗𝑘 is the level of  attribute 𝑘 for individual 𝑖 in alternative 𝑗, and 𝛽𝑘 is the utility weight of attribute 𝑘. Random utility theory has been very successful in explaining and predicting individual choice behavior and has been widely applied to measuring consumers’ choice behavior. A discrete choice experiment (DCE) is a practical method that implements random utility theory in empirical research on individuals’ choice behavior and has been extensively used in mathematical psychology, economics, transportation, marketing, environmental sciences, and, to a lesser extent, health care[109]. Preference Elicitation Methods Using Choice Experiment Discrete Choice Experiment (DCE) A DCE is a practical method to collect stated choice data when individuals are faced with a finite number of exclusive options that can be described by a finite group of attributes. Gathering stated choice data is particularly useful where access to revealed preference data is not practical or possible such as in health care or the assessment of a new 24  technology. In the last decade, statistical methods for the analysis of choice data have been widely developed to estimate the underlying preferences of individuals by analysis of stated choice data. Although DCEs have been relatively frequently used in health care research, there are few studies that have utilized this method in the context of genetic testing [57,110-113] despite the fact that a DCE offers unique advantages over other methods [102]. As such, this method will likely become an important tool for evaluation of genomic technologies in coming years. There are variety of statistical methods for the analyses of choice data ranging from simple conditional logit models to Bayesian mixed logit models[114] and Latent Class Analysis (LCA)[115]. In the next section, I discuss the statistical specification of choice model, and then will briefly discuss the main statistical methods that can be used for the estimation of choice models[116]. According to random utility theory, given a set of choices and their attributes, the choice that an individual makes correlates with their preference weights for different attributes. Therefore, the statistical analysis of choice data estimates those preference weights by looking at the choice data. Considering equation 1.6, 𝑃𝑖𝑗 (the frequency that alternative 𝑗 has been selected) and 𝑥𝑖𝑗𝑘 (the level of attribute 𝑘 for individual 𝑖 in alternative) are both directly observable given the choice data. Assuming a linear functional form for 𝑉𝑖𝑗 (Eq. 1.7), we can estimate 𝛽𝑘 using methods such as the conditional logit model.  Furthermore, estimation of an indirect utility function ( 𝑉𝑖𝑗 ) allows the estimation of the  25  Marginal Rates of Substitution (MRS) that theoretically represents how individuals tradeoff between the changes in attributes. For example, the MRS of attribute 𝑘2 for 𝑘1 is defined as:  𝑀𝑅𝑆𝑥𝑘1,𝑥𝑘2 =  𝜕𝑉 ⁄𝜕𝑥𝑘1 𝜕𝑉⁄𝜕𝑥𝑘2  (Eq. 1.7)  By inclusion of cost as an attribute in a DCE, we can calculate the Marginal Willingness to Pay (MWTP)[109] which can provide useful interpretations of the estimated utility weights. In essence, the MWTP is a special case of MRS where cost is selected as the reference attribute. Thus, the MWTP for any attribute 𝑥𝑘 can be calculates as the following:  𝑀𝑊𝑇𝑃𝑥𝑘 =  𝜕𝑉 ⁄𝜕𝑥𝑘 𝜕𝑉⁄𝜕𝑐  (Eq. 1.8)  where 𝑐 represents cost. MWTP theoretically indicates how much individuals, on average,  are willing to pay to receive a certain amount of change in attribute 𝑘. Best- Worst Scaling (BWS) Choice Experiment  A traditional DCE questionnaire generally consists of 8 to 16 choice tasks where, in each choice task, respondents have to choose between a few alternatives. Each alternative in a given choice task consists of several attributes, and each attribute in turn assumes a level within the possible range. As a simplified example, suppose a genetic test can be described by three attributes including sensitivity, specificity, and the cost of the genetic test. Furthermore, we assume each of the three attributes can take 3 possible levels 26  (Table 1.1). A sample choice task using traditional DCE method will be similar to Table 1.2, where a respondent is asked to choose either of alternatives given their attribute levels. Table 1.1 Examples of attributes and levels for a DCE Attribute Sensitivity Specificity Cost of genetic test  Levels 80%, 90%, %100 80%, 90%, %100 $2000, $1500, $1000  Table 1.2 Sample DCE questionnaire Attributes Sensitivity Specificity Cost of genetic test  Option A 100% 80%  Option B 80% 90%  $1000  $1500  √  The choice data then can be coded using effect coding or dummy coding (Betch_2005) for model estimation. The specification of the regression model will be the following model: 𝑉𝑖𝑗 = 𝛽0 + 𝛽11 𝑆𝑒𝑛𝑠𝑖𝑡𝑖𝑣𝑖𝑡𝑦100 + 𝛽12 𝑆𝑒𝑛𝑠𝑖𝑡𝑖𝑣𝑖𝑡𝑦90 + 𝛽21 𝑆𝑝𝑒𝑐𝑖𝑓𝑖𝑐𝑡𝑦100 +  𝛽22 𝑆𝑝𝑒𝑐𝑖𝑓𝑖𝑐𝑡𝑦90 + 𝛽31 𝐶𝑜𝑠𝑡1000 + 𝛽32 𝐶𝑜𝑠𝑡1500  (Eq. 1.9)  Regardless of the method that we use for coding the data (effect code or dummy code) [117] or the estimation method applied, all estimations will be relative to a reference scenario [118]. For instance, 80% sensitivity, 80% specificity, and $2000 cost comprise our reference scenario in this example and therefore the estimated 𝛽 coefficients indicate relative changes in overall utility compared to this reference scenario. In a traditional  27  DCE, even if the data is effect coded, the model intercept (𝛽0 ) includes utility of the reference scenario given that 𝛽13 , 𝛽23 , and 𝛽33 cannot be directly estimated in the  regression model. For example, 𝛽11 indicates the additional utility gain as a result of  increasing the sensitivity of the test from 80% to 100%, assuming that specificity and cost remain unchanged. With these limitations in mind, Best-Worst Scaling (BWS) is a novel method for conducting a DCE that offers several advantages over traditional DCE methods [109,118]. The format of the choice tasks in a BWS experiment allows for more efficient collection of individuals’ preference information. In a BWS experiment, each respondent chooses the most preferred and the least preferred items among a list of three or more items presented in a given choice task. This process is repeated with the next tasks containing a different set of items [118,119]. A sample BWS choice task has been shown in Table 1.3.  Table 1.3 A sample BWS choice task Best  √  Attribute-level  Worst  Sensitivity 100% Specificity 80% Cost of genetic test $1000  √  Similar to traditional DCE, the choice data resulting from a BWS choice experiment can be either effect coded or dummy coded, and the regression model for this example will have the following form: 28  𝑉𝑖𝑗 =  𝛽0 + 𝛽11 𝑆𝑒𝑛𝑠𝑖𝑡𝑖𝑣𝑖𝑡𝑦100 + 𝛽12 𝑆𝑒𝑛𝑠𝑖𝑡𝑖𝑣𝑖𝑡𝑦90 + 𝛽13 𝑆𝑒𝑛𝑠𝑖𝑡𝑖𝑣𝑖𝑡𝑦80 +  𝛽21 𝑆𝑝𝑒𝑐𝑖𝑓𝑖𝑐𝑡𝑦100 + 𝛽22 𝑆𝑝𝑒𝑐𝑖𝑓𝑖𝑐𝑡𝑦90 + 𝛽23 𝑆𝑝𝑒𝑐𝑖𝑓𝑖𝑐𝑡𝑦80 + 𝛽31 𝐶𝑜𝑠𝑡1000 +  𝛽32 𝐶𝑜𝑠𝑡1500  (Eq. 1.10)  The key advantage of BWS over traditional DCE is its ability to estimate coefficients of all attribute-levels relative to only one reference attribute- level. In our example, this means that we will be able to estimate 𝛽13 and 𝛽23 , and only 𝛽33 will be missing as a test costs of $2000 has been arbitrarily chosen as the reference attribute-level. This is due to the fact  that in a BWS choice task, comparisons across attributes are also possible, and therefore, additional preference information provides more degrees of freedom in the choice data. Given that all coefficients are estimated on the same scale, unlike traditional DCE results, the estimated utility weights across attributes are directly comparable. This property of BWS allows for estimation of relative importance of attributes in a common scale, which is not possible using a traditional DCE. Questionnaire for a DCE or BWS experiment that ensures unbiased and efficient preference estimation requires the use of a rigorous methodology. In most cases, the total number of distinct alternatives that can be constructed using the attributes and their possible levels in a choice experiment is very large resulting in a prohibitive number of possible choice tasks that can be generated. Therefore, an appropriate fractional factorial design is an essential part of questionnaire design. For example, a combination of 5 attributes each with 3 possible levels results in 35 (243) possible scenarios and 29  assuming 2 alternatives in each choice task can generate 29,403 (= 𝐶2243 ) potential choice tasks. However, a DCE or BWS questionnaire is usually limited to 8 to 16 choice tasks to  be amenable to completion. In DCE, the efficiency of the fractional factorial design can be tested by simulating the choices. The selection of an optimum design is done by generating a large number of possible designs and then selecting the design that provide the most precise coefficient estimates (i.e. smallest standard errors) and the best Defficiency for a given the sample size [109,120]. Alternatively, block designs such as balanced incomplete block designs (BIBD) or orthogonal main effect plans (OMEP) can be used to ensure an orthogonal fractional design[118]. Statistical Methods for Analysis of Choice Data Dubin and McFadden (1984) initially developed the statistical methods for analysis of choice data. Since then, substantial methodological progress has been made in this area and currently, several analytical methods are available[121]. These methods can be divided in to two large primary categories: 1) Generalized Linear Models (GLM), including the conditional logit model, the generalized extreme value model, and the mixed multinomial logit model [121](http://elsa.berkeley.edu/~train/software.html ); and 2) latent class models [122]. Mixed multinomial logit models, and particularly latent class models, provide more flexibility to incorporate various correlation structures that are usually present in the choice data. Therefore they are appropriate methods for analysis of choice data to capture potential heterogeneity of preferences.  30  1.2.2.3 Modeling and Simulation The essential role of mathematical models in health care is to aid the decision making process by combining evidence from clinical trials, observational studies, or other sources [123]. Modeling is a useful approach to extrapolate outcomes of interest, for example, over time, across populations, from one policy setting to a different setting, or from a clinical trial to real world settings [123]. A good modeling effort has to conform to generally accepted criteria such as, internal validity, face validity, convergent validity, and often predictive validity [123]. All models are built using certain assumptions and boundaries. Although extending boundaries of a model will generally result in a more realistic model and perhaps better accuracy, this comes at the price of additional complexity. Indeed, finding the right balance between accuracy and simplicity is always a challenge. Conditional on validity and sufficient accuracy, the simplicity of the model is a great advantage as it facilitates communication of assumptions and results to decision makers [124]. Maximizing simplicity also is helpful in choosing the modeling method. Some methods, such as decision tree analytic models, are widely accepted and used in health care analysis. The pitfall, however, is that using the same tools for every question can result in complex models. Brennan and colleagues [125] provide a taxonomy of the techniques that can be used for modeling in health care. My argument, in line with Brennan et al. [125], is that by extending the variety of methods in our modeling toolbox, we will be able to use the best tool for each specific problem. New methods can be used to construct  31  models with higher validity and accuracy which are often simpler for handling certain questions. These advantages can be illustrated when modeling the impacts of genomic technologies. The following sections describe several major types of modeling methods and advantages that each can offer. Finally, I recommend two methods for the modeling of genomic technologies based on their advantages and disadvantages Decision Tree Analysis and Markov Models Decision tree models are the most widely used models in health care. In this technique, mutually exclusive possible outcomes are mapped out using a decision tree. By assigning probabilities to each branch and defining the payoffs at the end of each branch, expected values can be analytically calculated, usually by starting from the terminal branches on the tree and rolling back to the initial node. Simulated decision tree analysis can incorporate uncertainty by assigning probability distributions to each of the parameters in the model, and then performing Monte Carlo simulations to propagate uncertainties of parameters to outcomes. Decision tree models are generally easy to implement and only require a moderate computational power. However, given that recursion is not possible, if the question at hand cannot be described in the format of decision trees, the size of decision tree can grow quickly and become unwieldy and impractical.  32  Table 1.4 Comparison of various modeling techniques 3 Individual level  Stochastic (vs.  Continuous  Time can  Interaction  Event driven  (vs. cohort  deterministic)  state (vs.  be  can be  (vs. time  discrete state)  modeled  modeled  driven)  level)  Decision tree rollback  √  Simulated decision tree  √  √  Deterministic Markov models  √  Simulated Markov models  √  √  System dynamics  √  √  √  Discrete event simulation  √  √  √  √  Agent based modeling  √  √  √  √  √  √  √  √  Recursion, however, can be modeled in Markov models. This possibility can significantly simplify modeling of pathways that require “looping” 4. Markov models can be constructed by defining discrete and mutually exclusive states, transition probabilities between those states, and defining fixed cycles. Cycles can represent fixed time intervals in the model and therefore time dependent variables essentially can be defined in  3 4  This table is constructed based on Borshchev & Filippov (2004), Caro (2005), and Brennan et al (2006). In programming, looping refers to performing recursive series of instructions to satisfy a condition.  33  Markov models. Monte Carlo simulation also can be used to incorporate the uncertainty of parameters and to facilitate the development of a stochastic model (Simulated Markov model). Markov models are also very popular in health care and are usually used in conjunction with decision tree models to create further flexibility in modeling. However, modeling of interactions is not feasible in Markov models [125-127]. System Dynamics System dynamics models are cohort simulation models with the extensive capability for modeling interactions. Consider an example where the prevalence of chicken pox in children can have a protective effect on the incidence of herpes zoster in the elderly. Modeling of this interaction can be easily done using a system dynamics model. Modeling of population trends as a function of time is another example where system dynamics model can be very useful. Building a system dynamics model usually includes the following steps: 1) determining causal structure of the model by defining all model variables (e.g. health states, probabilities, costs, etc); 2) describing each variable either as a stock variable (e.g. prevalence), flow variable (e.g. incidence), or auxiliary variable; 3) defining mathematical equations that describe existing causal relationships between different variables in the model; 4) assigning initial values to all stock variables; and 5) running the simulation and calibrating the parameters. In addition to providing the capability to model interactions, system dynamics models are particularly useful for modeling causal structures and continuous time-dependent variables. These models also are very computationally efficient, as a system dynamics model can be described by a set  34  of differential equations that can be efficiently solved using numerical approximation methods [126]. Discrete Event Simulation (DES) For some important questions in health care, time needs to be explicitly modeled in a continuous fashion. For example, modeling the waiting times in an emergency department or for surgery in a hospital requires individual-level, continuous time, stochastic models. The arrival time of a new patient, for instance, is randomly distributed over continuous time. Such questions can be modeled using discrete event simulation instead of Markov models that are unable to incorporate continuous time and individual level simulation [127]. Also, common resources such as the capacity of surgery rooms or the number of physicians in an emergency department put a constraint on the progression of individuals in different pathways in the model. Discrete event simulation is an individual level, continuous time, event driven, stochastic method that offers considerable flexibility for the modeling of diseases and interventions [126,127]. The advantages of DES are not limited to its capacity for modeling continuous time and queuing problems. In contrast to Markov models, there is no need to define states in a mutually exclusive way in DES. For instance, modeling a patient’s age and stage of cancer using a Markov model requires the definition of mutually exclusive discrete health states that can be constructed by all possible combinations of those variables. However, those attributes can be simply defined as two variables that an individual carries throughout a DES. Furthermore, any possible association between age and stage of cancer can be 35  incorporated with great flexibility. The capability of DES for individual level simulation provides a natural way to simulate actual pathways of individuals in hypothetical clinical trials or cohort studies, and essentially it can mimic the sequence of real events and pathways for a particular patient based on specific, assigned parameters [127]. Construction of a DES usually involves two stages: 1) generating hypothetical individuals and assigning characteristics (e.g. age, sex, disease stage) to each of them at the beginning of the simulation; 2) cloning those individuals into different arms of the model to ensure comparability of the results that is conceptually similar to clinical trials where we try to minimize confounding by randomization of individuals into different arms, 3) defining pathways, probabilities, or conditions that define the triaging of patients into different pathways, where these probabilities or conditions can be conditional on characteristics of each patient, the history of the individual in the model, time, or the value of other parameters in the model, and 4) running the simulation and model calibration. Overall, DES provides the ability to explicitly model time as a continuous variable, avoids the proliferation of health states, facilitates modeling of an individuals’ history in the model, and offers individual level outcomes that can be used to derive the nonparametric distribution of outcomes. Software packages such as ArenaTM provide great visual interfaces and real time simulations that can largely facilitate the debugging of the model. DES increases computation burden relative to cohort models, but this  36  disadvantage generally is negligible in the small and medium size models that are often required in the evaluation of health care interventions [128,129]. Stochastic Analysis and Value of Information Regardless of the method that is used for modeling, incorporating the uncertainty of parameters and quantifying their impact on the outcomes is an essential components of any economic evaluation 5 [130-133]. In addition to determining the mean effectiveness and mean costs that are traditionally reported as the main outcomes of cost-effectiveness studies for the different strategies, first-order uncertainty (random variation of the outcomes for different individuals) and second-order uncertainty (representing the uncertainty of model parameters) convey important information about model outcomes and need to be accounted for [123,132,134]. In the last decade, concepts such as credible intervals, cost effectiveness planes, and cost effectiveness acceptability curves have been widely utilized for the communication of the uncertainty in the costs and outcomes and explaining the implications for decision-making processes. Value of information (VOI) analyses, including the expected value of perfect information (EVPI) and the expected value of partially perfect information (EVPPI), are also valuable tools for evaluating the potential effect of uncertainty on health decisions [135,136]. Value of Information indicates the value of improved outcomes if we had additional information about uncertain parameters in a model. Despite the availability of several shortcut methods for the estimation of VOI [137], VOI has received little attention in  5  This also applies to any modeling effort for health technology assessment.  37  empirical cost-effectiveness studies or more so, in the decision-making process. Practical complexities involved in conducting VOI analysis and the substantial computational burden perhaps can be blamed for the infrequent application of VOI in cost-effectiveness studies. However, a number of recent studies have emphasized the benefits of VOI analysis in the context of genomics [94,138,139]. For example, VOI can be used to measure the value of identifying responders to a treatment in a given population. This value provides a threshold that represents the amount of public resources that can be justifiably invested in developing a new genetic test to identify responders to treatment. Alternatively, potential projects can be prioritized by comparing their respective estimated VOI during the development phase [2,94]. In the development of novel genomic tests, VOI can be used to select the candidate tests with the greatest potential for final validation, resulting in significant cost-savings in research and development. In most cases, numerous biomarker candidates survive early validation phases and enter into final validation procedures. Final validation efforts usually aim at finding the best candidates for commercialization by gathering further evidence on the proposed candidates’ performance (such as sensitivity, specificity, positive predictive value and negative predictive value). Considering that validation studies are expensive, the candidate biomarkers compete for the limited funding to perform large scale validation studies. VOI can help by ranking the biomarker candidates based on their performance and potential impact. In fact, VOI can replace subjective decisions that are generally made based on the perceived chance of success of a biomarker candidate in the later validation stages. For example, at the end of the early 38  validation phase, VOI may reveal that the specificity of the test has the highest influence on the overall success of the biomarker in the market and therefore, narrows the target for further validation studies.  1.3 Overview of Thesis Chapters Each of the following chapters of my thesis presents a practical example of the impact of genomics on patients, the public, health care providers, and decision-makers. As the nature of each research question is different, I have used four different methods for analysis. Each chapter introduces the implementation of a different method for answering a question about the health services evaluation of various genomic technologies. In chapter 2, using a discrete choice experiment, I measured and compared preferences of cancer patients as well as the public with respect to a hypothetical, genetically-guided treatment for cancer. In chapter 3, using a Best Worst Scaling (BWS) choice experiment, I estimated the relative importance of attributes which influence physicians’ decisions for integrating personalized medicine into their practices. Chapter 4 demonstrates using a system dynamics model that was used to evaluate the impact of three potential genomic/proteomic tests on the long term burden of COPD. Finally in chapter 5, I have shown an application of a discrete event simulation (DES) to measure the costeffectiveness of adding a new molecular diagnostic test (DX) to the current method of diagnosis of thyroid cancer. I have discussed the overall conclusions in the final chapter of my thesis (chapter 6). 39  Chapter 2: Genetic Testing to Determine Drug Response: A Discrete Choice Experiment (DCE) 2.1 Background Treatment options for cancer are mainly chosen based on the classification of the tumor and are usually based on the best knowledge of histogenesis, histological type, and stage of disease [140]. However, these criteria often fail to accurately differentiate among distinct subtypes of tumors, especially with respect to likelihood of response to treatment, forcing clinicians and patients to choose empirically. Thus, many patients end up experiencing significant side effects of chemotherapy without receiving clinical benefit [141]. Recent advances in genomics have created hope that genetic testing may help to identify patients who will likely respond to a particular drug and/or experience side effects. This information is valuable both for patients and physicians when choosing among possible treatment options and trading off between risks and benefits. Although such genetically guided treatments are not widely available at this time, swift developments in genomic sciences and related technologies are beginning to transform the field. For example, panitumumab (Vectibix®), a drug for the treatment of colon cancer, was initially shown to be effective only in 10% of cases. However, genetic testing revealed that response rates were much higher in those without a KRAS mutation[19]. Other examples are HER2 expression in breast cancer patients, which predicts response to trastuzumab  40  (Herceptin®)[142] and the BCR-ABL genotype in chronic myeloid leukemia, which predicts response to imatinib mesylate (Gleevec®)[143]. Despite some clear advantages for the use of genetic tests to predict response to therapy, there are also some limitations. Firstly, genetic tests often have a probabilistic relationship with drug response – people who exhibit a certain genotype have a probability of response that is higher than those without the genotype but it is rarely absolute[141]. As such, prediction error in genetic testing may lead to the misclassification of those that will respond. Secondly, the extent that an imperfect genetic test will be used in practice is affected by multiple factors. Patients and physicians consider various factors such as sensitivity and specificity of the test, invasiveness of the testing procedure, probability and severity of associated side effects of the testing procedure or the drug, and the overall costs before accepting the usefulness of a genetic test. Using a discrete choice experiment (DCE), we demonstrated how individuals make tradeoffs between different attributes of a hypothetical genetic test when deciding about their treatment options for cancer. In this study, we used a novel design to demonstrate how preferences of patients compare with those of the public for genetic testing and additionally, how individuals think differently when they are faced with an aggressive but curable cancer versus a non-aggressive and incurable cancer.  41  2.2 Methods 2.2.1 Questionnaire Design The theoretical foundations of DCE are based on Random Utility Theory, a wellestablished theory in economics and psychology [109] . DCE has been widely applied in mathematical psychology and economics to measure stated preferences, particularly in situations where access to revealed preferences is not feasible[109,144]. The DCE questionnaire consisted of 16 choice tasks, where in each choice task responders had to choose between two alternatives or a neither option. Each alternative in a given choice task consisted of 7 attributes, and each attribute in turn assumed a level within the possible range (Table 2.1). Choice of attributes and levels were primarily based on the opinion from experts who were in direct contact with cancer patients and a comprehensive search of the literature. A sample choice task consisting of three alternatives (including the neither option) has been shown in Table 2.2. Inclusion of the “neither” option in the choice tasks provided the possibility to opt-out whenever none of the presented alternatives was adequately attractive to the respondent. Thus, we avoided forcing non-demanders to choose an alternative and ensured estimation of unconditional rather than conditional preferences[109]. Potentially, a total of 8640 (= 43*33*51) distinct alternatives could be constructed using the 7 attributes and their possible levels which results in a prohibitive number of choice  42  tasks. Thus we implemented a fractional factorial design using 10 versions of the questionnaire. Table 2.1 Attribute and levels included in the DCE questionnaire Attribute Untreated responders*: Proportion of patients who could be cured by the new medication (responders) but will not receive it as a result of inaccurate genetic test result. Unnecessary treatment of non-responders†: Proportion of patients who would not benefit from the new medication (non-responders) but will receive it as a result of wrong genetic test result. Severity of side effects: The new medication may be associated with side effects such as nausea, hair loss, skin rash and fatigue. The potential levels of Side Effect Severity were: Likelihood of side effects: The side effects described in Attribute 3 will not necessarily occur for all individuals. Instead, they will occur with a particular percentage chance. Possible levels were: Genetic test turnaround time: The time required to obtain the genetic test results, after the test has been performed. Genetic test procedure: Type of the procedure that is needed for doing the genetic test. Genetic test cost: Please assume that you would be paying only for the genetic test out-of-pocket. *1-Sensitivity †1-specificity  Levels 5%, 20%, 35%, 50%  5%, 20%, 35%, 50%  Severe, Moderate, Mild  5%, 50%, 95%  2 days, 7 days, 12 days Mouth swab, Blood sample, Tumor biopsy, Bone marrow biopsy, Liver biopsy $50, $500, $1000, $1500  43  Table 2.2 A Sample choice task Attributes Untreated responders Unnecessary treatment of non-responders Severity of side effects Likelihood of side effects Genetic test turnaround time Genetic test procedure Cost of genetic test  Option A 5 out of 100 20 out of 100 Moderate  Option B 50 out of 100 5 out of 100 Mild  50out of 100  5 out of 100  2 days Liver biopsy $1000  7 days Blood sample $1500  Neither  The efficiency of the final design was assured by generating large number of possible designs and then selecting the design that provided the most precise coefficient estimates (i.e. smallest standard errors) and a better D-efficiency given the sample size[109,120]. Two fixed choice tasks, each containing a clearly dominant alternative, were included in all versions of the questionnaire to test for rationality and consistency of the answers given by each participant. The design of the web-based questionnaire, which facilitated direct data entry into our secured server, was done using the Choice Based Conjoint (CBC) application of Sawtooth (Sawtooth software Inc, SSI web version 6.6.6).  2.2.2 Procedure At the beginning of the questionnaire, we described one of the two possible scenarios to the participants (either fast-acting curable cancer or slow-acting incurable cancer shown, Table 2.3). Then we explained the attributes and levels in the DCE and asked participants to complete the 16 choice tasks in the DCE questionnaire. 44  Table 2.3 Scenarios for DCE Fast- acting curable cancer (Scenario for sample 1 and sample 2) Imagine that you have recently been diagnosed with a fast-acting but curable form of cancer. Currently, approximately 50 out of 100 (50%) of patients are cured after the first round of chemotherapy. If you are cured by this initial treatment, you will have a normal life expectancy; otherwise your life expectancy is approximately 1 year. In this case you will be given the second round of chemotherapy but your chance of being cured is about 10 out of 100 (10%). By adding a new medication to the first round of chemotherapy the cure rate increases from 50 out of 100 (50%) to 75 out of 100 (75%) . However, only some of individuals can benefit from the new medication (responders) and other individuals receive absolutely no benefit from adding the new medication to the standard chemotherapy (nonresponders). The downside of adding the new medication to the standard chemotherapy is that it increases the likelihood and severity of treatment side-effects.  Slow-acting incurable cancer (Scenario for sample 3) Imagine that you have recently been diagnosed with a slow-acting but incurable form of cancer. This means that the spread of the disease is usually slow, but treatments are only able to slow the spread further, and cannot cure the disease. Your life expectancy after being diagnosed with this type of cancer is approximately 10 to 13 years. You will receive treatment after you start experiencing symptoms, which may take several years after your initial diagnosis. Even if your treatment is successful, you are likely to experience numerous relapses, in which the disease returns after a period of improvement. These relapses will be treated until all options for treatment have been exhausted. By adding a new medication to the first round of chemotherapy your life expectancy can be increased by 2 years on average. However, only some of individuals can benefit from the new medication (responders) and other individuals receive absolutely no benefit from adding the new medication to the standard chemotherapy (non-responders). The downside of adding the new medication to the standard chemotherapy is that it increases the likelihood and severity of treatment sideeffects.  The extent that a genetic test will be used in practice is affected by perceived benefits and risks/costs of using the genetic test. As such, in the DCE questionnaire, participants needed to make a trade-off between the consequences of not taking the new drug when, in fact, it was beneficial (misclassification of responders as non-responders due to limited sensitivity of the genetic test) and, experiencing additional side effects of new chemotherapy without receiving any clinical benefit (misclassification of non-responders 45  as responders due to the limited specificity of the genetic test), the invasiveness of genetic testing procedure, the test turnaround time, and the cost of the genetic test. We did not specify the type of cancer, treatment, and the associated genetic test to increase generalizability of the results. Nonetheless, the sample of patients in this study were former and current lymphoma patients in British Columbia, and the disease descriptions provided in the DCE questionnaires were similar to aggressive and nonaggressive types of lymphoma. The descriptions at the beginning of the questionnaire explicitly stated that in the absence of a genetic test, all patients would be offered the new chemotherapy. As such, choosing the “neither” option in a choice task implied a respondent’s preference for opting-out from genetic testing and taking the new chemotherapy regardless of the predicted likelihood of response.  2.2.3 Study Sample Three distinct samples completed a single DCE. Sample 1 consisted of current or former lymphoma patients who had voluntarily agreed be contacted about research projects in British Columbia (BC), Canada. We initially contacted a list of 84 patients through email and 54 patients agreed to participate in this study. Two samples from the general public (sample 2 and sample 3), consisting of 588 and 578 individuals respectively, also participated in this study (Table 2.4).  46  Table 2.4 Characteristics of participants  Age (years) Mean (std) Education (%) Some high school High school College Bachelor degree Master degree Doctorate Gender Female Male Number of dependent children None 1 2 3 or more Description of current health situation Excellent Very good Good With some health problems Having serious health problems If knew anyone diagnosed with cancer Yes, very closely Yes No Household’s annual income (Can$) Less than 25000 25000 to 50000 50000 to 75000 75000 to 100000 100000 to 125000 More than 125000  Sample1 (Patients) N=50  Sample 2 (Public) N=588  Sample 3 (Public) N=578  58.6 (8.8)  48.5 (15.6)  47.9 (15.9)  0 3 (6.7%) 12 (26.7%) 16 (35.6%) 12 (26.7%) 2 (4.4%)  46 (7.9%) 247 (42.3%) 206 (35.3%) 71 (12.2%) 11 (1.9%) 3 (0.5%)  37 (6.5%) 257 (45.3%) 204 (36.0%) 46 (8.1%) 20 (3.5%) 3 (0.5%)  26 (52%) 24 (48%)  287 (48.8%) 301 (51.2%)  296 (51.2%) 282 (48.8%)  32 (72.7%) 2 (4.5%) 6 (13.6%) 4 (9.1%)  335 (57.2%) 84 (14.3%) 100 (17.1%) 67 (11.4%)  323 (56.8) 85 (14.9%) 92 (16.2%) 69 (12.1%)  5 (11.1%) 7 (15.6%) 12 (26.7%) 13 (28.9%) 8 (17.8%)  54 (9.3%) 173 (29.8%) 169 (29.1%) 167 (28.8%) 18 (3.1%)  66 (11.5%) 162 (28.2%) 191 (33.3%) 141 (24.6%) 14 (2.4%)  1 (2.3%) 13 (29.5%) 30 (68.1%)  108 (18.6%) 224 (38.5%) 250 (43.0%)  90 (15.7%) 237 (41.4%) 245 (42.8%)  1 (2.3%) 11 (25%) 7 (15.9%) 7 (15.9%) 3 (6.8%) 15 (34.1%)  88 (15.5%) 171 (30.1%) 127 (22.4%) 104 (18.3%) 42 (7.4%) 36 (6.3%)  76 (13.6%) 204 (36.4%) 131 (23.4%) 70 (12.5%) 52 (9.3%) 28 (5.0%)  47  We used the same choice questions for all three samples, but varied the underlying form of cancer described for sample 3 relative to sample 1 and sample 2.The preamble in the questionnaire described an aggressive, potentially curable cancer to participants in sample 1 and sample 2, and a non-aggressive, incurable cancer as the scenario for sample 3 (Table 2.3). Using this design, we sought to contrast the preferences of patients who had direct experience of disease with preferences of the public, and also to show how the type of cancer and its prognosis affected individuals’ preferences about the genetic testing. The samples from the general public (sample 2 and sample3) were recruited by Ipsos Reid (Vancouver, British Columbia) and were representative of the Canadian general population in terms of demographics and socio-economic characteristics. All subjects (patients and the general public) were invited to participate in this study through email. All participants were at least 19 years old and were able to read and write in English. In the initial letter, we provided a brief description of the study and invited individuals to participate. Once they agreed, each participant provided informed consent and then followed a link to the online questionnaire. Participants could choose not to answer any of the questions or withdraw at any point. The protocol for this study was reviewed and approved by the University of British Columbia - British Columbia Cancer Agency Research Ethics Board.  48  2.2.4 Statistical Analysis There are a variety of statistical methods for the analyses of DCE data that range from simple conditional logit models to Bayesian mixed logit models[114] and Latent Class Analysis (LCA)[115]. Critical assessment of these methods can be found elsewhere[116]. We chose LCA for analyses of the DCE data in this study, as it can effectively identify classes of individuals with similarities in their preferences. This aspect of LCA is particularly useful when one is interested in unraveling possible heterogeneities among participants’ preferences[115]. Analyses were performed using Latent Gold Choice version 4.5.0. The choice data were effect-coded for attributes with discrete values, with the exception of cost, which was modeled as a continuous variable[117]. An alternative specific variable was dummy coded and indicated the situations where “neither” was chosen [109,145]. Including cost as an attribute in the DCE enabled us to calculate Marginal Willingness to Pay (MWTP)[109] which can provide useful interpretations for estimated preference weights. MWTP indicates how much individuals on average are willing to pay to receive a certain amount of change in one of attribute levels.  2.3 Results 2.3.1 Sample Characteristics Mean age in the sample of current or previous lymphoma patients (sample 1) was 58.6 years, about 10 years higher than in the samples from the public (sample 2 and sample 3).  49  Also 34.1% of individuals in Sample1 reported a household income of ≥ Can$125,000 versus 6.3% and 5% of individuals in sample 2 and in sample 3 respectively. Table 2.4 has summarized the characteristics of the participants in the three samples. Response rates were 59.5%, 76%, and 79.5% in sample1, sample 2, and sample 3 respectively.  2.3.2 Estimation Results We used LCA to estimate a model with an identical specification in the three samples and compared the preference weights of attribute levels (Table 2.5). The MWTP associated with the levels in each attribute also have been reported in Table 2.5. The results showed that preference weight of “untreated responders: 5%” was significant and positive in all three samples (Table 2.5). This result suggests that patients and the public perceived a significant increase in their utilities by reducing the proportion of untreated responders from 50% to 5% (or equivalently, reducing sensitivity of the test from 50% to 95%). However, improving proportion of untreated responders from 50% to 35% or to 20% did not statistically significantly change the mean utility. This implies that even a 20% chance of failure to detect responders is still considered undesirable. The preference weight for having 5% untreated responders was notably larger amongst patients (0.833, p-value<0.0009) compared to both samples from the public (0.175, pvalue<0.0009 and 0.256, p-value<0.0009 respectively, in sample 2 and sample 3).  50  Table 2.5 Estimated utilities and Marginal Willingness to Pay (MWTP) in 3 samples Sample 1 (N=50)  Untreated responders 5% 20% 35% 50% Unnecessary treatment of non-responders 5% 20% 35% 50% Severity of side effects Mild Moderate Severe Likelihood of side effects 5% 50% 95% Genetic test turnaround time 2 days 7 days 12 days Genetic test procedure Mouth swab Blood sample Tumor biopsy Bone marrow biopsy Liver biopsy Neither (No test) Genetic test cost R2 R2 (0) Number of respondents Number of observations Log-likelihood (LL) BIC Prediction error  Sample 2 (N=588)  Sample 3 (N=578)  Coefficient  MWTP  Coefficient  MWTP  Coefficient  MWTP  0.8327 (0.1012) 0.0546 (0.1013) -0.1544 (0.1083) -0.7329  15,656 7,875 5,785  0.1752 (0.0262) 0.0363 (0.0261) -0.0413 (0.0272) -0.1702  1,151 688 430  0.2556(0.0266) 0.0297(0.027) -0.0462(0.0281) -0.2391  1,237 672 482  0.1344 (0.1027) 0.1983 (0.1062) 0.0146 (0.1046) -0.3473  4,817 5,456 3,619  0.1008 (0.0264) 0.0052 (0.027) 0.0015 (0.0268) -0.1075  694 376 363  0.2396(0.0266) 0.0356(0.0283) -0.1208(0.0276) -0.1544  985 475 84  0.3223 (0.08) 0.1 (0.0831) -0.4223  7,446 5,223  0.3232 (0.0202) 0.0769 (0.0207) -0.4001  2,411 1,590  0.3328(0.0207) 0.1422(0.0214) -0.475( )  2,020 1,543  0.2327 (0.0805) -0.0833 (0.0825) -0.1494  3,821 661  0.2473 (0.0201) -0.0217 (0.0206) -0.2256  1,576 680  0.2592(0.021) 0.0351(0.0211) -0.2943( )  1,384 824  0.0299 (0.0832) 0.0889 (0.0822) -0.1188  1,487 2,077  0.1238 (0.0207) -0.0072 (0.0206) -0.1166  801 365  0.1236(0.021) -0.0107(0.0215) -0.1129( )  591 256  0.4193(0.1173) 0.0864(0.1263) -0.1067(0.1279) 0.1063(0.123) -0.5053 -0.3176(0.124) -0.0001 (0.0001)  9,246 5,917 3,986 6,116  0.298 (0.0299) 0.2809 (0.0315) -0.0493 (0.0321) -0.2732 (0.0317) -0.2564 -0.9905 (0.0354) -0.0003 (0.0000)  1,848 1,791 690 -56  0.3053(0.0306) 0.3881(0.0326) -0.0823(0.0327) -0.3597(0.0335) -0.2514 -1.0254(0.0357) -0.0004(0.0000)  1,392 1,599 423 -271  0.1693 0.1856 50 659 -605.8 1282.0 0.43  -3,176 Ref  0.1109 0.1696 588 8766 -8213.8 16542.3 0.42  -3,302 Ref  -2,564 Ref  0.1522 0.2059 578 8661 -7821.6 15757.8 0.40  51  The preference weight of improving proportion of unnecessary treatment of nonresponders from 50% to 5% was not significant in the patients’ sample (0.134, p-value= 0.19). Conversely, corresponding preference weights in sample 2 and 3 were both significant, and more than two-fold larger in sample 3 compared to sample 2 (0.101, pvalue< 0.001, 0.240, p-value< 0.001, respectively). This implies that, compared with a fast-acting curable cancer, individuals put a higher weight on avoiding unnecessary treatments for a slow-acting incurable cancer. Individuals in the three samples had similar preferences with respect to severity and likelihood of side effects, and test turnaround times. Patients in sample 1 gained no significant utility with the shortening of turnaround time. While the utility for shortening of turnaround time form 12 days to 2 days was statistically significant in sample 2 and sample 3, the relatively small utilities gained implies a limited impact on overall utility (0.1238, p-value<0.001 and 0.1236, pvalue<0.001 in sample 2 and sample 3, respectively). In terms of the testing procedure, patients and the public preferences diverged when trading off between invasive procedures. Patients (sample 1) had a large negative utility for liver biopsy (-0.505). In contrast, the public had large negative utilities for bone marrow biopsy (-0.273 and -0.360 in sample 2 and sample 3 respectively), negative utilities for liver biopsy (-0.256 and -0.251 in sample 2 and sample 3 respectively), and smaller negative utilities for tumor biopsy (-0.049 and -0.082 in sample 2 and sample 3 respectively).  52  Overall, participants in the three samples showed negative preferences toward opting out from genetic testing to guide cancer therapy. However, patients tended to have less aversion to choosing the neither option (-0.318, p-value=0.01, -0.991, p-value<0.001, and -1.025, p-value<0.001 in sample 1, sample 2, and sample 3 respectively). Lastly, as expected, the coefficient of the cost attribute had a negative sign in the three samples, but its absolute magnitude was smaller for the patients (sample1) as compared to the other two samples.  2.3.3 Latent Class Analyses Using LCA, we were able to capture existing heterogeneities in preferences within each sample. By including individuals’ characteristics in the analysis, we examined characterizing classes based on these; however, none of those covariates contributed to class identification, suggesting that observed heterogeneities in preferences were solely due to latent variables. In sample 1, we identified three classes with distinct preference weights for improving proportion of untreated responders to 5% and choosing the “neither” option (Table 2.6). Class 1, which consisted of the majority of participants in sample 1 (64%), clearly preferred to take the genetic test and also had significant and positive preference weight for “untreated responders: 5%”. Class 2 (19%) showed strong preferences for “untreated responders: 5%” (3.235). In contrast, class 3 (16%) had a non-significant utility for this attribute level, and instead they had strong preferences for not taking the genetic test (6.794). 53  Table 2.6 Estimated utilities, latent class analysesa  Class size Untreated responders 5% 20% 35% 50% Unnecessary treatment of non-responders 5% 20% 35% 50% Severity of side effects Mild Moderate Severe Likelihood of side effects 5% 50% 95% Genetic test turnaround time 2 days 7 days 12 days  Class1  Sample 1 Class2  Class3  Sample2 Class1 Class2  Sample3 Class1 Class2  33 (64%)  9 (19%)  8 (16%)  467 (79%)  121 (20%)  415 (71%)  163 (28%)  0.6674 (0.1348) 0.1813 (0.1181) -0.1250 (0.1246) -0.7237  3.2347 (0.3937) 0.1813 (0.1181) -0.1250 (0.1246) -3.2910  -0.4963 (2.7957) 0.1813 (0.1181) -0.1250 (0.1246) 0.4400  0.1766 (0.029) 0.0411 (0.0272) -0.049 (0.0284) -0.1687  0.3636 (0.0747) 0.0411 (0.0272) -0.049 (0.0284) -0.3557  0.2803 (0.0277) 0.0309 (0.0281) -0.0528 (0.0289) -0.2584  0.2803 (0.0277) 0.0309 (0.0281) -0.0528 (0.0289) -0.2584  0.2287 (0.1223) 0.2247 (0.1243) -0.0056 (0.1195) -0.4478  0.2287 (0.1223) 0.2247 (0.1243) -0.0056 (0.1195) -0.4478  0.2287 (0.1223) 0.2247 (0.1243) -0.0056 (0.1195) -0.4478  0.1077 (0.0277) 0.0001 (0.0282) 0.009 (0.0281) -0.1168  0.1077 (0.0277) 0.0001 (0.0282) 0.009 (0.0281) -0.1168  0.2481 (0.0278) 0.0276 (0.0295) -0.1150 (0.0286) -0.1607  0.2481 (0.0278) 0.0276 (0.0295) -0.1150 (0.0286) -0.1607  0.4422 (0.0938) 0.0952 (0.0955) -0.5374  0.4422 (0.0938) 0.0952 (0.0955) -0.5374  0.4422 (0.0938) 0.0952 (0.0955) -0.5374  0.3632 (0.022) 0.0766 (0.0213) -0.4398  0.0785 (0.0633) 0.0766 (0.0213) -0.1551  0.3878 (0.0237) 0.1405 (0.0220) -0.5283  0.1578 (0.0500) 0.1405 (0.0220) -0.2983  0.2771 (0.0933) -0.0957 (0.0917) -0.1814  0.2771 (0.0933) -0.0957 (0.0917) -0.1814  0.2771 (0.0933) -0.0957 (0.0917) -0.1814  0.2817 (0.0219) -0.0193 (0.0213) -0.2624  0.0683 (0.0624) -0.0193 (0.0213) -0.049  0.2948 (0.0238) 0.0484 (0.0218) -0.3432  0.1311 (0.0483) 0.0484 (0.0218) -0.1795  0.0456 (0.0957) 0.1409 (0.0926) -0.1865  0.0456 (0.0957) 0.1409 (0.0926) -0.1865  0.0456 (0.0957) 0.1409 (0.0926) -0.1865  0.134 (0.0215) -0.0064 (0.0212) -0.1276  0.134 (0.0215) -0.0064 (0.0212) -0.1276  0.1308 (0.0217) -0.0090 (0.0222) -0.1218  0.1308 (0.0217) -0.0090 (0.0222) -0.1218  54  Table 2.6 Estimated utilities, latent class analyses - continued  Genetic test procedure Mouth swab Blood sample Tumor biopsy Bone marrow biopsy Liver biopsy Neither (No test) Genetic test cost Intercept R2 R2 (0) Number of respondents Number of observations Log-likelihood (LL) BIC Prediction error  Class1  Sample 1 Class2  Class3  Sample2 Class1 Class2  Sample3 Class1 Class2  0.5681 (0.1420) 0.0408 (0.1455) -0.1163 (0.1470) 0.1559 (0.1455) -0.6485 -4.2568 (0.6319) -0.0002 (0.0001) 0.865 (0.1991)  0.5681 (0.1420) 0.0408 (0.1455) -0.1163 (0.1470) 0.1559 (0.1455) -0.6485 1.6424 (0.3191) -0.0002 (0.0001) -0.3433 (0.2569)  0.5681 (0.1420) 0.0408 (0.1455) -0.1163 (0.1470) 0.1559 (0.1455) -0.6485 6.7945 (1.9938) -0.0002 (0.0001) -0.5217 (0.2651)  0.2894 (0.0333) 0.3105 (0.0333) -0.0601 (0.0336) -0.2442 (0.0341) -0.2956 -2.6783 (0.076) -0.0002 (0.0000) 0.6731 (0.0522)  0.2362 (0.0374) 0.3058 (0.0393) -0.0318 (0.0382) -0.2954 (0.0382) -0.2148 -2.7937 (0.1031) -0.0003 (0.0000) 0.4615 (0.0505)  0.5531 (0.0847) 0.3105 (0.0333) -0.0601 (0.0336) -0.5832 (0.104) -0.2203 0.9951 (0.0825) -0.0011 (0.0001) -0.6731 (0.0522)  0.56  0.29  0.29  0.56  0.34  0.34  50  588  578  659  8766  8661  -349.3  -6628.8  -6646.8  792.6  13423.4  13465.4  0.21  0.33  0.33  0.7531 (0.0680) 0.7579 (0.0756) -0.2967 (0.0879) -0.6849 (0.0932) -0.5294 0.4641 (0.0714) -0.0009 (0.0001) -0.4615 (0.0505)  a  We identified the coefficients that were not significantly different across classes (based on chi-squared probabilities), and re-estimated the model by restricting those to be equal across classes.  In sample 2, we found two classes: individuals in class 1 had larger utility for “severity of side effects: mild” and “likelihood of side effects: 5%” compared to those in class 2 (Table 2.6). Individuals in class 2 had a larger utility for “untreated responders: 5%”, a larger negative utility for cost, and a positive utility for choosing the neither option. Individuals in sample 3 also fell into 2 classes. The two classes in sample 3 mainly could be distinguished by their utilities for the severity of side effects, the likelihood of side effects, and the tendency for choosing neither option (Table 2.6). More specifically, class 1 had a  55  higher positive utility for “severity of side effects: mild”, and “likelihood of side effects: 5%”.  2.4 Discussion Patients and the public had different perceptions about the value of various aspects of genetic testing to guide cancer treatment. While patients were more concerned about improving sensitivity of the test (and presumably their survival chance), the public had a larger preference for decreasing severity of side effects, decreasing the likelihood of side effects, and switching to less invasive testing procedures. Our results suggest that although patients generally favor genetic testing to guide their treatment, they are more likely to select the neither option (opt-out) when the test lacks certain positive attributes. A plausible explanation is that sensitivity of the test was patients’ primary concern and as such, in the absence of an adequately sensitive test they preferred taking the treatment regardless of its side effects. This shows how some properties of the genetic test (e.g. sensitivity) could influence uptake. The type of cancer also influenced the preferences when we compared the results in the two samples from the public. Interestingly, unlike fast-acting curable cancer, for a slow-acting incurable cancer, individuals put similar emphasis on sensitivity and specificity and also had a larger negative preference toward the cost of genetic testing. The fact that for a slow-acting incurable cancer the change in the survival is ultimately small and is expected to be materialized after 13 years, leads to discounting the benefit of genetic testing in this scenario. 56  There is paucity of studies about preferences for genetic testing. The increasing number of new genetic tests ensuing from fast developments in genomic sciences suggests a need for further investigations in this area. In a study conducted by Griffith et al, willingness to pay for receiving breast cancer genetic services was estimated by conducting a DCE on 242 individuals with high, moderate, and low risk of developing breast cancer[113]. Using a DCE and following a rigorous methodology, Hall and colleagues [57] explored the factors that influenced participation in genetic carrier testing for Tay Sachs and cystic fibrosis among a sample from the general community and a sample of the Ashkenazi Jewish community. A recent study[111] also used DCE to estimate the tradeoffs among sensitivity, turnaround time, and cost of a postnatal genetic test to predict genetic abnormalities causing mental retardation in children. Finally, in a study done by Herbild et al[112], they elicited preferences of Danish general population for taking a pharmacogenetic test that could improve treatment of depression. The novel characteristic of our study is utilizing three distinct samples to demonstrate how preferences of patients differed from those of the public, and how type of cancer and its prognosis affected preferences for a genetic testing. Also, in contrast with previous studies, the results of our study are applicable to most genetic tests for guiding cancer treatment, as we did not specify the type of cancer, treatment, or the associated genetic test. Throughout this study, participants provided their choices considering the following assumptions: 1) if they decided to opt-out from genetic testing, they would receive the 57  new treatment regardless of its effect; 2) the new treatment was covered by their insurance policies. We acknowledge that under different circumstances in terms of the effect of genetic testing on the access to the new treatment, the current results may not apply. The larger standard errors around the estimated coefficients in sample 1 suggested that this sample was slightly underpowered. However, the sample size was restricted to a list of lymphoma patients in BC cancer agency’s contact list and willingness of those approached to participate and thus could not be increased. Despite this limitation, all of the point estimates in sample 1 were in line with our prior expectations in terms of the order of their magnitudes and corresponding signs. Moreover, sample 1 was not an archetypal sample of cancer patients in BC, as they had high income, high education level, and were 10 years older on average. This issue potentially limits the external validity of the results in sample 1. Indeed, smaller cost coefficients resulted in attaining substantially higher MWTP for patients and estimated MWTP’s cannot be directly compared between the samples of patients and the public. This study demonstrates how preferences of patients compare with those of the public for genetic testing. Additionally, the results show that how individuals think differently when they are faced with an aggressive but curable cancer versus a non-aggressive and incurable cancer. These results highlight the properties of genetic testing with a larger potential value for patients and the society. Also discrepancies between preferences of patients and the public emphasize the important issue of whose preferences should be considered in relevant healthcare decision making processes.  58  Chapter 3: Barriers to Integrating Personalized Medicine into Clinical Practice: A Best-Worst Scaling (BWS) Experiment. 3.1 Background With the availability of new technologies, the cost and time needed for complete sequencing of an individual’s genome is rapidly declining [146,147]. In 2007, Knom Incorportated announced the sequencing of an individual’s entire genome for US$350,000. Only three years later, in June 2010, Illumina announced providing the same service for less than US$9500, almost 35 times lower, using its new HiSeq2000 technology[14]. In addition to the rapidity with which the entire genome can be sequenced, it is expected that it will cost less than $1000 making it feasible for individuals to pay out of pocket to have their own personal genome sequenced [14,146,148,149]. Despite these advances, there remains the challenge of decoding and interpretation of the data in the DNA sequences that are generated. The application of these data to the practice of clinical medicine and integration into patient care, often referred to as ‘personalized medicine’[3], still requires a great amount of research in basic and applied sciences [4]. For example, well-designed clinical trials are still needed to establish the clinical validity of genetic tests due to the complexities of the interaction between the genes and the environment, and the interaction of multiple genes associated with different diseases.  59  Early signs of the therapeutic potential of personalized medicine are already visible in clinical practice and in developments led by the pharmaceutical industry. For example, Vectibix®, a drug for the treatment of colon cancer, was shown to be effective only in cases without KRAS mutation in the tumor. Thus, we could prevent inappropriate treatment of the other patients who are not expected to experience a benefit. As another example, Ziagen® (Abacavir), an anti-HIV drug, results in serious side effects among patients with the HLA-B5701 allele making prospective pharmacogenetic testing a realistic strategy in avoiding these adverse events [19,23,24]. These examples demonstrate how genetic testing facilitates tailored treatments leading to greater effectiveness and/or fewer adverse effects. Further, genetic testing may be useful for predicting genetic risk for different diseases – i.e. a surveillance tool. As advances in genomics makes genome sequencing more affordable, the demand for having these data, with the intent of using the information to inform clinical practice, will increase. This wave will hit front-line physicians who may be faced with a plethora of patients’ expectations for the integration of genomic data into clinical care. Discoveries based on human genome sequencing will increase the degree of complexity in diagnosis of diseases and their corresponding interventions. For example, different therapeutic approaches will be available depending on a patient’s genotype which will complicate the task of physicians and other health care providers who deliver diagnostic or treatment services.  60  Once the developments in genomics are advanced enough for application in a clinical setting, we need to assess physician’s readiness and willingness to use these new genetic tests. There are few studies that have investigated the role of primary care physicians in the development and application of personalized medicine to routine clinical care. Most of these studies, with the exception of a few surveys, are qualitative in nature (interviews, focus groups, etc) and have not been supplemented by quantitative data[13,78,79,8183,85,86,89,90,150]. The iGene study was intended to measure the relative importance of the barriers to integrating personalized medicine into practice from physicians’ perspective. The novelty of this study is its quantitative approach using Best Worst Scaling (BWS) choice experiment to answer this question and extending previous findings from the qualitative studies. As such, we combined strengths of qualitative and quantitative methods for hypotheses generation and hypothesis testing respectively.  3.2 Methods 3.2.1 Study Design This study was a cross sectional experimental design where a sample of physicians in British Columbia provided their opinions using a computer-administered questionnaire. We used Best Worst Scaling (BWS) choice experiment for estimating relative importance of attributes that might affect physician’s decision to utilize personalized medicine in their practice.  61  The attributes included in this experiment were identified through a qualitative study using several focus groups on 28 physicians in British Columbia (BC), Canada. Table 3.1 shows the list of attributes that emerged as important items during the discussions in the focus groups. Each attribute can assume a possible level (attribute-levels from now on). Choice experiments are particularly useful in measuring preferences about the options that are on the horizon of clinical innovation. Genetic tests that have not been implemented yet are a good example of this application[102]. Best-Worst Scaling (BWS), is a cutting edge method for conducting choice experiments and has several advantages over older methods for conducting choice experiments [109,118]. One of the key advantages of BWS is the ability to estimate the relative importance of all attributes on a common scale which is crucial for the purposes of our study.  3.2.2 Questionnaire Design Best Worst Scaling is a method to elicit individuals’ stated preferences for a specific set of attributes or characteristics associated with a decision or choice[109]. In BWS experiments, each respondent chooses the most preferred and the least preferred items among a list of three or more items presented in a given task. This process is then repeated with each subsequent choice task containing a different set of items. This method ensures obtaining a valid estimation of respondents’ underlying relative preferences for the items presented [118,119]. Considering this general framework, we designed a BWS experiment to measure preferences of physicians for the attributes of personalized medicine that may impact their decision to use it in their clinical practice. 62  Table 3.1 Attributes and levels included in the best worst questionnaire  Type of genetic tests:  Both types: Most of the available genetic tests would be intended to specify both patients’ predispositions to diseases and patients’ drug responses based on their genotypes.  Tests for disease predisposition: Most of the available genetic tests would be intended to specify patients’ predispositions to various diseases.  Tests for drug response: Most of the available genetic tests would be intended to specify more effective drugs and/or drugs with less adverse effects for specific diseases based on patients' genotypes. Training for the use of genetic testing:  Extensive training opportunities: 3 hours hands-on workshop for each new genetic test.  Some training opportunities: 1 hour in-service for each new genetic test.  No training opportunity: No particular training opportunity would be provided for physicians about the use of new genetic tests. Guidelines for the use of genetic testing:  Clear guidelines: There would be clear and comprehensive clinical guidelines for the use of genetic tests.  No guidelines: There would be any clinical guidelines for the use of genetic tests. Professional fee:  $100: A $100 professional fee would be defined in payment schedules (similar to the fee for performing a cardiovascular risk assessment) for using a genetic test.  $16: A $16 professional fee would be defined in payment schedules (similar to ECG tracing and interpretation fee) for using a genetic test.  $0: No professional fee would be added to payment schedules. Privacy protection laws:  Comprehensive protection laws: There would be comprehensive and effective laws and regulations to protect the privacy of patients' genetic information.  No protection laws: There would be no particular laws or regulations to protect the privacy of patients' genetic information. Cost of genetic tests:  $0: Medical insurance plans would cover 100% of the expense of the genetic tests that are shown to be cost-effective.  $250: A typical genetic test would cost about $250 and it usually would not be covered by health insurance.  $500: A typical genetic test would cost about $500 and it usually would not be covered by health insurance.  63  The questionnaire contained 16 tasks that had to be completed by each responder. Table 3.2 shows a sample task of the questionnaire. Prior to completing the choice questionnaire, respondents were provided with two pieces of background information: 1) a brief introduction regarding the concept of personalized medicine; and 2) how personalized medicine is expected to enhance a physicians’ ability to prevent the occurrence of disease, and to diagnose and treat disease once it occurs. Consequently, participants completed the tasks under the assumption that genetic tests for which there was evidence of clinical benefits were available to be used in their clinical practice. Responders chose the attribute- levels with the most positive effect and the most negative effect on their decision to use personalized medicine given the attributelevels presented in a given choice task. Table 3.2 A sample best worst choice task a Best  √  Worst Tests for disease predisposition No training opportunity Clear guidelines A $100 professional fee Comprehensive protection laws  √  Cost $250 a  Each questionnaire consisted of 16 choice tasks.  A Balanced Incomplete Block Design (BIBD) was used for the design of the consecutive tasks[151,152]. Balanced Incomplete Block Design ensures that all attributes and all pairs of attribute- levels are presented with a balanced frequency to the responders. This property affirms an unbiased estimation of attribute scales. Each questionnaire was 64  administered using a web-based survey and data were electronically collected on a secure server. Sawtooth® software (Sawtooth Software, Inc. Sequim, WA, USA) was used to generate the questionnaire, web page design, and facilitate the collection and recording of responses into a secure database.  3.2.3 Study Sample The choice experiment involved the administration of a questionnaire to a sample of physicians registered with the BC College of Physicians and Surgeons. Physicians were contacted via email and provided with a unique URL with password-protected identification embedded. There was a $50 incentive offered to participants upon completion of the questionnaire to compensate for the opportunity costs of participation in this survey which took approximately 30 minutes to complete. Physicians’ participation was optional, and they could choose to not answer any of the questions and/or withdraw at any time. The protocol for this study was reviewed and approved by University of British Columbia Human Ethics Board, division of human behavioral studies.  3.2.4 Statistical Analysis Best Worst Scaling data were coded for Latent Class Analysis (LCA) using Latent Gold Choice version 4.5.0. Conditional logit model can result in biased preference estimations if unknown correlation structures are present in the choice data. More sophisticated estimation methods such as latent class analysis or, alternatively, mixed logit are appropriate for modeling of choice data[115]. Latent Class Analysis can also indentify 65  segments of individuals with larger likelihood of fitting to a model and in our case, was an effective method to reveal heterogeneity of responders and their underlying preferences [115]. We also included the characteristics of physicians in the latent class analysis and tested for possible interaction of those characteristics with preference estimates. The conventional method for doing choice experiments, such as choice-based conjoint studies, facilitates the estimation of how moving between levels of an attribute affects overall utility. As the estimated model is essentially a difference model, estimated utilities only measure the utility of deviation from a reference level within each attribute. Therefore, it is not possible to compare the utility of levels between attributes. In contrast, BWS has a technical advantage that allows for the comparison of utilities across all levels of all attributes. Unlike choice based conjoint, the utility of all attribute-levels can be estimated relative to one reference attribute-level which is the only missing utility. As all utility estimates are then relative values in a common scale, it should be noted that shifting the horizontal axis in Figure 3.1, either upward or downward, will not affect the interpretation of the results. This study exemplifies a research question (i.e. relative importance of attributes) that cannot be properly answered using traditional methods of discrete choice experiment (DCE).  3.3 Results 3.3.1 Sample Characteristics Two hundred and forty four physicians were initially contacted and 197 physicians completed the experiment and fully responded to the questionnaire (response rate 81%). 66  The mean age of participants was 50.4 years (range 32 to 77 years) and on average, they had 22 years of experience in clinical practice (Table 3.3). Approximately one third of physicians who participated in this experiment were female (32%) and majority of them were practicing in urban areas in BC (85%). Table 3.3 Characteristics of participants (N=197) Age (years) Mean (Stda) Range (min-max) Practice experience (years) Mean (Std) Range (min-max) Gender (N) Female Male Practice field (N) Family physician Specialist Practice location (N) Rural Urban Practice type (N) Family physician group practice Multidisciplinary group practice Solo practice Specialized clinic Other Practice setting (N) Private office/clinic Hospital Community clinic/health centre Academic centre Nursing home Other a b  50.4 (9.2) 32 -77 22.0 (9.8) 0 - 40 63 (32%) 134 (68%) 182 (92.4%) b 11 (5.6%) 29 (14.7%) 165 (83.8%) 113 (57.4%) 11 (5.6%) 48 (24.4%) 10 (5.1%) 13 (6.6%) 139 (70.6%) 24 (12.2%) 16 (8.1%) 7 (3.6%) 6 (3.0%) 3 (1.5%)  Standard Deviation Sum of percentages do not sum up to 100% due to missing responses  67  3.3.2 Model Estimation Best Worst Scaling method allows estimation of coefficients for 15 attribute-levels relative to the one specific level of one attribute. Thus, we estimated the preference for each level of attributes relative to the out of pocket “cost of a genetic test = $500”. As such, each coefficient can be considered to be the ‘utility’ of that attribute level on a common underlying preference scale [151]. In addition, we calculated the impact of each attribute on the choices by averaging the coefficients of attribute-levels for that attribute. The calculated attribute impacts can be interpreted as the relative importance of attributes [118,119,151]. The estimated utilities of each attribute-level and the relative importance of each attribute are reported in Table 3.4. The availability of “Both types” of genetic tests (genetic tests intended to specify both patients’ predispositions to diseases and patients’ drug responses) and “Clear guidelines” were the most valued attribute-levels with mean utility estimates of 3.96 (p-value <0.001) and 3.79 (p-value <0.001), respectively. In comparison, utility estimates of “Extensive training opportunities” and a “$100” professional fee were slightly lower- 3.03(p-value <0.001) and 2.74 (p-value <0.001) respectively. Conversely, “No guidelines”, “No privacy protection laws”, and a cost of “$500” for a typical genetic test had the lowest utility estimates, 2.01 (p-value <0.001), 1.91 (p-value <0.001), and -1.74 (p-value=1) respectively). Utility estimates for “No training opportunity” and “No professional fee” were -1.52 (p-value <0.001) and -0.91(p-  68  Table 3.4 Estimation results from latent class analysis – conditional logit model Table 4: Estimation results from latent class analysis – conditional logit model Overall Class1 197 (100%) 135 (68%) Class size Coefficienta  Type of genetic tests Both types Tests for disease predisposition Tests for drug response Training for the use of genetic testing Extensive training opportunities Some training opportunities No training opportunity Guidelines for the use of genetic testing Clear guidelines No guidelines Professional fee $100 $16 $0 Genetic information Privacy protection laws Comprehensive protection laws No protection laws Cost of genetic tests $0 $250 $500 Intercept Female  2.99 3.96 2.69 2.31 1.23 3.03 2.16 -1.52 0.89 3.79 -2.01 0.78 2.75 0.50 -0.91 0.35 2.60 -1.91 0.00 1.95 -0.21 -1.74  R2 R2 (0) Number of respondents Number of observations  0.10 0.37  Log-likelihood BIC Prediction error  a  T- ratio 25.74 14.07 10.97 17.43 8.75 -28.51 24.28 -35.48 17.34 -5.05 -21.09 13.37 -34.88 7.52 -11.78 0.00  197 6304  Coefficient 3.32 4.40 2.94 2.63 1.21 3.24 2.27 -1.87 0.77 4.16 -2.61 0.56 2.07 0.46 -0.85 0.31 3.23 -2.60 -0.17 1.91 -0.43 -1.99  Class2 62 (32%)  T-ratio 21.76 12.59 11.03 14.86 7.90 -24.93 20.07 -30.65 6.22 -2.52 -12.93 14.53 -32.27 6.04 -10.57 0.00  Coefficient 2.29 3.06 2.17 1.65 1.25 2.60 1.95 -0.79 1.13 3.03 -0.77 1.24 4.16 0.60 -1.04 0.42 1.30 -0.46 0.36 2.02 0.26 -1.21  T-ratio 10.89 6.84 4.18 9.33 5.27 -10.94 10.26 -10.10 18.05 -1.37 -11.33 2.07 -8.61 5.75 -3.91 0.00  0.26  2.71  -0.26  -2.71  0.38  2.04  -0.38  2.04  0.02 0.40  135 4320  0.02 0.30  62 1984  -6430.4 10165.03 0.40  Indicates estimated utilities (relative preference weights)  value <0.001), respectively suggested their less detrimental effect on physicians’ decision to apply genetic testing.  69  3.3.3 Latent Class Analysis The second and third columns in Table 3.4 compare the utility estimates of two latent classes. The availability of “Both types” of genetic tests had the largest positive impact on the utility of respondents in class 1 while a “$100” professional fee had the largest positive impact in class 2. Furthermore, the lowest utility estimates in class 1 were “No guidelines” and “No privacy protection laws” that is in contrast to “$500” cost of genetic testing and “No professional fee” in class 2. In class 1, the positive impact of “Comprehensive privacy protection laws” was clearly larger when compared to class 2. The preference estimates in class 1 evidently covers a wider range, suggesting stronger preferences in class1 when compared to class 2. Figure 3.1 provides a visual illustration of the estimated coefficients in Table 3.4. Considering the relative importance of attributes that have been shown in Figure 3.2, “Type of genetic tests” was the most important attribute affecting physicians’ decision to use genetic testing in their practice in both classes, with a higher magnitude in class1 comparing to class2 (3.32 and 2.29 respectively). Importance of other attributes were similar in both classes with the exception of “Professional fee” and “cost of genetic tests” which had slightly larger importance in class 2.  70  Figure 3.1 Utility weight estimates for attribute-levels  Part-worth utility estimates for attribute-levels 5  Overall Class 1 Class 2  4 3 2 1 0 -1 -2  Genetic testing both types  Clear guidelines  Extensive training opportunities  A $100 professional fee  Genetic testing for disease pred  Comprehensive privacy protection  Genetic testing for drug respons  Some training opportunities  $0 test cost  A $16 professional fee  $250 test cost  No professional fee  No training opportunity  $500 test cost  No privacy protection laws  No guidelines  -3  As individuals’ characteristics (i.e. age, experience, practice type, location, setting, and filed) were not statistically significantly different between the two latent classes, class membership could not be explained based on their known characteristics. However, most female physicians belonged to class1 and the proportion of females in the two classes was statistically significantly different.  71  Figure 3.2 Average importance of attributes  Average importance of attributes  3  Overall Class 1 Class 2  2  1  0  Type of genetic tests  Training  Guidelines  Professional fee  Pivacy protection laws  Cost of genetic tests  -1  3.4 Discussion The current study measured relative importance of factors that affect the decision of physicians to incorporate new approaches of personalized medicine in their practice as they become available. The type of genetic tests was the most important attribute (2.99), suggesting that overall, availability of genetic tests has the largest influence on physicians’ decision to use personalized medicine. In comparison, privacy protection regulations and 72  out- of -pocket cost of genetic tests were the attributes with the least importance. Providing professional fee for using genetic tests had only a moderate impact on physicians’ decisions. Participants also indicated a very large importance for presence of clear clinical guidelines for the use of genetic testing (3.79). At the same time, lack of clinical guidelines had the lowest utility estimate (-2.01), suggesting that lack of guidelines was probably considered as the largest barrier for using genetic testing. In the feedback that physicians provided at the end of their choice experiment, they occasionally commented that providing online resources can be a practical alternative to formal training sessions about genetic testing. Our results suggests that the lack of training opportunities had smaller negative effect comparing to lack of guidelines, which essentially reflects their preference in this regard. In general, physicians in class 1 tended to be more willing to become pioneers in using new genetic tests- if they were accessible- and they perceived the cost of genetic tests, privacy protection laws, and professional fee as the less important obstacles against using personalized medicine. In contrast, physicians in class 2, which consist of a smaller group of mostly male physicians (32%), put less emphasize on the availability of genetic tests and more emphasize on professional fee and cost of genetic tests. A few studies have explored physicians’ viewpoint about integration of genomic medicine into clinical care using cross sectional surveys. Wideroff et al [79] used the results of a National Survey in United States to determine prevalence of using cancer susceptibility tests by physicians and to assess demographic variables associated with their use. 73  Freedman and colleagues[82] have surveyed 1251 physicians in the US and studied the factors that affect physicians’ opinion for using cancer susceptibility genetic tests. Suther et al[78] have shown how physicians’ perception about the characteristics of genetic tests influences the likelihood of adopting genomic medicine in their practice. Finally, Levy et al[90] have conducted a mail survey on 562 physicians and measured the importance of eight factors influencing physicians’ decision to use a genetic test to tailor smoking cessation treatment. For the first time, our study uses a quantitative approach that seeks to elicit the preferences of physician about personalized medicine. We used BWS choice experiment which is the state of the art method for measuring relative importance of attributes. In a choice experiment, responders are asked to indicate their trade-offs between different attributes and as such, estimated preferences reflect their underlying stated preferences [109]. This study exemplified how quantitative methods such as choice experiments are particularly useful for policy making and priority setting through measuring aggregated preferences. Choosing the appropriate levels of attributes to ensure full coverage of all possible attributes that might affect a physician’s decision to use genetic test was a challenge in the questionnaire designing stage. Although we tried to benefit from the discussion in focus groups and previous literature to address this issue, achieving a simple questionnaire and reducing the number of attribute-levels to the possible minimum forced us to make difficult trade-offs. This limitation is not unique to our study and most 74  choice experiments are confined in terms of maximum number of attributes and levels that are included in the design, mostly to avoid detrimental complexity of choice tasks. Nonetheless, we have utilized focus groups to indentify attributes with the largest influence on physicians’ decision and therefore we expect that inclusion of other attributes would not have significant effects on the estimated utility weights. Our entire sample is taken from BC; therefore, generalizability of these finding into other jurisdictions is questionable. Additionally, using a web-based sampling approach might introduce some biases; especially if the physicians who did not participate in the experiment were essentially a different class of physicals in terms of their underlying preferences. However, parallel findings of the focus groups and a relatively large sample size with a wide range of characteristics suggest that this bias should be negligible. The results of this study can inform decision makers that design guidelines for physicians and facilitating the use of personalized medicine in the coming years. Further, understanding of physicians’ preferences about personalized medicine will help prepare our health care system to respond in such a way that maximizes the potential benefits of relevant applications of genomics.  75  Chapter 4: Future Impact of Genomic and Proteomic Tests on the Burden of COPD: A System Dynamics Model. 4.1 Background The human and economic burden of chronic obstructive pulmonary disease (COPD) is substantial and is rapidly increasing worldwide [153]. COPD is now the third leading cause of mortality in the US [154] and the fourth leading cause globally [155]. Given  the enormous burden of COPD across the world, there is significant interest in  developing population-based programs to address this global crisis. It is widely  believed that smoking cessation is the single most effective intervention for COPD [156]. However, the impact of other strategies such as those that provide an early  diagnosis, reduce exacerbations, or increase the likelihood of appropriate therapy has not been systematically evaluated.  In the current study, we evaluated the impact of three potential interventions that targeted different stages of the disease. For this purpose, we first developed a dynamic model that projected the total burden of COPD (epidemiology, cost,  morbidity, and mortality) over the next 26 years using the population of Canada as a case study. Then, by measuring the clinical and economic outcomes, we estimated the incremental cost and incremental effectiveness of using each potential  intervention and compared the results across the three interventions. Finally, we  discuss the implications of our findings for prioritization of interventions to reduce the global burden of COPD.  76  4.2 Methods 4.2.1 Structure of the Model Built on the basis of current literature, the present study has synthesized existing evidence to predict the future burden of COPD using a system dynamics simulation model (Figure 4.1). The dynamic model, developed using Vensim® PLE Plus Version 5.10e (Harvard, MA, USA), projects the total population of 40 years or older (40-49, 50-59, 6069, ≥70 years) from 2011 to 2036 in Canada by taking into account projected annual rates of births, immigration, emigration, and mortality from Statistics Canada (Ref: "Population Projections for Canada, Provinces and Territories 2009 to 2036," 2010). The model also incorporates risks of disease progression among various sub-groups based on their smoking status (current smokers, previous smokers, non smokers), and their COPD stage as determined by their forced expiratory volume in one second (FEV1) values (no COPD, mild, moderate, or severe) based on spirometry. Given the heterogeneity in the prevalence of COPD across sub-groups and the performance characteristics of spirometry (i.e., sensitivity and specificity), the model also estimates sub-groups with a true positive (TP), false positive (FP), false negative (FN), and true negative (TN) diagnosis of COPD based on spirometry data. COPD, especially in asymptomatic patients, is often undiagnosed in the early stages of the disease and the early symptoms in smokers may be erroneously dismissed as smoking related bronchitis [157]. As such, the sub-group with undiagnosed COPD (UD), which consisted of COPD patients who have not received spirometry testing, is also included in the model.  77  Figure 4.1 Model structure  Background Mortality ( i, j, k, l, t) New Cohort (i, j, k, l, t)  COPD Specific Mortality ( i, j, k, l, t)  Cohort (i, j, k, l, t)  Cessation ( i, j, k, l, t)  Aging ( i, j, l, k)  Progression to Next Health State ( i, j, k, l, t)  Minor /Major Exacerbations  Cost  QALY Loss  i: Current Smoker, Past Smoker, Never Smoker j: No COPD, Stage I, Stage II, Stage III and higher k: 40-49, 50-59, 60-69, 70 years and older l: Current diagnosis as Undiagnosed, True Positive, True Negative, False Positive, and False Negative t: Time  Following the Global initiative for chronic Obstructive Lung Disease (GOLD) classification scheme, COPD patients were then divided into three levels of disease severity: mild (GOLD I), moderate (GOLD II), and severe (GOLD III or IV). All TP, FN, and UD patients were initially classified as mild COPD; however, UD and FN cases were assumed to 78  become TP if they progressed into the moderate or severe stages of disease (i.e., there were no FN or UD patients with moderate or severe COPD as symptoms would lead to a diagnosis of COPD). Progression rates across the different stages of the disease (mild, moderate, and severe) were calculated based on changes in the lung function related to aging and smoking status [158]. Exacerbations are major drivers of COPD morbidity and mortality. We modeled the likelihood of exacerbations according to disease severity (Table 4.2). The annual rates of minor (defined as an increase in symptoms requiring office visits and treatment with oral corticosteroids or antibiotics) and major exacerbations (defined as an increase in symptoms requiring an emergency visit or hospitalization) in each sub-group were projected based on published exacerbation rates and the number of prevalent cases at any given time with a specific level of disease severity [159,160].  4.2.2 Epidemiology There is significant variability in the estimated prevalence of COPD across various studies owing in large part to the use of different case definitions. For this analysis, we used the GOLD criteria for diagnosis [161].  79  Table 4.1 Mortality rates and prevalence of COPD by age, smoking status, and severity Current Smokers Background annual mortality rates (per 10,000) 40-49 50-59 60-69 70+ Smoking status among >40 y (%) Prevalence of moderate and severe COPD by age (%) 40-49 50-59 60-69 70+ Prevalence of COPD by severity (%) No obstruction Mild Moderate Severe Prevalence of mild, moderate, and severe COPD by age (%) 40-49 50-59 60-69 70+  Previous Smokers  Never Smokers  Overall  21.9 55.8 146.0 756.3 13.25  15.3 39.1 102.2 529.4 39.43  13.1 33.5 87.6 453.8 47.32  15.2 38.6 101.1 523.7  5.9 10.5 34.7 70.7  1.6 2.8 9.2 18.7  1.1 2.0 6.5 13.2  1.9 3.4 11.3 23.0 80.7 11.1 7.3 0.9  40.0 44.7 68.9 83.9  10.6 11.8 18.2 22.2  7.5 8.3 12.8 15.6  Reference Statistics a Canada[5]  b  Buist et al[11] Buist et al[11]c  Buist et al[11]d  Estimatede  13.0 14.5 22.4 34.1  a  rates were calculated based on relative risk of mortality per smoking status, 2002 Canadian life tables, and 2010 mortality estimates, Statistics Canada (See Appendix A) b Weighted averages based on the reported rates for men and women in Buist et al, table2 c,d Estimated based on rates in Buist et al (See Appendix A) e Estimated based on b, c and d  80  Table 4.2 Utilities, unit costs, and exacerbation rates by disease severity Utilities a 40-49 50-59 60-69 70-79 Chronic stage (all ages) Minor exacerbation (all ages) Major exacerbation (all ages) Direct costs( 2011 Can$)b Maintenance Per minor exacerbation episode Per major exacerbation episode Total direct cost per patient Indirect costs (2011 Can$) Maintenance  No COPD 0.874 0.864 0.828 0.79  Minor exacerbation episode Major exacerbation episode Total Indirect cost per patient Exacerbation rates (per patient year) Proportion of minor exacerbations Proportion of major exacerbations Probability of death per major exacerbation  Mild  Moderate  Severe  Reference Johnson et al [17]  0.81 0.72 0.519  0.72 0.658 0.447  0.67 0.475 0.408  Spencer et al [8] Spencer et al [8] Spencer et al [8]  144 161 6,501 572  430 161 6,501 1167  628 161 6,501 1796  Spencer et al [8] Spencer et al [8] Spencer et al [8] Estimated c  87  258  377  96 3,901 343 0.79 0.94 0.06 0.046  96 3,901 700 1.22 0.93 0.07 0.046  96 3,901 1078 1.47 0.9 0.1 0.046  Chapman et al, Spencer et al [8,12] Estimated c Spencer et al [8] Spencer et al [8] Spencer et al [8] Camp et al [14] d  a  These weights are based on EQ-5D scores All 2001 unit costs were inflated by 15.5% (based on 10 years change in CPI) as an approximation for 2011 Can$ amount. c Estimated based on incidence and proportion of minor/major exacerbation (Appendix A). d Estimated based on COPD specific mortality rate of 30.4 per 10,000 (Camp et al.) and probabilities of major exacerbations. b  Based on this definition, the prevalence of stage II COPD or higher was 10.1% (SE 4.8) in people 40 years of age or older according to a multicenter study across several countries [162]. This prevalence doubles if cases of mild (stage I) COPD are also included [162]. The numbers of patients initially assigned to each severity group were determined based on current prevalence estimates (Table 4.1).  81  In Canada, the prevalence of mild, moderate, and severe COPD is reported to be 11.1% (SE 1.2), 7.3% (SE 1.0), and 0.9% (SE 0.3) in adults 40 years of age or older, respectively [162]. The societal (direct and indirect) cost of a COPD patient is estimated to be $3,196 (Canadian dollars) per year [163]. In the United States the total economic cost of COPD was estimated at $42.6 billion (USD) annually with more than 60% of these costs attributed to hospitalizations [164].  4.2.3 Mortality The background mortality rates were assumed to be related to age and smoking status. The mortality rate from COPD is estimated to be 10,000 deaths per year in Canada (30.4 per 100,000 in the general population)[165], while in the US approximately 120,000 people die annually from COPD[163]. In the model, it was assumed that any COPD related death was associated with a major exacerbation. As such, the impact of risk factors for COPD related deaths (i.e., smoking status, age, and severity stage) was captured indirectly through increased risk of experiencing a major exacerbation.  4.2.3 Quality of Life The model captured COPD related morbidity and mortality by using quality adjusted life years (QALYs). We assumed that COPD had a negative impact on patients’ quality of life and that it was proportional to disease severity. We also assumed that exacerbations further reduced quality of life for the patient. The impact of COPD on patients’ quality of life was captured using the EQ-5D utility derived utility weights [166,167]. Age specific EQ-5D utility weights derived from the general Canadian population [168] were used as 82  reference utility weights to measure the area under the curve of QALYs lost due to COPD. Deaths were translated to equivalent QALY losses by assuming that the quality of life of patients dropped to zero at the time of death. The model calculated the overall QALYs lost at the population level, incorporating the number of COPD deaths, the number of COPD cases (mild, moderate, and severe), and the number of minor/major exacerbations over time (Table 4.2).  4.2.4 Costs The annual cost of COPD and the frequency of exacerbations were modeled differentially based on COPD disease severity. Using the previously derived data [160], we modeled the direct costs of maintenance therapy on the projected number of patients at each stage of the disease. Direct costs of exacerbations were also calculated based on the estimated unit cost per minor/major exacerbations[160] and the frequency of exacerbation episodes throughout the simulation. We assumed that the maintenance costs accrued only to patients who have been diagnosed as TP or FP. Thus, patients with mild COPD who were still undiagnosed (UD) did not generate any maintenance costs. However, we assumed that all COPD patients, including UD cases, were at risk of experiencing an exacerbation. The indirect cost of COPD is significant and accounts for approximately 37.5% of the societal cost of COPD[163]. As such, indirect costs were assumed to be 60% of direct costs for maintenance (mild, moderate, and severe) and exacerbations (Table 4.2).  83  4.2.5 Model Assumptions The time horizon of the simulation was from 2011 to 2036. We populated the model to simulate the burden of COPD in Canada, and as such, Canadian unit costs were used to estimate the economic burden of COPD. The effect of smoking cessation was modeled as moving from the current sub-group into similar previous smoker sub-group. Therefore, subsequent disease trajectories were modified according to parameters that corresponded to previous smokers. The costs and QALYs were not discounted in the base case scenario. However, the effect of an annual 3% discount rate on both costs and QALYs was captured in the sensitivity analysis. The unit costs reported in earlier studies [160], were adjusted using Consumer Price Index (CPI) in Canada to approximate 2011 unit costs.  4.2.6 Modeling of Interventions We estimated the outcomes for the following three hypothetical interventions. Hypothetical Intervention I: Screening Test for Early Detection of COPD Early detection of COPD may result in behavioral change in individuals that would prevent the development and progression of COPD [157]. Some studies suggest that repeated medical advice of physicians and educating patients on the risk of COPD related to cigarette smoking can significantly increase the probability of smoking cessation [164]. Further, it has been shown that patients, especially women [164],who are diagnosed with COPD are more likely to stop smoking. Thus, the effect of a hypothetical screening test for early COPD that could be used to foster smoking cessation was modeled. We 84  assumed that the intervention could reduce smoking rates by more than half (from 13. 25% to 6%), and that, each year, an additional 20% of current smokers would stop smoking. We also assumed that all current smokers would be screened to determine their predisposition to COPD. Hypothetical Intervention II: Predictive Test for Exacerbations Unscheduled physician visits and hospitalizations for exacerbations are responsible for more than 60% of COPD direct medical costs [163]. Prediction of exacerbations (or exacerbators) in advance could potentially reduce the risk of exacerbations through the use of targeted treatments. In this scenario, we assumed that a hypothetical intervention could reduce the frequency of exacerbations by 50% over a 5 year period. We also assumed that for this intervention to be implemented, physicians would have to “order” a diagnostic test once a year for all of their COPD patients. Hypothetical Intervention III: New Drugs to Avoid Progression into More Severe Disease Stages The effect of a hypothetical new drug that could reduce the rate of disease progression (based on the rate of decline of FEV1) by 50% was modeled. As such, all mild or moderate patients remained in their current stage of COPD longer under this scenario. The model also assumed that under this scenario, a test (e.g. spirometry) would have to be administered at least once every five years to monitor the drug’s effect on disease progression. We assumed all mild and moderate patients would take this test.  85  4.2.7 Model Validation and Sensitivity Analysis The model was calibrated to replicate population trends as predicted by Statistics Canada[169]. The causal structure of the model and input parameters were carefully discussed with expert clinicians and the model was validated by comparing the predicted epidemiological variables against the predictions in a number of independent studies. Furthermore, the results were examined under several extreme scenarios to assure robustness and internal validity of the model. Extensive one way sensitivity analyses were conducted to evaluate the effect of variation in input parameters on the results.  4.3 Results 4.3.1 Base Case Projections As it is shown in Figure 4.2, the population over the age of 70 in Canada will double to approximately 8 million in 2036. The model also predicts that the number of COPD patients will increase from 3.17 million in 2011 to 4.90 million in 2036. Based on these projections, the estimated annual societal cost of COPD is $4.24B ($2.6B in direct costs) in 2011, and will reach $7.27B ($4.5B in direct costs) per year in 2036. Over the next 26 years, COPD will be responsible for approximately $142.1B in societal costs ($95.1B in direct costs) and 13.09 million QALYs lost.  4.3.2 Cost-Effectiveness of Interventions In the base case analysis, we set the cost of the tests for all three interventions at zero to facilitate cross-comparison. However, as a sensitivity analysis, we repeated the simulation for all interventions assuming a cost of $100 per test. The impact of all three 86  interventions on the total societal costs is shown in Figure 4.2. The total societal costs, which were Can$142.1 B in 2036 in the base case, decreased to $137.1, $98.8, and $127.1 B with interventions I, II, and III, respectively. In the base case scenario, the total QALYs lost due to COPD were 13.1 million in 2036, which was reduced to 12.6, 11.1, 11.6 million with interventions I, II, and III, respectively (Figure 4.2). Intervention II resulted in significantly larger savings of QALYs and costs when compared to the other interventions. Figure 4.2 Total societal cost due to COPD  Total Societal Cost due to COPD 160 140  Billion Can$  120 100  Intervention III Intervention II  80  Intervention I  60  Base case  40 20 0 2011  2016  2021  2026  2031  2036  Despite a substantial reduction in the number of smokers in Intervention I (Figure 4.5), the overall number of COPD cases declined rather modestly (Figures 4.8, 4.9, 4.10, 4.11). We observed a large decline in the number of smokers with COPD as a result of  87  intervention I. However, the number of previous smokers or never smokers with COPD increased (Figures 4.13, 4.14, 4.15). Figure 4.3 Total QALYs lost due to COPD  Total QALYs Lost due to COPD 14 12  Million QALYs  10 Intervention III  8  Intervention II 6  Intervention I Base case  4 2 0 2011  2016  2021  2026  2031  2036  When we assumed a $100 per test, the aggregated societal cost in 2036 increased to $137.1, 109.1, and 129.1 billion for interventions I, II, and III, respectively. Even though predictive testing for exacerbations was assumed to be repeated annually and therefore resulted in the highest testing costs, the overall ranking of interventions in terms of total savings remained unchanged (Figure 4.19). Further, discounting both costs and QALY’s at a 3% annual rate did not change the order of savings associated with the three interventions (Figures 4.20 and 4.21).  88  Figure 4.4 Population trend in Canada  Population Trend in Canada (> 40) 9,000 8,000  Persons (millions)  7,000 6,000  40-49  5,000  50-59  4,000  60-69  3,000  70+  2,000 1,000 0 2011 2016 2021 2026 2031 2036  Figure 4.5 Number of smokers  Number of Smokers 3.0 2.5 2.0 Millions  Intervention III Intervention II  1.5  Intervention I  1.0  Base case  0.5 0.0  2011 2016 2021 2026 2031 2036  89  Figure 4.6 Number of previous smokers  Number of Previous Smokers 12.0 10.0  Millions  8.0  Intervention III Intervention II  6.0  Intervention I  4.0  Base case  2.0 0.0  2011 2016 2021 2026 2031 2036  Figure 4.7 Number of never smokers  Number of Never Smokers 14.0 12.0  Millions  10.0 Intervention III  8.0  Intervention II  6.0  Intervention I  4.0  Base case  2.0 0.0  2011 2016 2021 2026 2031 2036  90  Figure 4.8 Prevalence of COPD, all stages  Prevalence of COPD, All Stages 6  Million cases  5 4  Intervention III Intervention II  3  Intervention I  2  Base case  1 0  2011  2016  2021  2026  2031  2036  Figure 4.9 Prevalence of mild COPD  Prevalence of Mild COPD 3.5 3.0  Million cases  2.5 Intervention III  2.0  Intervention II  1.5  Intervention I  1.0  Base case  0.5 0.0  2011  2016  2021  2026  2031  2036  91  Figure 4.10 Prevalence of moderate COPD  Prevalence of Moderate COPD 3.0  Million cases  2.5 2.0  Intervention III Intervention II  1.5  Intervention I  1.0  Base case  0.5 0.0  2011  2016  2021  2026  2031  2036  Figure 4.11 Prevalence of severe COPD  Prevalence of Severe COPD 0.45 0.40  Million cases  0.35 0.30  Intervention III  0.25  Intervention II  0.20  Intervention I  0.15  Base case  0.10 0.05 0.00  2011  2016  2021  2026  2031  2036  92  Figure 4.12 Total deaths due to COPD  Total Deaths due to COPD 18.0 Thousand cases per year  16.0 14.0 12.0  Intervention III  10.0  Intervention II  8.0  Intervention I  6.0  Base case  4.0 2.0 0.0  2011 2016 2021 2026 2031 2036  * intervention ii is modeled to have a gradual implementation reaching maximum implementation by year 5.  Figure 4.13 Number of smokers with COPD  Number of Smokers With COPD 1.80 1.60 1.40  Proportion  1.20  Intervention III  1.00  Intervention II  0.80  Intervention I  0.60  Base case  0.40 0.20 0.00  2011  2016  2021  2026  2031  2036  93  Figure 4.14 Number of previous smokers with COPD  Proportion  Number of Previous Smokers With COPD 2.00 1.80 1.60 1.40 1.20 1.00 0.80 0.60 0.40 0.20 0.00  Intervention III Intervention II Intervention I Base case  2011  2016  2021  2026  2031  2036  Figure 4.15 Number of never smokers with COPD  Number of Never Smokers With COPD 1.80 1.60 1.40  Proportion  1.20  Intervention III  1.00  Intervention II  0.80  Intervention I  0.60  Base case  0.40 0.20 0.00  2011  2016  2021  2026  2031  2036  94  Figure 4.16 Proportion of smokers among all COPD cases  Proportion of Smokers Among All COPD Cases 0.45 0.40 0.35 Proportion  0.30  Intervention III  0.25  Intervention II  0.20  Intervention I  0.15  Base case  0.10 0.05 0.00  2011  2016  2021  2026  2031  2036  Figure 4.17 Proportion of previous smokers among all COPD cases  Proportion  Proportion of Previous Smokers Among All COPD Cases 0.39 0.38 0.37 0.36 0.35 0.34 0.33 0.32 0.31 0.30 0.29  Intervention III Intervention II Intervention I Base case  2011  2016  2021  2026  2031  2036  95  Figure 4.18 Proportion of never smokers among all COPD cases  Proportion of Never Smokers Among All COPD Cases 0.40 0.35  Proportion  0.30 0.25  Intervention III  0.20  Intervention II  0.15  Intervention I Base case  0.10 0.05 0.00  2011  2016  2021  2026  2031  2036  Figure 4.19 Total societal cost due to COPD, including testing costs  Total Societal Cost due to COPD Including Testing Costs 160.00 140.00  Proportion  120.00 100.00  Intervention III  80.00  Intervention II  60.00  Intervention I Base case  40.00 20.00 0.00  2011 2016 2021 2026 2031 2036  96  Figure 4.20 Total societal cost due to COPD, discounted  Total Societal Cost due to COPD (Discounted at 3% Annual Rate ) 80 70  Billion Can$  60 50  Intervention III  40  Intervention II  30  Intervention I Base case  20 10 0 2011  2016  2021  2026  2031  2036  Figure 4.21 Total QALYs lost due to COPD, discounted  Total QALYs Lost due to COPD (Discounted at 3% Annual Rate ) 10 9  Million QALYs  8 7 6  Intervention III  5  Intervention II  4  Intervention I  3  Base case  2 1 0 2011  2016  2021  2026  2031  2036  97  4.4 Discussion Our findings confirm the enormous financial burden that COPD will pose over the next 26 years. The best strategy to reduce the growing financial and patient burden of COPD is by reducing exacerbations. Smoking cessation, while it is the cornerstone of COPD prevention, has only a modest effect in attenuating the financial burden of COPD in next 26 years in Western countries such as Canada. This is due to three important factors in COPD pathogenesis. First, there is considerable lag time between exposure and the onset of COPD. Although smoking rates have fallen considerably over the past decade, many of these ex-smokers will develop clinically relevant COPD with aging. Second, independent of smoking, rates of COPD will increase with increasing average age of the population [162]. Third, there is growing evidence that COPD may continue to progress despite smoking cessation [170]. Similarly, interventions (even very effective ones) that reduce the rate of decline in FEV1 will have a modest effect in reducing the economic and human burden of COPD over the next 26 years. However, this is not a very effective strategy as clinically relevant outcomes such as symptoms, hospitalizations or mortality correlate only very loosely with FEV1. Indeed, there are many patients with very poor FEV1 who remain active and free of exacerbations, while there are others with relatively preserved FEV1, who are incapacitated by their disease and experience frequent exacerbations [159]. In our model, the most effective strategy was to target exacerbations. Exacerbations are expensive to treat and are fraught with considerable morbidity and mortality. The current 98  interventions for COPD reduce exacerbations by 20 to 30%. However, with improved understanding of the disease, in the near future, we may have therapies and diagnostic tests (to facilitate more “personalized care”) that may reduce exacerbations even further [171]. If so, these interventions may have a large impact on the financial and human burden of COPD in the Western world. To our best knowledge, this is the first dynamic population model that has compared the impact of population-based interventions that target different approaches to COPD management including primary (avoiding the development of disease), secondary (early disease detection), and tertiary (reducing the negative impact of an already established disease) prevention. The only other dynamic population model developed by Hoogendoorn et al [172,173] projected the burden of COPD in the Dutch population between 2000 and 2025. Using a comprehensive simulation model, they estimated the incremental cost and incremental QALYs of adding bupropion to minimal general practitioner counseling for smoking cessation. While the current analysis relied on Canadian data and parameters, this selection was done for illustration purposes and because of the availability of robust COPD-related cost and epidemiological data. We believe that the overall conclusions are widely applicable and we would expect similar results in other jurisdictions. There were some limitations to our study. There may be unpredicted changes in the future in various parameters that were modeled in this study, which may materially affect our projections. In another limitation, age was divided into only four categories. Having 99  smaller age groups and stratification by other characteristics such as gender could potentially increase the accuracy of the model. However, benefits of further stratifications in the model are generally restricted by availability of data. For example, in the current model, we relied on certain assumptions for deriving COPD prevalence and mortality rates stratified by age and smoking status (Appendix A). Smaller age groups would have required more extensive assumptions. Conducting probabilistic sensitivity analysis for dynamic models, unlike static models, entails additional complexities. For example, correlation between value of variables between different time points may lead to bifurcation [125,174]. As such, similar to previous studies based on dynamic models, we conducted extensive univariate sensitivity analyses to test the validity of the model and robustness of the results. However, we acknowledge that further work is warranted in this area. In conclusion, our results suggest that any intervention that can successfully reduce the number of exacerbations has an immediate and substantial impact on morbidity and costs of COPD and should be considered in conjunction with the ongoing efforts to reduce smoking rates.  100  Chapter 5: Cost-Effectiveness of Using A Molecular Diagnostic Test to Improve Pre-Operative Diagnosis of Thyroid Cancer: A Discrete Event Simulation (DES). 5.1 Background Thyroid nodules are common with a prevalence ranging from 4 to 7% in the general population [175]. Fortunately, only a small proportion of those nodules (approximately 5%) are cancerous [175,176]. Fine-needle aspiration biopsy (FNAB) is currently the standard of care for evaluating thyroid nodules [177] due to its low cost, low complication, and ready availability. The introduction of FNAB in the early 1980s has resulted in significant economic savings by reducing unnecessary surgeries and by increasing the detection rate of malignant thyroid nodules [178]. The major drawback of FNAB, however, is the large number of indeterminate, i.e. unable to be definitively classified as benign or malignant. Clinical decision-making following an indeterminate cytology result is challenging and may lead to either over or under treatment of thyroid nodules. The diagnostic uncertainty is further amplified by the subjective nature of thyroid cytology and different criteria that are used for the classification of nodules at different centers. The Bethesda System for Reporting Thyroid Cytopathology (TBSRTC) is a recent effort to achieve a consensus about the reporting of FNAB cytological results and their clinical significance [179]. The TBSRTC classification is particularly helpful for obtaining a clear definition of cytological outcomes and for the  101  management of indeterminate cases (i.e. Atypia of Undetermined Significance (AUS), Folicular Neoplasm (FN), and Suspicious cases). Even after following the currently recommended diagnostic algorithm, about 50% of resected nodules are eventually proved to be benign [179,180]. Therefore, the development of a new test that reduces the number of unnecessary surgeries would have a high clinical and economic value. A new molecular based diagnostic test (DX), when used in conjunction with FNAB, could potentially reduce the proportion of indeterminate diagnostic results, and, with high accuracy, identify malignant thyroid nodules. Validation studies suggest a very high sensitivity and specificity of the DX test. In the current study, we: 1) Evaluated the overall performance of FNAB in combination with the TBSRTC guidelines (current practice) in assigning patients to undergo surgery, and 2) Estimated the cost- effectiveness of using the new molecular test (DX) in conjunction with FNAB compared to current practice (FNAB in combination with the TBSRTC guidelines, hereafter referred to as NoDX). We repeated the analyses in steps 1 and 2 for patients with initial indeterminate FNAB cytology as well as for all patients who presented with thyroid nodules (i.e. including non diagnostic, benign, and malignant cases in the target population).  5.2 Methods 5.2.1 Model Deign We developed a discrete event simulation model using Arena version 13.0 (Rockwell Software, Inc., Milwaukee, WI) to calculate the incremental cost-effectiveness of using the DX test in conjunction with FNAB relative to current practice (NoDX) in two simulated 102  cohorts of 10,000 patients (10,000,000 random walks) with initial indeterminate FNAB cytology. The model was a patient level simulation in which each patient in the first cohort was assigned to the NoDX arm and an identical clone in the second cohort was assigned to DX (Fig. 1) and the model was run over a 10 year time horizon. Depending on the subsequent diagnostic results in each arm, patients received total thyroidectomy, hemithyroidecomy, or alternatively, were followed up in accordance with the TBSRTC guidelines. In particular, we simulated the number of malignant cases across diagnostic categories and then, by comparing the final cytology and histology of each patient, each patient was classified as either a true positive, false positive, true negative, or false negative. The incidence of major morbidity following surgery (i.e. permanent hypoparathyroidism and Recurrent Laryngeal Nerve Injury (RLNI)), and cancer recurrence were included in the model. Cancer related mortality was also incorporated into the model.  5.2.2 Data Sources and Assumptions In the simulation, following the TBSRTC classification of a nodule there were six possible outcomes subsequent to a FNAB [179]: 1) non-diagnostic or unsatisfactory (ND); 2) benign; 3) atypia or follicular lesion of undetermined significance(AUS), 4) follicular neoplasm (FN), 5) suspicious for malignancy, and 6) malignant. We defined indeterminate as AUS, FN, or suspicious for malignancy. Furthermore, based upon review of the literature, we used the proportion of patients in each category, and the probabilities of malignancy for each category, to inform the progression through the model (Table 5.1)  103  [179]. The possible pathways in the model were designed based upon the proposed management recommendations for each cytological category (Figure 5.1) [179]. Figure 5.1 Model structure Nondiagnostic (ND)  Repeat test  Benign  5 years Follow-up  FNAB/ DX+ FNAB  Fol. Neoplasm (FN)  Hemi thyroidecto my  Medullary carcinoma  Recurrent cancer  Hurthle cell carcinoma  Total thyroidecto my  Anaplastic carcinoma  Suspicious  Malignant  Papilary carcinoma  Follicular carcinoma  Post Surgery Histology/ Follow up  Atypia (AUS) Hypothetical Individual  Malignant (True Pos. OR False Neg.)  Total thyroidecto my Cancer related mortality Benign (True Neg. OR False Pos.)  Adverse events  Hypoparath yoidsim/ RLNI  Record outcomes  No adverse event  N  Repeat trial?  Exit from Study  Y  This is a simplified version of model.  Unlike the NoDx arm, the diagnostic results in the DX arm were confined to two possible diagnoses: benign or malignant. The overall prevalence of malignancy was assumed to be equal in the two arm of the model. Therefore, for a given sensitivity and specificity of the DX test, possible cytological and histological outcomes were simulated in the DX arm. Data from National Cancer Institute’s Surveillance, Epidemiology, and End Results (SEER) was utilized to characterize the major types of thyroid malignancy as well as the 104  proportions of different types of cancer conditional on having a malignant nodule (Table 5.1) [181]. Because there are large differences in the mean age of onset for the different types of thyroid cancer (e.g. individuals diagnosed with anaplastic carcinoma are, on average, greater than 20 years older than individuals diagnosed with papillary carcinoma) expected life years lost varies widely across the different types of thyroid cancer (Table 5.2). These differences are captured in the simulation as are the differences in the mortality rates associated with different thyroid cancers [182]. Permanent hypoparathyroidism and permanent RLNI were considered to be the two major complications of surgery. Despite the rates being low [183], the morbidity associated with these complications is generally persistent and thus, impacts clinical outcomes and costs. In the model we assumed that the failure of detection of malignancy delays necessary surgeries by 28 months on average [184,185] and therefore, increases the risk of cancer recurrence (RR=2.28) and cancer related mortality (RR= 2.11) [184].  105  Table 5.1 Model parameters Parameter Proportion of cytology results, first FNAB a Nondiagnostic (ND) Benign Atypia of undetermined significance (AUS) Follicular neoplasm (FN) Suspicious for malignancy Malignant Probability of malignancy conditional on cytology results Nondiagnostic (ND) Benign Atypia of undetermined significance (AUS) Follicular neoplasm (FN) Suspicious for malignancy Malignant Proportion of second cytology results if AUS  Point Estimate (%)  Range  Distribution (α, β)  11 65 4.5  (2.0, 20.0) (60.0, 70.0) (3.0, 6.0)  Beta (5,42) Beta (236, 127) Beta (34, 729)  7.3 7.2 5  (6.6, 8.0) (6.5, 7.9) (3.0, 7.0)  Beta (371, 4708) Beta (371, 4783) Beta (24, 450)  2.5 1.5 10  (1, 4) (0, 3) (5, 15)  Beta (11, 422) Beta (4, 258) Beta (14, 129)  22.5 67.5 98.5  (15, 30) (60, 75) (97, 99)  Beta (28, 95) Beta (105, 50) Beta (581, 9)  Benign Atypia of undetermined significance (AUS) Follicular neoplasm (FN) Suspicious for malignancy Malignant Proportion of second cytology results if ND  61.5 20.0  (55.4, 67.7) (18.0, 22.0)  Beta (153, 96) Beta (320, 1279)  6.9 6.8 4.7  (6.2, 7.6) (6.1, 7.5) (4.3, 5.2)  Beta (372, 5014) Beta (373, 5094) Beta (381, 7668)  Nondiagnostic (ND) Benign Follicular neoplasm (FN) Suspicious for malignancy Malignant Proportion of malignancy types  36.0 49.2 5.5 5.5 3.8  (32.4, 39.6) (44.3, 54.2) (5.0, 6.1) (4.9, 6.0) (3.4, 4.2)  Beta (256, 454) Beta (203, 209) Beta (378, 6456) Beta (378, 6556) Beta (385, 9777)  79.7 12.1 2.4 4.9 0.9  (76.8, 82.6) (9.7, 14.5) (1.3, 3.5) (3.3, 6.5) (0.2, 1.6)  Beta (607, 155) Beta (92, 669) Beta (18, 743) Beta (37, 724) Beta (7, 754)  Papillary Follicular Medullary Hurthle cell Anaplastic  Reference Cibas (2009)  Cibas (2009)  Cibas (2009)  Cibas (2009)  Davis (2010)  a  Point estimates represent the middle of the reported ranges in Cibas et al. Parameters for Beta distributions are calculated based on the reported ranges.  106  Table 5.2 Model parameters Parameter Conditional probability of mortality (%) Papillary Follicular Medullary Hurthle cell Anaplastic Average age of patients (years)a Papillary Follicular Medullary Hurthle cell Anaplastic Time between diagnosis and death in cancer related cases High risk malignancy (months) High risk malignancy (years) b Probability of cancer recurrence (%) Effect of false negative FNAB cytology on outcomes Average delay in diagnosis (months) Relative risk of mortality Relative risk of long term recurrence Probability of major side effects following thyroidecomy (%) Persistent hypoparathyroidism Permanent laryngeal recurrent nerve (LRN) injury  a  Point Estimate  Range  Distribution  7 15 24 25 86  (6.3, 7.7) (13.5, 16.5) (21.6, 26.4) (22.5, 27.5) (77.4, 94.6)  Beta(372, 4941) Beta(340, 1926) Beta(304,962) Beta(300,899) Beta(55,9)  47 50 48 49 70  (37.6, 56.4) (40.0, 60.0) (38.4, 57.6) (39.2, 58.8) (56.0, 84.0)  Gamma(0.47,100) Gamma(0.50,100) Gamma(0.48,100) Gamma(0.49,100) Gamma(0.70,100)  7 5 15.5  (4.0, 6.0) (5.6, 8.4) (8.3, 40.0)  Gamma(0.07,100) Gamma(0.05,100) Beta(3,17)  28 2.17 2.28  (25.2, 30.8) (2.0, 2.4) (2.1, 2.5)  Normal(28,1.4) Normal(2.17,0.11) Normal(2.28,0.11)  1.7 1  (1.5, 1.9) (0.9, 1.1)  Beta(393,22735) Beta(396, 39203)  Reference Shaha (2000)  Hundahl (2000)  Shaha (2000) Shaha (2000) Kebevew (2000) Sipos (2010)  Rosato (2004)  Estimated based on the age distribution of the malignant cases in Hundahl et al. Risk of recurrent cancer in 10 years follow-up  b  5.2.3 Quality of Life We included 10 different health states in the model: perfect health, surgery, pre-ablation, post ablation weeks 0 to 4, post ablation weeks 4 to 8, unilateral RLNI, bilateral RLNI, hypoparathyroidism, recurrence of differentiated carcinoma, and death (Table 5.3). A heath state utility value was assigned to each health state to facilitate the calculation of quality-adjusted life years (QALYs) as the final model outcome. For each individual in the 107  model, QALYs were calculated by multiplying the utility values of each health state with the time spent in that health state throughout the model. Table 5.3 Utility weights and unit costs Parameter Unit costs ($) Hemithyroidectomy, direct costs Hemithyroidectomy, indirect costsc Total thyroidecomy, direct costs Total thyroidecomy, indirect costs Ablation, direct costs Ablation, indirect costs Ultrasound guided FNAB Follow up, 5 years Utilities Surgery Preablation Postablation 0 to 4 weeks Postablation 4 to 8 weeks Unilateral RLNI Bilateral RLNI Hypoparathyroidism Cancer recurrence Well Duration of health states (days) Length of hospitalization for TTd Length of hospitalization for HT Time between surgery and ablation Post-ablation recovery period I Post-ablation recovery period II  Point Estimate  Range  Distribution for PSAa,b  Reference  $2390 $288 $3058  ($1912, $2868) ($230, $346) ($2446, $3670)  Gamma (23.9, 100) Gamma (2.9, 100) Gamma (30.6, 100)  Soria_2008 Soria_2008  $418 $3740 $784 $412 $1111  ($334, $502) ($2992, $4488) ($627, $941) ($330, $494) ($889, $1333)  Gamma (4.2, 100) Gamma (37.4, 100) Gamma (7.8, 100) Gamma (4.1, 100) Gamma (11.1, 100)  Soria_2008 Soria_2008 Wang_2010 Wang_2010 Khalid_2008 Shrime_2007  0.637 0.548 0.637 0.819 0.627 0.205 0.778 0.54 1  (0.57, 0.70) (0.49, 0.60) (0.57, 0.70) (0.74, 0.90) (0.56, 0.69) (0.18, 0.23) (0.70, 0.86) (0.49, 0.59) -  Beta (145, 82) Beta (180, 149) Beta (145, 82) Beta (72, 16) Beta (149, 88) Beta (318 , 1232) Beta (88, 25) Beta (183, 153) -  Mernagh_2010 Mernagh_2010 Mernagh_2010 Mernagh_2010 Kebevew_2000 Kebevew_2000 Kebevew_2000 Kebevew_2000 Definition  4.2 2.9 7 28 28  (3.4, 5.0) (2.3, 3.5) (5.6, 8.4) (22.4, 33.6) (22.4, 33.6)  Gamma (0.04,100) Gamma (0.03,100) Gamma (0.07,100) Gamma (0.28,100) Gamma (0.28,100)  Soria_2008 Soria_2008 Mernagh_2010 Mernagh_2010 Mernagh_2010  a  PSA: Probabilistic Sensitivity Analysis Parameters for Gamma and Beta distributions are calculated by assuming a standard deviation equivalent to 10% of point estimates. c Indirect costs for TT and HT were calculated based on lengths of hospitalizations and assuming an average income of $99 per day. The average income has been based on men and women annual income and assuming that 12% of patients were men. d TT: Total Thyroidectomy, HT: Hemithyroidectomy b  5.2.4 Costs The model included the cost of the DX test, total thyroidectomy, hemithyroidectomy, clinical follow-up (i.e. yearly physician visits for 5 years), the ablation procedure, and the 108  FNAB test (Table 5.3). As each simulated patient went through different stages in the model, the associated costs were assigned considering health resource utilization at each stage. We conducted a sensitivity analysis for a viable range of the DX test cost. The cost of the DX test was assumed to include costs of all related procedures including pathology, physician time, and specimen transport and processing. The effect of indirect costs due to productivity loss was also explored in the sensitivity analysis.  5.2.5 Model Outcomes Simulated outcomes for each patient were recorded for the 10 year time horizon of the model (e.g. initial and final cytology, occurrence of cancer, type of cancer, hemithyroidecomy, total thyroidecomy, completion thyroidecomy, mortality, and side effects). The health outcomes in terms of QALYs and costs for each patient were also accumulated throughout the model stages over the 10 years time horizon. In the base case analysis, only the direct costs were included and both QALYs and costs were discounted at 3% annually. However, sensitivity analyses were conducted by including indirect costs, and by changing the annual discount rate from 3% to 0%. Incremental net monetary benefits were calculated based on 𝜆∆𝑄𝐴𝐿𝑌 − ∆𝐶, where  ∆𝑄𝐴𝐿𝑌 and ∆𝐶 are, respectively, the incremental QALYs and incremental costs of DX  relative to NoDX, respectively, and λ is society’s willingness to pay for an additional QALY gained which we assumed to be $50,000. Finally, overall sensitivity, specificity, positive predictive value, and negative predictive value of the NoDx strategy (current practice) were also estimated. 109  5.2.6 Univariate Sensitivity Analysis For the base-case scenario, we generated 10,000 patients for each strategy, with each patient undergoing 1,000 random walks in order to minimize first level uncertainty (10,000,000 simulations per arm in total). Univariate sensitivity analyses were then conducted in order to capture the effect of changing the value of model parameters in their possible ranges on the overall outcomes of the model. For a given level of DX sensitivity (specificity), we re-ran the model by increasing specificity (sensitivity) from 80% to 100% by 2.5% increments steps and compared the resulting incremental costs and QALYs. Considering that the cost of DX is a onetime cost for each patient, the overall incremental cost changes linearly with the cost of the test. Thus, we excluded the cost of the DX test in the base case analysis, and the effect of the DX test cost on the outcomes was explored in univariate sensitivity analyses. We hypothesized that there is a non-negligible number of false negative and false positive cases amongst patients with benign and malignant FNAB cytology. As such, we also explored the benefits of expanding the target population of the DX test to include all individuals with palpable thyroid nodules.  5.2.7 Probabilistic Sensitivity Analysis (PSA) Values of the parameters in the base case model are point estimates drawn from various studies and are intrinsically uncertain. In order to capture the effect of parameter uncertainty on the outcomes, we conducted a probabilistic sensitivity analysis by assigning probability distributions to all model parameters [133]. For this purpose, we 110  sampled 1,000 times from the probability distributions of the input parameters and for each draw of the distributions we modeled 100,000 individuals in each arm. We plotted the scatter of incremental cost and incremental QALYs pairs on the cost- effectiveness plane, which reflects the distribution of incremental QALYs and incremental costs of DX versus NoDX, based upon the PSA results.  5.3 Results 5.3.1 Results of Base Case Analysis In the base case analysis, among 10,000 simulated indeterminate patients in the NoDX arm (current practice), 4,407 nonmalignant cases underwent surgery (false positives) and 116 malignant cases were not detected (false negatives). In comparison, assuming a DX sensitivity and specificity of 95%, the number of unnecessary surgeries was reduced to 323, and 175 malignant cases were not diagnosed (Table 5.4). Not surprisingly, the substantial reduction in the number of surgeries, was associated with a reduction in surgery- related adverse events (Table 5.4). Specifically, 129 and 29 patients experienced persistent hypoparathyroidism and permanent RLNI, respectively in the NoDX arm, compared to only 68 and 29 in DX arm, respectively. Overall, using the DX test resulted in gaining 0.046 (95% credible interval: 0.019 to 0.078) QALYs per patient, and a cost savings of $1,087 (95% credible interval: $691 to $1533) in direct costs per patient. Thus, if the cost of the DX test is less than $1,087, the DX test strategy would result in QALYs gained and lower costs relative to using the NoDX strategy.  111  As expected, varying the sensitivity and specificity of the DX test affected the results (Figure 5.2). Given a specificity of 95%, the incremental QALYs associated with using the DX testing strategy with a sensitivity of 87.5% or lower will be negative, making the DX testing strategy inferior to the NoDX strategy in terms of clinical outcomes. As it is shown in Figure 5.2, the incremental QALYs gained varies widely under different assumptions about the sensitivity of the DX testing strategy, while changes in specificity had a smaller effect on the incremental QALYs.  112  Table 5.4 Base case results, indeterminate cases Final cytology Non-diagnostic or Unsatisfactory (ND) Benign Atypia of Undetermined Significance (AUS) Follicular Neoplasm (FN) Suspicious for Malignancy Malignant Final histology True negative True positive False positive False negative Performance measures Sensitivity Specificity Positive Predictive Value (PPV) Negative Predictive Value (NPV) Malignant cases Papillary carcinoma Follicular carcinoma Medullary carcinoma Hurthle cell carcinoma Anaplastic carcinoma Major adverse events Persistent hypoparathyroidism Permanent laryngeal recurrent nerve injury Cancer related mortality Outcomes QALY loss Cost (Excluding DX Test Cost) Incremental QALY Incremental costa  a  No DX  DX  0 1479 454 3971 3987 109  0 6182 0 0 0 3818  1898 3579 4407 116  6007 3495 323 175  0.97 0.30 0.45 0.94  0.95 0.95 0.92 0.97  2924 503 172 75 21  2936 414 195 92 33  129 89 391  68 29 393  0.306 4638  0.266 3558 0.040 -1080  Negative cost indicates saving  Inclusion of indirect costs that are generally related to productivity loss during surgical procedures and inpatient hospital stays increased the estimated saving attributable to the DX testing to $1251 per patient (i.e. incremental cost equivalent to -$1251) under base case conditions.  113  Table 5.5 Results for all patients with palpable nodules Final cytology Non-diagnostic or Unsatisfactory (ND) Benign Atypia of Undetermined Significance (AUS) Follicular Neoplasm (FN) Suspicious for Malignancy Malignant Final histology True negative True positive False positive False negative Performance measures Sensitivity Specificity Positive Predictive Value (PPV) Negative Predictive Value (NPV) Malignant cases Papillary carcinoma Follicular carcinoma Hurthle cell carcinoma Medullary carcinoma Anaplastic carcinoma Major adverse events Persistent hypoparathyroidism Permanent laryngeal recurrent nerve injury Cancer related mortality Outcomes QALY loss Cost (Excluding DX Test Cost) Incremental QALY Incremental costa  a  No DX 448 7206 105 850 850 541  DX 0 8267 0 0 0 1733  7221 1312 1362 105  8185 1276 457 82  0.93 0.84 0.49 0.99  0.94 0.95 0.74 0.99  1126 176 76 31 8  1065 178 59 40 16  42 28 175  35 12 181  0.123 2341  0.098 2050 0.025 -291  Negative cost indicates saving  Implementation of the DX testing strategy was associated with a one-time cost per patient that is part of the overall direct cost in the DX arm. Figure 5.3 shows the net monetary benefit of using different costs for the the DX testing strategy under the assumptions of the base case scenario and using an arbitrary value of $50,000 per QALY gained as a threshold for willingness to pay (λ).  114  When we simulated 10,000 patients with indeterminate FNAB results in the NoDX testing strategy arm (current practice), 3695 patients had cancer and 3579 of those were assigned to surgery (either hemithyroidecomy or total thyroidecomy). However, 4407 patients with benign thyroid nodules also were assigned to receive an unnecessary surgery. Thus, we concluded that for indeterminate cases, sensitivity, specificity, PPV, and NPV of current practice (FNAB combined with the TBSRTC guidelines) are 97%, 30%, 45%, and 94%, respectively. When we considered all palpable nodules, the associated number were 94%, 84%, 49%, and 99%, respectively.  115  Figure 5.2 Effect of sensitivity and specificity of DX test on the outcomes  Incremental Cost and Effectiveness for Different Sensitivity and Specificty of DX Test 500 250 0  Incremental Cost ($)  -0.04  -0.02  0  0.02  0.04  0.06  0.08  -250 Fixed Specificity (95%)  -500 -750 Specificity =80%  -1000  Sensitivity=100%  Sensitivity=80%  -1250  Specificity =100%  Incremental QALY  Each consecutive point represent 2.5% incerase in sensetivity or specicifty  116  Figure 5.3 Net monetary benefit of DX test for different test costs  Net Monetary Benefit of DX Test for Different Levels of Cost per Test Net Monetary Benefit ($)  4,000 3,000  Sensitivity=95%, Specificity=95%  2,000 "Sensitivity=95%, Specificity=85%"  1,000 0  500  1000 1500 2000 2500 3000  (1,000)  "Sensitivity=85%, Specificity=95%"  (2,000) (3,000)  5.3.2 Results of the Sensitivity Analyses The results of the univariate sensitivity analyses suggest that the cost of hemithyroidectomy, the discount rate, and the risk of permanent hypoparathyroidism have significant effects on the incremental net monetary benefit (Figure 5.4). A higher probability of malignancy among patients with a cytological diagnosis of suspicious, FN, and AUS was associated with a higher incremental net monetary benefit for the DX testing strategy. However, when the prevalence of malignancy among benign cytology was set to the maximum of its range for the sensitivity analyses (3%), the incremental net benefit decreased slightly. When QALYs and costs were not discounted (0% discount 117  rate) the incremental net monetary benefit of DX versus NoDX increased to $3,370 compared with $3,055 at a 3% discount rate in the base case scenario. The results of univariate sensitivity analyses showed that neither of these changes in the input parameters could alter the overall outcomes and the DX test save QALY and cost under a wide range of assumptions about input parameters. We repeated the simulation for all patients with nodules. Sensitivity and specificity of the NoDX strategy was estimated to be 94% and 84%, respectively. Assuming that the DX test could maintain sensitivity and specificity of 95% in this scenario, it resulted in 0.025 incremental QALYs gained (i.e. 25 per 1,000 patients assessed) and $291 in cost-savings. The results of the probabilistic sensitivity analysis also suggest a high probability that the DX is the dominant strategy (Figure 5.5). All simulated points fall in the south east quadrant (the DX testing strategy being more effective but less costly than the NoDx strategy) if the cost per test was less than $500. Therefore, assuming 95% sensitivity and specificity, a DX testing strategy that costs less than $500 is the dominant strategy with 100% certainty.  118  2500 Relative risk of cancer recurrence delayed diag (2.5 to 2.1)  Ablation, direct costs (2992 to 4488)  Utility surgery (0.70 to 0.57)  Relative risk of mortality delayed diag (2.4 to 2.0)  Prob Bilateral RLNI (0.23% to 0.18%)  Total thyroidecomy, direct costs (3670 to 2446)  Probablity of cancer given benign cytology (0%, 3%)  Probablity of cancer given AUS cytology (15% to 5%)  Probablity of cancer given FN cytology (30% to 15%)  Probablity of cancer given suspicious cytology (75% to 60%)  Prob hypoparathyroidism (0.86% to 0.70%)  Discount rate (5% to 0%)  Hemithyroidectomy, direct costs (1912 to 2868)  Net Monetary Benefit ($)  Figure 5.4 Results of univariate sensitivity analysis  Univariate Sensitivity Analysis  3500  3250  3000  2750  119  Figure 5.5 Results of probabilistic sensitivity analysis  Incremental Cost Effectiveness Plane, DX versus NoDX 500 0  Incremental Cost ($)  -0.04 -0.02  0  0.02  0.04  0.06  0.08  0.1  0.12  -500 -1000 -1500 -2000  Incremental QALY  5.4 Discussion Given 95% sensitivity and specificity, utilization of the DX testing strategy to improve the preoperative diagnosis of indeterminate thyroid nodules can substantially reduce the number of diagnostic operations, and result in considerable QALY gains and cost savings. The expected cost-savings is inversely related to the cost of the DX testing strategy, but remains positive for any cost per DX test less than $1087. Provided a comparable sensitivity of the DX and NoDX strategies, the DX testing strategy remains cost-saving for any specificity >80%. In contrast, a small decrement in sensitivity of the DX test results in  120  considerable negative effect on incremental QALYs gained. This is due to: 1) FNAB in conjunction with TBSRTC guidelines already provides high sensitivity at the expense of very low specificity for indeterminate cases; and 2) the negative clinical consequences of missing a malignant case is much greater than the expected adverse events associated with unnecessary surgery for a benign case. The value of the DX testing strategy is largely due to its ability to prevent unnecessary surgical procedures (false positives) that currently comprise about half of thyroid operations [179]. When we simulated 10,000 patients with indeterminate FNAB results in the NoDX testing strategy arm, 4407 patients with benign nodules were predicted to undergo a surgical procedure that may not be necessary. In comparison, this number is reduced to 323 in the DX arm. In this analysis, we compared the overall performance of the DX testing to the NoDX testing strategy in terms of the number of undetected cancer cases and the number of patients with benign nodules who were assigned to surgery. We defined “positive” as a patient who has been referred for surgery based on the overall evaluation of the diagnostic strategy, as we believe this definition is more clinically relevant. Therefore, the estimated sensitivity, specificity, PPV, and NPV reflect the overall ability of the strategy to assign malignant cases to undergo surgery with high accuracy. For example, the reported sensitivity for the NoDX testing strategy indicates characteristics of FNAB in combination with guidelines rather than FNAB alone. This study is the first to evaluate the incremental cost-effectiveness of using the DX testing strategy relative to the current recommended diagnostic strategy, and it is the 121  first study to evaluate the overall performance of FNAB in combination with the TBSRTC management guidelines. Prior studies, however, have explored the performance of FNAB alone for the diagnosis of thyroid nodules. Gharib et. al. [186] suggested that depending on the definition of cytological outcomes, the sensitivity of FNAB can range from 65% to 98% and the specificity can range from 72% and 100%. In a retrospective analysis of 37,895 cases reported by Ravetto et al [187], the sensitivity and specificity of FNAB were estimated to be 91.8% and 75.5%, respectively. In a review by Tee et al [188] the “observed” sensitivity and specificity of FNAB that has been reported also has a wide range, suggesting that the estimated characteristics of FNAB depend on the criteria and the method of measurement. Similar findings have also been reported by several other studies [180,189-192]. Thyroid cancer is fatal in a very small number of patients despite receiving the best possible treatment. We included the effect of cancer-related mortality on QALYs, but the costs that can occur in final stages of disease (e.g. palliative care) have not been accounted for in our model. However, the number of individuals who develop end stage thyroid cancer were virtually identical in the two arms of the model, and therefore would have limited impact on incremental costs or outcomes. The impact of surgery on QALYs was calculated only for the inpatient periods after surgery. In reality, discomfort and limitation in daily activities usually lasts for a period of time after discharge from the hospital, and therefore we underestimated the negative effect of undergoing surgery on QALYs. Given a higher frequency of surgeries in the NoDX strategy, this assumption slightly favored the NoDX testing strategy. Finally, our results are measured in a single 122  cohort of patients followed for 10 years and the expected benefits of using the DX testing strategy would be higher if we consider the incidence in cohorts during subsequent years. Overall, the DX testing strategy appears to be the dominant diagnostic strategy (QALY saving and cost saving) compared to current practice if it is proved to have high sensitivity and specificity with a reasonable cost per test. The DX testing strategy should be considered as a new diagnostic tool to reduce the number of operations performed for benign pathology in patients with nodular thyroid disease and to establish the accurate detection of malignant nodules.  123  Chapter 6: Integrated Discussions 6.1 Important Findings and Implications The growing body of literature about the implications of genomic on health care and health care services reflects the relative importance of this area. However, the majority of previous studies have used conceptual frameworks to speculate about the effects of genomic technologies on our health care system. In my research, I have provided four case studies that demonstrate the impact of genomics on patients, the public, physicians, and health care decision-makers. In addition, I have introduced the application of several methods including best-worst scaling experiment, system dynamics, and discrete event simulation to genomics and have demonstrated their analytical advantage in the evaluation of genomic technologies. I have demonstrated the substantial analytical advantages and empiric data that can be achieved by application of these methods despite their unfamiliarity to this area. Figure 1.1 provides a visual description of different topics that have been covered in this thesis. Each chapter in my thesis has resulted in significant findings for improving clinical care and also has described a useful method for economic evaluation of genomic technologies. Furthermore, in a larger scale this thesis offered a systematic discussion about influence of genomic technologies on different stakeholders, the nature of research questions that will likely arise, and possible methodological challenges associated with evaluation of genomic technologies (Figure 1.1). Finally, particular advantages of four different methodologies have been explained in each case study. I am 124  hopeful that these case studies have shown feasibility and benefit of BWS, Dynamic Systems, and DES in this area and have encouraged future utilization of these methods in health research services and in assessment of genomic technologies in particular. Chapter 3: Barriers to Integrating Personalized Medicine into Clinical Practice: A Best-Worst Scaling Experiment Chapter 4: Future Impact of Genomic and Proteomic Tests on the Burden of COPD: A System Dynamics Model Chapter 5: CostEffectiveness of Using a Molecular Diagnostic Test to Improve Diagnosis of Thyroid Cancer: A Discrete Event Simulation Model  Health Care Providers  Decision Makers  Genomics  Patients  Chapter 2: Genetic Testing to Determine Drug Response: Measuring Preferences of Patients and the Public Using Discrete Choice Experiment (DCE)  The Public  Figure 1.1: The overall structure of my thesis. This figure highlights the impact of genomics on heath care providers, patients, the public, and decision makers. Furthermore, this figure illustrates the case studies in each chapter of this thesis to demonstrate the potential research questions in relation with heath care providers, patients, the public, and decision maker and shows the methods that have been used for answering those questions.  In chapter 2, I measured the preferences of cancer patients and the public for a hypothetical, genetic test-guided cancer treatment. The extent that a genetic test will be used in practice is affected by factors such as sensitivity and specificity of the test, invasiveness of the testing procedure, the probability and severity of associated side effects, and cost. Using a discrete choice experiment (DCE), I elicited preferences from cancer patients and the public for different attributes of a hypothetical genetic test for 125  guiding cancer treatment. Three distinct samples completed a single DCE (sample 1: 50 lymphoma patients in British Columbia, Canada; sample 2 and sample 3: 588 and 578 individuals from the public across Canada respectively). Samples 1 and 2 considered the test/treatment in the context of a fast-acting curable cancer while the scenario for sample 3 was based on a slow-acting incurable cancer. I conducted latent class analyses to identify heterogeneity in participants’ preferences. I demonstrated that patients and the public have different preferences about the various aspects of genetic testing to guide cancer treatment. Individuals have different preferences when they are faced with an aggressive but curable cancer versus a non-aggressive and incurable cancer. These findings also raise the important question of whose preferences should be considered in relevant healthcare decision-making processes given that patients and the general public demonstrated different results. In chapter 3, I measured relative importance of factors that affect the decision of physicians to incorporate personalized medicine into their practice. Using a Best Worst Scaling (BWS) choice experiment, I estimated the relative importance of attributes which influence physicians’ decisions for integrating personalized medicine into clinical. Six attributes were included in the BWS: type of genetic tests, training for genetic testing, clinical guidelines, professional fee, privacy protection laws, and cost of genetic tests. A total of 197 physicians in BC completed the experiment. Using latent class analysis (LCA), we explored physicians’ preferences’ and the underlying heterogeneity of such preferences. The results will be useful in designing the policies for supporting physicians and facilitating the use of personalized medicine in future. 126  In chapter 4, I evaluated the impact of three potential interventions based on the results of genomic/ proteomic biomarkers (a screening test for early detection of COPD, a predictive test for exacerbations, and a test to identify progression into more severe disease stages) on COPD burden. Using a dynamic simulation model, I projected the total burden of COPD (cost, morbidity, and mortality) from 2011 to 2036 using the population of Canada as a case study. The model indicated that annual societal cost of COPD is $4.24B ($2.6B in direct costs) in 2011, and will reach $7.27B ($4.5B in direct costs) in 2036. The estimated costs and QALYs suggested that the best single strategy to reduce the growing burden of COPD is by reducing the number of exacerbations. Smoking cessation, while it is the cornerstone of COPD prevention, has only a modest effect in attenuating the financial burden of COPD over the next 26 years in Western countries such as Canada. Therefore I concluded that any intervention that can reduce the number of exacerbations has a substantial impact on morbidity and costs of COPD and should be considered in conjunction with the ongoing efforts to reduce smoking rates. In chapter 5, I investigated the cost-effectiveness of adding a new molecular test (DX) as a companion to fine-needle aspiration biopsy (FNAB) to improve the diagnosis of malignant thyroid nodules. I constructed a patient level simulation model to estimate the clinical and economic outcomes of using the DX test versus the current practice (No DX) for diagnosis of thyroid nodules. I demonstrated that the sensitivity of the DX test, comparing to specificity, has a larger influence on the overall outcomes. The results of this costeffectiveness study suggested that, conditional on the proposed range for sensitivity,  127  specificity, and cost of the DX test, the use of the DX test seems to be a dominant intervention compared to No DX and should be considered by policy makers. Several studies have discussed methodological challenges in the evaluation of genomic technologies and emphasized on the need to utilizing more diverse and more sophisticated analytical tools. Flowers and Veenstra [95] developed a framework for evaluation of pharmacogenomic strategies that can be used by researchers, pharmacists, physicians, and policy makers. They highlighted several aspects that differentiate between genomic-based interventions and other interventions that should be considered in cost effectiveness studies. Jarrett et al. [99] reviewed the literature about economic evaluation of genomic technologies with the purpose to assess quality of those studies. They concluded that both quantity and quality of studies about economic evaluation of genomic technologies were not sufficient. They argued that although genomic technologies do not raise new types of questions for health economists, some aspects (e.g. externalities) need to addressed in assessment of genomic technologies which require further methodological development in health economics. Stallings et al. [97] also developed a model to measure cost reduction by using a personalized medicine approach in Asthma treatment. The parameters in their model were populated based on a large retrospective insurance claim data and their results showed how gene prevalence and the cost of genetic test could affect final estimations. The particular importance of this study was in providing an example of using claim data for evaluation of genomic technologies. Phillips et al. [8] reviewed the evidence on utilization, preferences, and economic value of genetic testing. They concluded that launching large prospective 128  studies, design of more accurate preference measurement techniques, and using more sophisticated models in order to capture complexities of genomics are important areas that need to be further explored. Grosse at al. [102] also recommended that health policy decision makers should use methods to evaluate the value of genetic information for patients and other stakeholders. In a commentary by Payne and Shabaruddin [98] in Future Medicine, the importance of improving quality of health economic studies in the context of personalized medicine has been discussed. Conti et al. [94], summarized the challenges and opportunities of integrating molecular medicine into clinical care that was discussed at the annual meeting of the Society for Medical Decision Making on 23 October 2007. Finally, several recent studies have focused on methodological challenges of using genetic markers as instrumental variables for assessment of association between modifiable factors and disease outcomes [193-195]. In this thesis, I have provided several examples about applications of genomics in clinical practice. I have shown how DCE and BWS can be used to measure preferences of patients, the public, and physicians about applications of genomics. I also have shown that how Dynamic Systems and DES can be used to enhance the ability to evaluate gnomically guided interventions. This is particularly important for evidence based decision making. In summary, this research has contributed into our understanding of patients and the public attitude toward genomic testing and the barriers for assimilation of genomics into medical practice by physicians. The suggested modeling tools also can be added to the 129  toolbox of traditional methods for evaluation of genomic technologies and should be considered by decision makers. The results of this research, if supplemented with further research, can be used for the design of necessary regulations and guidelines and will help prepare our health care system to respond in such a way that maximizes the potential health benefits of genomic technologies.  6.2 Limitations of My Research There are several limitations in the current research. Validity and feasibility of preference elicitation methods relies upon a clear understanding of the choice questionnaire by the respondents. In many cases, a substantial portion of responders may not be familiar with advances in genomics and its implications. Explaining different aspects of genetic testing and their associated health impacts, and explaining the concepts that require understanding of risks and probabilities by responders can be challenging and may impact the overall results of a choice experiment. In fact, comprehension and interpretation of the questions by respondents may vary depending on their familiarity with genetics. All choice experiments try to elicit the stated preferences of responders as a surrogate for their revealed preferences. However, stated preferences highly depend on how respondents envision the hypothetical scenarios that are presented to them. As such, stated preferences act only as a proxy of revealed preferences and the difference between the two should be determined in a real setting through observations. Nevertheless, this limitation should not prevent us from using stated preferences as the best feasible evidence for informing future decisions. Our sample recruitment in chapter  130  2 and chapter 3 was based on an Internet sampling methodology by a professional market research company. It is estimated that internet use is very high among Canadian households and individuals (about 84% of total population, http://www.internetworldstats.com/am/ca.htm). However, although the sample was closely matched to the demographics of the Canadian population, I acknowledge this sample does not completely represent general Canadian population. Internet users presumably are more amenable to adopting new technologies. However we believe that the advantages of sample recruitment using these methods (by shortening time needed for recruitment, less cost and logistic issues, accuracy of methods for data collection) can partly offset mentioned weaknesses. Overall, the evaluation of genomic technologies involves making substantial assumptions about efficacy, cost, and other characteristics since many of these are still unknown for many applications which make genomics a more challenging area for economic evaluation. Our case studies are not exceptions in this regard and my evaluations also relied on several hypothetical scenarios. Finally, I have not discussed several other methodological concepts such as prospect theory 6 and methods for economic analysis of observational data that were relevant to section 1.3. However, I believe that those are major topics with substantial applications in genomics and should be explored further in future studies.  6  Prospect theory proposed by Kahnemann and Tversky (Kahnemann 1973), is a descriptive utility theory that provides a general framework to explain observed deviations from expected utility theory.  131  6.3 Future Direction of Research Addressing the ethical, legal, and social issues that are associated with the increasing availability of genomic information requires further research. As the demand for genetic testing increases, individuals, physicians, government, and private companies will have access to information that did not exist before. This information, depending on who has access, can create numerous conflicting issues (10,51) including what main-stream economists classify as asymmetric information, moral hazard, adverse selection and externalities. In the rest of this section, I provide a list of potential research questions that can be targeted in future research. As previous studies have shown, people who are planning for pregnancy put more value on genetic information [57]. It is important to understand the extent and nature of the effect that genetic testing can potentially have on the reproductive decisions of families. Whether individuals incorporate information they learn from predictive genetic testing to adjust their insurance coverage is also an open question. The actual effect on the insurance market can be substantial as this genetic information can create a substantial adverse selection effect in the absence of appropriate policy measures. Access to genomic information, in a similar way, may affect time preferences and economic decisions of individuals such as trade-offs between consumption and savings, and between work and leisure. The results of predictive genomic tests also can affect individuals’ life styles, and influence fatalistic or preventative behavior among those who learn about their potential genetic risks.  132  6.4 Knowledge Transfer This research can be used to inform regulatory authorities and policy makers in Canada for future impacts of genomics on the health care system. I submitted the results in the form of abstracts, several presentations at scholarly conferences, and four manuscripts for publication in peer review journals to fulfill this goal. The results of each chapter of my thesis have direct impact on practitioners and decision makers. The of study in chapter 2 was conducted in close collaboration with researchers in BC cancer agency and provided useful information about preferences of patients and the public about using gnomically guided cancer treatments. This information will likely influence ongoing research on gnomically based diagnosis and treatments in cancer. In chapter 3, I studied the relative importance of the factors that affect physicians’ decision to use genomic testing in their practice. This was a high priority question for BC clinical genomic network that aims at expanding clinical application of genomics by physicians in BC. I presented the results of this study in several conferences including CADTH annual meeting and a continuing medical education (CME) workshop that was specifically designed to update physicians’ knowledge about recent applications of genomics in medicine. A version of this chapter has been accepted for publication in Genetics in Medicine [Najafzadeh M, Lynd LD, Davis JC, Bryan S, Anis A, Marra MA, Marra CA. Barriers for integrating personalized medicine into clinical practice: a best worst scaling choice experiment. (GIM-D-11-00182 R1)].  133  In Chapter 4, I developed the first dynamic model to extrapolate long term burden of COPD in Canada. I am planning to transform this model to an online application that can be accessed and run on the web. This will add the possibility of observing the model outcomes by changing the parameters and assumptions of the model. Furthermore, the main massage of this study- the significant impact of strategies that can reduce COPD exacerbations- emphasizes on the areas of COPD research that have higher potential benefits in the long term. For example, PROOF Center of Excellence already used this finding to prioritize their molecular biomarker research for COPD. This is an example where economic evaluation can help identify new technologies with greater clinical benefit and economic return for tax payers. Finally, the results of chapter 5, shows a clear example that a new technology can generate additional clinical benefit and be cost saving at the same time. I have communicated the results of this study with several specialists and surgeons in the field and this will likely contribute to decreasing the number of diagnostic thyroid surgeries in future.  134  References 1 2 3 4 5 6 7 8 9 10 11 12  13 14 15 16 17  18  Guttmacher AE, McGuire AL, Ponder B, Stefansson K: Personalized genomic information: Preparing for the future of genetic medicine. Nat Rev Genet 2010;11:161-165. Rogowski WH, Grosse SD, Khoury MJ: Challenges of translating genetic tests into clinical and public health practice. Nat Rev Genet 2009;10:489-495. Burke W, Psaty BM: Personalized medicine in the era of genomics. Jama 2007;298:16821684. Feero WG, Guttmacher AE, Collins FS: The genome gets personal--almost. Jama 2008;299:1351-1352. McGuire AL, Cho MK, McGuire SE, Caulfield T: Medicine. The future of personal genomics. Science 2007;317 Shurin SB, Nabel EG: Pharmacogenomics--ready for prime time? N Engl J Med 2008;358:1061-1063. Collins F: Has the revolution arrived? Nature 2010;464:674-675. Phillips KA: Closing the evidence gap in the use of emerging testing technologies in clinical practice. Jama 2008;300:2542-2544. Lee C, Morton CC: Structural genomic variation and personalized medicine. N Engl J Med 2008;358:740-741. Robertson GR, Grant DM, Piquette-Miller M: Pharmacogenetics of pharmacoecology: Which route to personalized medicine? Clin Pharmacol Ther 2009;85:343-348. Topol EJ: Genome scanning and cardiovascular disease. Heart 2008;94:1361-1363. Davey Smith G, Ebrahim S, Lewis S, Hansell AL, Palmer LJ, Burton PR: Genetic epidemiology and public health: Hope, hype, and future prospects. Lancet 2005;366:1484-1498. Scheuner MT, Sieverding P, Shekelle PG: Delivery of genomic medicine for common chronic adult diseases: A systematic review. Jama 2008;299:1320-1334. Genomeweb: As illumina cuts price of personal sequencing service, ceo says market growth hinges on analysis, 2010, Shendure J, Ji H: Next-generation DNA sequencing. Nat Biotechnol 2008;26:1135-1145. Collins FS, Green ED, Guttmacher AE, Guyer MS: A vision for the future of genomics research. Nature 2003;422:835-847. Kidd JM, Cooper GM, Donahue WF, Hayden HS, Sampas N, Graves T, Hansen N, Teague B, Alkan C, Antonacci F, Haugen E, Zerr T, Yamada NA, Tsang P, Newman TL, Tuzun E, Cheng Z, Ebling HM, Tusneem N, David R, Gillett W, Phelps KA, Weaver M, Saranga D, Brand A, Tao W, Gustafson E, McKernan K, Chen L, Malig M, Smith JD, Korn JM, McCarroll SA, Altshuler DA, Peiffer DA, Dorschner M, Stamatoyannopoulos J, Schwartz D, Nickerson DA, Mullikin JC, Wilson RK, Bruhn L, Olson MV, Kaul R, Smith DR, Eichler EE: Mapping and sequencing of structural variation from eight human genomes. Nature 2008;453:56-64. Kaiser J: DNA sequencing. A plan to capture human diversity in 1000 genomes. Science 2008;319:395.  135  19 20 21 22 23  24 25 26 27 28 29  30 31 32 33 34  35 36 37 38  Allison M: Is personalized medicine finally arriving? Nat Biotechnol 2008;26:509-517. Wilkinson GR: Drug metabolism and variability among patients in drug response. N Engl J Med 2005;352:2211-2221. Bayley H: Nanotechnology: Holes with an edge. Nature 2010;467:164-165. Mardis ER: A decade's perspective on DNA sequencing technology. Nature 2011;470:198203. Ikediobi ON, Shin J, Nussbaum RL, Phillips KA, Walsh JM, Ladabaum U, Marshall D: Addressing the challenges of the clinical application of pharmacogenetic testing. Clin Pharmacol Ther 2009;86:28-31. Shin J, Kayser SR, Langaee TY: Pharmacogenetics: From discovery to patient care. Am J Health Syst Pharm 2009;66:625-637. Phillips KA, Veenstra DL, Oren E, Lee JK, Sadee W: Potential role of pharmacogenomics in reducing adverse drug reactions: A systematic review. Jama 2001;286:2270-2279. Joshi VA, Kucherlapati R: Genetics and genomics in the practice of medicine. Gastroenterology 2008;134:1284-1288. Niederhuber JE: Translating discovery to patient care. Jama 2010;303:1088-1089. Ali-Khan SE, Daar AS, Shuman C, Ray PN, Scherer SW: Whole genome scanning: Resolving clinical diagnosis and management amidst complex data. Pediatr Res 2009;66:357-363. Janssens AC, Gwinn M, Bradley LA, Oostra BA, van Duijn CM, Khoury MJ: A critical appraisal of the scientific basis of commercial genomic profiles used to assess health risks and personalize health interventions. Am J Hum Genet 2008;82:593-599. Janssens AC, van Duijn CM: Genome-based prediction of common diseases: Advances and prospects. Hum Mol Genet 2008;17:R166-173. Wolfberg AJ: Genes on the web--direct-to-consumer marketing of genetic testing. N Engl J Med 2006;355:543-545. Bloss CS, Schork NJ, Topol EJ: Effect of direct-to-consumer genomewide profiling to assess disease risk. N Engl J Med 2011;364:524-534. Kuehn BM: Risks and benefits of direct-to-consumer genetic testing remain unclear. Jama 2008;300:1503-1505. McBride CM, Alford SH, Reid RJ, Larson EB, Baxevanis AD, Brody LC: Characteristics of users of online personalized genomic risk assessments: Implications for physician-patient interactions. Genet Med 2009;11:582-587. Foster MW, Sharp RR: Out of sequence: How consumer genomics could displace clinical genetics. Nat Rev Genet 2008;9:419. McGuire AL, Burke W: An unwelcome side effect of direct-to-consumer personal genome testing: Raiding the medical commons. Jama 2008;300:2669-2671. Issa AM, Tufail W, Hutchinson J, Tenorio J, Baliga MP: Assessing patient readiness for the clinical adoption of personalized medicine. Public Health Genomics 2009;12:163-169. Sanderson SC, Michie S: Genetic testing for heart disease susceptibility: Potential impact on motivation to quit smoking. Clin Genet 2007;71:501-510.  136  39  40 41 42 43 44 45 46  47  48 49 50 51 52 53 54 55  56  Palmer RC, Emmons KM, Fletcher RH, Lobb R, Miroshnik I, Kemp JA, Bauer M: Familial risk and colorectal cancer screening health beliefs and attitudes in an insured population. Prev Med 2007;45:336-341. Juengst E: Genetic testing and the moral dynamics of family life. Public Understanding of Science 1999;8:193-205. Ganz ML: Family health effects: Complements or substitutes. Health Econ 2001;10:699714. Patenaude AF: The genetic testing of children for cancer susceptibility: Ethical, legal, and social issues. Behav Sci Law 1996;14:393-410. Schlich-Bakker KJ, ten Kroode HF, Ausems MG: A literature review of the psychological impact of genetic testing on breast cancer patients. Patient Educ Couns 2006;62:13-20. Shaw C, Abrams K, Marteau TM: Psychological impact of predicting individuals' risks of illness: A systematic review. Soc Sci Med 1999;49:1571-1598. Heshka JT, Palleschi C, Howley H, Wilson B, Wells PS: A systematic review of perceived risks, psychological and behavioral impacts of genetic testing. Genet Med 2008;10:19-32. Wright AJ, Takeichi C, Whitwell SC, Hankins M, Marteau TM: The impact of genetic testing for crohn's disease, risk magnitude and graphical format on motivation to stop smoking: An experimental analogue study. Clin Genet 2008;73:306-314. Carpenter MJ, Strange C, Jones Y, Dickson MR, Carter C, Moseley MA, Gilbert GE: Does genetic testing result in behavioral health change? Changes in smoking behavior following testing for alpha-1 antitrypsin deficiency. Ann Behav Med 2007;33:22-28. Robins Wahlin TB: To know or not to know: A review of behaviour and suicidal ideation in preclinical huntington's disease. Patient Educ Couns 2007;65:279-287. Shiloh S: Genetic counseling: Psychological models in research and practice. Encyclopedia of Life Sciences 2006 Acheson LS, Crabtree BF: How do people interpret their family histories of diabetes, coronary disease, or cancer? Ann Fam Med 2004;2:532-533. Marteau TM, Croyle RT: The new genetics. Psychological responses to genetic testing. Bmj 1998;316:693-696. Marteau TM, Roberts S, LaRusse S, Green RC: Predictive genetic testing for alzheimer's disease: Impact upon risk perception. Risk Anal 2005;25:397-404. Woloshin S, Schwartz LM, Moncur M, Gabriel S, Tosteson AN: Assessing values for health: Numeracy matters. Med Decis Making 2001;21:382-390. Wang C, Gonzalez R, Merajver SD: Assessment of genetic testing and related counseling services: Current research and future directions. Soc Sci Med 2004;58:1427-1442. Ho SM, Ho JW, Chan CL, Kwan K, Tsui YK: Decisional consideration of hereditary colon cancer genetic test results among hong kong chinese adults. Cancer Epidemiol Biomarkers Prev 2003;12:426-432. Houfek JF, Atwood JR: Genetic susceptibility to lung cancer: Implications for smoking cessation. Medsurg Nurs 2003;12:45-49.  137  57  58 59 60 61  62 63 64 65 66 67 68 69  70 71 72 73 74 75 76  Hall J, Fiebig DG, King MT, Hossain I, Louviere JJ: What influences participation in genetic carrier testing? Results from a discrete choice experiment. J Health Econ 2006;25:520537. Neumann PJ, Hammitt JK, Mueller C, Fillit HM, Hill J, Tetteh NA, Kosik KS: Public attitudes about genetic testing for alzheimer's disease. Health Aff 2001;20:252-264. Lu L, Macdonald A, Wekvete C: Premium rates based on genetic studies: How reliable are they? Insurance: Mathematics and Economics 2008;42:319-331. Knoppers BM, Joly Y: Physicians, genetics and life insurance. Cmaj 2004;170:1421-1423. Viswanathan KS, Lemaire J, Withers K, Armstrong K, Baumritter A, Hershey JC, Pauly MV, Asch DA: Adverse selection in term life insurance purchasing due to the brca1/2 genetic test and elastic demand. 2007;74:65-86. Hoy M, Orsi F, Eisinger F, Moatti JP: The impact of genetic testing on healthcare insurance. Geneva Papers on Risk & Insurance 2003;28:203-221. Wilson D: Acquisition and disclosure of genetic information under alternative policy regimes: An economic analysis. Health Econ Policy Law 2006;1:263-276. Tabarrok A: Genetic testing: An economic and contractarian analysis. J Health Econ 1994;13:75-91. Holm HJ: Genetic information and investment in human capital. J Health Econ 1997;16:435-452. Hoy M, Witt J: Welfare effects of banning genetic information in the life insurance market: The case of brca1/2 genes. Journal of Risk and Insurance 2007;74:523-546. Hoy M, Polborn M: The value of genetic informationin life insurance market. Journal of Public Economics 2000;78:235-252. Jerry RH, 2nd: Life, health, and disability insurance: Understanding the relationships. J Law Med Ethics 2007;35:80-89. Godard B, Raeburn S, Pembrey M, Bobrow M, Farndon P, Ayme S: Genetic information and testing in insurance and employment: Technical, social and ethical issues. Eur J Hum Genet 2003;11:S123-142. Pokorski RJ: Genetic information and life insurance. Nature 1995;376:13-14. Raithatha N, Smith RD: Disclosure of genetic tests for health insurance: Is it ethical not to? Lancet 2004;363:395-396. Sankar P, Cho MK, Condit CM, Hunt LM, Koenig B, Marshall P, Lee SS, Spicer P: Genetic research and health disparities. Jama 2004;291:2985-2989. Aspinall MG, Hamermesh RG: Realizing the promise of personalized medicine. Harv Bus Rev 2007;85:108-117. Tracy EE: Are doctors prepared for direct-to-consumer advertising of genetics tests? Obstet Gynecol 2007;110:1389-1391. Fontanarosa PB, Pasche B, DeAngelis CD: Genetics and genomics for clinicians. Jama 2008;299:1364-1365. Levy H, Young JH: Perspectives from the clinic: Will the average physician embrace personalized medicine? Clin Pharmacol Ther 2008;83:492-493.  138  77  78 79  80 81 82  83  84  85  86 87  88  89 90 91 92  Teutsch SM, Bradley LA, Palomaki GE, Haddow JE, Piper M, Calonge N, Dotson WD, Douglas MP, Berg AO: The evaluation of genomic applications in practice and prevention (egapp) initiative: Methods of the egapp working group. Genet Med 2009;11:3-14. Suther S, Goodson P: Barriers to the provision of genetic services by primary care physicians: A systematic review of the literature. Genet Med 2003;5:70-76. Wideroff L, Freedman AN, Olson L, Klabunde CN, Davis W, Srinath KP, Croyle RT, BallardBarbash R: Physician use of genetic testing for cancer susceptibility: Results of a national survey. Cancer Epidemiol Biomarkers Prev 2003;12:295-303. Suther SG, Goodson P: Texas physicians' perceptions of genomic medicine as an innovation. Clin Genet 2004;65:368-377. Carroll JC, Brown JB, Blaine S, Glendon G, Pugh P, Medved W: Genetic susceptibility to cancer. Family physicians' experience. Can Fam Physician 2003;49:45-52. Freedman AN, Wideroff L, Olson L, Davis W, Klabunde C, Srinath KP, Reeve BB, Croyle RT, Ballard-Barbash R: Us physicians' attitudes toward genetic testing for cancer susceptibility. Am J Med Genet A 2003;120A:63-71. Gramling R, Nash J, Siren K, Culpepper L: Predictive genetics in primary care: Expectations for the motivational impact of genetic testing affects the importance family physicians place on screening for familial cancer risk. Genet Med 2003;5:172-175. Vadaparampil ST, Wideroff L, Olson L, Viswanath K, Freedman AN: Physician exposure to and attitudes toward advertisements for genetic tests for inherited cancer susceptibility. Am J Med Genet A 2005;135:41-46. Hindorff LA, Burke W, Laberge AM, Rice KM, Lumley T, Leppig K, Rosendaal FR, Larson EB, Psaty BM: Motivating factors for physician ordering of factor v leiden genetic tests. Arch Intern Med 2009;169:68-74. Kumar S, Gantley M: Tensions between policy makers and general practitioners in implementing new genetics: Grounded theory interview study. Bmj 1999;319:1410-1413. Schnoll RA, Shields AE: Physician barriers to incorporating pharmacogenetic treatment strategies for nicotine dependence into clinical practice. Clin Pharmacol Ther 2011;89:345-347. Shields AE, Levy DE, Blumenthal D, Currivan D, McGinn-Shapiro M, Weiss KB, Yucel R, Lerman C: Primary care physicians' willingness to offer a new genetic test to tailor smoking treatment, according to test characteristics. Nicotine Tob Res 2008;10:10371045. Ohata T, Tsuchiya A, Watanabe M, Sumida T, Takada F: Physicians' opinion for 'new' genetic testing in japan. J Hum Genet 2009;54:203-208. Levy DE, Youatt EJ, Shields AE: Primary care physicians' concerns about offering a genetic test to tailor smoking cessation treatment. Genet Med 2007;9:842-849. Fackler JL, McGuire AL: Paving the way to personalized genomic medicine: Steps to successful implementation. Curr Pharmacogenomics Person Med 2009;7:125. Meckley LM, Neumann PJ: Personalized medicine: Factors influencing reimbursement. Health Policy 2009;94:91-100.  139  93  94  95 96 97  98 99 100 101 102 103 104  105 106 107 108 109 110  111  Davis JC, Furstenthal L, Desai AA, Norris T, Sutaria S, Fleming E, Ma P: The microeconomics of personalized medicine: Today's challenge and tomorrow's promise. Nat Rev Drug Discov 2009;8:279-286. Conti R, Veenstra DL, Armstrong K, Lesko LJ, Grosse SD: Personalized medicine and genomics: Challenges and opportunities in assessing effectiveness, cost-effectiveness, and future research priorities. Med Decis Making 2010;30:328-340. Flowers CR, Veenstra D: The role of cost-effectiveness analysis in the era of pharmacogenomics. Pharmacoeconomics 2004;22:481-493. Issa AM: Evaluating the value of genomic diagnostics: Implications for clinical practice and public policy. Adv Health Econ Health Serv Res 2008;19:191-206. Stallings SC, Huse D, Finkelstein SN, Crown WH, Witt WP, Maguire J, Hiller AJ, Sinskey AJ, Ginsburg GS: A framework to evaluate the economic impact of pharmacogenomics. Pharmacogenomics 2006;7:853-862. Payne K, Shabaruddin FH: Cost-effectiveness analysis in pharmacogenomics. Pharmacogenomics 2010;11:643-646. Jarrett J, Mugford M: Genetic health technology and economic evaluation: A critical review. Appl Health Econ Health Policy 2006;5:27-35. Epstein RS, Teagarden JR: Comparative effectiveness research and personalized medicine: Catalyzing or colliding? Pharmacoeconomics 2010;28:905-913. Garber AM, Tunis SR: Does comparative-effectiveness research threaten personalized medicine? N Engl J Med 2009;360:1925-1927. Grosse SD, Wordsworth S, Payne K: Economic methods for valuing the outcomes of genetic testing: Beyond cost-effectiveness analysis. Genet Med 2008;10:648-654. Simon R: Clinical trials for predictive medicine: New challenges and paradigms. Clin Trials 2010;7:516-524. Crown WH: There's a reason they call them dummy variables: A note on the use of structural equation techniques in comparative effectiveness research. Pharmacoeconomics 2010;28:947-955. Simon RM: An agenda for clinical trials: Clinical trials in the genomic era. Clin Trials 2004;1:468-470. A different approach to health state valuation. Value Health 2009;12 Suppl 1:S16-17. Jehle GA, Reny PJ: Advanced microeconomic theory, ed 3rd. Prentice Hall, 2011. Mas-Colell A, Whinston MD, ; GJR, Whinston MD: Microeconomic theory Oxford University Press, 1995. Lancsar E, Louviere J: Conducting discrete choice experiments to inform healthcare decision making: A user's guide. Pharmacoeconomics 2008;26:661-677. Peacock S, Apicella C, Andrews L, Tucker K, Bankier A, Daly MB, Hopper JL: A discrete choice experiment of preferences for genetic counselling among jewish women seeking cancer genetics services. Br J Cancer 2006;95:1448-1453. Regier DA, Ryan M, Phimister E, Marra CA: Bayesian and classical estimation of mixed logit: An application to genetic testing. J Health Econ 2009;28:598-610.  140  112 113  114 115 116  117 118 119 120  121 122 123  124  125 126  127 128  129  Herbild L, Gyrd-Hansen D, Bech M: Patient preferences for pharmacogenetic screening in depression. Int J Technol Assess Health Care 2008;24:96-103. Griffith GL, Edwards RT, Williams JM, Gray J, Morrison V, Wilkinson C, Turner J, France B, Bennett P: Patient preferences and national health service costs: A cost-consequences analysis of cancer genetic services. Fam Cancer 2008;27:27. Train KE: Discrete choice methods with simulation. 2003 Greene WH, Hensher DA: A latent class model for discrete choice analysis: Contrasts with mixed logit. Transportation Research Part B: Methodological 2003;37:681-698. Louviere J: What you don’t know might hurt you: Some unresolved issues in the design and analysis of discrete choice experiments. Environmental and Resource Economics 2006;34:173-188. Bech M, Gyrd-Hansen D: Effects coding in discrete choice experiments. Health Economics 2005;14:1079-1083. Flynn TN, Louviere JJ, Peters TJ, Coast J: Best--worst scaling: What it can do for health care research and how to do it. J Health Econ 2007;26:171-189. Marley AAJ, Louviere JJ: Some probabilistic models of best, worst, and best-worst choices. Journal of Mathematical Psychology 2005;49:464-480. Louviere JJ, Islam T, Wasi N, Street D, Burgess L: Designing discrete choice experiments: Do optimal designs come at a price? Journal of Consumer Research: An Interdisciplinary Quarterly 2008;35:360-375. Train K: Discrete choice methods with simulation. Cambridge University Press, 2003. Hole AR: Modelling heterogeneity in patients' preferences for the attributes of a general practitioner appointment. Journal of Health Economics 2008;27:1078-1094. Weinstein MC, Toy EL, Sandberg EA, Neumann PJ, Evans JS, Kuntz KM, Graham JD, Hammitt JK: Modeling for health care and other policy decisions: Uses, roles, and validity. Value Health 2001;4:348-361. Sculpher M, Fenwick E, Claxton K: Assessing quality in decision analytic cost-effectiveness models. A suggested framework and example of application. Pharmacoeconomics 2000;17:461-477. Brennan A, Chick SE, Davies R: A taxonomy of model structures for economic evaluation of health technologies. Health economics 2006;15:1295-1310. Borshchev A, Filippov A: From system dynamics and discrete event to practical agent based modeling: Reasons, techniques, tools: The 22nd International Conference of the System Dynamics Society. Oxford, England, July 25 - 29, 2004., Caro JJ: Pharmacoeconomic analyses using discrete event simulation. Pharmacoeconomics 2005;23:323-332. Najafzadeh M, Marra CA, Sadatsafavi M, Aaron SD, Sullivan SD, Vandemheen KL, Jones PW, Fitzgerald JM: Cost effectiveness of therapy with combinations of long acting bronchodilators and inhaled steroids for treatment of copd. Thorax 2008;63:962-967. Lynd LD, Najafzadeh M, Colley L, Byrne MF, Willan AR, Sculpher MJ, Johnson FR, Hauber AB: Using the incremental net benefit framework for quantitative benefit-risk analysis in  141  130 131  132 133  134  135 136 137 138  139 140 141 142 143 144 145  146  regulatory decision-making--a case study of alosetron in irritable bowel syndrome. Value Health 2009;13:411-417. O'Brien BJ, Briggs AH: Analysis of uncertainty in health care cost-effectiveness studies: An introduction to statistical issues and methods. Stat Methods Med Res 2002;11:455-468. Claxton K, Sculpher M, McCabe C, Briggs A, Akehurst R, Buxton M, Brazier J, O'Hagan T: Probabilistic sensitivity analysis for nice technology assessment: Not an optional extra. Health Econ 2005;14:339-347. Claxton K: Exploring uncertainty in cost-effectiveness analysis. Pharmacoeconomics 2008;26:781-798. O'Hagan A, McCabe C, Akehurst R, Brennan A, Briggs A, Claxton K, Fenwick E, Fryback D, Sculpher M, Spiegelhalter D, Willan A: Incorporation of uncertainty in health economic modelling studies. Pharmacoeconomics 2005;23:529-536. Halpern EF, Weinstein MC, Hunink MG, Gazelle GS: Representing both first- and secondorder uncertainties by monte carlo simulation for groups of patients. Med Decis Making 2000;20:314-322. Felli JC, Hazen GB: Sensitivity analysis and the expected value of perfect information. Med Decis Making 1998;18:95-109. Claxton KP, Sculpher MJ: Using value of information analysis to prioritise health research: Some lessons from recent uk experience. Pharmacoeconomics 2006;24:1055-1068. Coyle D, Oakley J: Estimating the expected value of partial perfect information: A review of methods. Eur J Health Econ 2008;9:251-259. Walton SM, Schumock GT, Lee KV, Alexander GC, Meltzer D, Stafford RS: Prioritizing future research on off-label prescribing: Results of a quantitative evaluation. Pharmacotherapy 2008;28:1443-1452. Basu A, Meltzer D: Value of information on preference heterogeneity and individualized care. Med Decis Making 2007;27:112-127. Torpy JM, Lynm C, Glass RM: Jama patient page. Cancer: The basics. Jama 2010;304:1628. Mullighan CG: New strategies in acute lymphoblastic leukemia: Translating advances in genomics into clinical practice. Clin Cancer Res;17:396-400. Wolff AC: Liposomal anthracyclines and new treatment approaches for breast cancer. Oncologist 2003;8 Suppl 2:25-30. Capdeville R, Silberman S, Dimitrijevic S: Imatinib: The first 3 years. Eur J Cancer 2002;38 Suppl 5:S77-82. Louviere JJ, Hensher DA, Swait JD: Stated choice models, analysis and applications. Cambridge University Press, 2000. Kontoleon A, Yabe M: Assessing the impacts of alternative 'opt-out' formats in choice experiment studies: Consumer preferences for genetically modified content and production information in food. Journal of Agricultural Policy Research 2005;5:1-43. Service RF: Gene sequencing. The race for the $1000 genome. Science 2006;311:15441546.  142  147 148 149 150  151 152  153 154 155 156 157 158 159  160  161  162  Snyder M, Du J, Gerstein M: Personal genome sequencing: Current approaches and challenges. Genes Dev 2010;24:423-431. Bosch JR, Grody WW: Keeping up with the next generation: Massively parallel sequencing in clinical diagnostics. J Mol Diagn 2008;10:484-492. Podolak E: Sequencing's new race. Biotechniques 2010 48:105-111. Wilkins-Haug L, Erickson K, Hill L, Power M, Holzman GB, Schulkin J: Obstetriciangynecologists' opinions and attitudes on the role of genetics in women's health. J Womens Health Gend Based Med 2000;9:873-879. Marley AAJ, Flynn TN, Louviere JJ: Probabilistic models of set-dependent and attributelevel best-worst choice. Journal of Mathematical Psychology 2008;52:281-296. Louviere JJ, Street D, Burgess L, Wasi N, Islam T, Marley AAJ: Modeling the choices of individual decision-makers by combining efficient choice experiment designs with extra preference information. Journal of choice modelling 2008;1:128. Mannino DM, Buist AS: Global burden of copd: Risk factors, prevalence, and future trends. Lancet 2007;370:765-773. Minino AM, Xu J, Kochanek KD: Deaths: Preliminary data for 2008.; in National Vital Statistics Reports. Vol. 59 N (ed, 2008, Mathers CD, Loncar D: Projections of global mortality and burden of disease from 2002 to 2030. PLoS Med 2006;3:e442. Sin DD, McAlister FA, Man SF, Anthonisen NR: Contemporary management of chronic obstructive pulmonary disease: Scientific review. Jama 2003;290:2301-2312. Ramsey SD, Sullivan SD: Chronic obstructive pulmonary disease: Is there a case for early intervention? Am J Med 2004;20:3S-10S. Parkes G, Greenhalgh T, Griffin M, Dent R: Effect on smoking quit rate of telling patients their lung age: The step2quit randomised controlled trial. BMJ 2008;336:598-600. Hurst JR, Vestbo J, Anzueto A, Locantore N, Mullerova H, Tal-Singer R, Miller B, Lomas DA, Agusti A, Macnee W, Calverley P, Rennard S, Wouters EF, Wedzicha JA: Susceptibility to exacerbation in chronic obstructive pulmonary disease. N Engl J Med 2010;363:11281138. Spencer M, Briggs AH, Grossman RF, Rance L: Development of an economic model to assess the cost effectiveness of treatment interventions for chronic obstructive pulmonary disease. Pharmacoeconomics 2005;23:619-637. Rabe KF, Hurd S, Anzueto A, Barnes PJ, Buist SA, Calverley P, Fukuchi Y, Jenkins C, Rodriguez-Roisin R, van Weel C, Zielinski J: Global strategy for the diagnosis, management, and prevention of chronic obstructive pulmonary disease: Gold executive summary. Am J Respir Crit Care Med 2007;176:532-555. Buist AS, McBurnie MA, Vollmer WM, Gillespie S, Burney P, Mannino DM, Menezes AM, Sullivan SD, Lee TA, Weiss KB, Jensen RL, Marks GB, Gulsvik A, Nizankowska-Mogilnicka E: International variation in the prevalence of copd (the bold study): A population-based prevalence study. Lancet 2007;370:741-750.  143  163 164 165  166 167  168 169 170  171 172  173  174 175 176  177 178  179  Chapman KR, Bourbeau J, Rance L: The burden of copd in canada: Results from the confronting copd survey. Respir Med 2003;97:S23-31. Tashkin DP, Murray RP: Smoking cessation in chronic obstructive pulmonary disease. Respir Med 2009;13:13. Camp PG, Chaudhry M, Platt H, Roch M, Road J, Sin D, Levy RD: The sex factor: Epidemiology and management of chronic obstructive pulmonary disease in british columbia. Can Respir J 2008;15:417-422. Rabin R, de Charro F: Eq-5d: A measure of health status from the euroqol group. Ann Med 2001;33:337-343. Rutten-van Molken MP, Oostenbrink JB, Tashkin DP, Burkhart D, Monz BU: Does quality of life of copd patients as measured by the generic euroqol five-dimension questionnaire differentiate between copd severity stages? Chest 2006;130:1117-1128. Johnson JA, Pickard AS: Comparison of the eq-5d and sf-12 health surveys in a general population survey in alberta, canada. Med Care 2000;38:115-121. Population projections for canada, provinces and territories 2009 to 2036. Statistics Canada 2010 Gamble E, Grootendorst DC, Hattotuwa K, O'Shaughnessy T, Ram FS, Qiu Y, Zhu J, Vignola AM, Kroegel C, Morell F, Pavord ID, Rabe KF, Jeffery PK, Barnes NC: Airway mucosal inflammation in copd is similar in smokers and ex-smokers: A pooled analysis. Eur Respir J 2007;30:467-471. Hansel TT, Barnes PJ: New drugs for exacerbations of chronic obstructive pulmonary disease. Lancet 2009;374:744-755. Hoogendoorn M, Feenstra TL, Hoogenveen RT, Rutten-van Molken MP: Long-term effectiveness and cost-effectiveness of smoking cessation interventions in patients with copd. Thorax 2003;65:711-718. Hoogendoorn M, Rutten-van Molken MP, Hoogenveen RT, van Genugten ML, Buist AS, Wouters EF, Feenstra TL: A dynamic population model of disease progression in copd. Eur Respir J 2005;26:223-233. Liu JS, Chen R: Sequential monte carlo methods for dynamic systems. Journal of the American statistical association 1998;93:1032-1044. Mazzaferri EL: Management of a solitary thyroid nodule. N Engl J Med 1993;328:553-559. Borget I, Vielh P, Leboulleux S, Allyn M, Iacobelli S, Schlumberger M, de Pouvourville G: Assessment of the cost of fine-needle aspiration cytology as a diagnostic tool in patients with thyroid nodules. Am J Clin Pathol 2008;129:763-771. Layfield LJ, Cibas ES, Gharib H, Mandel SJ: Thyroid aspiration cytology: Current status. CA Cancer J Clin 2009;59:99-110. Hamberger B, Gharib H, Melton LJ, 3rd, Goellner JR, Zinsmeister AR: Fine-needle aspiration biopsy of thyroid nodules. Impact on thyroid practice and cost of care. Am J Med 1982;73:381-384. Cibas ES, Ali SZ: The bethesda system for reporting thyroid cytopathology. Am J Clin Pathol 2009;132:658-665.  144  180  181 182 183  184 185  186 187 188  189  190 191 192  193 194 195  Yassa L, Cibas ES, Benson CB, Frates MC, Doubilet PM, Gawande AA, Moore FD, Jr., Kim BW, Nose V, Marqusee E, Larsen PR, Alexander EK: Long-term assessment of a multidisciplinary approach to thyroid nodule diagnostic evaluation. Cancer 2007;111:508516. Davies L, Welch HG: Thyroid cancer survival in the united states: Observational data from 1973 to 2005. Arch Otolaryngol Head Neck Surg 2010;136:440-444. Shaha AR: Thyroid cancer: Extent of thyroidectomy. Cancer Control 2000;7:240-245. Rosato L, Avenia N, Bernante P, De Palma M, Gulino G, Nasi PG, Pelizzo MR, Pezzullo L: Complications of thyroid surgery: Analysis of a multicentric study on 14,934 patients operated on in italy over 5 years. World J Surg 2004;28:271-276. Sipos JA, Mazzaferri EL: Thyroid cancer epidemiology and prognostic variables. Clin Oncol (R Coll Radiol) 2010;22:395-404. Yeh MW, Demircan O, Ituarte P, Clark OH: False-negative fine-needle aspiration cytology results delay treatment and adversely affect outcome in patients with thyroid carcinoma. Thyroid 2004;14:207-215. Gharib H, Goellner JR: Fine-needle aspiration biopsy of the thyroid: An appraisal. Ann Intern Med 1993;118:282-289. Ravetto C, Colombo L, Dottorini ME: Usefulness of fine-needle aspiration in the diagnosis of thyroid carcinoma: A retrospective study in 37,895 patients. Cancer 2000;90:357-363. Tee YY, Lowe AJ, Brand CA, Judson RT: Fine-needle aspiration may miss a third of all malignancy in palpable thyroid nodules: A comprehensive literature review. Ann Surg 2007;246:714-720. Yang J, Schnadig V, Logrono R, Wasserman PG: Fine-needle aspiration of thyroid nodules: A study of 4703 patients with histologic and clinical correlations. Cancer 2007;111:306315. Morgan JL, Serpell JW, Cheng MS: Fine-needle aspiration cytology of thyroid nodules: How useful is it? ANZ J Surg 2003;73:480-483. Amrikachi M, Ramzy I, Rubenfeld S, Wheeler TM: Accuracy of fine-needle aspiration of thyroid. Arch Pathol Lab Med 2001;125:484-488. Cap J, Ryska A, Rehorkova P, Hovorkova E, Kerekes Z, Pohnetalova D: Sensitivity and specificity of the fine needle aspiration biopsy of the thyroid: Clinical point of view. Clin Endocrinol (Oxf) 1999;51:509-515. Fletcher JM: The promise and pitfalls of combining genetic and economic research. Health Econ 2011;20:889-892. Cawley J, Han E, Norton EC: The validity of genes related to neurotransmitters as instrumental variables. Health Econ 2011;20:884-888. von Hinke Kessler Scholder S, Smith GD, Lawlor DA, Propper C, Windmeijer F: Mendelian randomization: The use of genes in instrumental variable analyses. Health Econ 2011;20:893-896.  145  196  Rosenbaum WL, Sterling TD, Weinkam JJ: Use of multiple surveys to estimate mortality among never, current, and former smokers: Changes over a 20-year interval. Am J Public Health 1998;88:1664-1668.  146  Appendices Appendix A: Supplementary Material for Chapter 4 A.1  Interpolation of Rates in Subgroups  The subgroups in the dynamic model were defined based on smoking status (current smokers, past smokers, never smokers), COPD stage (no COPD, stage i, stage ii, stage iii or higher), and age group (40-49, 50-59, 60-69, ≥70 years). Therefore, input parameters for each subgroup in the model (e.g. prevalence, mortality rate, exacerbations) needed to be defined (3*4*4=48 possible subgroups). Prevalence rates for different smoking strata and prevalence rates for different age groups (i.e. marginal rates) are available [162]. However, prevalence rates stratified by smoking status and age group are not available. Therefore, we used a simple approximation method to interpolate the rates that were not available in the literature. This interpolation method ensured that the weighted average in each age stratum equate the observed marginal rate in that stratum. The following is an example of such approach: Prevalence rates of stage ii or higher COPD in different age groups are available in Buist et al. , Table 4.4 for men and women [162]. In each age group, overall prevalence was calculated by taking the weighted average of COPD prevalence rate for men (344) and women (483): (2.8%*344+ 1.3%*483)/ (344+483) = 1.9% (40-49 years old stratum) (6.4%*344+ 1.3%*483)/ (344+483) = 3.4% (50-59 years old stratum) 147  (12%*344+ 10.8%*483)/ (344+483) = 11.3% (60-69 years old stratum) (26.2%*344+ 20.7%*483)/ (344+483) = 23.0% (70+ years old stratum) Prevalence rate of stage ii or higher COPD also has been reported based on smoking status for men and women (Table 4.3, Buist et al [162]). We also calculated overall prevalence rates in each smoking stratum by calculating weighted average across men and women and then, estimated relative risk of having COPD in smokers and ex-smokers relative to never smokers (5.36 in smokers, 1.42 in ex-smokers, and 1 for never-smokers, by definition). In addition, we assumed that 13% of population are smokers, 39% are exsmokers, and 47% are never smokers [162]. Having the values in the margins of Table A.1 below (cells in grey), we would like to interpolate all of the unknown prevalence rates in Table A.1 (cells in white). Table A.1 Calculation of rates in subgroups  Prevalence rate by age (%) 1.9 3.4 11.3 23.0  Weighted RR Proportion RR by smoking Age 40-49 50-59 60-69 70+  0.71  0.56  0.47  0.13 5.36 Smokers  0.39 1.42 Ex-smokers  0.47 1.00 Never Smokers  5.9% 10.5% 34.7% 70.7%  1.6% 2.8% 9.2% 18.7%  1.1% 2.0% 6.5% 13.2%  We calculated the Weights RR by multiplying RR and the proportion of population in each smoking stratum (e.g. 5.36*0.13=0.71, 1.42*0.39=0.56, and 1.0*0.47=0.47). Then, 148  assuming that these weights are same for all age strata, we interpolated mortality rates for never smokers, as it follows: 1.9%/ (0.71+0.56+0.47) =1.1% (Never smokers, 40-49) 3.4%/ (0.71+0.56+0.47) =2.0% (Never smokers, 50-59) 11.3%/ (0.71+0.56+0.47) =6.5% (Never smokers, 60-69) 23%/ (0.71+0.56+0.47) =13.2% (Never smokers, 70+) And then, the prevalence rates for ex-smokers and smokers in each stratum were calculated by multiplying corresponding RR with the prevalence rates of never smokers in that stratum: 1.1%*1.42=1.6% (ex- smokers, 40-49) 1.1%*5.36=5.9% (smokers, 40-49) 2.0%*1.42=2.8% (ex-smokers, 50-59) and so on. Notice that the weighted average of prevalence rates in each row reflects the marginal prevalence rates in that row. Similar method was used to interpolate prevalence rates of COPD prevalence for stage i and higher COPD, and background mortality rates in Table 4.1.  149  A.2  Background Mortality Rates  According to the projections of Statistics Canada, total number of all-cause deaths will increase from 243,500 in 2010 to 375,400 in 2036 [169]. We used Canadian life tables to estimate age specific mortality rates for each age category (40-49, 50-59, 60-69, ≥70 years) using piecewise linear approximations. Then, we calculated the approximate number of total deaths that are attributable to each age category. The relative risk of background mortality has been reported as 1.7, 1.2, and 1 for current smokers, exsmokers, and never smokers, respectively[196]. These rates, in conjunction with prevalence of smokers (13.25%), previous smokers (39.4%), and never smokers (47.3%) were used to interpolate the number of deaths given smoking status within each age group (using the method that was explained above). Mortality rates are expected to decrease for population over 40 years old, as a result of prolonged life expectancy. The model also accounted for the declining mortality rates in future years.  A.3  Calculation of Direct and Indirect Costs  The estimated costs by Spencer et al[160] were used to calculate direct cost associated with a COPD patient at different stages of COPD. We adjusted those costs (cost of maintenance, minor exacerbations, and major exacerbations) using Consumer Price Index (CPI) such that they can represent 2011 Canadian currency value. Overall, direct cost for a patient with mild, moderate, or severe COPD consisted of maintenance costs and costs due to minor/major exacerbations. We assumed that maintenance costs are pertinent to 150  all patients in a given severity stage. However, exacerbations occur only in fraction of those patients and as such, we used frequency of minor/major exacerbations per each severity stage and calculated expected cost of exacerbations for a patients in the given severity stage. We assumed that direct costs and indirect costs consist 62.5% and 37.5% of total societal costs, respectively [163]. Therefore, we assumed that all indirect costs are 60% of the direct costs of maintenance and exacerbations at different severity stages. Unit costs are expected to increase in future years as a result of overall inflation rates. Therefore, we did not discount the outcomes in the base case analysis, as the effect of discount rate might be partly offset by inflation in the long term in a prospective simulation. However, the effects of non-zero discount rates were explored in the sensitivity analyses.  A.4  Progression Rates Between Different Stages of COPD  The rates of progression between mild to moderate and moderate to severe COPD mimicked the lung age curves that relate the changes in the lung function to aging and smoking status [158]. Annual progression rates from mild to moderate and from moderate to severe was initially set to 0.1 and 0.02 for current smokers. We assumed that those rates will be reduced by half among ex-smokers and non smokers. The dynamic model was mainly calibrated using these two parameters in order to maintain the current prevalence rates of mild, moderate, and severe COPD across different age groups. Therefore, prevalence rates of COPD (for a given age and smoking status) that 151  prevail in initial time point of the model were assumed to remain unchanged throughout the simulation. This limitation had been mentioned in the discussion, as unpredictable changes (e.g. in prevalence rates) might undermine this assumption in the following years. However, we conducted sensitivity analysis on these two parameters and tested robustness of the overall results under different assumptions for these parameters.  152  A.5 Model Structure in Vensim Figure A.1 Overall structure of the model  153  Figure A.2 Detailed structure of the Vensim model  154  N Mild4049 T Treated d  N Mild6069 Treated  N Mild5059 T t d Treated  <TP Smokers 4049> 4049 <FN Smokers S k 4049>  <TP TP Smokers S k 5059> <FN Smokers 5059>  N MildS4049  <UD Smokers 4049>  N Mild7099 Treated  <TP Smokers 6069> <FN Smokers 6069>  N MildS5059  <UD Smokers 5059>  <TP Smokers 7099> <FN Smokers 7099>  N MildS6069  <UD Smokers 7099>  <UD Smokers 6069> <TP XSmokers 5059>  <TP TP XSmokers 4049> <FN XSmokers 4049>  N MildXS4049  N Mild4049  <FN XSmokers 5059>  <UD XSmokers 4049>  <UD XSmokers 5059>  <TP NSmokers 4049>  <TP NSmokers 5059>  <FN NSmokers 4049> <UD NSmokers 4049>  N MildNS4049  <FN NSmokers 5059> <UD NSmokers 5059>  N MildXS5059  N Mild5059  <TP XSmokers 7099>  <TP XSmokers 6069> N MildXS6069 <FN XSmokers 6069>  N MildNS6069  <UD NSmokers 6069>  <FN XSmokers 7099>  N MildXS7099  <TP NSmokers k 7099>  <TP NSmokers 6069> <FN NSmokers NS k 6069>  N Mild6069  <UD XSmokers 7099>  <UD XSmokers 6069>  N MildNS5059  N MildS7099  <FN NSmokers 7099> 7099 <UD NSmokers 7099>  N MildNS7099  N Mild7099  Figure A.2 Detailed structure of Vensim model- continued  N Mild Treated  155  NMajorExac TPXS4049 <Mort Rate OHXS4049>  NExac N ac TPXS4049  <COPD XS5059>  <S iti it > <Sensitivity> <Test Uptake Rate>  Cessation TPXS4049  Prog TPXS4049 Prog Rate Mild XSmokers  NExac TPXS5059  <Sensitivity> <Test Uptake Rate>  NMajorExac TPXS6069  <Mort M t Rate Rt OHXS6069> OHXS6069  A i TPXS5059 Aging <COPD XS6069> XS6069  New TPXS5059 Cessation TPXS5059  <Sensitivity> Sensitivity  Prog TPXS5059 <Prog Rate Mild XS k > XSmokers>  NExac TPXS6069  <Mort Rate OHXS7099>  Mort Exac TPXS6069  Mort TPXS6069  TP XSmokers 5059  <Prop Major E Exac Mild>  <Exac E R Rate Mild Mild>  <Mort Mort Rate Major Exac>  Mort TPXS5059  Aging TPXS4049  New TPXS4049  <Propp Major j Exac Mild>  Mort Exac TPXS5059  <Mort Rate Major Exac>  TP XSmokers XS k 4049  <E <Exac R Rate t Mild>  NMajorExac TPXS5059  <Mort Rate OHXS5059>  M Exac Mort E TPXS4049  Mort TPXS4049  <COPD XS4049>  <Prop Prop Major Exac Mild>  <Exac Rate Mild>  <Test Uptake k Rate>  <Mort Rate Major Exac>  TP XSmokers 6069  <COPD CO XS7099>  Cessation TPXS6069  <Sensitivity> Sensitivity  P TPXS6069 Prog  <Prog Rate Mild XSmokers>  NMajorExac TPXS7099  <Test Uptake Rate>  NExac TPXS7099  Mort Exac TPXS7099  M t TPXS7099 Mort  Aging TPXS6069  New TPXS6069  <Exac Rate Mild>  <Mort Rate Major Exac>  TP XSmokers 7099  New TPXS7099 Cessation TPXS7099 Prog TPXS7099  <Prog Rate Mild XS k > XSmokers>  Figure A.2 Detailed structure of Vensim model- continued  <Prop Major Exac Mild>  156  <Exac Rate Mild>  NMajorExac FNXS4049 <Mort Rate OHXS4049>  NExac FNXS4049  <Mort Rate Major Exac>  FN XSmokers 4049  NMajorExac FNXS5059  Mort FNXS5059  Aging FNXS4049 <COPD XS5059>  New FNXS4049 Cessation FNXS4049  NExac FNXS5059  <Prog Rate Mild XSmokers>  <Mort Rate Major Exac>  <COPD XS6069>  Cessation FNXS5059 Prog FNXS5059 <PostDiag Cessation Rate>  Mort FNXS6069  Aging g g FNXS5059  New FNXS5059  <Test Uptake Rate>  NMajorExac NM j E FNXS6069  <Mort Rate OHXS6069>  NExac FNXS6069  <Prog g Rate Mild XSmokers>  <COPD XS7099>  Cessation C ti FNXS6069 Prog FNXS6069  <Test Uptake Rate> <PostDiag PostDiag Cessation Rate>  <Prog Rate Mild XS k XSmokers>  NE NExac FNXS7099 N S7099  Mort Exac FNXS7099 <Mort Rate Major Exac>  Mort FNXS7099  Aging FNXS6069  New FNXS6069  <Exac Rate Mild>  NMajorExac j FNXS7099  <Mort Rate OHXS7099>  <Mort Mort Rate Major Exac>  FN XSmokers 6069  <Sensitivity>  <Prop Major Exac Mild>  <Exac Rate Mild>  M tE Mort Exac FNXS6069  FN XSmokers 5059  <Sensitivity> y  Prog FNXS4049 <PostDiag C Cessation ti Rate> Rt >  <Prop Major Exac Mild>  Mort Exac FNXS5059  COPD S4049>  <Sensitivity> ensitivity>  <Exac Rate Mild>  <Mort Rate OHXS5059>  Mort Exac FNXS4049  Mort FNXS4049  est Uptake Rate>  <Prop Major Exac Mild>  FN XSmokers 7099  New FNXS7099 C Cessation i FNXS7099  <Sensitivity> y  Prog FNXS7099  <Test Test Uptake Rate> <PostDiag Cessation C i Rate> R  <Prog Rate Mild XS k XSmokers>  Figure A.2 Detailed structure of Vensim model- continued  <Prop P Major M j Exac Mild>  157  <Exac Rate Mild>  NMajorExac UDXS4049 <Mort Rate OHXS4049 OHXS4049>  <COPD XS4049>  <Mort Rate Major j Exac>  Aging UDXS4049  <COPD XS5059>  Cessation UDXS4049  <PostDiagg Cessation Rate>  <Mort Rate Major E Exac>  Aging g g UDXS5059  New UDXS5059  <COPD XS6069>  Cessation UDXS5059  <PostDiag Cessation Rate> <Prog Rate Mild XSmokers>  <Prog Rate Mild XSmokers>  <COPD XS7099>  Cessation UDXS6069 Prog UDXS6069  <Test Uptake Rate> <P tDi <PostDiag Cessation Rate> Rate  NExac UDXS7099  Mort Exac UDXS7099  Mort UDXS7099  Aging UDXS6069  New UDXS6069  <Sensitivity>  Prog UDXS5059  <Test Uptake Rate>  <Mort M R Rate OHXS7099> OHXS7099  <Mort Rate Major Exac>  UD XSmokers 6069  <Exac Exac Ra  NMajorExac UDXS7099  NExac UDXS6069  Mort o Exac ac UDXS6069  Mort UDXS6069  <Propp Major j Exac Mild>  <Exac Rate Mild>  NMajorExac j UDXS6069 <Mort Rate OHXS6069>  UD XSmokers 5059  <Sensitivity> Sensitivity  Prog UDXS4049 Test Uptake Rate>  NExac UDXS5059  Mort Exac UDXS5059  Mort UDXS5059  New UDXS4049  <Sensitivity> Sensitivity>  <Prop Prop Major Exac Mild> NMajorExac NM j E UDXS5059  <Mort Rate OHXS5059>  UD XSmokers 4049  <Exac Rate Mild> ild  NExac UDXS4049  Mort Exac UDXS4049  Mort UDXS4049  <Prop Major Exac Mild>  <Mort Mort Rate Major Exac>  UD XSmokers 7099  New UDXS7099 Cessation UDXS7099  <Sensitivity>  Progg UDXS7099  <Test Uptake Rate> <PostDiag ost ag Cessation Rate>  <Prog Rate Mild XSmokers>  <Prog Rate Mild XS k XSmokers>  Figure A.2 Detailed structure of Vensim model- continued  <Prop Major Exac Mild>  158  <Exac Rate Mild>  NMajorExac FPXS4049 <Mort Rate OHXS4049>  <Mort M R Rate M Major j Exac> Exac  NMajorExac j FPXS5059  Mort FPXS5059  Aging FPXS4049 <COPD XS5059>  New FPXS4049  NE NExac FPXS5059  Cessation FPXS4049  <PostDiag PostDiag Cessation Rate>  Mort FPXS6069  Aging FPXS5059 <COPD XS6069>  Cessation i FPXS5059  NExac FPXS6069  FP XS XSmokers k 6069  Prog FPXS5059  <Test Uptake Rate>  <Prog Rate Mild XSmokers>  FP XSmokers 7099  N FPXS7099 New  Cessation FPXS6069 Prog FPXS6069  <Test Uptake Rate> <PostDiag i Cessation Rate> Rate  <Mort Rate Major Exac>  <COPD XS7099>  New FPXS6069  <Sensitivity>  <Prog Rate Mild XSmokers>  NExac FPXS7099  Mort Exac FPXS7099  Mort FPXS7099  A i FPXS6069 Aging  <Exac Rate Mild>  NMajorExac FPXS7099  <Mort Rate OHXS7099>  <Mortt R <M Rate t M Major j Exac>  <Specificity> p y  <PostDiag Cessation Rate> <Prog Rate Mild XSmokers>  <Mort Rate Major Exac>  New FPXS5059  <Specificity>  NMajorExac FPXS6069  <M t R <Mort Rate t OHXS6069>  <Prop Major E Exac Mild> Mild  <Exac Rate Mild>  Mort Exac FPXS6069  FP XSmokers 5059  <Sensitivity>  Prog FPXS4049 <Test Uptake k Rate> Rate  <Prop Major E Exac Mild> Mild  Mort Exac FPXS5059  FP XSmokers 4049  <Sensitivity> Se s t v ty <Sensitivity> <Specificity> S ifi it  <Exac Rate Mild>  <Mort Mort Rate OHXS5059>  Mort Exac FPXS4049  Mort FPXS4049  <COPD XS4049>  NExac FPXS4049  <Propp Major j Exac Mild>  <S iti it > <Sensitivity>  Cessation FPXS7099  <Specificity>  Prog FPXS7099  <Test T U Uptake k Rate> Rate <PostDiag Cessation Rate>  <Prog P Rate R Mild XSmokers> XSmokers  Figure A.2 Detailed structure of Vensim model- continued  <Prop Major Exac Mild>  159  <Proportion i off Current Smokers>  Mort OHNS5059  Mort OHNS4049  <Proportion of XSmokers>  <RR Mort NSmokers NSmokers>  Mort Rate OHNS7099 Mort OHNS6069 M t OHNS7099 Mort  N OHNS4049 New  NSmokers OH4049  A i OHNS4049 Aging  <Prev COPD 4049>  A i OHNS5059 Aging  <RR COPD NSmokers>  NSmokers OH6069  <Prev COPD 5059>  A i OHNS6069 Aging  COPD NS6069 Prev COPD NS6069  Prev COPD NS5059  Prev COPD NS4049  RR COPD NSmokers  NSmokers OH5059  COPD NS5059  COPD NS4049 Proportion P ti off NS k NSmokers  <Mort Rate 7099>  <RR Mort NSmokers> Mort Rate OHNS6069  Mort Rate OHNS5059  Mort Rate OHNS4049 <Net Increase Pop4049>  <Mort Rate 6069>  <RR COPD NSmokers>  NSmokers OH7099  COPD NS7099 Prev COPD P NS7099  <Prev COPD 6069>  <RR RR COPD NSmokers>  <Prev COPD <P 7079 7079>  Figure A.2 Detailed structure of Vensim model- continued  <Mort Rate <RR Mort 5059> NSmokers>  RR Mort NSmokers  <Mort Rate 4049>  160  NMajorExac NM j E TPNS4049 <Mort Rate OHNS4049>  <Mort Rate OHNS5059>  Mort Exac TPNS4049  <Sensitivity> <Test T Uptake U k Rate> Rate  Aging TPNS4049 <COPD XS5059>  New TPNS4049 Cessation TPNS4049 Prog TPNS4049 Prog Rate Mild NSmokers  <Prop Major Exac Mild>  <Sensitivity> <Test Uptake p Rate>  NExac TPNS5059  Mort TPNS6069  Aging TPNS5059 <COPD XS6069>  New TPNS5059 Cessation TPNS5059  <Sensitivity> y  Prog TPNS5059 <Prog Rate Mild NSmokers>  NMajorExac TPNS6069  <Testt Uptake <T U t k Rate> e  <Exac Rate Mild>  NMajorExac NM j E TPNS7099  NExac TPNS6069 <Mort Rate OHNS7099 OHNS7099>  Mort Exac TPNS6069  <Mort Rate Major j Exac>  TP NSmokers 5059  <Prop Major j Exac Mild>  <Exac Rate Mild>  <Mort Rate OHNS6069>  M tE Mort Exac TPNS5059  Mort TPNS5059  Mort TPNS4049  TP NSmokers 4049  <Exac Rate Mild>  NMajorExac TPNS5059  NExac TPNS4049  <Mort Rate Major E > Exac>  <COPD XS4049>  <Prop Major Exac Mild>  <Exac Rate Mild>  <Mort Rate Major Exac>  TP NS NSmokers k 6069  <COPD XS7099>  N TPNS6069 New Cessation TPNS6069  <Sensitivity> y  P TPNS6069 Prog  <Prog Rate Mild NSmokers>  Mort Exac TPNS7099  Mort TPNS7099  A i TPNS6069 Aging S6069  <Test Uptake Rate>  NE NExac TPNS7099  <Mort Rate Major E ac> Exac>  TP NSmokers 7099  New TPNS7099 Cessation TPNS7099 P Prog TPNS7099  <Prog Rate Mild NSmokers>  Figure A.2 Detailed structure of Vensim model- continued  <Prop Major Exac Mild>  161  <Exac Rate Mild>  NMajorExac NM j E FNNS4049 <Mort Rate OHNS4049>  Aging g g FNNS4049 <COPD XS5059>  New FNNS4049 Cessation FNNS4049 Progg FNNS4049 Test Uptake Rate>  <PostDiag Cessation Rate>  <Prog Rate Mild NSmokers>  NExac FNNS5059  <Mort Rate Major j Exac>  <COPD XS6069>  Cessation FNNS5059  <Prog Prog Rate Mild NSmokers>  <COPD XS7099>  New FNNS6069 Cessation FNNS6069 Prog FNNS6069  <Testt U <T Uptake t k Rate> Rate <PostDiag Cessation Rate>  <Prog P Rate R Mild NSmokers>  NE NExac FNNS7099  Mort Exac FNNS7099  Mort FNNS7099  A i FNNS6069 Aging  <Exac E R Rate M  NMajorExac NM j E FNNS7099  <Mort Mort Rate OHNS7099>  <Mort Rate Major Exac>  FN NSmokers NS k 6069  <Sensitivity>  Prog FNNS5059 <PostDiag Cessation C ti R Rate> t  Mort FNNS6069  Aging FNNS5059  New FNNS5059  <Testt U <T Uptake t k Rate> Rate  NE NExac FNNS6069  M t Exac Mort E FNNS6069  FN NSmokers 5059  <S iti it > <Sensitivity>  <Prop Major <P M j Exac Mild>  <Exac Rate Mild>  NMajorExac FNNS6069  <Mort Rate OHNS6069>  Mort Exac FNNS5059  Mort FNNS5059  <COPD XS4049>  <Sensitivity> Sensitivity> y  <Prop M <P Major j Exac Mild Mild> NMajorExac FNNS5059  <Mortt Rate <M Rt OHNS5059>  <Mort Rate Major Exac>  FN NSmokers 4049  <Exac Rate Mild>  NExac FNNS4049  Mort Exac FNNS4049  Mort FNNS4049  <Prop Major Exac Mild>  <Mortt R <M Rate t M Major j Exac> Exac  FN NSmokers NS k 7099  N FNNS7099 New Cessation FNNS7099  <Sensitivity> S ii i  Prog FNNS7099  <Test Uptake Rate> <PostDiag Cessation Rate>  <Prog Rate Mild NSmokers>  Figure A.2 Detailed structure of Vensim model- continued  <Prop Major Exac Mild>  162  <Exac Rate Mild>  NMajorExac UDNS4049 <Mort Rate OHNS4049>  <Prop Major Exac Mild>  <COPD NS4049>  Mort UDNS5059  Aging UDNS4049 <COPD NS5059>  Cessation UDNS4049 Prog UDNS4049  <Test Uptake Rate>  <PostDiag Cessation Rate>  <Prog Rate Mild NS k > NSmokers>  <Mort Rate Major E > Exac>  UD NSmokers 5059  <COPD NS6069>  New UDNS5059 Cessation UDNS5059  <Sensitivity> y  <PostDiag Cessation Rate>  New UDNS6069  <Prog Rate Mild NSmokers>  <COPD NS7099>  Cessation C ti UDNS6069  <PostDiag Cessation Rate>  <Prog og Rate a e Mildd NSmokers>  <Mort Rate Major j Exac>  UD NSmokers 7099  New UDNS7099 C Cessation i UDNS7099  <Sensitivity> y  Prog UDNS6069  <Test Uptake p Rate>  NExac UDNS7099  Mort Exac UDNS7099  Mort UDNS7099  Aging g g UDNS6069  <Exac Rate Mi  NMajorExac j UDNS7099  <Mort Rate OHNS7099>  <Mort Rate Major Exac>  UD NSmokers 6069  <Sensitivity>  Prog UDNS5059  <Test Uptake p Rate>  NExac UDNS6069  Mort Exac UDNS6069  Mort UDNS6069  Aging UDNS5059  <Propp Major j Exac Mild>  <Exac Rate Mild>  NMajorExac j UDNS6069  <Mort Rate OHNS6069>  Mort Exac UDNS5059  New UDNS4049  <Sensitivity> S ii i <Sensitivity>  NExac UDNS5059  <Mort Rate OHNS5059>  <Mort Rate Major Exac>  UD NSmokers 4049  <Prop Prop Major Exac Mild> NMajorExac NM j E UDNS5059  NExac UDNS4049  Mort Exac UDNS4049  Mort UDNS4049  <Exac Rate Mild>  Prog UDNS7099  <Test Uptake Rate> <PostDiag Cessation Rate>  <Prog Rate Mild NSmokers>  Figure A.2 Detailed structure of Vensim model- continued  <Prop Major Exac Mild>  163  <E <Exac R Rate t Mild>  NMajorExac FPNS4049 <Mort Rate OHNS4049>  <Mortt Rate <M R t Major M j Exac>  FP NSmokers 4049  <Mort Mort Rate Major Exac>  FP NSmokers 5059  Aging FPNS5059  <Sensitivity>  Cessation FPNS4049  Cessation FPNS5059  <Specificity>  Progg FPNS4049 <PostDiag Cessation Rate>  <COPD NS6069>  New FPNS5059  New FPNS4049  <Prog Rate Mild Smokers>  NExac FPNS6069  <Mort Rate Major Exac>  FP NSmokers NS k 6069  <COPD NS7099>  N FPNS6069 New Cessation FPNS6069  <Sensitivity> y <Specificity>  Prog g FPNS5059  <Test T U Uptake k Rate> Rate <PostDiag C Cessation i Rate> R  <Prog Rate Mild Smokers>  Prog FPNS6069  <Test T U Uptake k Rate> Rate <PostDiag Cessation Rate>  <Prog P Rate R Mild Smokers> Smokers  NE NExac FPNS7099  M tE Mort Exac FPNS7099  Mort FPNS7099  A i FPNS6069 Aging  <Exac Rate  NMajorExac FPNS7099  <Mort M Rate R OHNS7099 OHNS7099>  M E Mort Exac FPNS6069  Mort FPNS6069  <Prop Major Exac Mild>  <Exac Rate Mild>  NMajorExac NM j E FPNS6069  <Mort Rate OHNS6069>  Mort Exac FPNS5059  Mort FPNS5059  Aging FPNS4049  NExac FPNS5059  <COPD NS5059>  <Sensitivity> <Sensitivity> <Specificity>  <Test Uptake Rate>  <Prop Major Exac Mild> NMajorExac FPNS5059  <Mort M Rate R OHNS5059> OHNS5059  <COPD COPD NS4049> NS4049  <Exac Rate Mild>  NExac FPNS4049  Mort Exac FPNS4049  Mort FPNS4049  <Prop Major E Exac Mild> Mild  <Mort Rate Major Exac>  FP NSmokers 7099  N FPNS7099 New  <Sensitivity> S ii i  Cessation FPNS7099  <Specificity>  Prog FPNS7099  <Test Uptake Rate> <PostDiag Cessation Rate>  <Prog Rate Mild Smokers>  Figure A.2 Detailed structure of Vensim model- continued  <Prop Major Exac Mild>  164  P Prop Major M j E Exac Mod NMajorExac M dS4049 ModS4049  Mort ModS4049  <Mort Rate OHS5059 OHS5059>  <Mort M Rate R Major M j Exac> Exac  Mod Smokers 4049  Aging g g ModS4049  <Prog P TPS5059 TPS5059>  Aging ModS5059  New ModS5059  <Prog FNS5059>  New ModS6069  Prog ModS5059  Prog ModS4049  Prog Rate Mod Smokers  <PostDiag Cessation Rate>  <Prog Rate Mod S k > Smokers>  <Mort Rate Major Exac>  Mod Smokers 7099  New ModS7099 <Prog TPS7099>  Cessation ModS6069  <P <Prog FNS6069>  Cessation ModS7099 <Prog FNS7099>  <Prog UDS6069>  Progg ModS6069  <P tDi <PostDiag Cessation Rate>  <Prog P R Rate M Mod d Smokers> Smokers  NExac ModS7099  Mort Exac ModS7099  Mort ModS7099  A i M Aging ModS6069 dS6069  <Exac Rate Mod>  NMajorExac M dS7099 ModS7099  <Mort Rate OHS7099>  <Mortt Rate <M R t Major M j Exac>  Modd Smokers M S k 6069  Cessation ModS5059  <Prog UDS5059>  <PostDiagg Cessation Rate>  NE NExac ModS6069  Mort Exac ModS6069  Mort ModS6069  <Prop Major Exac Mod>  <Exac Rate Mod>  NMajorExac NM j E ModS6069  <Mort Rate OHS6069>  <Mort Rate Major Exac>  Mod Smokers 5059  <Prop P Major M j Exac E Mod>  <Progg TPS6069>  Cessation C i ModS4049  Progg FNS4049> <Prog UDS4049>  NExac ModS5059  Mort Exac ModS5059  Mort ModS5059  New ModS4049 P Prog TPS4049> <Sensitivity>  <Exac Rate Mod>  NMajorExac j ModS5059  NExac ModS4049  Mort Exac ModS4049  <M t Rate <Mort Rt OHS4049>  <Prop Major Exac M d> Mod>  Exac Rate Mod  Progg ModS7099  <Prog Prog UDS7099 UDS7099>  <PostDiagg Cessation Rate>  <Prog Rate Mod Smokers>  Figure A.2 Detailed structure of Vensim model- continued  <Effect of Predictive T t on E Test Exac R Rate> t >  165  <NExac TPS4049> <NExac NExac FNS4049>  <NExac FNS5059> NExac MildS4049  <NExac FNS6069>  NExac MildS5059  <NExac TPS7099> <NExac FNS7099>  NExac MildS6069  <NExac UDS7099>  <NExac UDS6069>  <NExac NExac UDS5059>  NExac MildS7099  <NExac UDS4049 UDS4049>  <NExac FNXS4049>  NExac MildXS4049  <NExac FNXS5059>  NExac MildXS5059  NExac MildNS4049  <NExac FNNS5059> <NExac ModS5059> <NExac UDNS5059>  NE NExac Mild5059  <NExac <NE FNXS6069>  NExac NE MildXS6069  <NExac UDXS6069>  NExac MildNS5059  <NExac FNNS6069> <NExac UDNS6069>  N Exac Mild  <NExac TPXS7099> NExac Mild6069  <NE <NExac FNXS7099>  NExac MildXS7099  <NExac UDXS7099>  <NExac NE TPNS6069>  <NExac TPNS5059>  <NExac NExac TPNS4049>  <NExac UDNS4049>  NExac Mild4049  <NExac UDXS5059>  <NExac UDXS4049 UDXS4049>  <NExac <NE FNNS4049>  <NExac TPXS6069>  <NExac NExac TPXS5059>  <NExac TPXS4049>  NExac MildNS6069  <NExac NE TPNS7099> <NExac FNNS7099> <NExac UDNS7099>  NExac MildNS7099  NExac Mild7099  Figure A.2 Detailed structure of Vensim model- continued  <NExac TPS6069>  <NExac TPS5059 TPS5059>  166  NMajorExac ModXS4049 M dXS4049 <Mort Rate OHXS4049>  <Mort Rate OHXS5059>  <Mort Rate Major E > Exac>  Mod XSmokers 4049  <Progg TPXS5059>  New ModXS4049 <Prog P XS4049>  Cessation ModXS4049  <Prog DXS4049>  Prog ModXS4049 <PostDiagg Cessation Rate>  Progg Rate Mod XSmokers  NExac ModXS5059  <Mort M Rate R Major M j Exac>  Mod XSmokers 6069 <Prog TPXS6069> TPXS6069  C Cessation i ModXS5059 M dXS5059  Prog ModXS5059 <PostDiag Cessation Rate>  <Mort Rate OHXS7099> OHXS7099  NMajorExac ModXS7099 M dXS7099  <Prog Rate Mod XS k > XSmokers>  <Prog P FNXS6069>  Prog ModXS6069  <P tDi <PostDiag Cessation Rate> Rate  Mod XSmokers 7099 <P <Prog TPXS7099 TPXS7099>  C Cessation ti ModXS6069  <Prog UDXS6069>  <Prog Rate Mod XSmokers>  <Mort Rate Major Exac>  Mort ModXS7099  Aging ModXS6069  New ModXS6069  NExac ModXS7099 M dXS7099  Mort Exac ModXS7099  <Mort Rate Major Exac>  Mort ModXS6069  A i Aging ModXS5059  New ModXS5059  <Progg UDXS5059>  NE NExac ModXS6069  <Prop Major <P M j Exac E Mod> od  Mort Exac ModXS6069 M dXS6069  Mod XSmokers 5059  <Prog Prog FNXS5059>  <Exac Rate Mod>  <Exac Rate Mod>  NM j E NMajorExac ModXS6069  <Mort Rate OHXS6069> OHXS6069  M tE Mort Exac ModXS5059  Mort ModXS5059  Aging ModXS4049  <Prog P XS4049>  <Prop Major Exac Mod>  NMajorExac j ModXS5059  NExac ModXS4049  Mort Exac ModXS4049 Mort ModXS4049  <Exac Rate Mod>  <Prop Major Exac Mod>  <Exac Rate Mod>  New ModXS7099 M dXS7099  <Prog Prog FNXS7099>  C Cessation i ModXS7099  <Prog UDXS7099>  Progg ModXS7099  <PostDiag os g Cessation Rate>  <Prog Rate Mod XSmokers>  Figure A.2 Detailed structure of Vensim model- continued  <Prop Major Exac Mod>  167  <Exac Rate Mod>  NMajorExac ModNS4049 M dNS4049 <Mort Rate OHNS4049>  <Prog FNNS4049> <Prog og UDNS4049>  <Mort Rate OHNS5059>  <Mort Rate Major Exac>  M d NSmokers Mod NS k 4049 <Prog g TPNS4049>  Aging g g ModNS4049 <Prog TPNS5059>  Cessation M dNS4049 ModNS4049 Prog ModNS4049 Prog R P Rate t M Modd NSmokers NS o es  <Progg FNNS5059> <Prog UDNS5059>  NE NExac ModNS5059  Mort Exac ModNS5059 Mort ModNS5059 d S 0 9  New ModNS4049  <Prop Major <P M j Exac E Mod>  NMajorExac ModNS5059  NExac ModNS4049  Mort Exac ModNS4049 Mort ModNS4049  <Exac Rate Mod>  <Propp Major j Exac Mod>  <Mort Rate OHNS6069>  NMajorExac ModNS6069  Mod NSmokers 5059  New ModNS5059  Mod NSmokers 6069  <Prog TPNS6069>  Cessation M dNS5059 ModNS5059  <Prog FNNS6069>  Prog ModNS5059 <Prog Rate Mod NSmokers>  <Progg UDNS6069>  NMajorExac ModNS7099 M dNS7099 <Mort Rate OHNS7099>  <Mort M Rate R M Major j Exac> Exac  Mort ModNS6069  Aging ModNS5059  <Prop P Major M j Exac E Mod Mod>  NExac ModNS6069  Mort Exac ModNS6069 <Mort Rate Major Exac>  <Exac Rate Mod>  <Exac Exac Rate Mod> Mod  Cessation ModNS6069 Prog og ModNS6069  <Prog Rate Mod NSmokers>  <Mort Rate Major E > Exac>  Mod NSmokers 7099  <Prog P TPNS7099 TPNS7099>  New ModNS6069  Mort Exac ModNS7099 Mort ModNS7099  Aging ModNS6069  <Prog FNNS7099> <Prog UDNS7099>  NExac ModNS7099  New ModNS7099 C Cessation ti M ModNS7099 dNS7099 Prog ModNS7099 <Prog Rate Mod NS k NSmokers>  Figure A.2 Detailed structure of Vensim model- continued  <Prop Major Exac M d> Mod>  168  Exac Rate Sev  Prop Major Exac Sev  NMajorExac j SevS4049  <Mort Rate OHS4049 OHS4049>  <Mort Rate Major j Exac>  Sev Smokers 4049  <Prog Prog ModS4049>  NMajorExac SevS5059 <Mort Mort Rate OHS5059>  NMajorExac SevS6069  <M t Rate <Mort Rt OHS6069>  <Mort Rate Major E > Exac>  New SevS5059  <PostDiag Cessation C i Rate> R  <Prog Rate Mild Smokers>  <Mort Mort Rate Major Exac>  Sev Smokers 7099  New SevS7099 Cessation SevS7099 S S7099  <Prog ModS7099>  Prog SevS6069  <PostDiag Cessation C ti Rate> Rt >  NExac SevS7099  Mort Exac SevS7099  Mort SevS7099  Aging SevS6069  Cessation Se S6069 SevS6069  <Prog ModS6069>  Prog SevS5059  Prog SevS4049  <Mort Rate OHS7099>  New SevS6069  Cessation C ti SevS5059  NMajorExac j SevS7099  NE NExac S SevS6069 S6069  Sev Smokers 6069  Aging g g SevS5059  <Exac Rate Sev>  <Propp Major j Exac Sev>  <Mort Rate Major Exac>  Mort SevS6069  Sev Smokers 5059  <Prog ModS5059>  <Exac Rate Sev>  Mort Exac SevS6069  Aging SevS4049  Cessation SevS4049  <PostDiag Cessation Rate>  NExac SevS5059  Mort Exac S S5059 SevS5059  Mort SevS5059  New SevS4049  <Sensitivity>  <Prop Major Exac Sev>  <Exac Rate Sev>  NExac SevS4049  Mort Exac SevS4049  Mort SevS4049  <Prop Major Exac Sev>  <Prog P R Rate Mild Smokers>  Prog SevS7099  <PostDiag Cessation Rate>  <Prog Rate Mild Smokers>  Figure A.2 Detailed structure of Vensim model- continued  <Effect of Predictive Test on Exac Rate>  169  NMajorExac j SevXS4049  <Mort Mort Rate OHXS5059>  Mort SevXS4049  <Mort Rate Major Exac>  Sev XSmokers 4049  Cessation SevXS4049  <Mort Rate Major E > Exac>  Mort SevXS6069  Aging g g SevXS5059  New SevXS5059 <Prog Prog ModXS5059>  <PostDiag Cessation C ti Rate> Rt >  NExac SevXS6069  <Prog Rate Mild Smokers> k  New SevXS7099 Cessation SevXS7099 S XS7099  <Prog ModXS7099>  Prog SevXS6069  <PostDiag Cessation Rate>  <Mort Mort Rate Major Exac>  Sev XSmokers 7099  Aging SevXS6069  Cessation SevXS6069  <Prog ModXS6069>  NExac SevXS7099  Mort Exac SevXS7099  Mort SevXS7099  New SevXS6069  Cessation SevXS5059  <Exac Exac Rate S  NMajorExac SevXS7099  <Mort Mort Rate OHXS7099>  <Mort Rate Major Exac>  Sev XSmokers 6069  Prog SevXS5059  Prog SevXS4049 <PostDiag Cessation Rate>  NMajorExac SevXS6069  <Mort Rate OHXS6069>  <Prop P Major M j Exac Sev>  <Exac Rate Sev>  Mort Exac SevXS6069  Sev XSmokers 5059  New SevXS4049  Sensitivity>  NExac SevXS5059 S XS5059  Mort Exac SevXS5059 Se XS5059  Mort SevXS5059  Aging SevXS4049  <Prop Major Exac Sev>  <E <Exac R Rate t S Sev> >  NMajorExac SevXS5059  NE NExac SevXS4049  Mort Exac SevXS4049  <Mort Rate OHXS4049> OHXS4049  <Progg ModXS4049>  <Prop Major E Exac S Sev>  <Exac Rate Sev>  <Prog R <P Rate t Mild Smokers> Smokers  Prog SevXS7099  <PostDiag Cessation Rate>  <Prog Rate Mild Smokers>  Figure A.2 Detailed structure of Vensim model- continued  <Prop Major Exac Sev>  170  NMajorExac SevNS4049 <Mort Rate OHNS4049>  NExac SevNS4049  <Prop Major Exac Sev>  M tS Mort SevNS5059 NS5059  Cessation SevNS4049  <PostDiag Cessation Rate>  NExac SevNS6069  Aging SevNS5059  Sev NSmokers 6069  <Prog ModNS6069> M dNS6069  <Prog Rate Mild Smokers>  Mort SevNS7099  <Mort Rate Major Exac>  Sev NSmokers 7099  Aging g g SevNS6069  New SevNS7099  Cessation SevNS6069  Cessation SevNS7099  <Prog ModNS7099>  Progg SevNS6069  <PostDiag Cessation Rate>  NExac SevNS7099  Mort Exac SevNS7099 S NS7099  New SevNS6069  Cessation SevNS5059  <Exac Rate Sev>  NMajorExac SevNS7099  <Mort Rate OHNS7099>  <Mort Rate Major E > Exac>  Mort SevNS6069  Prog SevNS5059  Prog SevNS4049 <PostDiagg Cessation Rate>  <Mort Rate Major Exac>  New SevNS5059 <Progg ModNS5059>  NMajorExac S NS6069 SevNS6069  <Mort Rate OHNS6069>  <Prop Major Exac Sev>  <Exac E R Rate S Sev>  Mort Exac SevNS6069  Sev NSmokers 5059  Aging SevNS4049  New SevNS4049  nsitivity>  NExac SevNS5059  Mort Exac SevNS5059  <Mort Rate Major Exac>  Sev NSmokers 4049  <Prop Major Exac Sev>  <Exac Rate Sev>  NMajorExac SevNS5059  <Mort Rate OHNS5059>  Mortt E M Exac SevNS4049  M tS Mort SevNS4049 NS4049  <Prog NS4049>  OHS7099>  <Exac Rate Sev>  <Prog Rate Mild Smokers>  Progg SevNS7099  <PostDiag Cessation Rate>  <Prog g Rate Mild Smokers>  Figure A.2 Detailed structure of Vensim model- continued  <Prop Major E Exac S Sev> >  171  FUDeath Mild4049  Aging A i FUDMild4049  FUDeath Mild5059  Aging FUDMild5059  Mort FUDMild5059  Mort M FUDMild4049 <Mort M R Rate 4049>  NewFUD Mild6069  NewFUD Mild5059  NewFUD New U Mild4049 d 0 9  <Mort Rate 5059>  <NMort Mild7099>  <NMort N ot Mild6069>  <NMort Mild5059 Mild5059>  <Mortt Rate <M Rt 6069>  NewFUD Mild7099  FUDeathh FUD Mild6069  Aging FUDMild6069 Mort FUDMild6069 <Mort Rate 7099 7099>  FUDeath Mild7099 Mort FUDMild7099  Figure A.2 Detailed structure of Vensim model- continued  <NMort NM Mild4049>  172  FUDeath FUD th Mod4049  Aging FUDM d4049 FUDMod4049  FUDeath Mod5059  Aging A gi g FUDMod5059  Mort FUDM d5059 FUDMod5059  Mort FUDM d4049 FUDMod4049 Mort Rate 4049> 4049  NewFUD N FUD Mod6069  NewFUD M d5059 Mod5059  NewFUD Mod4049  <Mort M t Rate Rt 5059>  <NMort Mod7099> od7099  <NMort Mod6069 Mod6069>  <NMortt <NM Mod5059>  <Mort Rate 6069>  NewFUD N FUD Mod7099  FUDeath M d6069 Mod6069  Aging FUDMod6069 Mort FUDM d6069 FUDMod6069 <Mort M R Rate 7099>  FUDeathh Mod7099 M t Mort FUDMod7099  Figure A.2 Detailed structure of Vensim model- continued  <NMort Mod4049> M d4049>  173  <NMort NM t Sev5059>  FUDeath FUD th Sev4049  Aging g g FUDSev4049  FUDeath Sev5059  A i Aging FUDSev5059  Mortt M FUDSev5059  Mort FUDSev4049 <Mort Rate 4049>  NewFUD Sev6069  NewFUD Ne FUD Sev5059  NewFUD N FUD Sev4049  <Mort Rate 5059>  <NMort Sev7099>  <NMort Sev6069>  <Mortt Rate <M Rt 6069>  NewFUD S 7099 Sev7099  FUDeath S 6069 Sev6069  Aging FUDSev6069 M t Mort FUDSev6069 <Mort Rate 7099 7099>  FUDeath Sev7099 Mort FUDSev7099  Figure A.2 Detailed structure of Vensim model- continued  <NMort NMort Sev4049>  174  <FUDeath Mild4049>  <FUDeath Mild5059>  QALY MortMild4049  dQALY MortMild4049  U Mild> ild  <U OH4049>  <FUDeath Mild6069>  QALY MortMild5059  dQALY Q MortMild5059  <U Mild>  dQALY MortMild6069  <U Mild>  <U OH5059>  <FUDeath Mild7099>  QALY Q MortMild6069  <U OH6069>  QALY Q MortMild7099  dQALY Q MortMild7099  <U OH7099>  <U Mild> ild  dQALY ExacMild  <NExac NE Mild4049>  <NE ac <NExac Mild5059>  <Propp Major j Exac Mild>  dQALY Q ExacMild4049  QALY ExacMild4049  <Prop Major Exac Mild>  dQALY ExacMild5059  <NExac Mild6069> QALY Q ExacMild5059  dQALY ExacMild6069 <U MildExacMin>  <U MildExacMin>  ldE Mi ldExacMin> <U MildExacMaj>  <U OH4049>  <Prop Major Exac Mild>  <U MildExacMaj>  <U U MildExacMaj MildExacMaj> <U OH5059>  <NExac Mild7099> QALY ExacMild6069  <Prop Prop Major Exac ac Mild> d QALY ExacMild7099  dQALY ExacMild7099 <U MildExacMin> MildE Mi >  <U OH6069>  <U MildExacMaj>  <U OH7099>  d dQALY RegMild ild  <N Mild4049>  dQALY RegMild4049  175  U Mild>  QALY RegMild4049  U OH4049  dQALY R Mild5059 RegMild5059 <U Mild>  <N N Mild7099 Mild7099>  <N Mild6069>  <N Mild5059>  QALY RegMild5059 R Mild5059  U OH5059  dQALY RegMild6069 <U Mild>  QALY RegMild6069 R Mild6069  U OH6069  dQALY RegMild7099  QA QALY RegMild7099 R Mild7099  U OH7099 <U Mild>  Figure A.2 Detailed structure of Vensim model- continued  dQALY Q MortMild  <FUDeath Mild4049>  <FUDeath Mild5059>  QALY MortMild4049  dQALY MortMild4049  U Mild>  <U OH4049>  <FUDeath Mild6069>  QALY MortMild5059  dQALY MortMild5059  <U Mild>  dQALY M Mild6069 MortMild6069  <U Mild>  <U OH5059>  <FUDeath Mild7099>  QALY MortMild6069 M tMild6069  <U OH6069>  QALY MortMild7099  dQALY M tMild7099 MortMild7099  <U OH7099>  <U Mild>  dQALY Q ExacMild  <NExac Mild4049> ild4049  <NE <NExac Mild5059>  <Propp Major j Exac Mild>  dQALY ExacMild4049  QALY ExacMild4049  <Prop Major Exac Mild>  dQALY ExacMild5059  <NExac Mild6069> QALY ExacMild5059  dQALY E Mild6069 ExacMild6069 <U MildExacMin>  <U MildExacMin>  ldExacMin> <U MildExacMaj>  <U OH4049>  <Propp Major j Exac Mild>  <U MildExacMaj>  <U U MildExacMaj> MildE M j <U OH5059>  <NExac Mild7099> QALY ExacMild6069  <Prop Major Exac Mild> QALY ExacMild7099  dQALY E Mild7099 ExacMild7099 <U MildExacMin>  <U OH6069>  <U MildExacMaj>  <U U OH7099> OH7099  dQALY RegMild  <N Mild4049>  dQALY R Mild4049 RegMild4049  176  U Mild>  QALY RegMild4049  U OH4049 O 4049  dQALY RegMild5059 <U Mild>  <N Mild7099>  <N Mild6069>  <N Mild5059> ild 0 9  QALY RegMild5059 R Mild5059  U OH5059  dQALY RegMild6069 <U Mild>  QALY RegMild6069  U OH6069  dQALY RegMild7099  QALY RegMild7099 ild 099  U OH7099 <U U Mild Mild>  Figure A.2 Detailed structure of Vensim model- continued  dQALY Q MortMild  NExac Mod4049  <NExac ModNS4049>  <NExac ModXS5059>  NExac Mod5059  <NExac ModXS6069>  NE NExac M Mod6069 d6069  <NExac ModNS6069 ModNS6069>  <NExac ModNS5059>  <NExac ModXS7099>  NExac Mod7099  <NE <NExac ModNS7099> M dNS7099  N Exac Mod  <NExac SevS4049 SevS4049> <NExac NExac SevXS4049>  <NExac SevNS4049>  NE NExac S Sev4049 4049  <NExac SevXS5059> <NExac SevNS5059>  <NExac SevS7099 SevS7099>  <NExac SevS6069>  <NExac SevS5059> NExac Sev5059  <NExac SevXS6069>  <NExac SevNS6069>  NExac Sev6069  <NExac NExac SevXS7099> <NExac SevNS7099 SevNS7099>  N Exac E S Sev  NExac Sev7079  Figure A.2 Detailed structure of Vensim model- continued  <NExac ModXS4049>  <NExac ModS7099>  <NE <NExac ModS6069> M dS6069  <NExac NExac ModS5059>  <NExac ModS4049>  177  <FUDeath Sev4049>  <FUDeath Sev5059> Sev5059  dQALY SevMod4049  <U Sev>  QALY MortSev4049  <U OH4049>  <FUDeath Sev6069>  dQALY Q MortSev5059  <U Sev>  QALY MortSev5059  QALY Q MortSev6069  dQALY MortSev6069  <U OH6069>  <U S Sev> >  <U U OH5059 OH5059>  <FUDeath Sev7099>  dQALY MortSev7099 o tSev7099  QALY Q MortSev7099  <U OH7099>  <U Sev>  dQALY ExacSev E S  <NExac Sev4049>  <NExac <NE S 5059> Sev5059>  <Prop Prop Major Exac Sev Sev>  dQALY ExacSev4049 S E Mi > SevExacMin>  QALY Q ExacSev4049  <U OH4049>  <U SevExacMaj> S E M j>  <Prop Major Exac Sev>  dQALY ExacSev5059  <NExac Sev6069> QALY ExacSev5059  <U SevExacMin>  dQALY ExacSev6069 <U US SevExacMin> E Mi  <U OH5059>  <NExac Sev7079>  <Prop P M Major j E Exac S Sev> QALY ExacSev6069  dQALY ExacSev7099 <U SevExacMin>  <U U OH6069 OH6069>  <U SevExacMaj> S E M j>  <U SevExacMaj> S E M j>  <Prop Major <P M j E Exac Sev> S > QALY ExacSev7099  <U U OH7099 OH7099>  <U SevExacMaj> S E M j>  dQALY RegSev  <N Sev4049>  dQALY RegSev4049  178  <U Sev>  <N Sev6069>  <N Sev5059> QALY RegSev4049  <U U OH4049 OH4049>  dQALY Q RegSev5059 g <U Sev>  QALY RegSev5059  <U OH5059>  dQALY RegSev6069  <U U Sev> Sev  <N Sev7099> QALY Q RegMod6069 g 0  <U OH6069>  dQALY RegSev7099 <U Sev>  QALY Q RegSev7099 g  <U OH7099>  Figure A.2 Detailed structure of Vensim model- continued  dQALY MortSev  <dQALY MortMild>  <dQALY ExacMild> E acMild>  dQALY Q Exac  <dQALY dQALY RegMild>  <dQALY MortMod> M tM d  <dQALY ExacMod>  dQALY Mild  dQALY Reg  <dQALY RegMod>  <dQALY dQALY MortSev>  <dQALY RegSev>  <dQALY dQALY ExacSev>  dQALY Sev  dQALY Mod  <MyTime>  <Discount Discount Rate> Rate  Cost  QALY  dCost  dQALY  Discount Rate  <MyTime>  <Cost Mild>  <DirCost Mild>  <Cost Mod>  <DirCost Mod>  <Cost Sev>  <Di C t S <DirCost Sev> > Direct Cost  dDirCost  <Discount Rate>  <MyTime> y  Ti k Tick Indirect Cost dI dC t dIndCost  M Ti MyTime  179  Ticking <IndCost Mod> <I dC t Mild> <IndCost  <IndCost Sev>  Figure A.2 Detailed structure of Vensim model- continued  dQALY Q Mort  DirUCost R Mild RegMild  IndUCost ExacMild  IndUCost RegMild  IndCost RegMild g  DirCost ExacMild  DirCost RegMild g  IndCost ExacMild  DirUCost ExacMild  DirCost Mild  I dC t Mild IndCost  Cost Mild ild  <N Mod>  <N N Exac E M Mod> d  DirUCost RegMod  IndUCost ExacMod  IndUCost RegMod  I dC RegMod IndCost R M d  IndCost ExacMod  Di C t RegMod DirCost R M d  DirUCost ExacMod  DirCost ExacMod  Di C t Mod DirCost M d  IndCost Mod  Cost Mod  <N S Sev> >  ndUCost RegSev  IndCost RegSev  <N Exac Sev>  IndUCost ExacSev  IndCost ExacSev  DirUCost RegSev  DirCost RegSev g  DirCost Sev  IndCost dC S Sev  180 Cost Sev  DirUCost ExacSev  DirCost ExacSev  Figure A.2 Detailed structure of Vensim model- continued  <N Exac Mild> ild  <N Mild Treated>  <Mort Mort Rate 4049>  <Mort M R Rate 5059>  RR Mort Smokers  <Mort M 4049 4049> Mort Rate OHS4049 <New 4049>  <Cessation Rate OHS> OHS  Cessation Rate OHS  Nett Increase N I Pop4049 4049  Cessation OHS4049  <RR RR M Mort Smokers>  Mortt R M Rate t OHS5059  Cessation OHS5059  <RR Mort Smokers> Smokers Mort Rate OHS6069  <Cessation Rate OHS> <Mort Rate 7099>  <RR Mort Smokers>  Mort Rate OHS7099  Cessation OHS6069  Cessation OHS7099  Mort OHS6069 Mort OHS7099  New OHS4049  S ke Smokers OH4049  Aging g g OHS4049  Prev COPD S4049  RR COPD Smokers  Smokers OH5059  Aging OHS5059  COPD S5059  COPD S4049  Effect of Screening Test on Incidence  <Mort Mort Rate 6069>  Mort OHS5059  Mort OHS4049  <Aging g g 4049>  Proportion p of Current Smokers  <Cessation Rate OHS> OHS  Prev COPD 5059  Aging OHS6069  COPD S6069 Prev COPD S6069  Prev COPD S5059  <RR COPD Prev COPD 4049 P S k Smokers> This is prevalence p of ONLY Mild stagei g . All initial values used 1.17,, 1.31, nad 2.02 to adjust for total prevalences  S ke Smokers OH6069  <RR CO COPD Smokers>  S k Smokers OH7099  COPD S7099 Prev COPD S7099  Prev COPD 6069  <RR COPD Smokers> Smokers  Prev COPD 7079  Figure A.2 Detailed structure of Vensim model- continued  Effect of Screening Test on Prev  181  Prop p Major j Exac Mild  Exac Rate Mild  NMajorExac TPS4049  <Mort Rate OHS4049>  <Exac Exac Rate Mild> Mild <Prop Major Exac Mild> NMajorExac TPS5059  NExac TPS4049 <Mort Mort Rate OHS5059>  Mort Exac TPS4049  Mort TPS4049  <Prop Prop Major Exac Mild>  Mort Rate Major Exac  NExac TPS5059  Mort Exac TPS5059  Mort TPS5059  Mort TPS6069  COPD S4049> S4049  New TPS4049  <COPD S5059>  Cessation TPS4049 pecificity P Prog TPS4049  Test Uptake Rate PostDiag Cessation Rate  Prog Rate Mild Smokers  <COPD S6069>  Cessation TPS5059 <Sensitivity> <Test Uptake Rate> Rate  P TPS5059 Prog <PostDiag Cessation Rate>  <Progg Rate Mild Smokers>  <Mort Rate Major Exac>  TP Smokers 7099  Aging TPS5059 New TPS5059  NExac TPS7099  Mort Exac TPS7099  Mort TPS7099  TP Smokers 6069  Aging TPS4049  ensitivity  <Mort Rate OHS7099>  <Mort Rate Major E Exac>  <Exac Exac Rate Mild  NMajorExac TPS7099  NE NExac TPS6069  M t Exac Mort E TPS6069  <Mort M Rate R M Major j Exac>  TP Smokers S k 5059  TP Smokers k 4049  NMajorExac TPS6069  <Mort Rate OHS6069>  <Prop Major Exac Mild>  <E <Exac Rate R t Mild>  A i TPS6069 Aging New TPS6069  <COPD S7099>  C Cessation ti TPS6069  <Sensitivity> S ii i <Test Uptake p Rate>  Prog TPS6069  <PostDiag Cessation Rate>  <Prog Rate Mild Smokers>  New TPS7099 C Cessation ti TPS7099  <Sensitivity> <Test Uptake Rate> Rate  Prog TPS7099  <PostDiagg Cessation Rate>  <Progg Rate Mild Smokers>  Figure A.2 Detailed structure of Vensim model- continued  Effect of Predictive T t on Exac Test E R Rate t  182  Prop Major Exac Mild  Exac Rate Mild  NMajorExac TPS4049  <Mort Rate OHS4049>  <Exac Rate Mild> <Prop Major Exac Mild> NMajorExac j TPS5059  NExac TPS4049 <Mort Rate OHS5059>  Mort Exac TPS4049  Mort TPS4049  <Propp Major j Exac Mild>  M tR Mort Rate t M Major j Exac  NExac TPS5059  Mort Exac TPS5059  Mort TPS5059 S 0 9  Mort TPS6069  COPD S4049>  New TPS4049  <COPD COPD S5059 S5059>  Cessation TPS4049 pecificity Progg TPS4049  Test Uptake Rate PostDiag P Di Cessation Rate  Prog Rate Mild Smokers  <COPD CO S6069 S6069>  Cessation TPS5059 <S iti it > <Sensitivity> <Test Uptake p Rate>  Progg TPS5059 <PostDiag Cessation Rate>  <Prog Rate Mild Smokers>  <Mort Rate Major E > Exac>  TP Smokers S k 7099  Aging g g TPS5059 New TPS5059  NExac TPS7099  Mort Exac TPS7099  Mort TPS7099  TP S Smokers k 6069  Aging TPS4049  ensitivity  <Mort Rate OHS7099>  <Mort Rate Major Exac>  <Exac Rate Mild  NMajorExac TPS7099  NExac TPS6069  Mort Exac TPS6069  <Mort Rate Major Exac>  TP Smokers 5059  TP Smokers 4049  NMajorExac TPS6069  <Mort Rate OHS6069>  <Prop Major Exac Mild>  <Exac Rate Mild>  Aging TPS6069 New TPS6069  <COPD S7099>  Cessation TPS6069  <Sensitivity> <Test Uptake Rt > Rate>  Prog TPS6069  <PostDiag P Di Cessation Rate>  <Prog Rate Mild Smokers>  New TPS7099 Cessation TPS7099  <Sensitivity> S ii i <Test Uptake p Rate>  Prog TPS7099  <PostDiag Cessation Rate>  <Prog Rate Mild Smokers>  Figure A.2 Detailed structure of Vensim model- continued  Effect of Predictive Test on Exac Rate  183  <Exac Rate Mild>  NMajorExac UDS4049  <Mort Rate OHS4049>  <Mort Rate OHS5059>  <COPD S5059>  Cessation UDS4049  <PostDiag Cessation Rate>  <Prog Rate Mild Smokers>  Aging UDS5059 New UDS5059  <COPD COPD S6069 S6069>  Cessation UDS5059  <Sensitivity>  <PostDiag P Di Cessation Cessat o Rate> ate  UD Smokers 6069  <Prog Rate Mild Smokers>  <COPD COPD S7099> S7099  Cessation UDS6069 Prog UDS6069  <Test Uptake Rate> <PostDiag Cessation Rate>  <Prog og Rate ate Mildd Smokers>  NExac UDS7099  Mortt Exac M E UDS7099 <Mort Rate Major Exac> Exac  Mort UDS7099  Aging UDS6069  New UDS6069  <Sensitivity>  Prog UDS5059  <Test Uptake Rate>  <Mort Rate OHS7099> <Mort Rate Major Exac>  Mort UDS6069  <Exac Rate Mild>  NMajorExac UDS7099  NExac UDS6069  Mort Exac UDS6069  UD Smokers S k 5059  <Prop Major Exac Mild>  <Exac Rate Mild>  NMajorExac UDS6069  <Mort Rate OHS6069>  <Mort Mort Rate Major Exac>  Aging g g UDS4049  Prog UDS4049 Test Uptake Rate>  NExac UDS5059  Mort Exac UDS5059  Mort UDS5059  New UDS4049  <Sensitivity> Sensitivity>  <Prop Major E Exac Mild> NMajorExac UDS5059  <Mort Rate Major Exac>  UD Smokers 4049  <Exac Exac Rate Mild Mild>  NE NExac UDS4049  M t Exac Mort E UDS4049  Mort UDS4049  <COPD S4049>  <Prop Major Exac Mild>  UD S Smokers k 7099  New UDS7099 Cessation UDS7099  <Sensitivity>  Progg UDS7099  <Test Uptake Rate> <PostDiagg Cessation Rate>  <Prog Rate Mild Smokers>  Figure A.2 Detailed structure of Vensim model- continued  <Prop Major Exac Mild>  184  <Exac Rate Mild>  NMajorExac FPS4049  Mort FPS4049  <Exac Rate Mild> <Prop Prop Major Exac Mild> NMajorExac NM j E FPS5059  NExac FPS4049 <Mort Mort Rate OHS5059>  Mort Exac FPS4049  <Mort Rate OHS4049 OHS4049>  <Prop Major <P M j Exac Mild>  <Mort Rate Major Exac>  NExac FPS5059  M FPS5059 Mort  <COPD COPD S4049 S4049>  New FPS4049  <COPD S5059>  <Mort Rate Major E Exac>  New FPS5059  <COPD S6069>  Cessation FPS5059  Cessation FPS4049 <Specificity> Specificity  <PostDiag Cessation Rate>  <Prog Rate Mild Smokers>  P FPS5059 Prog <PostDiag Cessation Rate>  <Prog Rate Mild Smokers>  Mort Exac FPS7099 <Mort Rate Major j Exac>  Mort FPS7099  Aging FPS6069 New FPS6069  <COPD S7099>  New FPS7099  <Sensitivity>  <S iti it > <Sensitivity>  Cessation FPS7099  Cessation FPS6069 <Specificity>  <Specificity>  <Test Uptake Rate>  NE NExac FPS7099  FP Smokers 7099  Aging FPS5059  <Sensitivity>  P FPS4049 Prog Test Uptake p Rate>  <Mort Rate OHS7099>  FP Smokers 6069  Aging FPS4049  Sensitivity> <Sensitivity> Specificity>  Mort FPS6069  <Exac Rate Mild>  NMajorExac j FPS7099  NE NExac FPS6069  Mort Exac FPS6069 <Mort Rate Major Exac>  FP Smokers 5059  FP Smokers 4049  NMajorExac j FPS6069  <Mort M t Rate Rt OHS6069 OHS6069>  Mort Exac FPS5059  <Prop Prop Major Exac Mild>  <Exac Rate Mild>  Prog FPS6069  <Test Uptake Rate> <PostDiagg Cessation Rate>  <Prog Rate Mild Smokers>  Prog FPS7099  <Test Uptake p Rate> <PostDiag Cessation C ti Rate> Rt >  <Prog Rate Mild Smokers>  Figure A.2 Detailed structure of Vensim model- continued  <Prop Major Exac Mild>  185  New C N Cessation ti OHXS4049  Mort OHXS5059  Mort OHXS4049  <RR Mort XSmokers> XS k >  Mort Rate OHXS7099 Mort OHXS6069 Mort OHXS7099  New OHXS4049  XSmokers XS k OH4049  Aging OHXS4049  <Prev COPD 4049>  Aging OHXS5059  <RR COPD XSmokers>  XSmokers XS k OH6069  <Prev COPD 5059>  Aging OHXS6069  COPD XS6069 Prev COPD XS6069  Prev COPD P XS5059  Prev COPD XS4049  RR COPD XSmokers k  XSmokers XS k OH5059  COPD XS5059  COPD XS4049 Proportion of XSmokers  New Cessat Cessation o OHXS7099  <Mort Rate 7099>  <RR RR M Mort XSmokers> Mort Rate OHXS6069  Mort Rate OHXS5059  Mort Rate OHXS4049 <Net Increase Pop4049> p  <Mort Rate 6069>  <Cessation C i OHS7099>  New Cessation OHXS6069  New Cessation OHXS5059  <Mort Rate <RR Mort 5059> XSmokers>  RR Mort XSmokers  <Mort Rate 4049>  <Cessation C i OHS6069>  <Cessation OHS5059>  <RR COPD XSmokers>  XSmokers OH7099  COPD XS7099 Prev COPD XS7099  <Prev COPD 6069>  <RR COPD XSmokers> k  <Prev COPD 7079>  Figure A.2 Detailed structure of Vensim model- continued  <Cessation OHS4049>  186  <Mort M tE Exac TPS5059>  <Mort Exac TPS6069>  <Mort Exac TPS7099>  Mort Exac TPS  NMort Mild4049  <Mortt Exac <M E FNS4049>  <Mort Exac FNS5059>  <Mort Exac FNS6069>  <Mort Exac FNS7099>  NMort Mild7099  Mort Exac FNS  <Mort Exac UDS4049 UDS4049>  <Mort Exac UDS5059>  <Mort Exac UDS6069>  <Mort o t Exac ac UDS7099>  Mort Exac UDS <Mort Exac TPXS4049>  <Mort Exac TPXS5059>  <Mort Exac TPXS6069>  <Mort Exac TPXS7099>  Mort Exac TPXS <Mort Exac FNXS4049>  <Mort Mort Exac FNXS5059>  <Mort Exac FNXS7099>  <Mort Mort Exac FNXS6069>  Mort Exac FNXS <Mort Mort Exac UDXS4049>  <Mort Exac UDXS5059>  <Mort Exac UDXS6069 UDXS6069>  <Mort Exac UDXS7099> NMort Mild6069  Mort Exac UDXS  <M t Exac <Mort E TPNS4049>  <Mort Exac TPNS5059>  <Mort Exac TPNS6069>  <Mort Mort Exac TPNS7099>  Mort Exac TPNS  NMort Mild5059  <Mort Exac FNNS4049> FNNS4049  <Mort Exac FNNS5059>  <Mort Exac E ac FNNS6069>  <Mort M t Exac E FNNS7099>  Mort Exac FNNS  187  <Mort Exac UDNS4049 UDNS4049>  <Mort Exac UDNS5059 UDNS5059>  <Mort Exac UDNS6069>  Mort Exac UDNS  <Mort Exac UDNS7099>  Figure A.2 Detailed structure of Vensim model- continued  <Mort M t Exac E TPS4049>  <Mort Mort Exac M dS5059> ModS5059>  <Mort Exac ModS6069> ModS6069  <Mort Exac ModS7099>  M tE Mort Exac ModS M dS  NMort Mod7099  Mort o Mod4049 od 0 9  <Mort Exac ModXS4049 ModXS4049>  <Mort Exac ModXS5059>  <Mort Exac ModXS6069 ModXS6069>  <Mort Exac ModXS7099 ModXS7099>  Mort Exac ModXS  ort Mod5059  NMort Mod6069  <Mort Exac ModNS4049>  <Mort M Exac E ModNS5059>  <Mort M Exac E ModNS6069>  Mort Exac M dNS ModNS  <Mort Exac ModNS7099>  Figure A.2 Detailed structure of Vensim model- continued  <Mort Mort Exac M dS4049> ModS4049>  188  NM t Sev4049 NMort S 4049  <Mort Exac SevXS4049 SevXS4049>  <Mortt Exac <M E SevS5059>  <Mort Exac SevS6069>  <Mort Exac SevS7099>  M tE Mort Exac S SevS S  NM S NMort Sev7099 7099  <Mort Exac <Mort Exac SevXS6069 SevXS5059 SevXS6069> SevXS5059>  <Mort Exac SevXS7099 SevXS7099>  Mort Exac SevXS NMort Sev6069  NMort Sev5059  <Mortt Exac <M E SevNS4049>  <Mort Mort Exac S NS5059> SevNS5059>  M t Exac Mort E S SevNS NS  <Mort Exac SevNS6069> S NS6069  <Mort Exac SevNS7099>  Figure A.2 Detailed structure of Vensim model- continued  <Mortt Exac <M E SevS4049>  189  <Smokers Smokers OH5059>  <Smokers Smokers OH6069>  <Smokers OH7099>  N OHS N Smokers <N MildS> <N ModS> <N N SevS SevS> <XSmokers OH4049>  <XSmokers OH5059>  <XSmokers <XSmokers OH6069> OH7099>  N OHXS  OH  Test Total Pop <N MildXS>  N XS XSmokers k  <N NM ModXS> dXS <N SevXS> <NSmokers OH4049>  <NSmokers <NSmokers <NS k OH5059> OH6069> OH5059  <NSmokers <NS k OH7099> OH7099  N OHNS <N MildNS> <N ModNS>  190  <N SevNS>  N NSmokers  Figure A.2 Detailed structure of Vensim model- continued  <Smokers OH4049>  N MildS  <N MildXS6069> <N MildXS7099>  <N MildXS4049><N MildXS5059>  N Mild  N MildXS  <N MildNS4049> <N MildNS5059><N MildNS6069>  <N N MildNS7099 MildNS7099>  N MildNS  <Mod Smokers <Mod Smokers <Mod Smokers 4049> 6069> 5059>  <Mod Smokers 7099>  N ModS <Mod XSmokers 4049>  <Mod XSmokers 5059>  <Mod M d XSmokers XS k 6069 6069>  <Mod XSmokers 7099>  N Mod  N ModXS M dXS Mod NSmokers <Mod NSmokers <Mod NSmokers <Mod <Mod <M d NS NSmokers k 7099> 6069> 6069 5059> 4049>  N ModNS <Sev Smokers 4049>  <Sev Smokers 5059>  <Sev Sev Smokers 6069>  <Sev Smokers 7099>  N SevS <Sev XSmokers 4049>  <Sev XSmokers <Sev Sev XSmokers 5059> 6069> 6069  <Sev Sev XSmokers 7099> 7099  N SevXS  191  <Sev Sev NSmokers 4049> 4049  <Sev NSmokers 5059>  N SevNS  <Sev Sev NSmokers 6069> 6069  <Sev NSmokers 7099>  N Sev  N COPD CO  Figure A.2 Detailed structure of Vensim model- continued  <N MildS4049> <N MildS5059> <N MildS6069> <N MildS7099>  Appendix B: Supplementary Material for Chapter 5 Figure B.1 Overall structure of discrete event simulation model in Arena  Pa t e i n t w it h N o d u le s  T h y r o id  M a lig n a n t D ia g n o s is  I n it ia liz e  T o t a l T h y r o id e c t o m M a lig n a n t  Page 3  0  As s g i n Pa r a m F NAB  F NAB  e t e r s  0  Pa r a m  Page 2  T o t a l T h y r o id e c t o m 0 F S u s p ic io u s  L o b e c t o m S u s p ic io u s S u s p ic io s D ia g n o s is  Su r g  t ype  y  F o lic u a l r F N  V si it Y e a r ly f o r Ye a r s Su s p  O  0  r  ig in a l  D u p lic a t  0 L o b e c t o m  1st  e  y H is  t o lo g y  F o lic u la r C a r c in o m  H is t o lo g y  a  5  Ty pe  of  Ca n c e r p e r p e r p e r p e r  c e n t c e n t c e n t c e n t  Re p e a t T h y r o id e c t o m  _ f c _ h c _ m c _ a cH u  Su r g e r y  A b la t io n  r t h u le C e l C a r c in o m a  y  M a lig n a n t L a t e  T h y r o id e c o m  y  0  u e  F N D ia g n o s is  F a ls e  M e d u la r y C a r c in o m a  p _ m a lig _ f n a b p _ s u s p _ f n a b p _ a u s _ f n a b p _ n d _ f n a b p _ b e n ig n _ f n a b  A b la t io n  0 A t y p ia o r U n s a t is f a c o r y  Re p e a t AUS AUS  F NAB  E ls e  0 0 F o lo w u p  D ia g n o s is V is it  Y e a r ly Ye a r s  f o r  H is t o lo g y  p _ m a lig _ a u s f n a b p _ s u s p _ a u s f n a b p _ f n _ a u s f n a b p _ b e n ig n _ a u s f n a b  T r  F a ls e T r  u e  B e n ig n  F a ls e  A n a p la s t ic C a r c in o m a  T r u e  N e g a t iv e  F a ls e  e  5  H y p o p a r a t h y r o id is m Pe r m a n n t  0  A s s ig n P a r a m F NAB DX  e t e r s  F NAB  ND  Ag e  p _ m a lig _ n d f n a b p _ s u s p _ n d f n a b p _ f n _ n d f n a b p _ b e n ig n _ n d f n a b  0  0 L o b e c t o m  y  a x ( t t t _ s u r  _ s u r g e r g e r y , h t  y , h t _ s u r  C a r c in o m a  _ s u r g e r e r y ) T r u e  0g  y )  M o r t a lit y ?  Re c o r d  u e  ND  Page 7  a * m a x ( t  T r a c k E ls e  T r  N D H is t o lo g y  0  L a r y n g e a l Ne r v e P a r a ly s is  0  Page 6  Ef f e c t s p _ h y p o p a r p _ L R N I * m  F N A B N D R e s u lt s  E ls e  C a r c in im a S p e c if ic M o r t a lit y  F a ls e  S id e Re p e a t No n d a i g n o s t ci U n s a t si f a c t o r y  Page 8  P a p ila r y C a r c in o m  a  DX  F o lic u la r C a r c in o m  a  DX  Page 9  H u r t h u le C a r c in o m  H is t o lo g y M a ligT nr au ne tC a n c e r DX D ia g n o s is D X  of  Ca n c e r p e r p e r p e r p e r  H is t o lo g y E ls e  0  e s  AE  0 0 0  Ce l a DX  0  T r  u e  M a lig n a n t  DX  0  T r u e  P o s it iv e  T r ia l  Re s e t  0 M e d u la r y C a r c in o m a  V a r ia b le s  DX  Ca n c e r  _ f c _ h c _ m c _ a c  u e  F a ls e  DX  c e n t c e n t c e n t c e n t  T r  F a ls e  0  Ty pe  0  y  u t c o m  En d No  En d DX  Su r g e r y  T o t a l T h y r o id e c t o m M a lig n a n t DX  O  F a ls e  F a ls e  A b la t io n  M a lig n a n t D ia g n o s is DX  Re p e a t  N e g a t iv e  P o s it iv e  F a ls e  u e  A U S H is t o lo g y  AUS  0  0  B e n ig n  Su r g e r y  B e n ig n  F N A B A U S R e s u lt s  H is t o lo g y B e nTigr nu D ia g n o s is  F NAB a n d M o le c u la r DX  Page 5  u e  F a ls e  F NA B  0  E ls e  T r  T r  Re c u r r e n c e  u e  F a ls e  Ca n c e r  N e o p la s m  T r  0  E ls e  0 C lo n in g 1  0  P o s it iv e  Ca n c e r  0  a  0  _ s u s p _ s u s p  E ls e  T r u e P a p ila r y C a r c in o m  F a ls e  H is t o lo g y S u s pTicr iouues D ia g n o s is  S u s p ic io u s p _ t t p _ h t  A s s ig n C h a r a c t r s it ic s F NAB  Page 1  H is t o lo g y M a ligT nr au ne t D ia g n o s is  y a ls e  0  e t r iz e 0  Page 4  y  0  T r u e R e c u r r e n c e RDeXp e a t  T h y r o id e c t o m  Ab a l t io n y D DX X  Page 10  Re p e a t  F a ls e  DX  F a ls e  F a ls e  N e g a t iv e  L a t e T h y r o id e c o m DX DX  y  A n a p la s t ic C a r c in o m a _ D X A s s ig n C h a r a c t r s it ic s F NAB DX  1st  F NA B DX p _ m  F a ls e a lig _ f  P o s it iv e  DX  n a b  H y p o p a r a t h y r o id is m Pe r m a n n t DX  E ls e  S id e  Ef f e c t s  DX  p _ h y p o p a r p _ L R N I * t t  0 Su r g e r y Be n g i n DX  a * t t _ s u r  C a r c in im a S p e c if ic M o r t a lit y DX  T r  B e n ig n  V is it Y e a r ly f o r Ye a r s DX  5  H is t o lo g y B e nTigr nu D ia g n o s is D X  e  0  T r  T r ia l D x  u e  F a ls e  En d L a r y n g e a l Ne r v e P a r a ly s is DX  F a ls e  T r u e  N e g a t iv e  Ag e DX  T r a c k  T r  u e  M o r t a lit y  DX ?  Re s e t DX  V a r ia b le s  F a ls e  F a ls e  No  AE  DX  0  C a r c in o m a  DX  0 0  u t c o mE neds  DX  0 D ia g n o s is  O  u e  H is t o lo g y  0  0  Be n g i n DX  0 Re c o r d DX  _ s u r g e r y _ d x g e r y _ d x  E ls e  DX  192  Figure B.2 Detailed structure of discrete event simulation model in Arena  193  194  195  196  197  198  199  200  201  

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:
http://iiif.library.ubc.ca/presentation/dsp.24.1-0072612/manifest

Comment

Related Items