Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Dental considerations in minimizing osteoradionecrosis in head and neck cancer patients : Delphi-derived… Cramer, Carl Kimberly 1997

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

Item Metadata


831-ubc_1997-0519.pdf [ 22.78MB ]
JSON: 831-1.0087778.json
JSON-LD: 831-1.0087778-ld.json
RDF/XML (Pretty): 831-1.0087778-rdf.xml
RDF/JSON: 831-1.0087778-rdf.json
Turtle: 831-1.0087778-turtle.txt
N-Triples: 831-1.0087778-rdf-ntriples.txt
Original Record: 831-1.0087778-source.json
Full Text

Full Text

D E N T A L CONSIDERATIONS IN MINIMIZING OSTEORADIONECROSIS IN H E A D A N D N E C K C A N C E R PATIENTS: DELPHI-DERIVED FACTORS FOR A DECISION A N A L Y S I S by C A R L K I M B E R L Y C R A M E R B.Sc. (Hons.) 1974, D.M.D. 1979 University of British Columbia, Vancouver, B.C., Canada A THESIS SUBMITTED IN PARTIAL F U L F I L L M E N T OF THE REQUIREMENTS FOR THE DEGREE OF M A S T E R OF SCIENCE in THE F A C U L T Y OF G R A D U A T E STUDIES (Department of Health Care and Epidemiology) We accept this thesis as conforming to the required standard THE UNIVERSITY OF BRITISH C O L U M B I A August 1997 © Carl Kimberly Cramer, 1997 In presenting this thesis in partial fulfilment of the requirements for an advanced degree at the University of British Columbia, I agree that the Library shall make it freely available for reference and study. I further agree that permission for extensive copying of this thesis for scholarly purposes may be granted by the head of my department or by his or her representatives. It is understood that copying or publication of this thesis for financial gain shall not be allowed without my written permission. Department of / / ^ g e ^ - Z ^ <^^-€, cZ-^U-'<^a-The University of British Columbia Vancouver, Canada Date A A B S T R A C T Osteoradionecrosis (ORN), a potential complication in head and neck cancer patients irradiated at high doses, can have severe consequences on patient quality of life and costs of treatment. Teeth can never be extracted from irradiated bone without the risk of osteoradionecrosis, but diseased teeth carry a greater ORN risk. The management of teeth so as to minimize the occurrence of ORN, while preserving maximal quality of life for all patients at risk, is a day-to-day problem for dental staff in many oncology treatment centers. Guidelines for the extraction of teeth prior to radiotherapy published by the National Institute of Health and elsewhere in the literature are not exact,1 so additional information, ideally a validated clinical decision analysis on the dental minimization of ORN, or an algorithm or clinical policy deriving from it, would be highly useful. This thesis is a step towards these clinical goals, producing data of immediate clinical interest in addition to data specially suited to ORN epidemiology and decision analysis. [i.e., they do not fully explain the decision process or the weighting of the factors related to the decisions] (Barker and Barker 1990). References cited in the Abstract are Barker, Bruce and G. Barker "Oral Complications of Management of Radiation Therapy to the Head and Neck." Northwest Dentistry 69 (5, Sept.-Oct. 1990): 19-23. National Institute of Health. "National Institutes of Health Consensus Development Conference: Oral Complications of Cancer Therapies: Diagnosis, Prevention, and Treatment." National Cancer Institute Monographs 1990 (9). 184 pp. A D A Oral Healthcare Guidelines. Head and Neck Cancer Patients receiving Radiation Therapy (Chicago: American Dental Association, 1989). Barker, Gerry, B. Barker, and R. Gier, and editors Peter Stevenson-Moore, Ernest Glass, et al. Oral Management of the Cancer Patient A Guide for the Health Care Professional 4th ed. (Kansas City, Missouri: University of Missouri, June 1992). [This represents consensus of experts including the members of the Executive Committee of the International Society for Oral Oncology.] ii The thesis is also justified on methodological grounds. Formal, traditional clinical decision tree analysis of complicated dental treatment dilemmas are very rarely found in the literature, and the incorporation of survey research is uncommon. Thus, this research is exploratory with respect both to the survey process used and its proposed outcomes. Official and semi-official documents indicating appropriate oral management of patients at risk for osteoradionecrosis (NIH 1990; A D A 1989; Barker, Barker, and Gier 1992) do not appear to have utilized formal consensus methodologies to reach their conclusions, and none appeared to provide an opportunity for anonymous dissent from the positions initially presented. This study does not, however, seek to duplicate the implied objective of NIH (1990) (to produce in a living document, general guidelines for clinicians). It primarily seeks, using formal consensus methodology, to examine systematically the assumptions made during preradiotherapy dental intervention decisions, point out weaknesses, indicate considerations relevant to a theoretical decision analytic framework, suggest a theoretical clinical decision analysis, and indicate research necessary to quantify further the decision process. By achieving consensus, the way to create validated decision analyses useful for clinician and patient education and decision support is made clear. Measures proposed in this research are also a step toward validated patient-derived utilities that would facilitate the implementation of decision analysis, cost-benefit, and quality control frameworks. The focus of the eventual theoretical decision analysis, and primary research question for which data are to be identified using a modified Delphi technique, is: "Under what conditions should teeth be extracted within the high dose radiation volume in order to minimize the frequency and severity of ORN?" The Delphi consensus approach was chosen for the Survey because it permitted analysis of responses from a iii select group of experts, provided anonymity required to minimize intimidation, and it did not require simultaneous geographical proximity of the experts and involvement of a trained panel-discussion facilitator. Open- and close-ended items, scenarios, and rating scales for agreement with Delphi statements and confidence in responses were presented to volunteer experts. 12 of 22 experts solicited from the U.S. and Canada (55%) participated in the Delphi survey, and 4 of 16 in Europe and Australia. (In other words, 28 of 38 experts (74%) solicited for participation in the Delphi Survey agreed to participate. 16 of the 38 (42%) actually returned the 60-page long survey.) The inclusion of the close-ended items in the survey was made practical by the body of literature on the clinical problem. This permitted combination of Delphi rounds one and two and useful comparison of responses; the comparisons generally implied high content validity for the close-ended items. Expert opinion consensus revealed that major risk factors for ORN were Radiation (particularly dose, dose per fraction, the timing of radiation, the osseous volume irradiated to high dose, and the presence or absence of interstitial therapy); Periodontal Status (as assessed using conventional static indicators, e.g. pocket depth); and Endodontic status. Anatomical site { a) mandible; b) posterior part of mandible; perhaps c) proximity of roots to surgical cuts} also were considered by respondents to be highly relevant to ORN risk. The oral hygiene status of the dentition, projected caries rate, patient compliance, and salivary status were considered relevant to ORN risk but are presumed to be mathematically dependent and to need to contribute to the decision analysis as a multifactorial integral indicating their combined ORN risk. Surgical trauma, particular metabolic states, anticipated mandibular perfusion capability, and mucosal friability were considered by respondents to be relevant to ORN risk or the decision to extract, but either they, or their effects, were regarded as difficult to quantify. {Factors proposed for inclusion in the decision analysis were assessed and characterized as to their suitability (i.e., quantifiability, controllability, and degree of independence). This approach was indicated by the natures of the dental variables potentially relevant to the clinical problem.} Only a minority of respondents indicated that hyperbaric oxygen or chemotherapy affected ORN prognosis. These findings may be significant, or artifacts of the survey's design. Close-ended items were not present to prove the reliability of these responses. A decision analysis on ORN-prophylactic tooth extraction should include factors related to the desirability of retaining particular teeth irradiated at high dose. Factors proposed as relevant to this proposed multifactorial utility sub-variable included caries rate, tooth mechanical soundness /restorability, and tooth functionality. The median minimum expert-recommended extraction healing time prior to radiotherapy was 7 days; the median desired time, 14 days (mean, 11.6). Factors integrated in the scales proposed in this study for ORN-associated pain and function were supported as related to quality of life. Pain was regarded the most important factor affecting quality of life, function and pathological fracture were regarded as secondary. (Other factors were suggested and merit further investigation.) Function levels in various ORN categories clearly appear to be associated negatively with pain, suggesting that quality of life utilities could, i f only validated pain data were available at some particular time in the future, be related (as an approximation) to pain alone, at first. The clinimetric scale outcome measures suggested in this thesis could be incorporated in frameworks for quality improvement or cost-benefit analysis, although conclusions drawn would be limited by their ordinality and lack of validation. Although all data reported are subject to reliability and validity constraints outlined in the thesis, qualitative data supplied on important topics were quite consistent from item to item, and relatively few (and non-critical) items were associated with low consensus or other proposed indicator of poor data quality. Most quantitative estimates of ORN risk were judged to lack adequate quality and validity, but pain and function assessed in two ways for categorized ORN demonstrated consistency, and seemingly remarkable consistency was present in the extreme ORN categories- ORN3b (progressive ORN with jaw fracture,) and ORNO (patient at risk, no ORN). Cross-impact analysis and other statistical manipulations designed to correct for potential confounders generally were inadvisable due to the small numbers of responses present. A theoretical, introductory decision analysis on the dental minimization of osteoradionecrosis is presented at the conclusion of the thesis. Further research is required before it can be validated and used clinically. Actions advocated to enable this, and to enhance the data base for all future studies on the dental management of ORN, are: 1. Adoption of an identical definition for osteoradionecrosis of the jaws, in future studies 2. Application to the Canadian Institute of Health Information to recommend the adoption by the "International Classification of Diseases" of a specific diagnostic code for osteoradionecrosis of the jaws. 3. Initiation of ongoing collection of epidemiological data for ORN of the jaws from non-experimental data bases employing such definitions. 4. Initiation of further prospective data gathering in patients at risk for ORN, with detailed recording of the treatments employed and the ORN prognostic factors outlined above, and ORN outcomes 5. Adoption of a common standard for categorization of ORN outcomes 6. Integration of existing data on the distribution of categorized ORN severity 7. Assessing expert consensus on the relationships of chemotherapy and hyperbaric oxygenation to ORN rates and healing, anonymously 8. Creation of validated patient-derived outcome scales for osteoradionecrosis 9. Consideration of new decision trees, e.g. on prosthetic or physiological rehabilitation. 10. Refinement of existing decision trees, once the data base improves 11. Validated clarification of interactions and effects of potentially dependent ORN risk variables. vii T A B L E OF CONTENTS ABSTRACT ii. TABLE OF CONTENTS viii. LIST OF TABLES xi. LIST OF FIGURES xiv. ACKNOWLEDGMENTS xvi. I. INTRODUCTION 1 1. T H E C L I N I C A L P R O B L E M 1 2. D E C I S I O N A N A L Y S I S . 4 A . OVERVIEW 4 B . DECISION TREES 7 C . UTILITIES 11 D . SENSITIVITY ANALYSIS 13 E . DATA FOR PROBABILITIES AND UTILITIES 15 F . DECISION ANALYSIS AND ... THE PATIENT 17 G . DECISION ANALYSIS AND ... BAYESIAN REASONING 19 H . DECISION ANALYSIS AND ... DENTISTRY 20 3. O R N - P R O P H Y L A C T I C D E C I S I O N A N A L Y S I S A N D T H I S T H E S I S 22 II. LITERATURE REVIEW 25 0. E X E C U T I V E S U M M A R Y 25 1. D E L P H I C O N S E N S U S T E C H N I Q U E ; O P I N I O N D A T A 28 2. U T I L I T I E S A N D O U T C O M E V A R I A B L E S 38 3. C L I N I C A L P R O B L E M 45 A . ORAL CANCER EPIDEMIOLOGY A N D PREDISPOSING FACTORS 45 B . ORAL SQUAMOUS CELL C A . LESIONS A N D TREATMENT STRATEGIES.... 52 EXTERNAL IRRADIATION 56 BRACHYTHERAPY 58 C . PATHOLOGY OF RADIATION INJURY 61 D . OSTEORADIONECROSIS PATHOGENESIS, DEFINITION, DIAGNOSIS, CLINICAL PRESENTATION, A N D TREATMENT 67 HYPERBARIC OXYGEN, OSTEORADIONECROSIS TREATMENT, A N D DENTAL CARE OF EXISTING LESIONS 76 E . OSTEORADIONECROSIS-PROPHYLACTIC DENTAL SCREENING STRATEGIES (STRESS: PERIODONTAL, ENDODONTIC, A N D INFECTION STATUS) 82 F . OSTEORADIONECROSIS EPIDEMIOLOGY 87 4. F A C T O R S T H A T M A Y P R E D I S P O S E O S T E O R A D I O N E C R O S I S : S U M M A R Y 145 viii III. METHODOLOGY 166 T H E D E L P H I P R O C E S S 1. R A T I O N A L E A N D S T R A T E G Y 166 2. T H E S A M P L E 168 3. R E S P O N S E R A T E 169 4. C R E A T I O N O F S U R V E Y O U T C O M E S C A L E S 171 5. A N A L Y S I S A . QUANTITATIVE DATA 175 B . QUALITATIVE DATA 177 6. P R E S E N T A T I O N O F D A T A 183 IV. RESULTS: SURVEY FINDINGS 186 P R E F A C E 186 1. D I R E C T A S S E S S M E N T O F O R N R I S K F A C T O R S 188 A . GENERAL AND DENTAL 188 B . IRRADIATION To HIGH DOSE, A N D RELATED 202 C . ENDODONTICS IN GREATER DETAIL 204 D . RELATION OF O R N RISK TO TIMING OF DENTAL EXTRACTIONS AND IN RELATION TO O R N RATES REPORTED BY RESPONDENTS....206 E . RELATION OF ORAL-SURGICAL, MEDICAL, AND PROSTHETIC TECHNIQUE TO O R N RISK 210 2. A S S E S S M E N T O F C H A R A C T E R I S T I C S O F O R N " R I S K F A C T O R S " . . 2 1 3 A . PERCEIVED RELEVANCE TO ORN-PROPHYLACTIC TREATMENT PLANNING 213 B . FACTOR PRACTICALITY FOR DECISION ANALYSIS, E.G. QUANTIFIABILITY....216 C . CLINICAL CONTROLLABILITY OF ORN RISK FACTORS 217 O R A L H Y G I E N E S T A T U S A N D ITS P R E D I C T I O N 221 3. F A C T O R S R E L E V A N T T O E X T R A C T I O N O F T E E T H I N T H E H I G H D O S E R A D I A T I O N V O L U M E 223 A . GENERAL 223 B . SPECIFIC 231 C . QUALITY OF POTENTIAL INDICATORS FOR ORN-PROPHYLACTIC EXTRACTION 234 4. Q U A N T I T A T I V E E S T I M A T E S O F O R N R I S K ...235 5. O U T C O M E S . A N D F A C T O R S R E L A T E D T O O U T C O M E S 237 A . SEVERITY OF O R N LESIONS 237 B . ORN-ASSOCIATED FACTORS THAT M A Y AFFECT A PATIENT'S QUALITY OF LIFE 237 C. ANATOMICAL LOCATION AND O R N CLFNICALPRESENTATION 239 D . FACTORS SKEWING O R N SEVERITY DISTRIBUTIONS 242 E . OSTEORADIONECROSIS SEVERITY DISTRIBUTIONS 245 F. O R N SEVERITY CATEGORY ASSOCIATIONS WITH PAIN A N D FUNCTION 245 6. S U R V E Y : P O T E N T I A L S O U R C E S O F B I A S 257 A . DEFINITION OF ORN 257 B . RESPONDENT AND PRACTICE CHARACTERISTICS 258 7. R E S P O N S E S O F L O W D A T A Q U A L I T Y 260 ix V. DISCUSSION 261 1. I N T E R P R E T A T I O N O F S U R V E Y F I N D I N G S 261 2. P O T E N T I A L S O U R C E S O F B I A S 282 A . OVERLAP OF DELPHI RESPONDENTS WITH CITED AUTHORS 282 B . DEFINITION OF OSTEORADIONECROSIS 283 C RESPONDENT PERSONAL AND PRACTICE CHARACTERISTICS 286 3. I N T E G R A T I O N : F A C T O R S U I T A B I L I T Y F O R D E C I S I O N A N A L Y S I S 292 4. R E L I A B I L I T Y 297 5. V A L I D I T Y C O N S I D E R A T I O N S A N D L I M I T A T I O N S 299 6. B IAS , R E L I A B I L I T Y , & V A L I D I T Y C O N C L U S I O N S & APPLICATIONS. . . 304 VI. CONCLUSIONS 307 1. C O N C L U S I O N S 307 2. L I M I T A T I O N S ; D A T A Q U A L I T Y 315 3. R E C O M M E N D A T I O N S 318 BIBLIOGRAPHY 327 APPENDIX 1 LARGE TABLES 364 APPENDIX 2 LETTERS AND FORMS SENT TO RESPONDENTS 374 APPENDIX 3 DELPHI SURVEY 385 APPENDIX 4. RESPONSES OF LOW DATA QUALITY (SOME DETAILS).. 446 APPENDIX 5. CLASS OF DATA GENERATING UNDEFINED OUARTTLES. .448 X LIST OF TABLES: 1 . 1993 B.C. CANCER INCIDENCE RATES (from B.C. Cancer Agency Annual Report) 47 2. MAJOR RISK FACTORS ASSOCIATED WITH ORAL CANCER (after table of same name in Anon./Douglass (1996, 3) 50 3. "CAUSES" OF OSTEORADIONECROSIS [AMONG 71 CASES] (from Daley, Drane and MacComb (1972) 89 4. RESULTS OF PUBLISHED ORN INCIDENCE STUDIES Table Format and First 11 Rows after Morton (1986) Table I "The Results of Published ORN Incidence Studies" 94 5. TUMOR SITES AND INCIDENCE [RATE] OF ORN Modification of Morton (1986) Table III "The Incidence of ORN at Different Tumor Sites" 94 6. TUMOR STAGE AND INCIDENCE OF ORN Modification of Morton (1986) Table IV "Tumor Stage and Incidence of ORN" 95 7. TREATMENT AND INCIDENCE OF ORN Modification of Morton (1986) Table V "Treatment and Incidence of ORN" 95 8. "RISK FACTORS FOR LATE COMPLICATIONS: RESULTS OF MODEL FITTING" {Reproduces Table 3 of same name, Withers et al (1995a)} 98 9. "RESULTS OF MULTIVARIATE ANALYSIS (COX MODEL)" (Reproduction of part of Table 4 of same name, Withers et al 1995b) 99 10. SOME OSTEORADIONECROSIS CASE DATA, SHffiUYA et al (1993) STUDY 102-103 11. "ASSOCIATION BETWEEN DENTAL DISEASE & RADIATION NECROSIS OF THE MANDIBLE IN 46 PATIENTS IRRADIATED FOR MALIGNANT TUMORS OF FLOOR OF MOUTH AND RETROMOLAR TRIGONE" [Title & raw data from Table 1, Murray et al (1980b) p. 101; algebraic equivalents added.] 108 12. SOME OSTEORADIONECROSIS CASE DATA (PERNOT ET AL (1995) STUDY) 110 13. OSTEORADIONECROSIS CASE DATA (SUMMARY OF HERZOG, SADER, DEPPE & ZEILHOFER (1995) STUDY).. 113-114 14. OSTEORADIONECROSIS CASE DATA (BEUMER, HARRISON, SANDERS AND KURRASCH 1984 STUDY) 119-120 15. OSTEORADIONECROSIS CASE DATA (SUMMARY OF LEGROS ET A L 1984 STUDY)... 123 16. A SAMPLE OF THE EXCEL MASTER SPREADSHEET 178 17. A SAMPLE TABULATION OF QUALITATIVE DATA (ITEM 1050) 181 18. POTENTIAL ORN RISK FACTORS. A L L ITEM 1140 RESPONSES 364-365 19. ITEM 1140. INDIVIDUAL POTENTIAL ORN RISK FACTORS 189 20. ITEM 1140. POTENTIAL ORN RISK FACTORS GROUPED BY CONCEP 189 21. ITEM 3200 EVALUATION OF FACTORS AS PROGNOSTICATORS FOR ORN FREQUENCY 191 22. ITEMS 3240-3420. POTENTIAL ORN RISK FACTORS 194 23. ITEM 3450. PERIODONTAL STATUS PREDICTORS OF ORN RISK 197 xi LIST OF TABLES, continued: 24. ITEM 3460. PERIODONTAL PREDICTORS OF ORN RISK, BY IMPORTANCE RANK... 199 25. ITEM 3470. OCCLUSAL FACTORS AS ORN PREDICTORS 199 26. ITEM 3440 INFECTION PREDICTORS OF ORN RISK 200 27. ITEM 1200 RADIOTHERAPEUTIC CHARACTERISTICS REDUCING ORN RISK 202 28. ITEM 3180. RADIATION VARIABLES AS ORN RISKS 202 29. NORMALLY RECOMMENDED AND ABSOLUTE MINIMUM RECOMMENDED PRE-RADIOTHERAPY DAYS OF HEALING, VS. REPORTED ORN RATES...207& 366 30. ITEMS 1520 AND 1320,1330, &1340. ABSOLUTE MINIMUM RECOMMENDED PRE-RADIOTX TOOTH SOCKET HEALING TIMES VS. REPORTED ORN FREQUENCY 366 31. ITEMS 1510 AND 1320, 1330, &1340. NORMALLY RECOMMENDED PRE-RADIO-TX TOOTH SOCKET HEALING TIMES VS. REPORTED ORN FREQUENCY 366 32. ITEM 1230 DENTAL EXTRACTION TECHNIQUES TO MINIMIZE ORN 210 33. SUMMARY OF RESPONSE ANALYSES FOR SOME ITEMS 215 34. ITEM 1690. DENTAL EXTRACTION FACTORS NOT LARGELY UNDER THE CONTROL OF THE DENTIST 218 3 5. EXCEPTIONS TAKEN TO POTENTIAL FACTOR BEING BEYOND DENTIST'S CONTROL 220 36. ITEM 1240. PREDICTORS OF FUTURE ORAL HYGIENE PERFORMANCE 222 3 7. ITEM 1250. PREDICTORS OF COMPLIANCE WITH FLUORIDE, ETC, HOME CARE 222 38 . ITEM 1130. HIGH-DOSE-VOLUME TOOTH EXTRACTION MAJOR INDICATIONS 227 39. ITEM 1180. NON-RADIOTHERAPEUTIC INDICATIONS FOR TOOTH EXTRACTION 229 40. ITEM 1190. RADIOTHERAPEUTIC FACTORS AFFECTING DECISION TO EXTRACT MANDIBULAR TEETH IN THE PRE-RADIOTHERAPY WINDOW...230 41. ITEM 1890 SERIES. FACTORS INDICATING EXTRACTION (BY DECREASING AGREEMENT CONSENSUS- - ONLY DATA OF ADEQUATE CONSENSUS REPORTED) 367 42. ITEMS 2110-3030. [BY DESCENDING AGREEMENT]: SUITABILITY OF FACTORS AS CLASSIC INDICATIONS FOR EXTRACTION.234 43. ITEM 3610 SCENARIO: RECOMMENDED CHANGES IN MANAGEMENT 368 44. ITEM 3640 SCENARIO: RECOMMENDED CHANGES IN MANAGEMENT 368 45. ITEM 3670 SCENARIO: RECOMMENDED CHANGES IN MANAGEMENT 368 46. ITEM 3700 SCENARIO: RECOMMENDED CHANGES IN MANAGEMENT 368 47. ITEM 3730 PART A. SCENARIO: RECOMMENDED CHANGES IN MANAGEMENT 368 48. ITEM 3790 SCENARIO: RECOMMENDED CHANGES IN MANAGEMENT 369 49. ITEM 3760 SCENARIO: RECOMMENDED CHANGES IN MANAGEMENT 369 xii LIST OF TABLES, concluded: 50. ITEM 3820 SCENARIO: RECOMMENDED CHANGES IN MANAGEMENT 369 51. ITEM 3730 SCENARIO: PART B. PERCENTAGE CHANGES IN ORN IF INTERVENTIONS OCCUR OR DO NOT OCCUR 369 52. ITEM 1040: FACTORS FOR CLASSIFYING ORN SEVERITY 237 53. ITEM 1050 ORN FACTORS AFFECTING PATIENTS' QUALITY OF LIFE, BY DESCENDING PRODUCT ' 23 8 54. ITEM 1060 DIFFERENCES BETWEEN ORN PRESENTATIONS IN MAXILLA AND MANDIBLE 240 55. ITEM 1070 DIFFERENT EFFECTS ON QUALITY OF LIFE OF MAXILLARY AND MANDIBULAR ORN 240 5 6. ITEM 1090 DIFFERENCES BETWEEN ORN PRESENTATIONS WITHIN MANDIBLE 241 57. ITEM 1110. DIFFERENT EFFECTS ON QUALITY OF LIFE OF ORN LESIONS IN DIFFERENT PARTS OF MANDIBLE 242 58. ITEMS 4040-4090. PAIN & FUNCTION CATEGORIES "OFTEN ASSOCIATED WITH" CATEGORIZED ORN 246 59. ITEMS 4040-4090. PAIN & FUNCTION CATEGORIES "OFTEN ASSOCIATED WITH" CATEGORIZED ORN (PAIN AND FUNCTION DATA SIDE BY SIDE)....246 60. ITEMS 4160- 4260: DATA AND CHARACTERISTICS OF DATA FOR PAIN AND FUNCTION DISTRIBUTION IN CATEGORIZED ORN 370-371 61. # 4040-4090. RESPONDENT ASSOCIATIONS OF CATEGORIZED PAIN AND FUNCTION WITH CATEGORIZED ORN, AS PERCENTAGES 372 62. #4160-4260 PERCENTAGE DISTRIBUTION OF CATEGORIZED PAIN AND FUNCTION IN ORN CATEGORIES ( 5 RESPONDENTS) 372 63. "MEANS" OF #4040-4090 AND #4160-4260 DATA PERCENTAGES 372 64. ITEM 1010. "CRITERIA FOR DEFINING ORN" 257 65. ITEM 1020. PERCEIVED ETIOLOGIC/ PATHOPHYSIOLOGIC FACTORS 258 66. ITEM 1360 RADIOTHERAPEUTIC FACTORS THAT M A Y VARY AND INFLUENCE CLINICIAN PERCEPTION 258 67. SOME CHARACTERISTICS OF RESPONDENTS' PRACTICES 259 68. HYPOTHETICAL USAGE IN DECISION ANALYSIS OF FACTORS THAT ITEM 3200 RESPONDENTS INDICATE AS AFFECTING ORN RATE 266. 69. MODIFYING DATA IN SUBSEQUENT ITEMS, FOR FACTORS POSTULATED FOR INCLUSION IN DECISION ANALYSIS IN TABLE 68 294-96 70. SUMMARY OF MEASURES OF ASSOCIATION AND DISPERSION, IN ITEMS ASSOCIATED WITH RELIABILITY CALCULATIONS 373 71. DATA PAIRS FOR RELIABILITY COMPARISONS 373 72. ITEM 1770 DATA, TYPIFYING THAT GENERATING UNDEFINED QUARTILES 448 xiii LIST OF FIGURES: 1. "THE DECISION A N A L Y T I C APPROACH" after (Weinstein & Fineberg 1980, Figure 1-1, 5) 5 2. A GENERIC MEDICAL-SURGICAL DECISION TREE modified after Figure 6-7 (Sox et al 1988, 156) "The complete decision tree for comparing medical and surgical treatment of a hypothetical disease" 8 3. A SIMPLE ORN-PROPHYLACTIC TOOTH EXTRACTION DECISION TREE 9 4. "ESTIMATED R E L A T I V E FIVE-YEAR PERCENTAGE S U R V I V A L FOR SELECTED TYPES OF C A N C E R B Y R A C E , UNITED STATES, 1988" after Figure 15, (White, Caplan, and Weintraub 1995, 54) . 48 5. RESULTS CHART 1. RECOMMENDED MINIMUM NUMBER OF DAYS POST-EXTRACTION PRE-RADIOTHERAPY HEALING TIME VS. REPORTED ORN RATE 208 6. RESULTS CHART 2 NORMAL RECOMMENDED NUMBER OF DAYS POST-EXTRACTION PRE-RADIOTHERAPY HEALING TIME VS. REPORTED ORN RATE 208 7. RESULTS CHART 3 (ITEMS 4040-4090) PAIN AND FUNCTION, FREQUENCY OF ASSOCIATION WITH ORN CATEGORY, COMPARED 246 8. RESULTS CHART 4 (ITEMS 4160-4260) MEAN DISTRIBUTIONS OF PAIN IN CATEGORIZED ORN 249 9. RESULTS CHART 5 (ITEMS 4160-4260) MEAN DISTRIBUTIONS OF FUNCTION IN CATEGORIZED ORN 249 10. RESULTS CHART 6. #4040 FREQUENT ASSOCIATIONS, & #4160 DISTRIBUTION, OF PAIN IN ORN 3b 251 11. RESULTS CHART 7. #4040 FREQUENT ASSOCIATIONS, & #4160 DISTRIBUTION, OF FUNCTION IN ORN 3b 251 12. RESULTS CHART 8. #4040 FREQUENT ASSOCIATIONS, & #4160 DISTRIBUTION, OF PAIN IN ORN 3a 252 13. RESULTS CHART 9. #4040 FREQUENT ASSOCIATIONS, & #4160 DISTRIBUTION, OF FUNCTION IN ORN 3a 252 14. RESULTS CHART 10. #4040 FREQUENT ASSOCIATIONS, & #4160 DISTRIBUTION, OF PAIN IN ORN 2 253 xiv LIST OF FIGURES, concluded: 15. RESULTS CHART 11. #4040 FREQUENT ASSOCIATIONS, & #4160 DISTRIBUTION, OF FUNCTION IN ORN 2 253 16. RESULTS CHART 12. #4040 FREQUENT ASSOCIATIONS, & #4160 DISTRIBUTION, OF PAIN IN ORN lb 254 17. RESULTS CHART 13. #4040 FREQUENT ASSOCIATIONS, & #4160 DISTRIBUTION, OF FUNCTION IN ORN lb 254 18. RESULTS CHART 14. #4040 FREQUENT ASSOCIATIONS, & #4160 DISTRIBUTION, OF PAIN IN ORN la 255 19. RESULTS CHART 15. #4040 FREQUENT ASSOCIATIONS, & #4160 DISTRIBUTION, OF FUNCTION IN ORN la 255 20. RESULTS CHART 16. #4040 FREQUENT ASSOCIATIONS, & #4160 DISTRIBUTION, OF PAIN IN ORN 0 256 21. RESULTS CHART 17. #4040 FREQUENT ASSOCIATIONS, & #4160 DISTRIBUTION, OF FUNCTION IN ORN 0 256 22. A FIRST MODEL FOR AN ORN-MINIMIZING TOOTH EXTRACTION THEORETICAL DECISION TREE 324 23. A MODIFIED REDUCED FRAMEWORK FOR A N ORN-MINIMIZING DECISION ANALYSIS 325-326 xv Acknowledgments I would like to acknowledge the contributions of: • Martin Schechter, Thesis Supervisor. Professor, Department of Health Care and Epidemiology, and Head, Division of Epidemiology and Biostatistics • James Busser, Thesis Advisor. Clinical Assistant Professor, Department of Medicine, and MSI Director, Vancouver Hospital • Joel Epstein, Thesis Advisor. Clinical Professor, Department of Oral Medical and Surgical Sciences, Faculty of Dentistry; Head, Dental Programs, Vancouver Hospital; Oral Medicine Specialist, Division of Dentistry, B.C. Cancer Agency • Sam Sheps, Thesis Advisor. Professor and Head, Department of Health Care and Epidemiology. • Dr. Frances Wong, who along with Dr. Epstein reviewed the draft survey, and also Drs. Ruth Milner and Michael MacEntee, who made helpful suggestions prior to that time. • Participants in the Survey, who must remain anonymous. • My wife, Gillian, who transcribed my Literature Review dictation. • M.B . , J.H., and J.T. of the Faculty of Dentistry, for their moral support and advice. xvi To • My parents Carl F. and Aurora, my wife Gillian, and my brother Marc. • My renal transplant and orthopedic surgeons and physicians, particularly Doctors Michael Moriarty, E.C. "Burt" Cameron, Ted Reeve, Paul Keown, Clive Duncan, and John Price, and all the exemplary allied personnel associated with the Vancouver General Hospital Renal Programs, especially Lois Watson; and my family physician, Edward S. Robinson. xvii CHAPTER I. INTRODUCTION 1. THE CLINICAL P R O B L E M "The incidence, predisposing factors, and clinical course of [osteoradionecrosis] are subject to considerable variability and are dependent on many complex inter-relating factors" (Beumer, Harrison, Sanders and Kurrasch 1984 ). Patients treated for oral cancer face a range of consequences. Surgical therapy may directly compromise oral function, while high dosage irradiation may cause complications such as xerostomia (decreased salivation), taste loss, mucositis, dermatitis, trismus (muscle spasm), infections, dental caries, growth and developmental abnormalities, soft tissue necrosis, and osteoradionecrosis (NIH 1990). Osteoradionecrosis ("ORN") is necrosis of bone that may follow therapeutic irradiation for head or neck cancer in the absence of recurrent or metatstatic disease. Diagnosis is primarily based on clinical signs ~ ulceration or necrosis of the mucous membrane, with sustained exposure to the oral environment of the underlying necrotic bone. Osteoradionecrosis can severely affect patient quality of life (NIH 1990). It may result in severe, constant pain (NIH 1990), pathologic fracture of the mandible in 23% of cases, progressive necrosis in 19%, with resolution only in 15% of cases (Epstein et al 1987b). As ORN can be refractory to treatment (Stevenson-Moore and Epstein 1993), prevention is highly desirable. 1 Dental pre-radiotherapy interventions have been shown to favorably influence rates of complications, including osteoradionecrosis (NIH 1990). Teeth can never be extracted from irradiated bone without the risk of osteoradionecrosis, but diseased teeth carry a greater ORN risk. The timing of necessary tooth extractions also is important. For instance, Prophylactic extraction of teeth before radiation therapy has been suggested to prevent the development of dental complications that may require extraction and stimulate the development of [osteoradionecrosis . ...twice the risk of osteonecrosis is seen in those in whom teeth were extracted after irradiation (2.2 vs. 1.1%). A comparison of the number of cases of necrosis in relation to the number of patients who had teeth removed shows a higher percentage (7.1 vs. 5.4%) of necrosis in patients with teeth extracted after radiotherapy. These findings demonstrate a reduced risk of developing necrosis when selected teeth are extracted before radiotherapy..." (Epstein et al 1987a, 52). Societal expectations for retention of teeth (Hand et al 1988) and dental contributions to quality of life noted in the Literature Review section of this thesis underscore the importance and desirability of retention and maintenance of functional teeth in irradiated sites. Thus, there is a day-to-day clinical problem and dilemma whether functional teeth that contribute to quality of life should be sacrificed in order to minimize the small risk of ORN and its much greater diminution in quality of life in the few who eventually develop ORN. The question, "Under what conditions should teeth be (Epstein, et al 1987b; Murray et al 1980a; Murray et al 1980c; Gehrig 1969) 2 extracted within the high dose radiation volume in order to minimize the frequency and severity of osteoradionecrosis^ and its impacts on quality of life?" is a primary focus of the formal theoretical decision analysis for which this study will gather data. Theoretical decision analysis critically examines the present basis for clinical decision making and furthers the eventual production of a validated clinical decision analysis. The latter will provide a quantitative basis for making the necessary clinical decisions with minimal bias. Decision analyses of the clinical problem do not now exist. Discussion of osteoradionecrosis risks and associations with various therapeutic and patient4 factors, the epidemiology of head and neck cancer, and other aspects of the clinical problem will be found in the Literature Review. specifically related to native bone. Persons interested in osteoradionecrosis in neovascularized bone or the use of osseo-integrated implants in mandibular reconstruction could consider Mirante et al (1993) as an introduction to these topics. However, head-and-neck cancer treatment-planning "decision analyses" are now available. Weiss, Harrison and Isaacs (1994, "Use of Decision Analysis in Planning a Management Strategy for the Stage N 0 Neck") is a conventional decision analysis, apparently of high quality, suggesting for the average patient that when the probability of occult metastasis is greater than 20%, treatment rather than observation becomes preferable, and that of the treatment options considered, elective irradiation appeared to be slightly preferable (as assessed by "folding back" branches) to optional elective neck dissection. In contrast, Coulthard (1993), using a less conventional format, describes in qualitative terms some diagnostic and therapeutic maneuvers appropriate to Stages 1,2,3 and 4 carcinomas of the oral cavity. Schein (1989, 48-61) discusses considerations related to decision-making for the management of cancers of the lip, oral pharynx, salivary glands, and paranasal sinus, (e.g., compliance with recommended home care procedures) 3 2. DECISION ANALYSIS A . OVERVIEW Diagnosis and treatment planning are important aspects of the practice of any physician, dentist, or surgeon. The validity with which practitioners are able to identify health problems and treatment options, and the consequences in each case, will help to determine the potential quality of the health care services provided. Support for these diagnostic or treatment planning decision making processes will be provided by the application of formal clinical decision analysis5 and the data associated with individual clinical decision analyses. Decision analyses, and the clinical algorithms or practice policies incorporating them, could speed the physician decision-making process and assist in quality or cost-benefit ratio assessments of the services provided. They also could assist patient education and informed decision making by patients (Merz 1993). Figure 1, a reproduction of Figure 1-1, p. 5, of (Weinstein and Fineberg 1980), illustrates "The Decision Analytic Approach". The scientific basis of decision analysis derives from operations research and game theory (Last 1988). Its formal application in medical science (specifically, medical diagnosis) can be traced to (Ledley and Lusted 1959). A l l available choices and their potential outcomes in a clinical dilemma are identified (Hagen 1992). Best available sources of data are located for the probabilities and the measure of the desirability6 ("utility") that will be assigned to each potential outcome. 5 Simpson et al, quoted by McCreery and Truelove (1991, Part 1), defined clinical decision analysis as a "structured approach to guide a person to workable solutions of a problem, to make plans, and to evaluate data." 6 for instance, the preference of the "average" patient for the outcome 4 FIGURE 1 "THE DECISION A N A L Y T I C APPROACH" after (Weinstein & Fineberg 1980, Figure 1-1, 5) I. Identify and Bound the Decision Problem • Alternative Actions • Additional Information Possible • Treatment Options • Possible Clinical Information Obtained • Possible Clinical States of the Patient at Different Points in Time • Other Considerations II. Structure the Decision Problem • Decision Tree Representing Logical and Temporal Sequence of Clinical Problem • Clinical Starting Point • Choices • Probabilistic Events • Outcomes III. Characterize the Information Needed • Uncertainties • Valued Outcomes IV. Choose a Preferred Course of Action • Synthesis of Structures and Available Information • Quantification as a Means to this End • Sensitivity Analysis of Conclusions 5 Decision trees are, at the present time, the most popular type of decision analytic framework employed in formal decision analyses (Higgins and Martin 1988). Others proposed have included flow charts, the discrimination net proposed by Feigenbaum (Feigenbaum 1959), and the Markov model (Sonnenberg and Beck 1993). The "decision matrix" technique is mentioned in (Douglass and McNeil 1983), who also provide a good review of basic decision analysis concepts as they apply to dentistry. Clinical decision analyses are usually based on assessments of clinical risk and non-monetary desirability of clinical outcomes. However, decision analyses also can incorporate cost-benefit analysis, as described methodologically in Sox et al (1988).8 Schoenbaum, McNeil, and Kavet (1976) "The Swine-Influenza Decision" illustrates an application of decision-analytic cost-benefit evaluation to public health. Both quality of life assessment and economic evaluation is utilized in reference to cancer clinical trials in Morris and Goddard (1993), while Hilden and Habbema (1990) describe "The [statistical] Marriage of Clinical Trials and Clinical Decision Analysis". Fulton (1985) incorporated policy analysis (alluding to probabilities and utilities as beliefs and values.) Despite any potential limitations in the data for an individual decision analysis (for instance, one ultimately deriving from this thesis, additional research, and the literature), the exercise in clinical reasoning required to frame and explicitly structure the analysis can itself be a valuable exercise exploring the assumptions otherwise informally made by clinicians during the decision making associated with a clinical problem. Thus it highlights the nature of the decision process and areas in need of further research, while the effects of variables that are as yet poorly quantified can be modeled using sensitivity analysis. 7 and clinical epidemiological principles as these apply to dental diagnosis 8 Analogously, these also have application in the commercial world (Ungson and Braunstin 1982). 6 B. DECISION TREES Barnoon and Wolfe (1972) state that a conventional decision tree representation of the clinical decision process requires for each disease classification: "a) a list of tests and therapeutic measures which comprise the action space b) a list of outcomes c) a probability association associating the outcomes in each stage with the actions leading to these outcomes d) the utility associated with each of the terminal outcomes" Examples are illustrated in Figure 2 ("A Generic Medical-Surgical Decision Tree") and Figure 3 ("A Simple ORN-Prophylactic Tooth Extraction Decision Tree"). Formal decision analysis is based on an understanding of the concept of expected value defined as a weighted average of all possible outcomes of a decision (Weinstein 1980b). Given weights of probabilities associated with each outcome and the expected values associated with all the outcomes, the decision maker should choose the option that on average yields the best expected value. Probabilities and utilities are generally expressed as decimal values between 0 and 1. A health-based quality of life utility of" 1.00" would be associated with ideal health; of 0.00, with death; and negative decimal fractions, with fates worse than death, e.g. torture or excruciating, unrelieved, disabling pain. 7 FIGURE 2 A GENERIC MEDICAL-SURGICAL DECISION TREE modified after Figure 6-7 (Sox et al 1988,156) "The complete decision tree for comparing medical and surgical treatment of a hypothetical disease", K E Y : after (Sox et al 1988, 152) Decis ion Node : Where several choices are possible Probability Node: I1 Where chance determines which . outcome will occur; probabilities within a branch totaling 1.0. Perform Surgery i Start | Treat Medically Disease Present p = 0.1 p = 0.9 Disease Absent Disease Absent p = 0.9 p = 0.1 Disease Present Survive Repeat Surgery for Cure I p = 0.9 p = 0.1 Operative Death Operative Death Palliate p = 0.02 p = 0.98 Survive Operative Death p = 0.01 p = 0.99 Survive Cure p = 0.1 p = 0.9 No Cure No Cure p = 0.1 p = 0.9 Cure Cure p = 0.1 p = 0.9 No Cure Utility Utility 2 Utility Utility 5 Utility 6 Utility 7 Utility 8 Utility 9 Utility 10 Utility 11 8 FIGURE 3 A SIMPLE ORN-PROPHYLACTIC TOOTH EXTRACTION DECISION TREE Average probability of ORN r "Yes": ! Extract Tooth occurrence, given "Yes," = a Average probability of no ORN, given "Yes," = 1 - a | r Should Tooth ' Be Extracted? ! r "No": ! Do Not Extract Tooth Utility b Average quality of life, given "Yes" & ORN, = b Utility b' Average quality of life, given "Yes" &no ORN, = b' Average quality of life, Average probability of ORN Utility given "No" occurrence, given "No," = c d & ORN, = d Average quality of life, Average probability of no ORN, Utility given "No" given "Yes," = 1 - c d' & noORN, = d' KEY: after (Sox et al 1988, 152) Decis ion Node : Where several choices are possible Probability Node: Where chance determines which outcome will occur "a" and "c" are ORN probability variables, and "b" and "d" are utility (i.e., outcome) variables, and the average relative desirability for the choice "Yes" branch of the decision tree is (a multiplied by b) plus [(1-a) multiplied by b' ] and for the choice "No" branch of the decision tree is (c multiplied by d) plus [(1-c) multiplied by d'.] 9 In Figure 3, the only example of a "decision node" is "Extract?" However, with multiple decision nodes, or with multi-attribute frequency or utility variables, the relative desirability of the two "branches" of the "decision tree" may reverse for certain values of the variables a, b, c, and d. (Sometimes a ,b, c, and d themselves may be multi-attribute variables with their values determined in turn by the values of contributory variables.) Sometimes a single decision can have multiple outcomes that are divorced from any future influence of the care-giver. For instance, the decision "Yes" might generate some chance of death, some chance of survival with unavoidable functional deficit, and some chance of survival with potential for complete prosthodontic rehabilitation. The choice "Yes", having multiple potential outcomes, then would be termed a "probability node." Each probability node displays all clinically relevant potential events at that particular point in the decision tree. A l l probabilities at each chance node must add up to one, and thus the consequences or branches at each chance node must be mutually exclusive. The relative desirability of the branches "Yes" and "No" could reverse i f "Yes" and "No" generate different outcome probability distributions. "Sensitivity" and "threshold" analyses (to be discussed later) would permit the identification of the circumstances under which the relative desirability of the two choices for the decision whether to extract teeth in the high dose radiation volume reverses. Traditional decision trees assume that clinical choices are associated with statistically independent, non-recursive effects that can form mutually exclusive "branches". If clinical states are found to interact to produce different outcomes or probabilities, such interacting states must be considered as assumptions for different decision trees. 10 C. UTILITIES Utilities are the preferences or states of desirability associated with each particular outcome. Although various approaches to utility assessment can be considered, decision analyses often pursue patient-derived utilities as the ultimate goal compatible with patient-centered health. (Health-care-provider-chosen measures such as survival times, etc. might not reflect the total concerns of the patient.) For many types of studies, patient-derived utilities ideally can be generated in one of several standard forms: quality of life ("QOL "), health-related quality of life ("HRQOL") , Declining Exponential Approximation of Life Expectancy ("DEALE"), quality adjusted life expectancy/ Quality-Adjusted Life Years ("QALY"), etc. (Sox et al 1988; Hagen 1992). Quality-adjusted life years9 are most often used (Torrance and Feeny 1989). An alternative to quality adjusted life years has been proposed, the healthy-years equivalent ("HYEs") 1 0 . Surrogate scales, e.g. measures associated in some way with quality of life, may be substituted as an interim measure when validated measures for quality of life are lacking. Patient-derived dental outcome measures of various types are discussed in Slade (1996) and associated articles". Patient-centered utility determinations using the scientifically-favored (Mehrez and Gafni 1989) standard reference gamble technique seem to have been reported in only two dental publications, Fyffe (1992) and an abstract on orthodontic tooth extraction decision analysis, Lee and Miller (1993). Bass et al 9 also proposed as an epidemiologic measure for dentistry in the article Haugejorden and Klock (1992) 1 0 The advantages and relative complexity and limitations of "HYE's" are discussed by Fryback (1993), its references, and Mehrez and Gafni (1989). 1 1 in the Symposium on Self-Reported Assessments of Oral Health Outcomes (International Association for Dental Research Meeting, March 17,1996, San Francisco) reported in J. Dent. Educ. 60 (6, June, 1996). 11 (1994) found that patient-derived standard gamble values were significantly greater than rating scale values. Methods other than the Standard Reference Gamble for estimating patient-derived utilities exist; these include the time-tradeoff technique (Sox 1988), the willingness-to-pay technique, categorical scaling, and the use of visual analogue or other rating scales. Maas and Stalpers (1992) describe another approach to utility determination, "additive conjoint measurement", in which "the treatment with the higher utility is determined from pair comparisons among outcomes that vary in quality and quantity of life". They advocate this for utility determinations associated with the choices between radiation therapy and laryngectomy in laryngeal cancer patients12. Patient-derived utilities reflect directly the values and feelings of the person/patient for whom either the care-giver is responsible, or who under an emerging paradigm should be an autonomous decision maker ethically responsible within limits for his/her own fate. A n alternative utilizes physician-derived rating-scale values as surrogates for patient quality of life. Physician-derived and patient-derived utilities can be highly correlated. For instance, Boyd et al (1990) found that physicians, and patients involved with colostomies, tended to predict similar and associated utilities for colostomies in comparison to patients who did not themselves have a colostomy. 1 2 claiming that their method of comparing outcome pairs allows better testing of underlying axioms, and avoids difficulties inherent with, other, "risk-based", assessment procedures. 12 D. SENSITIVITY A N A L Y S I S The values of a particular variable in a decision analysis may not be a constant, or it may be a "constant" that can be determined at the time of analysis only in approximate terms. In such cases 1. the value of this variable of interest should be varied over its entire clinically relevant 13 range , while the other variables are held at their assumed values; 2. the impact of these variations should be assessed by "folding back" the decision tree branch involved and valuing the branch over the range of variable values 3. a determination should be made as to whether the relative desirability/undesirability of the branch involved, in comparison with other branches associated with the decision node, changes as the variable value is altered. (If this occurs, then the decision recommendations change) (Hughes and Hughes 1990; Sox et al 1988). The identification of critical values of variables which result in different decision recommendations is termed "threshold analysis". The purposes of sensitivity analysis are: 1. to determine the variable values over which the decision considered is valid 2. to assist in validating the decision analytic model (particularly important when " 'soft data' are used, for instance in subjective estimations of probability and utility"); this also encourages acceptance and utilization by clinicians (Critchfield 1986).14 The ranges are usually determined by reference to the literature and expert opinion. 1 4 as quoted in Hughes and Hughes (1990) 13 "In one-way sensitivity analysis, the value for one probability or outcome measure is varied, and the others remain constant. In two-way sensitivity analysis, pairs of variables are changed simultaneously. In three-way sensitivity analysis, two-way sensitivity analysis is repeated for several values of a third variable" (Sox et al 1988, 303). "N-way" sensitivity analysis is described as a technique suitable for evaluating decision trees when none of the parameters are known with any degree of certainty (Hughes and Hughes 1990). However Hughes and Hughes (1990) cite Shortliffe (1979) as arguing that such decision analysis is invalid because it depends heavily on subjects' possibly inaccurate probability and utility estimates. Techniques of sensitivity analysis for subjective probabilities described in Weinstein, Fineberg and Elstein (1980, 180-3) explore the stability of decision analysis over a range of estimates. 14 E. D A T A FOR PROBABILITIES A N D UTILITIES The five stages of Cooper's 1982 "Integrative Research Review" described in Smith, Smith and Stullenbarger (1991) are relevant to efforts in gathering data for risk estimate probabilities in decision tree branches: 1. problem formulation 2. data collection 3. data evaluation 4. analysis and interpretation 5. public presentation. Additional principles of systematic literature review outlined in other references15 have theoretical relevance but are too lengthy to outline. One also could consider techniques of meta-analysis, i.e. statistical synthesis of related articles.16 Evaluation of literature must include an evaluation of its quality. Obviously, data from controlled clinical trials and prospective cohort studies are among the most desirable sources for probabilities that will be assigned to the branches in a decision tree. Types of flaws potentially associated with case series, cohort studies, case control studies, etc. include confounding, non-generalizability, co-intervention biases, diagnostic suspicion biases, and competing risks, etc. (Jaeschke and Sackett 1989). "Non-experimental" data systems sometimes play a role in clinical epidemiological studies, which in turn can contribute to clinical decision analyses. Roos 1 5 (Wood-Dauphinee and McPete 1986; Schechter and Leblanc 1986; Jaeschke and Sackett 1989; Burdette and Gehan 1970) - the latter being particularly relevant to epidemiological studies. 16 (Smith, Smith and Stullenbarger 1991; Eddy, Hasselblad, and Shacter 1990; Droitcour, Silberman, and Chelimsky 1993). 15 (1989) describes the use of non-experimental data (e.g. hospital data bases, cancer registries) in planning of randomized trials and in clinical decision analysis. Retrospective analyses can be performed with the goal of estimating probabilities of different outcomes and determining interventions worthy of further research to examine their impact on outcomes and etc. Parenthetically, Roos (1989) advocates the maintenance and improvement of "well organized high quality data bases" as a priority for "funding agencies interested in technology assessment, cost control, and quality of care". Lacking sufficient high-quality data, the decision analyst, when attempting to provide an analytic framework, to isolate significant criteria, or to quantify inputs and outputs, can face problems similar to those of individual practitioners partly dependent upon clinical impressions and anecdotal reports. This is the case for the clinical problem of the dental management of osteoradionecrosis, for which validated quantitative epidemiological data for the effects of assumed prognostic variables is scant. Validated 1 n quality-of-life measures for late oral complications of cancer are essentially nonexistent. When limited research data are available, the decision analyst may utilize formal consensus methodologies (e.g. nominal group technique, Delphi technique) intended to improve the reliability (if not necessarily, the validity) of relevant expert clinical opinion (Fink, Kosecoff, Chassin, Brook 1984). These can then be integrated with data of higher quality to provisionally identify factors and quantify values suitable for a decision analysis. This information can be abstracted into a framework for analysis (e.g. a decision tree), and thus contribute to the decision and sensitivity analyses. 1 7 For details and references, please see Literature Review. 16 F. DECISION A N A L Y S I S A N D THE PATIENT Although both patients and clinicians can benefit from recognition of the treatment alternatives that can be generated by the application of a suitable decision .analysis, a question sometimes presents itself: "Is the decision analysis truly applicable to the particular patient at hand?" Decision analyses can create clinical policies, decision rules, protocols, and algorithms that can be applied to classes of patients (Littenberg and Sox 1988); when implemented, these should improve the consistency and overall quality of patient management (Eddy 1990a). However, specialized assumptions may be necessary for particular patients with unusual conditions (e.g. a particular metabolism, family history, or psychological state). Patient-derived utilities may be individualized to particular patients using validated measures, "chair-side" standard gamble interviews, or other means. The value that clinical decision making places on individual patient preferences fits quite well with recent societal, legal, medical, and dental trends. The current literature recognizes "...the evolution to patient-centered care in many areas of medicine: patient care, health-related law, medical education, research, and quality assessment" (Laine and Davidof 1996), the "shift from clinician-patient paternalism to self-determination..." (Smith 1990), and that the patient now may be regarded as an "informed consumer" (Niessen 1994, 1327) requiring informed consent to an extent perhaps greater than traditionally assumed by the health professions {Canterbury vs. Spence, 1972, cited in Murphy (1976) and Kauffman (1983)). 17 Kay, Brickley, and Knilljones (1995) conclude that "an over-concentration on the biomedical model of dental health may cause dentists to make decisions which are inappropriate to their patients' values and preferences". Others advocate a move from traditional dental practice based on clinical experience to one based upon an evidence-based approach requiring the use of scientifically unbiased data and "specific rules of evidence to quantify their recommendations" (Levine and Shanaman 1995; McGuire and Newman 1995). Decision analysis can integrate successfully with these trends and in fact has been advocated by Merz (1993) as a new paradigm for treatment planning and legal informed consent. Clancy, Cebul and Williams (1988) concluded from their randomized, controlled trial that individualized decision analysis can influence the clinical decisions taken by "knowledgeable and interested patients". However, not all patients are willing to participate in clinical decision making. For instance, one study indicated that only 64% of the public thought that they would want to select their own treatment if they developed cancer, while 59% of cancer patients thought similarly (Degner, and Sloan, 1992). Interventions exist, however, to foster patient-involved decision making (Neufeld, Degner, and Dick 1993; Carter 1992). 18 G. DECISION ANALYSIS AS A N AID TO B A Y E S I A N REASONING Several theories of decision making exist. They differ in that some are prescriptive (i.e., how decisions ought to be made) while others are descriptive (i.e., how decisions actually are made). Perhaps the Bayesian approach could fall into the former category. Were a physician to think in Bayesian terms, his or her estimate of the "posterior probability" of a disease being present would be based on knowledge of two factors, the prior probability of the disease, and the true- and false-positive rates of the test(s) involved. New data would be incorporated into reconsiderations of the problem in a manner that would overcome forms of bias perhaps otherwise present (e.g. representativeness, availability, regression to the mean, and anchoring, discussed in the Literature Review). In essence, decision trees could be considered Bayesian. However, Bayesian logic may not model the way that humans behave during the clinical decision making process. "In everyday clinical reasoning, levels of risk tend to be treated categorically rather than on a continuous probability scale... Intuitive decision making generally processes risk factors sequentially rather than simultaneously. A case is scanned first for evidence concerning one risk and if that does not yield a decision the second [risk] is evaluated. Decisions may vary depending on which risk is considered first" (Elstein, et al 1992). Slovic, Rorer and Hoffman (1971) describe three potential categories of reasons why humans may not display Bayesian reasoning: misperception, misaggregation (inability to utilize information presented in a Bayesian format), and artifact (e.g. distortions in the processing of extreme probabilities).18 1 9 In contrast, Bergus has proposed that "the Bayesian approach is intuitive to physicians and can mirror their internal diagnostic processes" (Bergus, in press, cited in Murray and Bergus 1995). Further perspectives are offered by (Edwards 1968) and psychological decision theory, heuristics and behavioral decision theories described in (Thomas, Wearing and Bennett 1991, 17-28.) This section cannot detail all concepts and methodology of decision analysis or clinical epidemiology. Readers interested in these may consider (Sackett, Haynes, Guyatt, and Tugwell 1991; Thomas, 19 Whether or not Bayes' theorem is considered a description of how humans analyze data or how they should analyze data, conventional decision trees do employ Bayesian logic and should enable routine use of Bayesian thinking. H. DECISION ANALYSIS A N D DENTISTRY Decision analysis has been applied to dentistry in matters of diagnosis, clinical judgment, patient management and treatment, and formation of health policies. Tables 1 through 5 of McCreery and Truelove (1990, Part 2) list dental decision-analysis-related literature on diagnosis, treatment planning, disease prediction, computer applications and policy. Table 2, "Publications on Treatment Planning" lists 25 papers, of which one article in each of restorative dentistry and radiography, endodontics, oral surgery, and oral medicine feature decision trees. (Another in oral medicine involves a Markovian model.) A good historical overview of decision making in dentistry is provided in the article, "Decision Making in Dentistry. Part I: A historical and methodological overview" (McCreery and Truelove 1991). The Journal of Dental Education did not advocate the incorporation of clinical decision analysis into dental school curricula until December of Wearing and Bennett 1991; Sox et al 1988; and a slightly dated, eminently readable and authoritative book written at a higher level, Weinstein and Fineberg 1980). 20 1992 (Bolender 1992). Although dental decision-related analyses appear to be of relatively high quality, they are relatively few compared to medicine and most fail to approach closely the standard "formal" decision tree protocol advocated in the textbook Sox (1988). For instance, Stockstil, Bowley, and Atinasio (1992) consider fixed prosthodontics decision trees only in non-probabilistic terms, even though Knoebel (1986 advocated the incorporation of probabilistic decision making analyses into clinical practice. One dental article, (Mileman and Kievit 1992), is structured according to the "traditional" decision tree analysis principles outlined in Sox et al (1988). It applies decision tree and cost-benefit analysis to oral radiology, conducting three-way cost effectiveness sensitivity analysis for variation in the sensitivity and specificity of radiographs. (Radiographs are considered as a diagnostic test, given a varying a priori chance of a peri-apical lesion20.) The article varies from traditional decision analysis, however, in that it does not assign patient-derived utilities to the final outcomes. The article Brickley, Kay and Shepherd (1995, "Decision Analysis for Lower-third-molar Surgery") is a potentially useful dental example of semi-formal decision tree analysis. It surveys the literature data to ascertain probabilities for various outcomes following extractions of lower third molars, suggests preliminary utility values, 21 determines clinical relevance, and performs sensitivity analyses. 2 0 A periapical lesion is a lesion around a tooth root tip, e.g. inflammation or infection. It frequently results from pathology of the tooth pulp and root canal soft tissue. 2 1 However, in this paper bleeding outcomes are not well defined. 21 3. ORN-PROPHYLACTIC DECISION ANALYSIS A N D THIS THESIS It was noted above that when limited research data are available, the decision analyst may utilize formal consensus methodologies to improve the reliability of relevant expert clinical opinion, subsequently integrating these opinions with other relevant data of higher quality to provisionally identify factors and quantify values suitable for a decision analysis. The goal of the present thesis research is to produce opinion data that can augment data from the literature for a theoretical decision analysis focused on the research question, "Under what conditions should teeth be extracted within the high dose radiation volume in order to minimize the frequency and severity of osteoradionecrosis and its impacts on quality of life?" Toward this goal a highly modified Delphi Survey of 2 2 suitable content was constructed. The thesis— background, methodology, findings, etc.— assumes the following outline: Chapter 2: Literature Review 1. Delphi Consensus Technique. Opinion Data. 2. Utilities and Outcomes 3. Clinical Problem (Oral Cancer Epidemiology, Treatment Strategies; Radiation Effects; ORN Pathogenesis, Definition, etc.; ORN; ORN Epidemiology and Predisposing Factors) The modifications of traditional Delphi technique employed, and rationale for same, will be discussed in subsequent sections. As described in the beginning of Results, the Delphi survey identifies radiological, dental, and medical potential ORN risk factors, then characterizes their importance, relevance, practicality, and controllability as contributory factors to an ORN-prophylactic dental clinical decision analysis. Factors relevant to extraction of teeth in the high dose radiation volume are considered. Outcome-related-factors (ORN anatomic location associations with clinical presentation, ORN severity associations, factors skewing ORN severity distributions, and ORN pain and function and potential Quality Of Life associations) are then presented. 22 Chapter 3: Methodology Chapter 4: Results 1. Delphi Process Rationale, 2. Sample, 3. Response Rate; 4. Creation of Survey Outcome Scales; 5. Analysis. 1. Identification and Direct Assessment of ORN Risk Factors (a. General and Dental; b. Irradiation To High Dose, etc. c. Endodontics; d. Timing of Dental Extractions e. Relation of Oral-surgical, Medical, and Prosthetic Technique to ORN Risk) 2. Assessment of Characteristics of ORN "Risk Factors" a. Perceived Relevance to ORN-Prophylactic Treatment Planning Factor Practicality for Decision Analysis , e.g. b. Quantifiability, c. Clinical Controllability 3. Factors Relevant to Tooth Extraction 4. Quantitative Estimates of ORN Risk. 5. Outcomes, and Factors Related to Outcomes a. Severity of ORN Lesions b. ORN-Associated Factors That May Affect a Patient's Quality of Life c. Anatomical Location and ORN Clinical Presentation d. Factors Skewing ORN Severity Distributions e. Osteoradionecrosis Severity Distributions f. ORN Severity Category Associations with Pain & Function 6. Survey: Potential Sources of Bias a. Definition of ORN 248 b. Respondent and Practice Characteristics....250 7. Responses of Low Data Quality 23 Chapter 6: Discussion Chapter 7: Conclusions Bibliography Appendixes: 1. Potential Sources of Bias a. Overlap of Delphi Respondents With Cited Authors b. Definition Of Osteoradionecrosis c. Respondent Personal and Practice Characteristics 2. Interpretation of Survey Findings 3. Integration: Factor Suitability for Decision Analysis 4. Reliability 5. Validity Considerations and Limitations 6. Bias, Reliability, and Validity Conclusions and Applications 1. Large Tables 2. Letters and Forms sent to Respondents. 3. Delphi Survey 4. Details: Responses of Low Data Quality 5. Class of Data Generating Undefined Quartiles 24 CHAPTER II. LITERATURE REVIEW 0. "EXECUTIVE SUMMARY" I. DELPHI CONSENSUS TECHNIQUE; OPINION DATA Delphi technique opinion data may be useful but must be accepted with reservations. Significant reliability may be found or developed, but unrecognized sources of bias may persist, and validity will be unproved. II. UTILITIES AND POTENTIAL OUTCOME VARIABLES Utilities suited to the proposed decision analysis do not exist at the present time. The basis for the creation of outcome scales for the survey (as described in the Methodology section) are factors related to quality of life, as described in this section; information about the clinical problem, provided in section 3. of the Literature Review; and articles describing the clinical experience of head and neck cancer.1 III. CLINICAL PROBLEM. 1. ORAL CANCER EPIDEMIOLOGY AND PREDISPOSING FACTORS Squamous cell oral cancer tends to be a disease of the elderly and of multifactorial origin and risk factors, slowly declining in incidence from about 2 to 8 (locally =3.5) 7100,000 persons/year, with about a 50% 5-year post-treatment survival rate. e.g., Dudgeon, DeLisa, and Miller (1980) and Dhillon, Palmer, Pittam, and Shaw (1982) 25 2. ORAL SQUAMOUS CELL CARCINOMA LESIONS AND TREATMENT STRATEGIES 90% of all carcinomas of the oral cavity are squamous. Common locations for oral squamous carcinoma are the oral portion of the tongue, the mandibular gingiva, and the floor of the mouth. Treatment often involves irradiation with high doses, and surgery. Irradiation sources may be external or intra-oral. Irradiation with intra-oral sources is termed brachytherapy, and one form of this, "interstitial" therapy, involves placement of the sources within the tissues. A l l forms of cancer therapy have side-effects, and the approach best suited to balance favorable prognosis with a patient's personal definition of quality of life may be difficult to determine. 3. P A T H O L O G Y OF RADIATION INJURY The oral complications of radiotherapy may include injuries to the salivary glands, oral mucosa, oral musculature, alveolar bone, periodontium, and to a lesser extent, teeth. Side-effects may include xerostomia, rampant dental caries, mucositis, taste loss, osteoradionecrosis, infection, dermatitis and trismus (NIH 1990). 4. OSTEORADIONECROSIS PATHOGENESIS, DEFINITION, DIAGNOSIS, CLINICAL PRESENTATION, A N D TREATMENT Osteoradionecrosis follows partial vascular/ microvascular obliteration (a condition to which the mandible is particularly sensitive), relative hypoxia, and cellular death. Although radiography may assist its diagnosis, most authors consider the diagnosis of ORN to be based primarily on clinical signs and symptoms of ulceration or necrosis of the mucous membrane, with exposure of necrotic bone for more than 3 months (Epstein, et al 1987a, 48). Since ORN lesions can have severe consequences on quality of life and can be extremely refractory to treatment, the preferred health 26 strategy is prevention rather than cure. The use of hyperbaric oxygenation as prophylaxis is good in theory but some question its universal effectiveness. Conservative therapy of existing ORN lesions usually has far fewer negative consequences than radical, but radical surgery sometimes is necessary. 5. ORN-PROPHYLACTIC D E N T A L SCREENING STRATEGIES (STRESS: PERIODONTAL, ENDODONTIC, A N D INFECTION STATUS) Risk for osteoradionecrosis can be reduced by effective pre-irradiation-treatment oral assessment, and interventions affecting ORN prognostic variables (NIH 1990). 6. OSTEORADIONECROSIS EPIDEMIOLOGY The literature describing osteoradionecrosis (e.g. rates, potential prognostic factors, location, timing, severity, and consequences) is reviewed. ORN rates cited in the literature have ranged from 0 to 100%, reflecting not only clinical experience but artifacts of definition and diagnostic and other types of bias. The comparability (and thus, the external validity) of some ORN literature can be questioned because clinical protocols and patient case distributions differ. The "causes" of osteonecrosis are dis-cussed in the literature, but in some cases these may simply be associations. Causes and variables identified in a few articles have been incorporated into ORN risk rate prediction formulas. If a separate diagnostic code for ORN of the jaws could be created in the International Classification of Diseases", the basis of many registries, this will widen the future ORN data base available for epidemiological analysis. IV. P O T E N T I A L F A C T O R S P R E D I S P O S I N G O S T E O R A D I O N E C R O S I S : S U M M A R Y This section gathers most hypothesized ORN risk factors and extraction indications for final review, and complements previous discussions with additional perspectives. 2 7 1. DELPHI CONSENSUS TECHNIQUE; OPINION DATA In the absence of prospective, randomized, controlled scientific studies, analyses and patient care guidelines can be based on consensus methodologies. Four formal consensus methodologies have been described: the nominal group technique (Horton 1980) the Delphi technique, the Glaser "State of the art technique" (Glaser 1980), and a model developed by National Institutes of Health. The "Delphi Technique" and the "Nominal Group Technique" are the two with the longest history (Fink et al 1984). Many consensus methodologies assume that opinions of experts grouped under special 2 circumstances can reveal a closer approximation of the objective truth than would be achieved through conventional pooling of expert opinion, although some question this notion (French 1986). General characteristics of the Delphi Technique as described in Whitman (1990) and Pil l (1971) include anonymity, multiple iterations with controlled feedback, and statistical summaries. Typically, an open-ended survey is sent to empaneled experts who in turn provide responses that allow the Delphi experimenter to frame items for subsequent close-ended surveys. Each subsequent survey will incorporate statistical summaries of item responses from the previous version so that each respondent can compare these with his/her own thoughts. (Each open- or close-ended version of the (providing guidance, e.g., in decision analyses), 28 survey is termed an "iteration" or a "round".) Responses to subsequent rounds should tend to converge. Rowe, Wright and Bolger (1991) explain the original concept of Delphi proposed by Dalkey as one in which holdout experts most confident in their opinions tend to move the group median toward the area containing the true answer. They describe Delphi as a two step procedure involving first, an interacting stage seeking to debias individual judgments, and then one or more stages with feedback3'4. Linstone and Turroff (1975) characterize Delphi as a method for structuring a group communication process such that it is effective in allowing the group to deal with complex problems. The advantages claimed for the Delphi Technique include: 1. Participation without geographic, and to a lesser extent, temporal, constraints 2. Anonymity during participation, thought to reduce bias from deference to the more prestigious members of the panel 3. Lack of need for a trained interview facilitator (Fink 1984; Millholland 1973). as in Rowe, Wright and Bolger (1991, 238 & 243, Figures 1 & 2). (Please Note: "Traditional" Delphi technique is explained particularly well by this article.) Although Rowe, Wright and Bolger (1991) state, citing Parente and Anderson-Parente (1987) in particular, that there is a general trend towards more valid judgments with multiple iterations, they also interpret several papers to indicate that panelists "may produce more valid judgments over rounds in the absence of additional information of the other group members' estimates". This is contrary to the just-explained traditional interpretation of why Delphi works. Rowe, Wright and Bolger (1991) conclude that the process of iteration itself may be playing a part in allowing panelists to reconsider their original thoughts and move their own opinions closer to a reconsidered, final position. They also cite Hill and Fowles (1975) as explaining that concurrence may be due to "bandwagon" and "fatigue" effects as well as reasoned consideration of arguments. 29 Present official and semi-official documents indicating appropriate oral management of patients at risk for osteoradionecrosis5 do not appear to have utilized formal consensus methodologies to reach their conclusions. The NIH Consensus Conference relied on "consensus panel" experts' reports, as modified by audience input, to create a consensus; it then published the revised papers with recommendations on pretherapy dental interventions, management of complications, and areas needing future research. Thus, the NIH (1990) methodology6 may involve a theoretical risk that some audience members may have been too intimidated to question the positions presented for their approval. The Delphi technique is compatible with a Bayesian information processing framework, and is useful when data available are vague or not reliable (Sahal 1975).7 Delphi studies in health usually predict or prescribe the future. British Columbia Ministry of Health (1991) is a typical Delphi— an initial questionnaire for issue identification, a second questionnaire for future priority setting, and a third questionnaire for final ranking of the issues. A dental example, Kaldenberg, Becker, and Hallen (1990, "Dentistry in the Year 2000: Assessments from a Delphi Panel") differs from classical Delphi Technique in the sense that full scale multiple iteration was not employed. 1) Consensus Development Conference on Oral Complications of Cancer Therapies: Diagnosis, Prevention, and Treatment". { 1989; published in 1990 and referenced in this thesis as " (NIH 1990)"}. 2) ADA (1989): "American Dental Association Oral Health Guidelines Head and Neck Cancer Patients Receiving Radiation Therapy". (This was developed by the ADA "Council on Community Health, Hospital, Institutional and Medical Affairs" "primarily as a resource tool rather than [as] an attempt to set specific standards of care"). 3) Barker, Barker, and Gier (1992), representing consensus among the authors, three consultants, and the five members of the "Executive Committee" of the "International Society for Oral Oncology". and ADA (1989) methodology not reviewed here Sahal (1975) also recommends that Delphi studies incorporate cross-impact analyses to ascertain interacting variables. 30 Milholland, Wheeler, and Heieck (1973, "Medical Assessment by a Delphi Group Opinion Technic") use the Delphi technique in a manner somewhat similar to this thesis research. A related example, in which a Delphi-related consensus methodology is used to produce a list of clinical indications for a surgical procedure, is Merrick, et al (1987). Merrick, et al (1987) "used a two-round consensus panel method to derive and rate the appropriateness of comprehensive sets of detailed clinical indications for performing carotid endarterectomy." It found that ratings derived in this process were reliably reproduced six to eight months after the completion of the process, were consistent with those in the literature, and that a statistical analysis demonstrated that they followed a logical rationale. Merrick, et al (1987) concluded that consensus methods that do not force agreement can be used with panels of physicians to produce detailed, reliable, and valid indications, as well as identifying areas of disagreement appropriate for further study. Fink, Kosecoff, Chassin, Brook (1984) state that Delphi reliability increases with the number of rounds, and an early school of thought advocated three Delphi close-ended item rounds in addition to an initial open-ended survey of the experts. However, Sweigert and Schabacker (1974, "The Delphi Technique: How Well Does It Work In Setting Educational Goals") concluded that "if ranking is the major concern, one round may be enough". Fink, Kosecoff, Chassin, Brook (1984) admit that respondent fatigue can set in after two or three rounds. Sahal (1975) cites his references Gettys, Kelly and Peterson (1973), Peterson (1973), and Martino (1972), as stating that rounds beyond two may not result in any advantage, a position echoed by Linstone and Turoff (1975), and Brockhoff (1975). 8 Milholland et al (1973) found that while uncertainties remained, there was a marked increase in the consistency of answers from the first to the second round. 31 Fink, Kosecoff, Chassin, Brook (1984) assert that Delphi reliability increases with the size of the group, but that too large a group becomes difficult to manage. Millholland et al (19739) state that the quality of a response improves with increasing group size up to about 13. Fink, Kosecoff, Chassin, Brook (1984, 982) recommend that desired consensus must be defined in advance of a study. Consensus in Delphi is very often measured through comparison of measures of central tendency such as a median (Rowe, Wright and Bolger 1991). Milholland, Wheeler, Stanley, and Heieck (1973) consider the median of responses to be the best single number for describing group consensus. Kendall (1977), like Milholland (1973) used changes in interquartile ranges as indicators for convergence from one round to the next. Differences between rounds also can be assessed using Spearman Rho rank correlation coefficients (Sweigert and Schabacker (1974). In contrast, Dajani (1979) advocates that a Chi-squared test should be used to determine the stability of responses between rounds. When population criteria are met, regression analysis may be used (Salancik 1973). Heifer (1971) and Salancik, Wenger, Heifer (1971) consider that information return is maximized when Delphi statements are neither overly specified nor vague, i.e. the correct degree of complexity— about 20-25 words for events of average familiarity. In contrast, Laurent (1972, in Sahal 1975) suggests that long statements may actually enhance reporting performance because the respondents take cues from the information complexity iii the item, so additional, and more accurate, information may be elicited. and its reference 1 32 Curley, Young and Yates (1989, 116) regard ambiguity of opinion, whether due to "lack of evidence, the presence of conflicting evidence, unreliable evidence, or some other source of inherent uncertainty in the clinical case" as important because "it reflects the inadequacy of a point probability judgment." They found that a confidence rating for a respondent's evaluation of a survey item appeared to best reflect the construct of ambiguity as they defined it, and suggested that interview subjects should be asked "not only for their evaluation of an answer to a question but their certainty regarding it." Thus, confidence ratings should be included in survey responses, and in surveys that duplicate the findings of Curley, Young and Yates (1989), would be very important. Methods other than Likert scale and confidence scale ratings may be used to provide probabilistic judgments (Poses, Cebul and Centor 1988). Sahal (1975) quotes his reference Martino (1970) as stating that assessors tend to prefer to evaluate probabilities in terms of ratios and that a diversity of probability responses expressed as ratios may tend to follow a log- normal distribution. Data transformations for medians of year estimates and probabilities also may be considered (Kendall 1977). Variants of Delphi are described in Kendall (1977), Rauch (1979) and Webler et al (1991). In the group Delphi variant Webler et al (1991), a two-part questionnaire using open questions and Likert scales was iterated among rotating sub-groups to build consensus and to define disagreement. Kendall (1977) suggested other modifications of the Delphi Technique with the objective of making forecasts more precise. 33 Bias may be present in improperly conducted Delphi technique studies. Decisions derived through Delphi can be influenced by the circumstances of the study, such as each participant's awareness of other individuals' knowledge and talents (Christensen 1993). Information shared before, as well as during a Delphi study, would tend to invalidate it. One early assumption was that group consensus would tend to be more conservative that individual opinion. In contrast, however, subsequent observations have tended to promote a perception that groups tend to be upwardly biased, or more optimistic than individuals — a phenomenon termed "risky shift". However, the data for this hypothesized group dynamic of "risky shift" may be controversial (Sahal 1975, esp. footnotes 166-67). Delphi technique, when abused through fraudulent manipulation of statistical feedback from one iteration to the next, will influence responses in the direction of the bias introduced (Nelson 1978): This finding underlines the need for careful survey design to minimize known sources of bias, and serves as a reminder that properly conducted consensus experiments may retain unrecognized sources of bias. The potential biases in surveys of any type are well discussed in Woodward, Chambers, and Smith (1991). Forms of bias mentioned in the Discussion also could apply. Cognitive biases in interpretation also exist. Shafer (1986) implies that an assumption that a respondent used Bayesian logic could lead to interpretation bias, and also that the dynamics of the formation of opinions by groups cannot be confused with the formation of opinions by individuals. Genest and Zideck (1986, "Rejoinder", 147-48) suggest that the challenge of the theory of combining [subjective opinion data] 34 distributions is that it is not always clear when a "given situation calls for compromise, summarization, or consensusalization." He suggests that combination of opinion should be guided by the decision context. Opinion data, e.g. in Delphi surveys, are subject to many forms of bias. Potential errors in probability assessment include three principles that are utilized to generate subjective probability estimates: representativeness, availability and anchoring (Taversky and Kahnemann10). Under conditions of uncertainty each of these may lead to biased judgments about probability assessments. The principle of "representativeness" is illustrated when an estimate of a certain datum is referred to, or derived from, a larger class of data in which the assessment is better known. "Availability" 1 1 is employed by accessing memories that are the most prominent— e.g. due to a series of different reasons including unusualness, meaningfulness, recent events vs. those long past, or consequences (i.e. consequences of a particularly serious condition). "Anchoring" is a tendency of survey or interview respondents to choose particular probabilities. In many cases, these would tend to be midrange probabilities. For others, or in other circumstances, people may falsely anchor rare probabilities as extremely rare (approaching zero), or probable events as closer to a probability of one, than reality would dictate. Waltert (1989) cites his reference Winterfeldt and Edwards (1986) as stating that extreme probabilities are best identified in a logarithmic fashion for survey respondents and also states that estimates of extreme probabilities are especially prone to effects of availability bias. 1 0 Reference 260 of Weinstein and Fineberg (1980, Chapter 6, 176-77) 1 1 References 31-35 of Woltert (1989) 35 Kahneman and Tversky (1996) believe that frequency judgments as well as probability judgments can be susceptible to large and systematic biases. Redelmeier (1991) commented that physicians may be swayed in evaluations of diagnostic judgments not only by availability bias, but by wishful thinking. Another cognitive bias that may influence physicians' medical decision making when facing complex situations has been described by Redelmeier and Sharir (1993). They found that increasing the number of options paradoxically can increase the "probability of maintaining the status quo, selecting a default option, or delaying the decision". Along with the anchoring and availability biases already mentioned, Shafer (1986) mentions framing, conjunctive bias, and disjunctive bias as cognitive illusions that lead respondents to make biased or inconsistent probability estimates. Framing bias can arise from inadequate ordering and phrasing of items designed to elicit responses, leading, for instance, experimental subjects to over-estimate probabilities of moderately likely events. "The effect of this bias would depend on whether one inquires about the probability of the event'' or that of its complement (). It is usually advisable to ask about both" (Woltert 1989). Hil l and Fowles (1975) suggest that the Delphi Panel should contribute the phrasing for many event statements if the experimenter is going to avoid potential phrasing bias in his survey and eventual findings. Conjunctive and disjunctive bias is explained by Woltert (1989) as follows: "Subjects tend to overestimate the joint probability of independent events (e.g. winning three times successively in the game of craps) and to underestimate the probability of the union of disjunctive events (e.g., losing on the first, second or third try)." 36 Experts apparently are less influenced by these cognitive biases when dealing within their specialty (Winkler 1986; Woltert 1989 reference 42). Physician determinants of opinion, especially two main categories of care giver characteristics classified as institutional and person dependent, must be taken into account during consensus studies (Liberati, 1986). Variations in clinician opinion not only 12 influence clinical treatment planning, but proven variations in treatment planning could influence a clinician's range of clinical experiences, which in turn might bias his/her opinions. Liberati cites Palmer and Reilly (1979) as stating that although predictors of therapeutic behavior are not fully supported in the literature, the ones most frequently considered are characteristics that may be classified as "institutional"-- e.g. the size of the hospital, volume of patients, and availability of ad-hoc programs or facilities— and personal— e.g. the amount of training after medical school, specialization, age and graduation time. Other factors as-yet-unquantified may apply (Liberati 1986). The survey-based restorative decision making analysis by El-Mowafy and Lewis found variations in restorative dentistry treatment planning that could be associated with age of practitioner, sex, university of training, amount of continuing education, and type of practice (El-Mowafy and Lewis 1994). Grembowsky, Melgrom, and Fisset (1989) stated that reasons for variations in treatment patterns include different sources of education in practitioners of various ages. Dolan et al (1992) concluded that a patient's age influences general dentists treatment planning decisions, "perhaps limiting the treatment options offered to older adults", and Johnson (1993) concluded that non-dental factors can influence dentists' decisions to extract teeth in older adults. Of the latter, the most influential and pertinent non-dental factors, relevant in 13-17% of cases, were patient/family requests, financial limitations, and inability to care for one's teeth. Dental factors cited for the extraction of teeth in older adults were, in decreasing frequency of citation: 1. non-restorability of the tooth (54%), 2. caries (46%) 3. prosthetic considerations (45%), and 4. periodontal disease (40%). The economic structure of the delivery of services also may influence choice of treatment (Liberati et al (1986), citing Hornbrook and Berki (1985) 37 2. UTILITIES AND POTENTIAL OUTCOME VARIABLES The concept of utility determination and application is explored in Lane (1987). Utilities are founded on modern utility theory, a "normative, rational model of decision making under uncertainty" (Torrance 1987). Utilities can be evaluated and compared for validity and acceptability to subjects, reliability precision and ease of use (Torrance 1987). The relatively better validated methods of determining utility assessment are discussed in the article Stiggelbout et al (1994). Techniques used for determining utilities include standard gamble, time trade off, and rating scales. Quality of life assessments are an ultimate goal for utilities and for validation of outcomes in a decision analysis. It is interesting to note parenthetically that quality of life has been correlated with prognosis for survival duration (Coates 1993). Health-care providers often prefer not to evaluate quality of life in its entirety, but to focus with greater sensitivity and practicality on health-related quality of life. Generic instruments for assessing general health status can, in turn, be supplemented with disease-specific supplements or batteries, which by themselves are specific and sensitive to the effects of a particular disease. "The preferred strategy depends on project aims, methodological concerns and practical constraints. Generic measures are necessary to compare outcomes across different populations and interventions, particularly for cost-effectiveness studies. Disease-specific measures assess the special states and concerns of diagnostic groups . Specific measures may be more sensitive for the detection and quantification of small changes that are important to clinicians or patients" (Patrick and Gale 1989). 38 Although the measurement of outcomes as qualities of life is not practical in this research, some relevant literature should be mentioned. Concepts of quality of life in oncology are reviewed in the references (Osoba 1992; Osoba 1991a; Osaba 1991b; Morris 1994; Lindley 1992; Aaronson 1988) and books listed on page 270S of "Controlled Clinical Trials" Volume 12. Liberati et al (1986) suggest a growing role in oncology for clinical decision analysis utilizing patient-derived outcome measures: "...In assessing the efficacy of anti-cancer measures, clinicians and public policy makers have traditionally given overriding consideration to the effect of therapeutic procedures on survival. However, and this is particularly true for solid tumors, almost all standard and experimental treatments have now reached a plateau of success, mortality 13 becoming an insensitive end point for the definition of risk-benefit profiles." "As a partial byproduct of this situation, patient-oriented end-points (i.e. quality of life assessments) are now being considered among the variables essential to the therapeutic decision making process"14 Similarly, Castillo (1994) stated, "Flexibility is needed in selection of therapy to allow effective treatment of the tumor, sparing anatomy when demanded, maintaining respect of the patient's own definition of quality of life". 1 5 1 3 Liberati et al (1986), citing McNeil et al (1981 & 1982). 1 4 Liberati et al (1986), citing Pauker and McNeil (1981). 1 5 Degner and Sloan (1992) found that in the case of cancer, 64% of the public thought that they would want to be highly involved in treatment decision making. 39 Reviews of quality of life assessment in head and neck cancer include Gotay and Moore (1992), and Hassan and Weymuller (1993). Head-and-neck-cancer-specific quality of life measures include the efficient and validated Browman et al (1993) "Head And Neck Radiotherapy Questionnaire", a morbidity/quality of life instrument16 for clinical trials of radiation therapy in locally advanced head and neck cancer. The Cella (1993a) scale version of the efficient, validated, and very general Cella et al (1993b) "Functional Assessment of Cancer Therapy Scale" includes a sub-scale adapted for head and neck cancer. It consists of 28 general items, 11 specific items and 6 experimental items. A now-validated thirty-item "EORTC Core Quality of Life Questionnaire" and a "Diagnosis-Specific Module for Head and Neck Cancer Patients" also exists (Bjordal and Kaasa 1992). Validated quality of life assessments for oral complications of head and neck cancer therapy are essentially lacking (Browman 1993; Epstein 1997). Lacking validated 17 scales for quality of life in these patients, clinimetric scales for dimensions associated with quality of life are alternatives, at least as surrogates. Oncology outcomes assessed for measurability, relevance and applicability by Pater (1994) include survival, response, symptoms, toxicity, quality of life and cost. Osaba (1993) stated that there has been in oncology an overwhelming emphasis on collection and collation of quantitative biological data such as "survival, disease-free intervals, time to progression, and response rates", with a relative lack of emphasis on (Dimensions: oral stomatitis, throat, digestive, skin, energy, and psycho-social) 1 7 (Feinstein 1987) 40 collection of symptom rating data, because these "soft" data were not considered amenable to quantitative evaluation. He further suggests that "our capability to evaluate fully the perceived benefits accruing from cancer therapy" would be enhanced if quantitative evaluation of symptom control was incorporated routinely into evaluations of health outcomes, even if these are not directly relatable to quality of life in the simplified format of data gathering employed as a step towards validated patient-derived quality of life estimates." Osaba's symptom check list for patients with recurrent head and neck cancer undergoing dose-intensive chemotherapy includes items related to the presence of the following symptoms in a four unit rating scale ranging from 1 = not at all, to 4 = very much. The symptoms rated by patients include nausea , vomiting, anorexia, tiredness, difficulty swallowing solid and liquid food, shortness of breath, difficulty opening the mouth, pain, medication for pain, and residual pain following medication. Other surrogate scales exist for dimensions related to quality of life in head and neck cancer patients, and following therapy for head and neck cancer, or complications of same. These include the List, Ritter-Sterr and Lansky (1990) performance status scales for head and neck cancer patients, the stomatitis-measuring tools described by Hyland (1986), and measures for pain (Epstein and Stewart 1993). List, Ritter-Sterr and Lansky (1990) provide a clinician-rated assessment tool containing three sub-scales: normalcy of diet, understandability of speech, and the ability to eat in public. "Results indicate that the scale is reliable across cross-raters and sensitive to functional differences across a broad spectrum of head and neck cancer". 41 Prosthodontic status also influences quality of life (as measured by a "feeling thermometer" approach) (Jacobson et al 1990). Characteristics of the prostheses found to be most important to quality of life were, in descending order of importance, comfort, function, and appearance. Murry et al (1994) relate the ability to swallow as a key issue in the quality of life. (The bedside swallow assessment was used in combination with the Browman et al (1993) "Head And Neck Radiotherapy Questionnaire").18 Oral status also is associated with systemic health quality of life and economic productivity (Hollister and Weintraub 1993). Further, "Both systemic health and quality of life are compromised when edentulousness, xerostomia, soft tissue lesions, or poorly fitting dentures affect eating and food choices. Conditions such as oral clefts, missing teeth, severe malocclusion, or severe caries are associated with feelings of embarrassment, withdrawal and anxiety. Oral and facial pain from dentures, tempero-mandibular joint disorders, and oral infections affect social interaction and daily behaviors. Dental disease accounts for many loss of work and school days. Lower wage earners and minorities are disproportionately affected" (Hollister and Weintraub 1993). Objective assessment of swallowing function in head and neck cancer patients can be provided by scintigraphy (Muz 1991). 42 Laurin (1994) found that the nutritional status of elderly individuals with poor masticatory performance is impaired. The Hollister and Weintraub (1993) review article found poor mastication especially associated with reduced Vitamin C intake, and further consider the literature to indicate that xerostomia is associated with a lower total dietary intake of calories, particularly for institutionalized patients, and also with impaired intake of protein, vitamins A and C, thiamin and riboflavin. They cite Rhodus and Brown (1990) as stating the xerostomia is so highly associated with nutritional deficiency that it should be considered a clinical indicator for nutritional deficiency. The xerostomic individual tends to have a very sore mouth with difficulty swallowing and potential changes in taste perception which in turn can decrease the desire to eat. Coulter, Marcus and Atchison (1994) disagree with arbitrary separations sometimes made between general health and oral health, to the diminishment of the latter, explaining impacts of poor oral health on general physical and psychological well-being, and associations with stress and economic and social dysfunction. Hollister and Weintraub (1993) cite literature which suggests the possibility of an association in the debilitated patient between teeth (most likely teeth of poor periodontal health) and aspiration pneumonia. They then conclude that the effects of oral disorders on quality of life include "pain, poor oral and facial aesthetics, impairments to eating, chewing and speaking, a decreased desire to interact socially, and/or a poor sense of well-being."1 9 Truyn et al (1986, in (Hollister and Weintraub 1993)) found that aesthetics and food intake impairments may cause withdrawal from society, and feelings of anxiety, depression, low self-esteem and sorrow. Employment, social and sexual relationships were all made less comfortable. Impairments of masticatory function were associated with missing anterior teeth, less than 24 teeth, and needs for extractions. 43 Dental and facial pain affects quality of life, and effects may depend on the duration and severity of the pain (Hollister and Weintraub 1994). Others conclude that patient subjective utilities, at least those involving pain, appear to bear more strongly on final, rather than intermediate, subjective intensities, with duration less important (Redel-meier and Kahneman 1996; Frederickson and Kahneman 1993; Varey and Kahneman 20 1992). Confirmation of the applicability of these findings to in vivo chronic conditions (e.g. ORN, or more typical radiotherapy experiences) seems lacking at the present time, but Epstein and Stewart (1993) found that pain often increases throughout the course of 21 radiation and persists following treatment, in some cases for six to twelve months. Varey and Kahneman (1992) proposed in their study that utility integration over time could be considered a normative rule for the evaluation of extended episodes. To investigate this, they asked subjects to assign global utilities to different symptomology histories. One goal was to test whether the integration of disutilities over time would be associated with an integration of symptomology, such as pain, over time. They did not find this to be the case. The subjects showed extreme sensitivity to symptomology trends at the end of the sequence of symptomology and a disregard for the length of time of discomfort, etc. Thus, subjective utilities, at least those involving pain, would appear to bear more strongly on final outcome than intermediate subjective states. Redelmeier and Kahneman (1996) found similar results in association with two short term medical procedures. Epstein and Stewart (1993), in their article "Radiation Therapy and Pain in Patients with Head and Neck Cancer." employed a portion of the McGill pain questionnaire along with visual analogue scales (VAS) for pain, recording of medications employed, their effect, and the location of the pain. Though the pain frequently required systemic analgesics in addition to oral rinses, patients reported generally mild pain over typical durations of one to three months or three to six months, rising as described above. Two asides worth noting are: 1. Oral pain could not be considered an indicator for oral disease because many types of oral disease do not cause pain in the early stages (Hollister and Weintraub 1993). 2. Future research into pain outcomes could incorporate the work of Moore (1996, "Combining Qualitative and Quantitative Research Approaches in Understanding Pain.". 44 3. CLINICAL PROBLEM A . O R A L C A N C E R EPIDEMIOLOGY A N D PREDISPOSING FACTORS "[Osteoradionecroses]... result from the aggressive treatment of cancer, many would not occur if cancers could be detected and treated at an early phase. The emphasis of [the NIH Consensus Development Conference on osteoradionecro-sis] on the prevention and treatment of complications should not detract from the basic goal ofprevention and early detection of cancer" (NIH 1990). The oral cavity may be defined as the anterior two thirds of the tongue, the floor of the mouth, the buccal mucosa, gingiva, retro-molar trigone, and hard palate (Crooke and Esche 1993). Yasumoto et al (1995) defines four anatomical regions of the oro-pharynx: 1. anterior: base of the tongue and vallecula 2. lateral: tonsilar and faucial pillars 3. superior: soft palate and uvula 4. posterior: posterior pharyngeal walls. Baden (1987) uses the term "oro-pharyngeal cancer" to mean "oral cancer plus pharyngeal cancer". Other authors appear to use the term "oro-pharyngeal cancer" to refer to "oral pharynx cancer". Thus, the term "cancer of the oral cavity/pharynx" will be substituted below when Baden (1987) uses the term "oro-pharyngeal cancer". Cancers of the lip are excluded from most discussions below because their epidemiological characteristics, and the potential of treatment to produce osteoradionecrosis, tend to be different from other cancers of the oral cavity. A multi-factorial pathogenic mechanism is suggested for cancer of the oral cavity/ pharynx (Baden 1987), and squamous cell carcinoma of the oral cavity may be 45 associated with an increased risk for cancers of the upper digestive and respiratory tracts (Shibuya 1987). 95% of oral cancer cases are found in people over 40 years of age (Nikiforuk 1991). 4% of all cancers in males and 2% in females in 198722 were cancers of the oral cavity and pharynx, accounting for approximately 2% of all male cancer deaths and 1% of all female cancer deaths. A worldwide male to female prevalence ratio 23 has changed from 6:1 in 1950 to 2:1 in 1987 (Baden 1987). Oro-pharyngeal cancer incidence rates exhibit great geographical variation, and in 1971 accounted for 47% of all admissions to the Cancer Hospital of Bombay (Baden 1987, citation). The sex ratio for mouth cancer mostly ranges between 4:1 and 1.5:1 male:female, being highest in southwestern, southeastern, and central Europe. Contributing factors for geographic variation in incidence and death rates may include ethnicity and exposure variability (Smith, Pindborg and Binnie 1990, 10). For instance, the chewing of betel nut confers additional risk in India (Baden 1987). American age-adjusted oral-and -pharyngeal mortality rates between 1973 and 1987 • ranged from ~8 to -10 / 100,000 for American Black males, without clear secular trend, • declined from ~ 6 to ~ 4.5 / 100,000 for American white males, • held stable at ~ 2 / 100,00 for black females, and • very slowly declined from ~ 2 for white females ((White et al 1994), cited in (I.O.M. 1995,66)) apparently worldwide possibly in association with changed female and male exposures to tobacco. 46 Also, National Cancer Institute age-adjusted death rates for oral and pharyngeal cancer would appear to have declined by 14% among persons over age 65, and by 21% for persons under age 65, between 1973 and 1990 (Beardsley 1994). However, assessment of incidence and prevalence rates of oral or oro-pharyngeal cancers reflect diagnostic capabilities, and any diagnostic bias present in secular trends over time may distort the apparent rates of oral (or oro-pharyngeal) cancer even if these rates are adjusted for the increasing age of the population. Sex- and age-specific incidence rates for cancers of the tongue and mouth in B.C. in 1993 are listed in Table 1. TABLE 1. 1993 B.C. CANCER INCIDENCE RATES (B.C. Cancer Agency Annual Reporf) AGE CATEGORY: (YEARS) INCIDENCE RATES fNew CANCER of the MOUTH MALES | FEMALES Cases per 100,000 Persons per Year): CANCER of the TONGUE MALES f FEMALES 0-34 0.3 i i 0.2 0.1 i 0.0 35-44 0.7 1 i 0.3 1.4 1 i 0.7 H 45-54 4.7 " T~ 1 4.3 2.8 " T " 1 f.6—' 55-64 7.1 " T~ 1 9.9 8.3 " T " 1 7.3 1 65-74 22.0 1 1 4.9 13.0 1 1 3 - 5 — n 75-84 15.0 1 1 15.1 5.0 1 1 2.3 85-124 35.7 " T~ 1 46.4 28.6 " T " 1 l 7.1 0-124 3.8 1 3.3 2.6 1— 1 1.3 White, Caplan, and Weintraub (1995) use American Cancer Society 1993 data to illustrate survival for head and neck cancers, relative to other types, in their Figure 15, "Estimated Relative Five-year Percentage Survival for Selected Types of Cancer By Race, United States, 1988", p.54, reproduced as Figure 4. 47 FIGURE 4. "ESTIMATED RELATIVE FIVE-YEAR PERCENTAGE S U R V I V A L FOR SELECTED TYPES OF C A N C E R B Y R A C E , UNITED STATES, 1988" after Figure 15, (White, Caplan, and Weintraub 1995, 54) C A N C E R S : E S T . 5 Y R . % S U R V I V A L S 1 2 3 4 5 6 7 8 9 10 11 12 13 TYPE OF CANCER T Y P E OF CANCER: KEY: SERIES: American Pancreas 1 1 Whites -Lung. Leukemia 7 3 2 .Blacks Non-Hodgkin's Lymphoma 4 A L L SITES = 5 O R A L C A V I T Y A N D P H A R Y N X = 6 Colon 7 Cervix 8 Prostate 9 Bladder 10 Breast 11 Hodgkin's Lymphoma Melanoma of the Skin 12 13 Tobacco use, nutritional status, alcohol intake, the presence or absence of neck dissection, irradiation dose, use of hyperbaric oxygen, cancer stage, and other variables have been examined for relevance to the risk, prognosis, and/or outcomes of head and 48 Tobacco use, nutritional status, alcohol intake, the presence or absence of neck dissection, irradiation dose, use of hyperbaric oxygen, cancer stage, and other variables have been examined for relevance to the risk, prognosis, and/or outcomes of head and neck cancer and/or osteoradionecrosis. Beumer et al (1979a) considers the prognoses for tumor and patient to be important in dental treatment planning for head and neck cancer patients at risk for osteoradionecrosis. These observations justify a brief review of cancer prognosis. Oral and/or pharyngeal cancer incidence rates rise in association with exposures to tobacco smoking, snuff dipping, reverse smoking, tobacco and betel nut quid chewing, and excessive alcohol consumption. Increased incidence of squamous cell salivary carcinomas and other cancers of the lip are associated with ionizing radiation, and solar actinic radiation (e.g., wavelengths of 320 to 400 nanometers)(multiple references, cited by Baden (1987, 50-55). Individual factors such as genetic makeup, biochemical status, and immune deficiency may be some of the factors contributing to a possible hypothetical individual susceptibility to oral or pharyngeal cancer (Baden 1987). Evidence is scanty and/or contradictory regarding whether nutritional deficiencies, occupational risks, biological risks such as dental trauma or oral sepsis, or micro-organisms cause oral or pharyngeal cancer (Baden 1987, 54-56, and references.). In the absence of exposure to betel nut, Roush, Holford, Schymura, and White (1987, 47) associate relative risk factors of 2-10 with tobacco use, 2-6 with alcohol use, and about 20 in comparison with use of neither alcohol nor tobacco. Table 2 reproduces a table, "Major Risk Factors Associated with Oral Cancer" from (Anon./Douglass, 1996). 49 T A B L E 2. MAJOR RISK FACTORS ASSOCIATED WITH O R A L C A N C E R (after table of same name in (Anon./Douglass 1996, 3) 0 D D S R A X T Q S ^ : RISK FACTORS: MALES FEMALES COMBINED SEXES Smoking and Alcohol • 1.5-37.7 5.1 - 107.9 Alcohol (between 5 & 30 drinks/wk) • 1.7-8.8 1.3-9.1 Smoking , (20 to > 40 years of age) • 1.9-3.6 2.9-5.0 (< 20 to > 40 cigarettes/ day ) • 1.2-2.8 1.8-6.2 Leukoplakia "M 12.7 4.3 Genetic Factors— brother with oral cancer <•> 7.4 Human Papillomavirus Types 16 and 18 • 6.2 Types 6 and 11 • 2.8 . Mouthwash (High Alcohol Content) 1.6 1.9 (Low Alcohol Content) 0.7 0.8 Edentulousness* 1.7 PROTECTIVE FACTORS: Cessation of Smoking (<10yrs) 1.1 1.8 (>20 years<») 0.7 0.4 Fruit -0 containing vitamins A, C, and E<> 0.2-0.7 0.5-0.8 • (Blot, McLaughlin, Winn, et al. 1988) W, (Winn, Blot, McLaughlin, et al. 1991) * (Day, Blot, Austin et al. 1993) • (International Agency for Research in Cancer Monographs. 1995) • (Winn 1995) Consider also: (Spitz 1994) and (Blot, Devesa, McLaughlin, and Fraument 1994) Prevention of osteoradionecrosis ultimately would be assisted by prevention of the same cancers that contribute to osteoradionecrosis as a side effect of treatment. One may suggest that cessation or reduction of tobacco and alcohol use (and, in alcoholics, supplementation of diet with proteins, vitamins and trace metals to decrease the promoter Exposure to betel nut is not reflected in this table. "An odds ratio expresses the risk of developing oral cancer associated with any specific factor. Values larger than 1.0 indicate increased risk; values less than 1.0 represent a decreased risk or protective effect. Values are usually statistically adjusted to reduce the influence of other risk variables associated with the disease." 50 effect of alcohol) should tend to reduce the incidence and prevalence of cancer of the oral 26 cavity or pharynx (Baden 1987), and thus its complications. Stell (1991) found that the primary predictor of survival, both from initial presentation and from the date, of recurrence, was the site, irrespective of other host factors. Roland, Caslin, Nash and Stell (1992) found through retrospective analysis of a series of 3,294 patients that site was closely associated with grading (poorly differentiated tumors being more common in the pharynx, and well-differentiated tumors being more common in the mouth and larynx.) Rates of metastasis were 46% (near metastases) and 3.4% (distant metastases) in patients with poorly differentiated tumors, and 28% and 1.8% in patients with well-differentiated tumors. Survival rates were 27% and 33%, respectively, for poorly and well differentiated tumors, and recurrence rates 27% and 25% respectively, for poorly and well differentiated tumors at the primary site, and 30% and 26% respectively, in the lymph nodes (Roland, Caslin, Nash and Stell 1992). Umeda (1992) also observed a correlation between histologic grade of malignancy and the prevalence of neck metastasis, and Cusumano and Mark Persky (1988) that survival correlated best with the T N M stage of disease at initial presentation, although Zatterstrom et al (1991) criticized T N M classification and histopathologic grading systems as only semi-quantitative and subjective, then went on to find that histologic grade, clinical stage, tumor size, and patient age did not correlate with prognosis. Other proposed prognostic variables for oral squamous cell carcinoma include D N A ploidy (Tytor 1989), and for squamous cell carcinoma of the mobile tongue, tumor thickness (Nathanson 1989). Barker et al (1992) lists resources for promotion of tobacco cessation and nutritional enhancement. 51 B. ORAL SQUAMOUS CELL CARCINOMA LESIONS AND TREATMENT STRATEGIES 2 7 90% of all carcinomas of the oral cavity are squamous. A number of variants of squamous cell carcinoma of the oral cavity exist. These include verrucous, adenoid squamous, spindle cell, and basaloid (Cadier et al 1992). Common locations for oral squamous carcinoma are the oral portion of the tongue, the mandibular gingiva, and the floor of the mouth. Floor-of the-mouth lesions are characteristic of alcoholics. Many squamous cell carcinomas arise from malignant transformation of premalignant lesions, as described in Silverman (1987). Thus, excisional biopsy is indicated in many cases to prevent malignant transformation (Silverman 1987). Squamous cell carcinomas of the oral cavity themselves are infiltrative, aggressive, and generally require aggressive therapy including irradiation (McGaw and Pan 1996). Therapeutic irradiation can cause residual as well as transient pathology in the tissues irradiated to high dose. Surgery to remove affected tissue and potential spread is usually involved. Mortality from oral cancer is high. (Less than one half are cured.) Of the 10,000 deaths per year in the U.S. (2.5% of all cancer related deaths), the survival rate increases dramatically when the cancer is detected early (i.e., when the lesion is less than 3 cm in diameter and there is no regional cervical lymph node involvement (Silverman 1987). in overview. The relative merits of different techniques (types of irradiation, chemotherapy subtypes, etc.) as squamous cell carcinoma tumor control measures will not be detailed, since the treating dentist would accept these for a particular case or a particular decision analysis as assumptions given to him/her by others. Since osteoradionecrosis arises from radiation therapy (usually treating squamous cell carcinoma), this treatment modality will be emphasized. 52 Neck dissection often is undertaken to remove metastases. Concern has been expressed, however, that this when elective ("just in case the lymph nodes are involved"), might needlessly compromise the tissues or blood supply. Mendenhall (1988) found no difference whether radiotherapy alone was administered, or an elective neck dissection also took place, with regard to rates of disease control at the primary site or ability to surgically manage patients who developed primary site recurrences of head and neck squamous cell carcinoma. Lampe (1971) considers squamous cell carcinoma to be a radioresponsive and radiocurable neoplasm. Super-voltage radiation may be used in-and-around bone to minimize acute and chronic cutaneous radiation reactions and spare bone tissues, but low voltage irradiation may be used for superficial lip carcinomas and orthovoltage radiation for larger or older lip lesions or broader involvements. These extra-oral approaches may be combined with brachytherapy (e.g. low intensity radium needles) to provide additional impact on the neoplasm. Cure often occurs but some tumors may not be eradicable because of metastasis or tumor radio-resistance to dose ranges that are tolerable by normal structures. (The tongue is well vascularized and tolerates high radiation doses.) Radiotherapeutic doses necessary for tumor control are "almost inevitably" followed by osteoradionecrosis in cases where the squamous cell carcinoma has invaded 28 the mandible ( Henk 1985; Henk and Langdon 1995) McGregor and MacDonald (1988) found that invasion of the non-irradiated edentulous mandible most frequently is through the residual alveolar occlusal ridge, but found that the irradiated mandible 28 Radiation doses associated with the "almost inevitable" ORN arising in this circumstance were undefined. They might be implied, from the Henk and Langdon (1995) discussion of Bedwinek (1976), to be >7000 cGy. {A cited reference, (Carter 1980), does not define the relevant doses.)) 53 tended to be prone to much more variable routes of tumor entry. McGregor and MacDonald (1989) reported that two modes of spread of squamous cell carcinoma in the mandible are observed, "spread in relation to the inferior alveolar nerve, and spread in the spaces between cancellous bony trabeculi" Hyperbaric oxygenation is frequently employed to treat, and less commonly to prevent, osteoradionecrosis. An in-vitro experiment by Sklizovic (1993) suggests that squamous cell carcinoma should not grow at a higher rate in the presence of hyperbaric oxygenation. Bradfield, Kinsella, Mader, Bridges, and Calhoun (1996) conclude that this finding is typical of results reported for transplanted tumor cells. However, Bradfield, Kinsella, Mader, Bridges, and Calhoun (1996) provide a few anecdotes associating "rapid progression of head and neck squamous carcinoma after hyperbaric oxygenation", and suggest that some evidence reported in the literature would appear to support "a role for hyperbaric oxygen in enhancing growth of persisting tumors." 30 Bernstein et al (1993) describes carcinogenesis as a potential late complication of cancer therapy. He states that intermediate doses of radiation, for instance 200 to 1000 cGys, appear to be more tumorigenic than larger doses, "perhaps because larger doses of radiation result in cell death, whereas intermediate doses are more likely to result in mutations capable of producing tumors". However, Bernstein et al (1993) cite their 29 McGregor and MacDonald (1989) reported that the pattern of spread in cancellous bone in irradiated mandibles and the incidence of nerve-related spread was not significantly different in irradiated or non-irradiated mandibles. In contrast to the alternative of spread through the spaces between cancellous bony trabeculi, a tendency for nerve-related spread was identified as more frequent in the edentulous mandible than in the partially dentate non-irradiated mandible. Totsuka et al (1991) reported that most squamous cell lesions with erosive bone defects showed an "expansive" pattern histologically. They found that the extent of the radiologically detected bone defect approximated the size of the histologic involvement. 30 a comprehensive review of the biology of chronic radiation effect on tissues and wound healing. 54 reference (Hellman 1985) as stating that it is very difficult to demonstrate increases in the incidence of cancers after therapeutic irradiation except in children,31 and (Friedman et al 1988) states that irradiation to high dose would appear "neither to protect against, nor induce multiple primaries, within head and neck radiation portals". Guillamondegui (1993) states "It has been clearly established that adequate, individualized selection of treatment is the most significant factor in the treatment of oral cavity cancer. Surgical and functional rehabilitation of the patient represents a further step in the decision making sequence. Every patient and his tumor should be considered as components of a unique equation. The most suitable form of therapy should be selected in each individual case." 32 The Summary of (Harrison and Fass 1990) reflects the philosophical approach of one institution: "Radiation therapy using both external beam and brachytherapy is one of the mainstays of treatment for oral cavity cancer. For early lip, tongue, and floor of mouth lesions, radiation alone is highly effective and produces an excellent functional result. More advanced lesions are frequently treated with combined therapy or radiation alone with surgical salvage.' "Buccal mucosa and gingiva lesions are generally treated with surgery but can be managed by radiation therapy if they are early. More advanced lesions are treated by surgery plus post operative radiation.' "Retromolar trigone lesions can be managed by radiation alone if they are small, but larger lesions are generally treated by combined therapy.' "Clearly, the goal of treatment of squamous cell cancer of the oral cavity is cure of a disease with optimum function results. These goals could be achieved only by cooperative interdisciplinary evaluation and management by the surgeons, radiation oncologists, and dentists involved in the care of these patients" (Harrison and Fass 1990). (However, Bernstein et al (1993) references Shore (1984), Shore (1990) found that infants given irradiation averaging 2,250 cGy to treat presumed thymic enlargement demonstrated a significantly higher incidence rate for skin cancer than their control siblings. Bernstein et al (1993) also cite references for "numerous cases of basal cell carcinoma arising within radiodermatitis...". an excellent review of intra-oral procedure for radiation therapy for oral cavity cancer 55 A comprehensive overview of current techniques and indications for radiotherapy of head and neck cancer patients is provided in Ang, Anders and Peters (1994). E X T E R N A L I R R A D I A T I O N X and gamma rays are high energy photons of very short wave length produced in an electrical device such as a linear accelerator that accelerates electrons to a high energy then stops them on a target of tungsten, gold, etc., thus producing the x-rays. On the other hand, gamma rays are emitted from natural sources as unstable radioactive nuclei prone to achieve lower energy states (Bernstein, et al 1993). Bernstein et al (1993) state that until 1950 most external beam radiotherapy involved kilovoltage (orthovoltage) machines generating x-rays at voltages up to about 300 kilovolts peak (KVP). Gehrig (1969) observed one mild case of osteoradionecrosis in a patient who received only 2500 cGy of orthovoltage, but cited another author (Meyer) as indicating that in general a minimum of 4000 cGy was required. Unless otherwise specified, future discussions in this thesis should be assumed to refer to the newer super-voltage x-radiation which, relative to orthovoltage, is bone-sparing and is associated with higher dose thresholds for onset of osteoradionecrosis. Criteria for selection of type of irradiation therapy includes the origin of the Irradiation voltages used in external irradiation may include 1. ortho-voltage (150- 250 KVP)(Rubin and Doku 1976) (150-400 KVP)(Behesti and Javid 1978); 2. super voltage (500 KeV to 8 MeV) (Buschke and Parker 1972) in (Rubin and Doku 1976) a) x-rays produced by an x-ray therapy machine, or b) 60Cobalt, which emits high energy photon gamma rays with an average energy of 1.2 MeV (Rubin and Doku 1976, who stress this is not megavoltage.) 3. mega-voltage x-rays (20-MeV, capable of greater penetration), OBehesti and Javid 1978) and 4. electron beam (electrons in range of 8 to 50 MeV) (Behesti and Javid 1978), produced by linear accelerators, that are less penetrating than megavoltage x-rays [as illustrated in Figures. 1, 2 and 3 of Bernstein et al (1993),which illustrate the effect of increasing beam energy on the penetration of a beam ofphotons (X or gamma rays in tissue), the penetration of a spectrum of electron beam energies in tissue, and the skin-sparing dose build-up effects seen with MV irradiation.], and thus are appropriate for treating superficial lesions. (Bernstein et al 1993) 56 tumor, the size and location of the tumor, and the sensitivity to irradiation. X-rays are composed of photons which interact with living tissue through a combination of photoelectric and Compton effects. The amount of energy absorbed by tissues through photo-electric effect is approximately dependent on the cube of the tissue's atomic number: 5.9 for adipose tissue, 7.4 for muscle, and 13.8 for bone.34 Therefore, since the ratio of the atomic numbers of bone in relation to soft tissue is about 2, the amount of energy absorbed by bone through photoelectric effect would be approximately 2 , or eight fold, that of soft tissue. A second way in which radiation is absorbed is the Compton Effect, which is proportional to the electron density of the living tissues. At the energy range of 250 K V P , more Compton effect is present, so the ratio of energy absorbed through all methods in bone is 2:1 overall, rather than the 8:1 ratio that would exist i f only the photoelectric effect was present. Radiation in the 1 to 10 MeV range (e.g. 6 0Cobalt, 1.2 MeV) tends to have radiobiologic effects dominated by the 5. higher energy beams capable of deeper tissue penetration, produced by linear accelerators (Bernstein et al 1993). Betatron irradiation consists of two types: beta rays (negatively charged particles) and photons (x-rays). Electrons of energies from 3 to 45 MeV, and x-rays of over 30 MeV may be delivered (Rubin and Doku 1976). Linear accelerators are capable of emitting x-rays and electrons in the range of 4 to 40 MeV. Neutrons may be produced in a cyclotron. These lar