Medical Decision Making

The purpose of this study was to analyze the management of individual patients with unruptured intracranial aneurysms (UN-ANs) using a decision-analytic approach. Tran sition probabilities among Glasgow Outcome Scale (GOS) categories were estimated from the published literature and data from patients who had been treated at Kitasato University Hospital. Utilities were obtained from 140 health providers based principally on the GOS. Baseline analysis for a healthy 40-year-old man with an anterior UN-AN less than 10 mm in diameter showed that the quality-adjusted life expectancies for preventive operation and follow-up were 15.34 and 14.66 years, respectively. For a follow-up strategy to be preferred, the annual rupture rate had to be as low as 0.9%. These results were sustained through extensive sensitivity analysis. The results sup port preventive operation for UN-ANs, and identify problems that can be clarified with a well-designed stratified clinical trial. Key words: decision analysis; Markov model; unruptured intracranial aneurysms; Glasgow Outcome Scale; utility; preventive oper ations. (Med Decis Making 1998;18:357-364)

Objective. The authors developed a Markov decision model to evaluate the health implications of testing for mutations in the BRCA1 and BRCA2 breast-ovarian cancer susceptibility genes. Prophylactic measures considered included various combinations of immediate and delayed bilateral mastectomy and oophorectomy or taking no action. Methods. The model incorporated the likelihood of developing breast and/or ovarian cancer, survival, and quality of life. Parameter values were taken from public data bases, the published literature, and a survey of cancer experts. Outcomes considered were additional life expectancy and quality-adjusted life years (QALYs). Results are reported for 30-year-old cancer-free women at various levels of hereditary risk. Results and conclusions. The vast majority of women will not benefit from testing because their pre-test risks are low and surgical prophylaxis is undesirable. However, women who have family histories of early breast and/or ovarian cancer may gain up to 2 QALYs by allowing genetic testing to inform their decisions. Key words: BRCA1; BRCA2; ge netic testing; breast cancer; ovarian cancer; decision analysis. (Med Decis Making 1998;18:365-375)

The optimal management strategy for men who have localized prostate cancer remains controversial. This study examines the extent to which suggested treatment based on the perspective of a group or society agrees with that derived from individual patients' preferences. A previously published decision analysis for localized prostate cancer was used to suggest the treatment that maximized quality-adjusted life expectancy. Two treatment recommendations were obtained for each patient: the first (group-level) was derived using the mean utilities of the cohort; the second (individual-level) used his own set of utilities. Group-level utilities misrepresented 25-48% of individuals' pref erences depending on the grade of tumor modeled. The best kappa measure achieved between group and individual preferences was 0.11. The average quality-adjusted life years lost due to misrepresentation of preference was as high as 1.7 quality-adjusted life years. Use of aggregated utilities in a group-level decision analysis can ignore the substantial variability at the individual level. Caution is needed when applying a group- level recommendation to the treatment of localized prostate cancer in an individual patient. Key words: decision analysis; utility assessment; prostate cancer; patient pref erences. (Med Decis Making 1998;18:376-380)

This study used a single binary-gamble question per health state per respondent to obtain societal preferences for the health states intermittent claudication and major amputation and compare those with Health Utilities Indices obtained from patients, to test the feasibility of this method, and to investigate whether the utility depends on the presentation of a vignette as generic vs disease-specific. A random sample of the general U.S. population (n = 1,003) was randomly divided into ten subgroups. In tel ephone interviews, subjects answered one binary-gamble question in a standard-gam ble format for each of two health states. The risks of death varied across subgroups but not between health states. Mean utility was estimated by the area above the pro portional distribution of responses indicating acceptance of the gamble. The method is based on the binary-choice method used in contingent-valuation studies of willing ness to pay. The health states were alternatively described by generic and disease- specific vignettes in two subsamples. The results suggest that the binary-gamble ques tion can be used to obtain societal preferences for health states, and that disease-specific descriptions yield lower utilities compared with generic descriptions of health states. Key words: peripheral arterial occlusive disease, intermittent claudica tion ; quality of life; standard gamble; utility assessment. (Med Decis Making 1998;18: 381-390)

Objective. Temporary health states cannot be measured in the traditional way by means of techniques such as the time tradeoff (TTO) and the standard gamble (SG), where health states are chronic and are followed by death. Chained methods have been developed to solve this problem. This study assesses the feasibility of a chained TTO and a chained SG, and the consistency and concordance between the two meth ods. Patients and methods. Seventy female early-stage breast cancer patients were interviewed. In using both chained methods, the temporary health state to be evaluated was weighed indirectly with the aid of a temporary anchor health state. The patients were asked to evaluate their actual health states, a hypothetical radiotherapy scenario, and a hypothetical chemotherapy scenario. Results. Sixty-eight patients completed the interview. The use of the anchor health state yielded some problems. A significant difference between the means of the TTO and the SG was found for the anchor health state only. For the other health states, the results were remarkably close, because the design avoided some of the bias effects in traditional measurements. Conclusion. The feasibility and the consistency of the chained procedure were satisfactory for both methods. The problems regarding the anchor health state can be solved by adapting the methods and by the use of a carefully chosen anchor health state. The chained method avoids biases present in the conventional method, and thereby the TTO and the SG may be reconciled. Moreover, there are several psychological advantages to the method, which makes it useful for diseases with uncertain prognoses. Key words: utility assessment; time tradeoff; standard gamble; breast cancer; chemotherapy; ra diotherapy. (Med Decis Making 1998;18:391-399)

The purpose of this study was to compare time-tradeoff values and standard-gamble utilities obtained during telephone interviews with those obtained through face-to-face interviews. Sixty-five patients with peripheral arterial occlusive disease completed both interviews. One week prior to the telephone interview, the patients received by mail a questionnaire in which the value and utility measures were presented in writing. The face-to-face interviews used the same questions, but the interviewer used visual aids. The mean time-tradeoff values were 0.84 (SD 0.20) vs 0.86 (SD 0.17) for the telephone and face-to-face interviews, respectively (p = 0.31). The mean standard-gamble utilities were 0.93 (SD 0.16) vs 0.92 (SD 0.17) for the telephone and face-to-face interviews, respectively (p = 0.26). In conclusion, telephone interviews yield similar time-tradeoff values and standard-gamble utilities compared with face-to-face interviews, suggesting that telephone interviews can replace face-to-face interviews. Key works: utility as sessment ; time tradeoff; standard gamble; telephone interviews. (Med Decis Making 1998;18:400-405)

Objective. To determine the attitudes of obstetricians and gynecologists toward hor mone replacement therapy (HRT), and the beliefs and intuitions that affected those attitudes. Design. A questionnaire was sent to 1,000 gynecologists in the United States; 328 replies were received. The questionnaire asked about effects of HRT, practices concerning HRT, and decisions in hypothetical scenarios. Results. The re spondents strongly favored HRT, and they were well informed about its effects on osteoporosis, cardiovascular disease, and breast cancer. They were aware of conflict ing findings concerning breast cancer. The strength of their recommendation of HRT was sensitive to patient differences in risk factors. The respondents also showed four biases hypothesized to cause resistance to HRT: omission bias (more concern about harmful acts than harmful omissions); proportionality bias (attention to relative risk rather than risk differences); naturalness bias (preference for the natural); and ambi guity (avoiding options with missing information). Proportion bias, naturalness bias, and (weakly) omission bias were related to less favorable attitudes toward HRT. Con clusion. Although specialists are highly favorable toward HRT in general, some neg ativity toward HRT may result from decision biases. Key words: hormone replacement; estrogen; decision making; biases. (Med Decis Making 1998;18:406-411)

Background. Information order can influence judgment. However, it remains unclear whether the order of clinical data affects physicians' interpretations of these data when they are engaged in familiar diagnostic tasks. Methods. Of 400 randomly selected family physicians who were given a questionnaire involving a brief written scenario about a young woman with acute dysuria, 315 (79%) returned usable responses. The physicians had been randomized into two groups, and both groups had received the same clinical information but in different orders. After learning the patient's chief com plaint, physicians received either the patient's history and physical examination results followed by the laboratory data (the H&P-first group) or the laboratory data followed by the history and physical examination results (the H&P-last group). The results of the history and physical examination were supportive of the diagnosis of UTI, while the laboratory data were not. All physicians judged the probability of a urinary tract infection (UTI) after each piece of information. Results. The two groups had similar mean estimates of the probability of a UTI after learning the chief complaint (67.4% vs 67.8%, p = 0.85). At the end of the scenario, the H&P-first group judged UTI to be less likely than did the H&P-last group (50.9% vs 59.1 %, p = 0.03) despite having identical information. Comparison of the mean likelihood ratios attributed to the clinical information showed that the H&P-first group gave less weight to the history and phys ical than did the H&P-last group (p = 0.04). Conclusions. The order in which clinical information was presented influenced physicians' estimates of the probability of dis ease. The clinical history and physical examination were given more weight by phy sicians who received this information last. Key words: diagnosis; urinary tract infec tions ; judgment; primary care; clinical decision making. (Med Decis Making 1998;18: 412-417)

The author presents evidence for the descriptive adequacy of the quality-adjusted life years (QALY) model as applied to health profiles. One important assumption of the model is preferential independence: if two profiles have the same health state during year X, then preference between them does not switch if the level of health changes during year X. In experiment 1, 30-year health profiles were used to perform 27 em pirical tests of independence with 98 subjects. Independence was reliably satisfied in all 27 tests. In experiment 2, 15 additional tests were conducted. These tests had been specifically designed to be more sensitive to independence violations, but indepen dence was still mostly satisfied. In both experiments, the conclusions about indepen dence hold regardless of what discount rate is used. These results act as a "lower bound" on the validity of the QALY model for health profiles. Key words: quality-ad justed life years; QALY model; health profiles; preferential independence. (Med Decis Making 1998;18:418-428)

Random variability of blood pressure complicates the diagnosis and subsequent treat ment of hypertension. To evaluate the importance of the number of blood pressure measurements in the correct diagnosis and control of hypertension, the authors used a Bayesian model to estimate the true average blood pressure of a group of newly diagnosed hypertensives, then calculated the diagnostic error that would result from monitoring methods using 24 daytime measurements or from using only three random monitoring measurements. The study population consisted of 129 individuals with newly diagnosed mild hypertension according to standard criteria, who were also eval uated with an ambulatory blood pressure monitor. In true normotensives (daytime di astolic blood pressure <90 mm Hg), the negative predictive value with three measure ments was 0.92, and it rose to 0.96 with monitoring methods. In mild hypertensives (90-104 mm Hg), the positive predictive value was 0.64 with three measurements and 0.84 with monitonng methods, thus reducing the rate of false mild hypertensives from 35% to 15%. Finally, in patients with moderate or severe hypertension (>104 mm Hg), the positive predictive value improved from 0.26 with three readings to 0.61 with mon itoring methods. Similar results were observed with daytime systolic pressure mea surements. As the number of measurements increased, the diagnostic error due to the random variability of blood pressure became progressively smaller. In cases of hyper tension, the large improvement in predictive values may justify using monitoring meth ods to confirm standard diagnosis. Key words: ambulatory blood pressure monitoring; Bayesian analysis; blood pressure; diagnostic error; hypertension; random variability. (Med Decis Making 1998;18:429-435)

A fully Bayesian approach to a general nonlinear ordinal regression model for ROC- curve analysis is presented. Samples from the marginal posterior distributions of the model parameters are obtained by a Markov-chain Monte Carlo (MCMC) technique— Gibbs sampling. These samples facilitate the calculation of point estimates and cred ible regions as well as inferences for the associated areas under the ROC curves. The analysis of an example using freely available software shows that the use of nonin formative vague prior distributions for all model parameters yields posterior summary statistics very similar to the conventional maximum-likelihood estimates. Clinically im portant advantages of this Bayesian approach are: the possible inclusion of prior knowl edge and beliefs into the ROC analysis (via the prior distributions), the possible cal culation of the posterior predictive distribution of a future patient outcome, and the potential to address questions such as: "What is the probability that a certain diagnostic test is better in one setting than in another?" Key words: ROC curve; diagnostic test; ordinal regression; Bayesian methods; MCMC; Gibbs sampling; maximum likelihood (Med Decis Making 1998;18:436-443)