Until the mid-1980s, most economic analyses of healthcare technologies were based on decision theory and used decision-analytic models. The goal was to synthesize all relevant clinical and economic evidence for the purpose of assisting decision makers to efficiently allocate society's scarce resources. This was true of virtually all the early cost-effectiveness evaluations sponsored and/or published by the U.S. Congressional Office of Technology Assessment (OTA) (15), Centers of Disease Control and Prevention (CDC), the National Cancer Institute, other elements of the U.S. Public Health Service, and of healthcare technology assessors in Europe and elsewhere around the world. Methodologists routinely espoused, or at minimum assumed, that these economic analyses were based on decision theory (8;24;25). Since decision theory is rooted in—in fact, an informal application of—Bayesian statistical theory, these analysts were conducting studies to assist healthcare decision making by appealing to a Bayesian rather than a classical, or frequentist, inference approach. But their efforts were not so labeled. Oddly, the statistical training of these decision analysts was invariably classical, not Bayesian. Many were not—and still are not—conversant with Bayesian statistical approaches.

Interest in Bayesian analyses has increased recently, in part as a response to policy makers wanting sound scientific bases for health technology assessments, and associated healthcare funding decisions. This paper provides a brief and simplified description of Bayesian reasoning. Bayes is illustrated in a clinical setting of an expert helping a woman understand the potential risk of passing on an inheritable disease (hemophilia) to her next child, based on disease occurrence in two living children. The illustration describes fundamental concepts and derivations, such as Bayes theorem, likelihood functions, prior probability, and posterior probability. A second illustration shows the use of Bayes for interpreting clinical trial results. The uncertainty in the clinical effect before and after the trial analyses has been completed is characterized by the Bayes prior and posterior probabilities, respectively. Techniques are also shown for estimating the potential loss (e.g., in lives lost) for making the wrong decision with and without knowledge of the trial results, an estimation that cannot be carried out using techniques of hypotheses testing associated with the frequentist school of statistics. Information from Bayes analysis then may be used to help policy makers decide, or justify, whether the analyses provides a sufficient basis for making a treatment recommendation, or whether there remains a need to request more information. Subsequent papers in this volume offer additional examples and clarification of the use of Bayes in clinical practice and in interpretation of clinical studies.

The objective of this paper is to illustrate the advantages of the Bayesian approach in quantifying, presenting, and reporting scientific evidence and in assisting decision making. Three basic components in the Bayesian framework are the prior distribution, likelihood function, and posterior distribution. The prior distribution describes analysts' belief a priori; the likelihood function captures how data modify the prior knowledge; and the posterior distribution synthesizes both prior and likelihood information. The Bayesian approach treats the parameters of interest as random variables, uses the entire posterior distribution to quantify the evidence, and reports evidence in a “probabilistic” manner. Two clinical examples are used to demonstrate the value of the Bayesian approach to decision makers. Using either an uninformative or a skeptical prior distribution, these examples show that the Bayesian methods allow calculations of probabilities that are usually of more interest to decision makers, e.g., the probability that treatment A is similar to treatment B, the probability that treatment A is at least 5% better than treatment B, and the probability that treatment A is not within the “similarity region” of treatment B, etc. In addition, the Bayesian approach can deal with multiple endpoints more easily than the classic approach. For example, if decision makers wish to examine mortality and cost jointly, the Bayesian method can report the probability that a treatment achieves at least 2% mortality reduction and less than $20,000 increase in costs. In conclusion, probabilities computed from the Bayesian approach provide more relevant information to decision makers and are easier to interpret.

Current health economic literature does not provide clear guidelines on how uncertainty around cost-effectiveness estimates should be incorporated into economic decision models. Bayesian analysis is a promising alternative to classical statistics for incorporating uncertainty in economic analysis. Estimating a loss function that relates outcomes to societal welfare is a key component of Bayesian decision analysis. Health economists commonly compute the loss function based on the quality-adjusted life-years associated with each outcome. However, if welfare economics is adopted as the theoretical foundation of the analysis, a loss function based in cost-benefit analysis (CBA) may be more appropriate. CBA has not found wide use in health economics due to practical issues associated with estimating such a loss function. In this paper, we present a method based in conjoint analysis for estimating the CBA loss function that can be applied in practice. We illustrate the use of the methodology using data from a pilot study.

A framework is presented that distinguishes the conceptually separate decisions of which treatment strategy is optimal from the question of whether more information is required to inform this choice in the future. The authors argue that the choice of treatment strategy should be based on expected utility, and the only valid reason to characterize the uncertainty surrounding outcomes of interest is to establish the value of acquiring additional information. A Bayesian decision theoretic approach is demonstrated through a probabilistic analysis of a published policy model of Alzheimer's disease. The expected value of perfect information is estimated for the decision to adopt a new pharmaceutical for the population of patients with Alzheimer's disease in the United States. This provides an upper bound on the value of additional research. The value of information is also estimated for each of the model inputs. This analysis can focus future research by identifying those parameters where more precise estimates would be most valuable and indicating whether an experimental design would be required. We also discuss how this type of analysis can also be used to design experimental research efficiently (identifying optimal sample size and optimal sample allocation) based on the marginal cost and marginal benefit of sample information. Value-of-information analysis can provide a measure of the expected payoff from proposed research, which can be used to set priorities in research and development. It can also inform an efficient regulatory framework for new healthcare technologies: an analysis of the value of information would define when a claim for a new technology should be deemed substantiated and when evidence should be considered competent and reliable when it is not cost-effective to gather any more information.

The objective of this paper is to discuss why Bayesian statistics are not used more in healthcare decision making and what might be done to increase the use of Bayesian methods. First, a case is made for why Bayesian analysis should be used more widely. Serious weaknesses of commonly used frequentist methods are discussed and contrasted with advantages of Bayesian methods. Next, the question of why Bayesian methods are not used more widely is addressed, considering both philosophical differences and practical issues. Contrary to what some might think, the practical issues are more important in this regard. Finally, some steps to encourage increased use of Bayesian methods in healthcare decision making are presented and discussed. These ideas are straightforward but are by no means trivial to implement, largely because it is difficult to fight tradition and make major paradigm shifts quickly. The primary needs are improved Bayesian training at the basic level (which means textbooks and other materials as well as training of those who teach at the basic level), procedures to make Bayesian analysis easier to understand and use (better software and standard methods for displaying and communicating Bayesian outputs will help here), and the education of decision makers about the advantages of Bayesian methods in important healthcare decision-making problems.

The aim of this paper is to discuss the use of Bayesian methods in cost-effectiveness analysis (CEA) and the common ground between Bayesian and traditional frequentist approaches. A further aim is to explore the use of the net benefit statistic and its advantages over the incremental cost-effectiveness ratio (ICER) statistic. In particular, the use of cost-effectiveness acceptability curves is examined as a device for presenting the implications of uncertainty in a CEA to decision makers. Although it is argued that the interpretation of such curves as the probability that an intervention is cost-effective given the data requires a Bayesian approach, this should generate no misgivings for the frequentist. Furthermore, cost-effectiveness acceptability curves estimated using the net benefit statistic are exactly equivalent to those estimated from an appropriate analysis of ICERs on the cost-effectiveness plane. The principles examined in this paper are illustrated by application to the cost-effectiveness of blood pressure control in the U.K. Prospective Diabetes Study (UKPDS 40). Due to a lack of good-quality prior information on the cost and effectiveness of blood pressure control in diabetes, a Bayesian analysis assuming an uninformative prior is argued to be most appropriate. This generates exactly the same cost-effectiveness results as a standard frequentist analysis.

A desirable element of cost-effectiveness analysis (CEA) modeling is a systematic way to relate uncertainty about input parameters to uncertainty in the computational results of the CEA model. Use of Bayesian statistical estimation and Monte Carlo simulation provides a natural way to compute a posterior probability distribution for each CEA result. We demonstrate this approach by reanalyzing a previously published CEA evaluating the incremental cost-effectiveness of tissue plasminogen activator compared to streptokinase for thrombolysis in acute myocardial infarction patients using data from the GUSTO trial and other auxiliary data sources. We illustrate Bayesian estimation for proportions, mean costs, and mean quality-of-life weights. The computations are performed using the Bayesian analysis software WinBUGS, distributed by the MRC Biostatistics Unit, Cambridge, England.

Bayesian statistics provide effective techniques for analyzing data and translating the results to inform decision making. This paper provides an elementary tutorial overview of the WinBUGS software for performing Bayesian statistical analysis. Background information on the computational methods used by the software is provided. Two examples drawn from the field of medical decision making are presented to illustrate the features and functionality of the software.

Decision making in health care has become increasingly reliant on information technology, evidence-based processes, and performance measurement. It is therefore a time at which it is of critical importance to make data and analyses more relevant to decision makers. Those who support Bayesian approaches contend that their analyses provide more relevant information for decision making than do classical or “frequentist” methods, and that a paradigm shift to the former is long overdue. While formal Bayesian analyses may eventually play an important role in decision making, there are several obstacles to overcome if these methods are to gain acceptance in an environment dominated by frequentist approaches. Supporters of Bayesian statistics must find more accommodating approaches to making their case, especially in finding ways to make these methods more transparent and accessible. Moreover, they must better understand the decision-making environment they hope to influence. This paper discusses these issues and provides some suggestions for overcoming some of these barriers to greater acceptance.

Objective: To enhance the validity of a well-known expert panel process, we used data from patient surveys to identify and correct rating errors.

Methods: We used the two-round RAND/UCLA panel method to rate indications of harmful (presence of problems), hazardous (at risk for problems), and nonhazardous (no known risks) drinking in older adults. Results from the panel provided guidelines for classifying older individuals as harmful, hazardous, or nonhazardous drinkers, using a survey. The classifications yielded unexpectedly high numbers of harmful and hazardous drinkers. We hypothesized possible misclassifications of drinking risks and used the survey data to identify indications that may have led to invalid ratings. We modified problematic indications and asked three clinician panelists to evaluate the clinical usefulness of the modifications in a third panel round. We revised the indications based on panelist response and reexamined drinking classifications.

Results: Using the original indications, 48% of drinkers in the sample were classified as harmful, 31% as hazardous, and 21% as nonhazardous. A review of the indications revealed framing bias in the original rating task and vague definitions of certain symptoms and conditions. The modified indications resulted in classifications of 22% harmful, 47% hazardous, and 31% nonhazardous drinkers.

Conclusions: Analysis of survey data led to identification and correction of specific errors occurring during the panel-rating process. The validity of the RAND/UCLA method can be enhanced using data-driven modifications.

Objectives: We were interested in health coverage in women's magazines in the United States and how it compared with articles in medical journals, women's health interests, and women's greatest health risks.

Methods: We examined 12 issues of Good Housekeeping (GH) and Woman's Day (WD) and 63 issues of the New England Journal of Medicine (NEJM) and the Journal of the American Medical Association (JAMA). We tallied the most common health questions of women presenting to the University of Michigan's Women's Health Resource Center (WHRC) during the same period.

Results: Less than a fifth of the magazine articles dealt with health-related topics. Of those, a third dealt with diet, with the majority emphasizing weight loss rather than eating for optimal health. Few of the articles cited research studies, and even fewer included the name of the journal in which the study was published. In JAMA and NEJM, less than one-fifth of original research studies dealt with women's health topics, most commonly pregnancy-related issues, hormone replacement therapy, breast and ovarian cancer, and birth defects. At the same time, the most common requests for information at the WHRC related to pregnancy, fertility, reproductive health, and cancer.

Conclusion: The topics addressed in women's magazines do not appear to coincide with the topics addressed in leading medical journals, nor with women's primary health concerns or greatest health risks. Information from women's magazines may be leading women to focus on aspects of health and health care that will not optimize risk reduction.

Objectives: Time costs borne by women when undergoing cervical cancer screening have rarely been elucidated, although such costs may pose substantial barriers to care. The purpose of this project was to quantify the opportunity costs associated with cervical cancer screening in young women attending Planned Parenthood Clinics.

Methods: We conducted a self-report survey of 105 women from six clinics to measure travel, waiting, and exam times associated with cervical cancer screening. Respondents recorded their time of arrival and departure, length of time in the waiting room, age, income level, and hours per week they worked outside of the home. Time costs were valued three ways: through self-reported hourly wage, age- and gender-adjusted minimum earnings, and national age- and gender-adjusted hourly wages.

Results: Respondents were on average 24 years old, worked 29 hours per week outside the home, and earned less than $20,000 per year. Mean time for one-way travel was 18.7 minutes; waiting room time was 16.9 minutes; and exam time was 50.8 minutes. Time costs were estimated to be $14.08 per visit based upon the self-reported hourly wage; $16.46 per visit based upon age- and gender-adjusted minimum earnings; and $19.63 per visit based upon age- and gender-adjusted national wage rates.

Conclusions: Time costs associated with cervical cancer screening represent an important opportunity cost and should be considered in studies attempting to identify barriers to screening adherence. Our results indicate that time costs accounted for up to 25% of cervical cancer screening costs. Time costs should be identified, measured, valued, and included in cost-effectiveness analyses of cervical cancer screening.