Medical Decision Making

The Cochrane Collaboration is an international organization that aims to help people make well-informed decisions about health care by preparing, maintaining, and ensur ing the accessibility of rigorous, systematic, and up-to-date reviews (and, where pos sible, meta-analyses) of the benefits and risks of health care interventions. This article describes the structure of the Cochrane Collaboration and the processes used to carry out the task of research transfer. The authors focus on recent achievements of the Collaboration, emphasize the current challenges and barriers the Collaboration is fac ing to live up to the expectations it has generated, and highlight the strategies that the organization is using to meet those challenges. Key words: research transfer; decision making; evidence-based medicine; systematic reviews; meta-analysis; education; in formatics ; Internet; communications. (Med Decis Making 1998;18:2-9)

Facilitating the successful and consistent use of research results to support health care decisions is a formidable task. Barriers to effectively transferring the results of research into the decision-making process have been created between practitioners and researchers, who traditionally have worked in isolation from each other. The need for them to work cooperatively to break down these barriers is paramount as changes within the health care environment increase. The Ontario Health Care Evaluation Net work (OHCEN) and the Critical Care Research Network (CCR-Net) have attempted to address these concerns by bringing together teams of researchers, practitioners, and administrative personnel with the purpose of equipping them with tools to meet on coming health care challenges. Key words: research transfer; decision making; evi dence-based medicine; systematic reviews; meta-analysis; education; informatics; In ternet; communications. (Med Decis Making 1998;18:9-16)

Physicians sometimes order diagnostic tests to reduce the risk of malpractice liability. The authors develop an expected-utility model that links a rational physician's concerns about malpractice liability to increases in the use of diagnostic tests and use this model to assess the effects of defensive testing on patients' interests. To do so, they adapt the threshold approach to clinical decision making to incorporate the physician's inter ests, focusing on 1) the effect of the physician's expected liability risks and 2) the effect of any expected liability reduction due to diagnostic testing. Surprisingly, the mere existence of liability risks is often sufficient to widen the range of disease probabilities for which diagnostic testing is the preferred clinical strategy. If testing reduces the physician's expected liability risks, the testing range is widened further. For some dis ease probabilities, testing is preferred by the physician even though it is not in the patient's best interests. When tests are performed in such instances, utility is trans ferred from the patient to the physician and the physician's insurer. Although the de fensive use of diagnostic tests improves clinical outcomes for some patients, it worsens clinical outcomes for others. Moreover, defensive testing worsens the expected out comes of all patients whose clinical strategies are changed. Physicians should realize that defensive testing necessarily reduces the overall quality of patient care. Key words: decision analysis; decision threshold; defensive medicine; diagnostic testing; expected utility; malpractice liability. (Med Decis Making 1998;18:19-28)

Studies comparing more than two competing therapies are common in several fields, but standard meta-analytic methods can make only pairwise comparisons. The meth ods proposed in this article, a generalization of current meta-analytic methods, allow for any number of competing therapies and include both fixed- and random-effects models. Key words: meta-analysis; random-effects models; multiple treatment studies. (Med Decis Making 1998;18:37-43)

Objectives. To define clinical outcomes and prevailing patterns of care for the initial hospitalization of infants at greatest risk for respiratory distress syndrome (RDS); to estimate direct medical care costs associated with the initial hospitalization; and to introduce and demonstrate a simulation technique for the economic evaluation of health care technologies. Method. Clinical outcomes and usual-care algorithms were determined for infants with RDS in three birthweight categories (500-1,000 g; >1,000- 1,500 g; and >1,500 g) using literature- and expert-panel-based data. The experts were practitioners from major U.S. hospitals who were directly involved in the clinical care of such infants. Using the framework derived from the usual care patterns and outcomes, the authors developed an itemized "micro-costing" economic model to sim ulate the costs associated with the initial hospitalization of a hypothetical RDS patient. The model is computerized and dynamic; unit costs, frequencies, number of days, probabilities and population multipliers are all variable and can be modified on the basis of new information or local conditions. Aggregated unit costs are used to estimate the expected medical costs of treatment per patient. Results. Expected costs of initial hospitalization per uncomplicated surviving infant with RDS were estimated to be $101,867 for 500-1,000 g infants; $64,524 for >1,000-1,500 g infants; and $27,224 for >1,500 g infants. Incremental costs of complications among survivors were esti mated to be $22,155 (500-1,000 g); $11,041 (>1,000-1,500 g); and $2,448 (>1,500 g). Expected costs of initial hospitalization per case (including non-survivors) were $100,603; $72,353; and $28,756, respectively. Conclusions. An itemized model such as the one developed here serves as a benchmark for the economic assessment of treatment costs and utilization. Moreover, it offers a powerful tool for the prospective evaluation of new technologies or procedures designed to reduce the incidence of, severity of, and/or total hospital resource use ascribed to RDS. Key words: respiratory distress syndrome; low birthweight; technology assessment; economics; costs. (Med Decis Making 1998;18:44-51)

The purpose of this study was to develop a model that predicts the outcome of su pervised exercise for intermittent claudication. The authors present an example of the use of autoregressive logistic regression for modeling observed longitudinal data. Data were collected from 329 participants in a six-month exercise program. The levels of the polytomous outcome variable correspond to states they defined in a Markov de cision model comparing treatment strategies for intermittent claudication. Autoregres sive logistic regression can be used to fit multistate transition models to observed longitudinal data with standard statistical software. The technique allows exploration of alternative assumptions about the dependence in the outcome series and provides transition probabilities for different covariate patterns. Of the alternatives examined, a Markov model including two preceding responses, time, age, ankle brachial index, and duration of disease best described the data. Key words: longitudinal data analysis; autoregressive models; logistic regression; Markov models; peripheral arterial occlu sive disease; intermittent claudication; exercise. (Med Decis Making 1998;18:52- 60)

The theory of reasoned action, which postulates that personal attitudes and external social influences predict intentions to undertake a behavior, was used as a conceptual framework for developing a questionnaire to elicit beliefs and attitudes associated with the decision to participate in a hypothetical cancer chemotherapy trial. After completing the questionnaire, two-thirds of the 150 respondents indicated they would enroll in such a trial if it were available. Considerable variation existed in both "universal" and "trial- specific" beliefs and attitudes underpinning their intentions. A substantial amount of the variance in "intention" to participate was explained by "attitude" alone (75%). So cial influences, although statistically significant, made a mere 1% additional contribu tion. One interpretation is that subjective expected-utility theory, which essentially pre dicts beliefs or "attitude," is a better model. The authors conclude that both theories may be criticized regarding how well they capture the rationality and nuances of de cision behavior. Key words: attitudes; beliefs; clinical trials; oncology. (Med Decis Mak ing 1998;18:61-69)

Neural networks were developed to predict perioperative cardiac complications with data from 567 vascular surgery patients. Neural network scores were based on cardiac risk factors and dipyridamole thallium results. These scores were converted into like lihood ratios that predicted cardiac risk. The prognostic accuracy of the neural networks was similar to that of logistic regression models (ROC areas 76.0% vs 75.8%), but their calibration was better. Logistic regression overestimated event rates in a group of high-risk patients (predicted event rate, 64%; observed rate 30%; n = 50, p < 0.001). On a validation set of 514 patients, the neural networks still had ROC similar areas to those of logistic regression (68.3% vs 67.5%), but logistic regression again overesti mated event rates for a group of high-risk patients. The calibration difference was reflected in the Hosmer-Lemeshow chi-square statistic (18.6 for the neural networks, 45.0 for logistic regression). The neural networks successfully estimated perioperative cardiac risk with better calibration than comparable logistic regression models. Key words. neural networks; logistic regression; likelihood ratio; Bayes' theorem; cardiac risk. (Med Decis Making 1998;18:70-75)

Objective. Elicited preferences for health states vary among scaling methods, manners of describing health states, and other features of the elicitation process. The authors examined the effects of changing the search procedure for a subject's utility on mean utility values. Methods. A randomized controlled trial of two search procedures (titration and "ping-pong") using two otherwise identical computer programs that describe health states related to Gaucher's disease, then measuring subjects' preferences. Setting. Paid, healthy volunteers recruited from the community through advertisements. Re sults. The mean time tradeoff (TTO) and standard gamble (SG) utility values for life with severe anemia and splenomegaly and life with chronic bone pain from Gaucher's disease were between 0.10 and 0.15 higher with the titration search procedure than with the ping-pong procedure. Effects of the search procedure were additive with var iability due to scaling methods, resulting in mean differences in utility ratings for the same health state of as much as 0.28 among procedures and scaling methods. Effects of search procedures on utility values persisted on repeated testing at week 2 and week 3; there was no evidence of convergence to a single "true" utility value over time. Conclusions. The procedure used to search for subjects' utility values strongly influences the results of preference-assessment experiments. Effects of search pro cedures persist on repeated testing. The results suggest that utility values are heavily influenced by, if not created during, the process of elicitation. Thus, utility values elic ited using different search procedures may not be directly comparable. Key words: search procedures; utility elicitations; patient preferences; titration; ping-pong. (Med Decis Making 1998;18:76-83)

Anecdotal evidence suggests that patients who have life-threatening conditions often choose to undergo high-cost, high-risk treatments for them. This kind of risk-seeking behavior seems irrational because most patients are risk-averse. The Health Stock Risk Adjustment (HSRA) model seeks to explain this phenomenon. The model is based on the concept of relative health stock—the ratio of patients' expected quality-adjusted life years (QALYs) after a diagnosis to their expected QALYs before the diagnosis. The model predicts risk-averse patients will behave in a risk-seeking manner as their relative health stocks deteriorate. The HSRA model can help physicians better under stand why some seriously ill patients seek high-risk treatments while others elect to forgo treatment. State legislatures and insurers are attempting to appropriately design insurance benefits for patients with life-threatening conditions. The HSRA model can help predict which patients will most likely take advantage of these benefits. Key words: patients' preferences; decision making; expected utility theory; treatment choice. (Med Decis Making 1998;18:84-94)

Measures of decision sensitivity that have been applied to medical decision problems were examined. Traditional threshold proximity methods have recently been supple mented by probabilistic sensitivity analysis, and by entropy-based measures of sen sitivity. The authors propose a fourth measure based upon the expected value of perfect information (EVPI), which they believe superior both methodologically and prag matically. Both the traditional and the newly suggested sensitivity measures focus en tirely on the likelihood of decision change without attention to corresponding changes in payoff, which are often small. Consequently, these measures can dramatically over state problem sensitivity. EVPI, on the other hand, incorporates both the probability of a decision change and the marginal benefit of such a change into a single measure, and therefore provides a superior picture of problem sensitivity. To lend support to this contention, the authors revisit three problems from the literature and compare the results of sensitivity analyses using probabilistic, entropy-based, and EVPI-based mea sures. Key words: sensitivity analysis; expected value of perfect information. (Med Decis Making 1998;18:95-109)

The authors propose a new generalized method for ROC-curve fitting and statistical testing that allows researchers to utilize all of the data collected in an experimental comparison of two diagnostic modalities, even if some patients have not been studied with both modalities. Their new algorithm, ROCKIT, subsumes previous algorithms as special cases. It conducts all analyses available from previous ROC software and provides 95% confidence intervals for all estimates. R_OCKIT was tested on more than half a million computer-simulated datasets of various sizes and configurations repre senting a range of population ROC curves. The algorithm successfully converged for more than 99.8% of all datasets studied. The type I error rates of the new algorithm's statistical test for differences in A_z estimates were excellent for datasets typically en countered in practice, but diverged from alpha for datasets arising from some extreme situations. Key words. receiver operating characteristic (ROC) analysis, maximum-like lihood estimation; partially-paired data; missing data. (Med Decis Making 1998;18: 110-121)