Medical Decision Making

In an attempt to replicate the findings of a classic study of medical decision making, the authors studied decision making in modern pediatrics practice. They prepared case scenarios and surveyed pediatricians for three common clinical decisions: tympanostomy tube place ment, radiography orders, and emergency room referrals. Initial reviewers rated the cases according to the likelihoods that they would take the clinical actions mentioned. Subsequently, other physicians presented with a subset of scenarios in which the initial reviewers were least likely to act tended to be more active in the tympanostomy (p = 0.004) and radiography (p = 0.076) decisions. In these cases the physicians appeared to have a bias toward action. For a subset of scenarios in which the initial reviewers were most likely to act, subsequent reviewers were neither more nor less likely to act than the initial reviewers. (Med Decis Making 1991 ;11 :154-158)

A prospective cohort study was done to assess the effects of value bias and the inappropriate use of the availability heuristic on physicians' judgments of the probability of bacteremia. Subjects of the study were 227 medical inpatients in a university hospital who had blood cultures done. Estimates of the probabilities that individual patients would have positive blood cultures were collected from the house officers who ordered the cultures. Clinical data and culture results were also obtained. Based on the data the authors calculated "value varia bles," reflecting doctors' assessments of the risks that individual patients would die in the hospital if they were to have bacteremia. "Recalled experience variables" reflected the doctors' recollections of recent experiences with patients with bacteremia. The physicians significantly overestimated the likelihood of bacteremia for most of their patients. Their ROC curve for this diagnosis showed moderate discriminating ability (area = 0.687, SE = 0.073). Two recalled experience variables were significantly associated with the physicians' prob ability estimates. The value variables were significantly inversely associated with them. These relationships were independent of several clinical variables and measures of disease severity. The physicians' intuitive diagnostic judgments were thus influenced by the availability heu ristic and by wishful thinking, a form of the value bias. The availability heuristic may mislead physicians by causing them to believe that random variations in the prevalence of a non- epidemic disease represent real trends. Wishful thinking may lead physicians to underes timate the likelihood of a disease for patients most at risk for its consequences. Teaching physicians to develop better judgmental strategies may improve the quality of their judgments and hence their patient care. Key words: diagnosis; probability; septicemia; availability heu ristic ; value bias. (Med Decis Making 1991;11:159-168)

Rare cases are a central problem when an expert system is constructed from example cases with machine learning techniques. It is difficult to make a decision support system (DSS) to cover all possible clinical cases. An inductive learning program can be used to construct an expert system for detecting cases that differ from routine cases. The ID3 algorithm and the pessimistic pruning algorithm were tested in this study: a DSS was built directly from the data of patient records. A decision tree was generated, and the cases misclassified by the decision tree as compared with the classifications of a clinician were listed on a checklist, which formed the feedback to the clinician. In clinical situations about 5-10% of functional thyroid disorders may be misclassified. At this error level, the method found over 90% of the errors with a specificity of 95%. In simple medical classification tasks this dynamic self- learning system can be used to create a DSS that can assist in the quality control of clinical decision making. Key words: expert systems; automated data processing; computer-assisted diagnosis; medical decision making; thyroid disorders. (Med Decis Making 1991;11:171- 175)

Physician decisions concerning allocation of health care resources to patients are highly variable and poorly understood. Psychological androgyny theory (PAT) has been employed as a model of the interpersonal and task activities required of physicians for care of their patients. Several studies have successfully predicted physician resource utilization using measures derived from PAT. Using a sample of 97 first-year medical students, the authors explored the relationship between PAT and risk preference in loss-framed gambles in order to elucidate the process whereby variables derived from PAT predict resource utilization. As hypothesized, students selecting the certain loss had significantly higher mean androgyny scores than did students selecting uncertainty. Research involving these constructs is in tegrated in the context of a theoretical "causal model," which highlights issues deserving of future research. Key words: androgyny; medical resource utilization; medical students; risk preference. (Med Decis Making 1991;11:176-179)

The study purpose was to determine whether differences in the weights assigned to various dimensions of health by 90 women in three subgroups (benign breast disease, breast cancer receiving chemotherapy, and breast cancer receiving other therapies) were associated with differences in self-reported health status in these dimensions. Two methods, one direct and the other indirect, were used to elicit values for mobility, depression, and social support. Two different scales also provided self-reports of health status in each of these dimensions. These measures, in conjunction with sociodemographic variables, were used to test for status- value relationships. No statistically significant association between health values and health status was observed. The absence of any detectable association may have been a result of methodologic difficulties in assessing broadly defined dimensions of health. A possible solution would be to use "individualized" dimensions that are uniquely important to the individual, and to take into account such factors as possible influences of past health status and values, and possible gaps between expected health status and health status actually experienced. Key words: health status measurement; patients' values; utility assessment. (Med Decis Making 1991 ;11 :180-188)

The authors sought to explain regional differences in physicians' accuracies in diagnosing pneumonia by prospectively studying emergency department patients at three sites and analyzing differences in physicians' diagnostic strategies and patient characteristics. They enrolled 1,119 Illinois patients, 150 Nebraska patients, and 142 Virginia patients presenting with fever or respiratory symptoms for whom physicians ordered a chest radiograph because of suspicion of pneumonia. Emergency department physicians recorded patients' clinical findings and estimated the probability that a chest radiograph would show pneumonia. A measure of accuracy, the correlation between physicians' probability estimates and actual outcomes, was 0.41 (95% Cl 0.36-0.46) at Illinois, 0.66 (95% Cl 0.54-0.75) at Nebraska, and 0.55 (95% Cl 0.42-0.65) at Virginia. Physicians' strategies at the three sites differed markedly in their weightings of asthma, signs of consolidation, cough, tachypnea, age, and gender. These differences in weighting paralleled differences in the optimal clinical strategies derived from patient data at the three sites. Differences in diagnostic accuracy were best explained by differences in the difficulties of diagnosing pneumonia in the populations. Phy sicians at each site used clinical findings in a way that was close to optimal for their location. This type of analysis provides a new tool for understanding the sources of regional variations in clinical practice. Key words: judgment analysis; variation research; pneumonia. (Med Decis Making 1991 ;11 :189-197)

In the last four years an active debate has been conducted in this journal about single-parameter ROC curves and methods for correcting curves that have been subjected to verification bias. The major discussants in this debate have produced these three commentaries on a study recently published in MDM. As a researcher I conclude that Diamond's single-parameter ROC, while conceptually attractive and easy to use, is not free from the verification bias that can be found in many diagnostic assessment studies. As Editor I judge that enough space has been devoted to this problem for the time being. Let the scientific community sort through the material already presented on this topic. May it lead to clinical studies wherein single- parameter ROCs can be profitably applied.—JRB

The author describes a new methodology to solve medical decision problems involving a choice between alternatives under conditions of risk and uncertainty when imperfect empirical information from diagnostic technologies is available. The main new concept is the gener alized ROC (GROC) curve, which extends considerably the scope of analysis of the traditional ROC curve as well as the threshold approach to medical decision making. Both techniques become special cases of the new general approach. The author shows how to apply the technique and derive comprehensive clinical guidelines to solve the medical decision problem involving one patient in the most general situation. The larger problem of evaluating the performances of diagnostic technologies for a population of patients with varying prior prob abilities of disease is then addressed. A new measure of performance is proposed that goes beyond the well-accepted area under the ROC curve index. This measure is used to compare different technologies for a population of patients, and a new methodology is proposed to carry out such comparisons. Key words: technology assessment; decision theory; diagnostic information; ROC analysis. (Med Decis Making 1991;11:208-220)

This study investigated the influence of positive affect, induced by report of success on an anagram task, on medical decision making among third-year medical students. The subjects were asked to decide which one of six hypothetical patients, each of whom had a solitary pulmonary nodule, was most likely to have lung cancer. They were asked to verbalize their clinical reasoning as they solved the problem. The positive-affect and control groups did not differ in the tendency to make a correct choice, but subjects in the positive-affect condition were significantly earlier in identifying their choices. These subjects were also significantly more likely to go beyond the assigned task, expressing interest in the cases of the other patients and trying to think about their diagnoses, even though that task was not assigned. The positive-affect subjects also showed evidence of configural or integrative consideration of the material to a reliably greater extent than did control subjects, and there was significantly less evidence of confusion or disorganization in their protocols than in those of controls. These findings are compatible with earlier work suggesting a different organizational process and greater efficiency in decision making among people in whom positive affect had been induced, and with recent work suggesting that positive affect facilitates flexibility and inte gration in problem solving. They also indicate that these effects may apply to the problem- solving strategies of professionals in clinical probem-solving situations. Key words: positive affect; medical decision making; problem solving; diagnostic reasoning. (Med Decis Making 1991;11:221-227)