'A Look at the Chief Complaints' revisited. Current obstacles and opportunities for decision analysis

Advances in Quantitative Techniques for Making Medical Decisions

Articles relating decision theory to medical problems have appeared in the medical literature over the past thirtyyears. Mostof the work important to recent advances has been performed in the past 10 to 15 years. We describe the progress which occurred between 1978 and 1988: an increased acceptance of quantitative approaches, advances in analytical techniques, simplification of older methods to enhance accessibility, and a better understanding of the interface between methods that prescribe how decisions should be made and those that describe how decisions actually are made. We also discuss problems that have not been overcome and those which reflect the "growing pains" of success, including increased scrutiny by non-methodologist medical content experts.

Prophylactic replacement of Björk-Shiley convexo-concave valves at risk of strut fracture. Björk-Shiley Study Group

OBJECTIVE

Prophylactic replacement of Björk-Shiley convexo-concave valves (Shiley, Inc., Irvine, Calif.) has been advised for selected groups of patients. If prophylactic replacement is considered, risks of postoperative morbidity and mortality have to be weighed against benefits of replacement. Here we report the results of prophylactic replacement of Björk-Shiley convexo-concave valves at risk of strut fracture in The Netherlands.

METHODS

We reviewed medical records of 36 patients undergoing prophylactic replacement of their Björk-Shiley convexo-concave valves before August 1995. Replacement was judged to be prophylactic if the risk of strut fracture outweighed that of death from reoperation, or the patient wished to have the valve replaced although it was not recommended. The procedure was also considered to be prophylactic if a concomitant pathologic condition, not likely to require cardiac surgery in the near future, was present or if preoperative examination revealed an unexpected cardiac pathologic condition.

RESULTS

Twenty-two 70-degree and 16 60-degree Björk-Shiley convexo-concave valves and one spherical valve were replaced (25 aortic and 14 mitral, including three double-valve replacements). Early mortality was 2.8% (1/36) (exact 95% confidence interval [CI] 0.1 to 14.5). Mean follow-up was 33 months. One- and 3-year survivals were 94% (95% CI 79% to 99%) and 91% (95% CI 74% to 97%), respectively. All three deaths were sudden.

CONCLUSIONS

If special care is taken in selecting patients, the risk of prophylactic replacement is comparable to that of primary valve replacement. More data are needed to assess whether the risk of sudden death is possibly increased.

The discrepancy between daily practice and the policy of a decision-analytic model: the management of fever of unknown origin

The optimal treatment of children with fever of unknown origin is controversial, in spite of two decision analyses that advise treatment with antibiotics for all such children. The aim of this study was to analyze the differences between pediatricians' diagnostic and therapeutic decisions and the outcomes of the decision-analytic models. Thirty-six pediatricians were asked to evaluate 30 patient cases and to give their diagnostic and therapeutic judgments. In addition, the pediatricians were asked questions about the epidemiology of fever of unknown origin. Analyses showed that the differences in policy between pediatricians and the models could not be explained by the reasons mentioned in the literature, i.e., 1) differences in epidemiologic data used, 2) differences in the weighting of clinical information, and 3) differences in the evaluation of outcomes. The differences in policy might be due to a difference between the objective of the models and pediatricians' aim. In a curative setting, pediatricians are not trying to prevent meningitis (or another serious disease) by treating possible occult bacteremia, but rather aim to detect meningitis in an early stage. A decision analysis determining the most cost-effective strategy for early detection of meningitis might therefore be more easily accepted by pediatricians.

Decision analysis in the clinical neurosciences: a systematic review of the literature

Clinical decision analysis can be a useful scientific tool for individual patient management, for planning of clinical research and for reaching consensus about clinical problems. We systematically reviewed the decision analytic studies in the clinical neurosciences that were published between 1975 and July 1994. All studies were assessed on aspects of clinical applicability: presence of case and context description, completeness of the analysed strategies from a clinical point of view, extendibility of the analyses to different patient profiles, and up-to-date-ness. Fifty-nine decision analyses of twenty-eight different clinical problems were identified. Twenty-eight analyses were based on the theory of subjective expected utility, twelve on cost-effectiveness analysis. Four studies used ROC analysis, and fifteen were risk-, or risk-benefit analyses. At least six studies could have been improved by more elaborately disclosing the context of the clinical problem that was addressed. In eleven studies, the effect of different, yet plausible assumptions was not explored, and in eighteen studies the reader was not informed how to extend the results of the analysis to patients with (slightly) different clinical characterisitics. All studies had, by nature, the potential to promote insight into the clinical problem and focus the discussion on clinically important aspects, and gave clinically useful advice. We conclude that clinical decision analysis, as an explicit, quantitative approach to uncertainty in decision making in the clinical neurosciences will fulfill a growing need in the near future.

Truce on the battlefield: a proposal for a different approach to medical informed consent

What is informed consent in medicine? For more than a generation, this deceptively simple question has vexed the law, discomfited medicine, and generated much inspired, provocative, and even contentious commentary.

The question has also spawned several lawsuits. On one side stand patients who claim that, at the time of consent, they were ignorant of a particular risk; who state that, with more or different information, they would have chosen a different treatment; and who argue that, because of an adverse outcome, they now deserve remuneration. On the other side, doctors uneasily watch the lengthening list of suits. Some, troubled by the law's expectations, have reacted by variously describing informed consent as a myth, a fiction, an unattainable goal, or a snare to entrap physicians. They point to the legal commentary condemning informed consent law as ill-defined, diffuse, and fraught with inconsistency, hazy at its best and virtually indecipherable to physicians at its worst: and lacking a fair standard to determine when a patient has sufficient knowledge to give effective consent.

Decision analysis--a helpful tool for clinicians to establish diagnostic-therapeutic guidelines?

In this paper we focus on the question: Does decision analysis provide a framework to assess the value of diagnostic tests in clinical practice and how can it be used by clinicians in establishing diagnostic-therapeutic guidelines. To study this question we performed two analyses concerning the use of pelvic lymphadenectomy and pedal lymphography for staging prostate cancer. Both analyses yielded similar results as far as the preferred strategy was concerned, yet the approach and set up of the two analyses were different. The first analysis was performed in accordance with the textbooks on decision analysis. However, using this traditional approach we encountered some difficulties: in structuring the decision tree, in eliciting values for the quality of life parameters, and in interpreting the results. These difficulties urged us to modify the approach, presented in the second analysis. In this second analysis, the decision problem was split into several consecutive decision problems which corresponded to the questions posed by the clinicians. Longevity and quality of life were considered separately and the consequences of treatment and testing, which affect the quality of life of the patients, were indicated by just two parameters. Finally, the result of the analysis was expressed in clinically meaningful terms. The second analysis is compared with different approaches presented in the literature for analyzing decision problems involving diagnostic tests. Despite some unresolved methodological problems it is concluded that decision analysis provides a good framework for clinicians to structure and analyze complex decision problems.

Use of formal methods in medical decision making: a survey and analysis

Apparent low usage of formal decision techniques by general clinicians has raised questions about dissemination methods and about the techniques' perceived usefulness. Two literature searches examined whether use of formal decision techniques among clinicians had indeed failed to increase from the 1970s to the 1980s. A general MEDLINE search for the period 1983-87 relative to 1973-77 indicated that usage of formal decision techniques had more than doubled. This increase, however, was due to increased coverage of formal decision techniques in specialist methods journals. A manual search of seven major clinical journals and a MEDLINE search restricted to the clinical journals of the manual search disclosed no increase in overall usage for the same time periods. MEDLINE detected only a small subset of the actual instances of formal method usage found by the manual search. Individual medical subspecialties were found to utilize different formal decision techniques to different degrees. The authors suggest interventions that may increase the usage of formal decision techniques among general clinicians.

Decision analysis in oncology: panacea or chimera?

A review of the literature on the use of decision analysis in clinical oncology shows that, although these techniques have been available for more than 25 years, they have not been widely applied: only 19 decision analyses of therapeutic management in clinical oncology were found. The main disadvantages concern the difficulty of accurately assessing probabilities and defining measures of outcome. Time-consuming analysis may produce results that are either equivocal or simply confirm the expectations of common sense. If the basic design fails to include all relevant factors then any conclusions will be of little value. The main advantages are that, by demanding that problems be explicitly stated and analysed in a logical fashion, deficiencies in current knowledge, belief and practice are identified. The usefulness of these techniques lies in formulating management guidelines, either for treatment or for follow-up. They have only a limited role in decision making for individual patients.

Utility and cost of carcinoembryonic antigen monitoring in colon cancer follow-up evaluation. A Markov analysis

The use of carcinoembryonic antigen (CEA) to monitor patients after intentionally curative colon cancer resection can have advantages (improved life expectancy as a result of early detection of recurrences) as well as disadvantages (false-positive CEA rises and early detection of incurable recurrences in asymptomatic patients). This study estimated how the favorable and unfavorable effects of CEA monitoring affect life expectancy and quality of life of colon cancer patients. These effects were simulated using a Markov analysis for which the variables had been defined on the basis of data found in literature. The influence of CEA monitoring on quality adjusted life expectancy appears to be modest and varies, according to the data used, from an average increase of +7 days (+0.3%) to an average decrease of -5 days (-0.09%). This value is dependent, among other things, on patient related variables; the adverse effects especially dominate in older patients with a favorable Dukes' stage of the primary tumor and if operative mortality exceeds 2%. The total cost of CEA monitoring, including diagnosis and therapy performed as a result of true- or false-positive CEA rise, is considerable. High cost and low return leads to a high marginal cost-effectiveness ratio, which varies from $22,963 to $4,888,208 per quality adjusted life year saved. It is concluded that CEA monitoring should not be used for routine following of colon cancer patients, as its advantages have so far been demonstrated insufficiently.

Guiding individual decisions: a randomized, controlled trial of decision analysis

In early 1983, all 1,280 faculty and resident physicians at one hospital who were eligible to be vaccinated against hepatitis B were divided randomly into three groups: Group 1 physicians received general information about the risks and benefits of alternative vaccine decisions; Group 2 physicians were additionally invited to provide personal information for an individualized decision analysis (12.6 percent responded); and Group 3 physicians, who served as controls, were not contacted. In one year's follow-up, 20 percent of physicians were screened for hepatitis B antibody or vaccinated. More Group 2 physicians whose decision analyses recommended screening or vaccination took these actions (39 percent) than any other group. Group assignment remained significantly associated with vaccine decisions after analyzing results by the "intention to treat" principle, and after adjusting for training status, exposure to blood and blood products, and pre-study intentions about the vaccine. Despite the low overall vaccine acceptance rate, it is concluded that individualized decision analysis can influence the clinical decisions taken by knowledgeable and interested patients.

Clinical guidelines for using two dichotomous tests

Building on the threshold model developed by Pauker and Kassirer for a single test, the authors describe a decision analytic model for two tests with dichotomous outcomes. The model includes ten decision strategies that differ depending on which tests are performed, whether the tests are performed together or in sequence, and the definition of a positivity criterion used to make the treatment decision when the test results disagree. Formulas derived from the model are used to compute the preferred option as a function of disease probability and to calculate test and test-treatment thresholds. General guidelines developed from the model can be used without calculation to identify relative preferences for alternative options and to predict threshold effects.