The paper by Fiscella and colleagues1 in this issue of the Journal is an important addition to the literature on the variations in medical care provided by family physicians and general internists. This careful study documents, as have many others, that the care provided by family physicians is less costly than that provided by general internists. Its new contribution is the finding that interspecialty differences in risk aversion are predictive of differences in resource utilization. Previous studies have shown that physicians who are more risk averse are more likely to order diagnostic tests, refer to specialists, hospitalize, and generate higher overall costs. The study by Fiscella and colleagues suggests that some of the difference in medical care resource utilization between family physicians and internists may be explained by psychological differences in risk aversion. What are some of the other factors that may contribute to the difference in resource utilization between physicians in these 2 primary care specialties? Why are general internists more risk averse than their family physician counterparts?
The threshold approach to clinical decision making first described by Pauker and Kassirer2 is a useful conceptual framework for thinking about variations in physician decision making and medical resource utilization. For a given disease (D), for which there is a given treatment and a given diagnostic test, this model prescribes a no treatment-test threshold (P1) and a test-treatment threshold (P2). If the patient’s probability of disease, P(D), is less than P1, then the model prescribes that the patient should not be tested or treated; if P(D) is greater than P2, then the patient should not be tested but simply treated. If P(D) is greater than P1 but less than P2, then the diagnostic test should be performed. The optimal P1 and P2 thresholds can be determined using decision analysis, taking into account the costs, risks, and benefits of these 3 strategies. A large number of variables must be included in the decision tree, such as the efficacy of treatment and the placebo effect; the probability of treatment side effects; the sensitivity and specificity of the test; the outcomes of treated disease, untreated disease, and treatment side effects; and the costs of tests, treatment, and adverse outcomes of disease and treatment. Although these kinds of analyses have been published for some combinations of diseases, treatments, and tests,3 physicians often make these decisions using subjective threshold estimates and subjective estimates of disease probability based on personal experience.
Variability in physicians’ subjective estimates of the P1 and P2 thresholds,4-5 along with variability in subjective estimates of disease probability,6 may explain a significant amount of the variation in physician decision making and resource utilization.7 Physician discomfort with uncertainty and risk aversion may well have a significant unconscious effect on subjective P1 and P2 threshold estimation. The anticipated effect would be to lower the probability of disease at which it is acceptable not to treat or test (P1) and to increase the probability of disease at which it is acceptable to treat without testing (P2). A risk averse physician, therefore, would be more likely to test before deciding to do nothing or to treat, increasing the use of diagnostic and consultant resources.
Physician training may also have an impact on threshold estimation. Internists spend much more of their training time in the inpatient setting than family physicians. They are schooled in the art of differential diagnosis and the importance of ruling out possible, though unlikely, diagnoses. Family physicians are typically taught that common things are common. This could mean that internists subjectively set the P1 threshold lower and the P2 threshold higher, so they are more likely to test for unlikely diagnoses than family physicians.
The threshold approach prescribes a patient management strategy based on the physician’s subjective estimate of the patient’s probability of disease. An underestimation of the probability of a likely disease would result in unnecessary testing, as would an overestimation of the probability of an unlikely disease. A general internist trained mostly in a tertiary care setting would be more likely to underestimate the probability of common diseases and overestimate the probability of uncommon diseases than a family physician trained mostly in a primary care setting.
Questions For Future Research
There are many unanswered questions regarding the difference in medical care resource utilization between general internists and family physicians. Are there systematic differences between no treatment-test and test-treatment threshold estimates for various diseases between family physicians and general internists? Are there systematic differences in subjective estimates of disease probability? How much of the difference in resource utilization relates to the personality traits of persons who choose each specialty? How much is because of differences in learned approaches to differential diagnosis and clinical problem solving, and how much is due to experiences in caring for different populations of patients? Is risk aversion a personality trait that a physician brings to training, or is it a behavior that is learned during training? If it is a personality trait, can it be modified? If it is a learned behavior, can it be unlearned? Do physicians trained in community-based residency programs use fewer medical care resources than physicians trained in university-based programs? Do general internists trained in primary care tracks use fewer resources than those trained in traditional tracks? These questions provide significant opportunities for future research on this subject.