Incorporating Bayesian reasoning into physician testing and treatment decisions

University Of Maryland Baltimore, Baltimore MD

Investigators

Linked publications, trials & patents

Abstract

A significant, but under-recognized problem in clinical medicine is the accuracy of test interpretation and decision-making regarding treatment. Fewer than 20% of physicians understand the frequency of false positives in testing. When given an example of testing for a rare disease with a 95% specific test, most physicians answered that 95% of positive tests were true positive (vs. ~2% in reality). Similar results were found in real life examples of mammograms with practicing gynecologists, in which the majority incorrectly reported a positive test was >80% likely to be a true positive (the correct answer was only 7%). Our group and others have found, when asked about the benefits of common treatments for hypertension, myocardial infarction and stroke prevention, physicians overestimate the benefits of treatments, with less than 25% accurately identifying the absolute benefit of treatment (and many estimating an effect 2-10 times larger than reality). These misunderstandings lead to overdiagnosis, overtreatment and overuse, which may account for up to one third of medical care services. The benefits of medical care could be achieved with less patient harm and more satisfaction if test interpretation and decision-making were improved. Diagnostic probabilities underlie nearly all physician testing and treatment decisions. Bayesian reasoning is thought to be the most rational way to assess this. One application of Bayesian reasoning is in diagnosis. Given an initial assessment of the likelihood of a disease and clinical test results, Bayesian reasoning is the mathematically valid method to quantify the probability that the disease is truly present. Psychological techniques for online training in Bayesian reasoning or displaying the results of such reasoning in visual aids has improved performance on physician testing but we know of no interventions that have applied training or visual aids to improve clinician ability to interpret tests or understand treatment effects for actual patients. To reach our overall goal of reducing harms of medical care while maintaining benefits, we propose 3 objectives:1) explore physician factors associated with accurate Bayesian reasoning, 2) develop educational training and visual aids to improve physician understanding of tests and treatment, and 3) pilot these tools in a stepped wedge cluster trial. The PI is well suited for this innovative research based on a non-traditional background in psychology, epidemiology, quality improvement and clinical medicine and has assembled experts in evidence based medicine, visual presentation, qualitative and quantitative research methods. This project has the potential to transform clinical medicine, making it more evidence based with more patient involvement while improving patient outcomes and healthcare costs.