Probability estimates are often used by clinicians during a diagnosis as they narrow down their initial range of possible diagnoses based on test results, pre-test probabilities, and the sensitivity and specificity of the tests. However, when it comes to explicitly calculating probabilities, it has been found that physicians’ estimates can be inaccurate, generally too conservative or prone to errors. So there could be potential to improve quality of diagnosis, and thus medical care, through teaching medical students how to make more accurate estimates.

Brush and colleagues in Canada recently conducted a randomised controlled trial [1] looking at two different methods of teaching – receiving explicit instruction regarding testing and Bayesian revision (concept group), or being exposed to repeated examples of cases with feedback (experience group) – to see whether there was any improvement in students’ abilities to update diagnostic probabilities when compared to controls.

The authors randomised 61 medical students to either the control group or one of the two teaching methods and compared their subjective post-test probability to a Bayesian calculation of post-test probability. The results showed that the students in the concept group demonstrated a significant improvement in probability estimates compared to the other two groups (p<0.001) – there was a mean (SE) discrepancy of 0.4% (0.7%) between the subjective score and the Bayesian score, compared to 3.5% (0.7%) in the experience group and 4.3% (0.7%) in the control group. It seems, therefore, that there is therefore a benefit in teaching Bayesian reasoning to medical students in order to improve their diagnostic accuracy.

We thank Gus Hamilton for drawing our attention to this paper.

Richard Lilford, ARC WM Director

Reference:

1. Brush Jr JE, Lee M, Sherbino K, et al. Effect of Teaching Bayesian Methods Using Learning by Concept vs Learning by Example on Medical Students’ Ability to Estimate Probability of a Diagnosis. A Randomized Controlled Trial. JAMA Netw Open. 2019; 2(12): e1918023.