Shuhan He, MD

Kian Samadian, MD

Diagnostic uncertainty remains a major challenge in clinical care. This is in part because it’s difficult for clinicians to measure how much individual clinical features — such as symptoms, exam findings, demographics, and diagnostic tests — narrow down a diagnosis.

In a recent study, Mass General Brigham researchers explored how a mathematical measure called “Shannon entropy” can be used to calculate how much diagnostic doubt is removed by any given finding.

Using this measure, they analyzed 405 features from 23 reviews to discover how much each factor reduced diagnostic uncertainty. They then compared these insights to those gleaned from traditional metrics such as Youden’s index (a combined measure of sensitivity and specificity that reflects overall accuracy) and positive predictive value (the probability that a patient truly has a condition when a test is positive), finding them strongly correlated.

The study, led by Shuhan He, MD, and Kian Samadian, MD, from the Mass General Brigham Department of Emergency Medicine, found that most features removed little ambiguity, with nearly half of the features reducing diagnostic uncertainty by less than 20%. However, certain features like ultrasound‑based tests were far more informative. The authors conclude that entropy‑based analysis may help clinicians — especially those in emergency medicine — identify which features provide the greatest diagnostic value at the point of care.

Paper Cited:
Samadian, K. D., Chua, E., Peng, B., Coleska, A., Hassan, A., Chong, P., Locke, B., Liebovitz, D. M., Rohlfsen, C., & He, S. (2025). Entropy removal of clinical features. Scientific reports, 15(1), 45182. https://doi.org/10.1038/s41598-025-29069-0 |
Read the paper

Summary reviewed by: Shuhan He, MD, senior author