Entropy removal of clinical features

Abstract

Abstract Interpreting clinical findings is fundamental to diagnosis and care. However, the contribution of individual features to reducing diagnostic uncertainty remains unclear. Information theory’s Shannon entropy offers a way to quantify how much a finding narrows diagnostic possibilities. We analyzed 405 symptoms, physical signs, demographic factors, and tests drawn from 23 reviews to calculate entropy reduction from diagnostic tables and compared them to established accuracy measures, including Youden’s index and predictive values. Most features yielded modest uncertainty reductions, with nearly half removing less than one-fifth of uncertainty, while a subset of high-performance findings reduced uncertainty by more than 40%. Entropy reduction correlated strongly with Youden’s index and positive predictive value. Entropy analysis may enhance evaluation by highlighting features that offer greater informational benefit.