Entropy reduction and F-score: Applications of information theory in diagnostic test evaluation

This paper proposes an approach based on information theory and machine learning, applying the F-score and entropy reduction to analyze the flow of information among the patient, the diagnostic test used to evaluate the disease, and the evaluators who interpret the results, thereby enriching the critical assessment of diagnostic tests. The F-score, by integrating positive predictive value and sensitivity, offers a synthetic measure that is less dependent on prevalence. Entropy reduction is a comprehensive metric that quantifies the decrease in uncertainty and informational gain of a diagnostic test, allowing accurate comparisons between different tests using a single measure of accuracy. For practical application, we designed a specific calculator that integrates these indicators. Its implementation would facilitate the interpretation of diagnostic studies and improve clinical decision-making.