When Clinical Prediction Models Meet Clinical Judgment: A Framework for Classifying Prediction Approaches.

Abstract

Background and objective: There is an ever increasing interest in using clinical prediction models to improve diagnostic and prognostic accuracy. However, the literature often conflates distinct prediction approaches, resulting in ambiguous findings and slowing scientific progress. This commentary aims to clarify the various ways in which physicians use clinical prediction models and highlight the implications of these approaches for prediction accuracy by drawing on an existing framework from the judgment and decision-making literature.

Methods: We promote a framework distinguishing between four distinct prediction approaches: clinical judgment, mechanical judgment, clinical synthesis, and mechanical synthesis. We apply this framework to recent studies and illustrate how it distinguishes between previously conflated approaches.

Results: Our analysis demonstrates that current research often obscures meaningful differences between the prediction approaches, thereby impeding improvement in patient care. The framework provides conceptual clarity and highlights clinical synthesis, in which the physician incorporates the model output into their own judgment, as a practical yet under-researched approach.

Conclusions: The prediction accuracy achieved in practice depends on how physicians use clinical prediction models. Greater recognition and use of the proposed framework can improve future research by properly classifying prediction approaches, fostering more accurate research syntheses, and supporting the more effective and fair application of clinical prediction models in practice.