Enhancing Clinical Prediction Performance by Incorporating Intuition

Excerpt

Widely used clinical prediction scores can assess a patient’s risk to help clinicians implement a variety of preventive strategies. Such scores typically come in the form of equations that generate a predicted risk probability for a given patient at any point in time. Such equations are formed by applying a machine learning algorithm to a longitudinal data source that contains patient characteristics and outcomes, including laboratory values, demographic details, comorbidities, and genetic information. In this brief technical report, I hypothesize a mechanism to collect intuition feedback from clinicians and incorporate the feedback as a novel additional covariate to train machine learning algorithms. This may improve the performance of prediction scores and enhance clinical decision-making support. …