Resolving Public Expert Model Disagreement in Medical Risk Evaluation

A. Grichnik, M. Taylor, C. Nikolopoulos, J. Mason

Medical Risk Stratification (MRS) models capture the relationships between modifiable and unmodifiable factors to help us understand how various risk factors jointly contribute to the likelihood of contracting a disease in the future. While multiple MRS models exist for diseases like cardiovascular disease (CVD) and diabetes, these models often conflict with one another when applied to real-world populations outside their original study groups. In this paper we examine the conflict between public MRS models of CVD and diabetes and quantify the disagreement using both simulated and real-world populations. A process to resolve these conflicts is presented and the resulting improvement in predictive power is quantified through Bayesian Posterior Probability (BPP) analysis of a population prior to disease onset. By producing improved MRS models, we provide valuable knowledge to those charged with improving the health of populations under their care.

PDF full paper