A Machine-Learned Comorbidity Index

2026-06-15

AI Summary

Traditional comorbidity indices, such as Charlson and Elixhauser, are mortality-centric and rely on linear assumptions, limiting their generalizability across diverse clinical outcomes and their ability to capture nonlinear risk associations. This work proposes the first machine learning–based comorbidity index optimized for multiple outcomes, which maps diagnosis codes end-to-end into a single scalar by maximizing the normalized Hilbert–Schmidt Independence Criterion (nHSIC) to model nonlinear dependencies with various clinical endpoints. Evaluated on multiple electronic health record (EHR) benchmark datasets, the proposed method significantly outperforms strong existing baselines, achieving both theoretically grounded and information-rich risk stratification for hospitalization within a unified framework.

Abstract

Traditional comorbidity scores (e.g., Charlson and Elixhauser) are widely used for risk adjustment and patient stratification, but they have two key limitations: (i) they are largely mortality-centric and do not align well with other clinical outcomes, and (ii) their linear, rule-based structure cannot capture nonlinear, outcome-specific risk relationships. We propose a Machine-Learned Comorbidity Index (MLCI) that maps diagnosis codes to a single scalar by maximizing the normalized Hilbert-Schmidt Independence Criterion (nHSIC) between the learned score and multiple clinical outcomes. MLCI captures nonlinear risk-outcome dependence and is supported by a theory that characterizes when a unified, informative admission-level ordering can be achieved across outcomes. Empirical results on multiple benchmark electronic health record (EHR) datasets show that MLCI outperforms strong baselines across multiple evaluation metrics.

Problem

• comorbidity index
• clinical outcomes
• nonlinear risk relationships
• risk adjustment
• patient stratification

Innovation

• Machine-Learned Comorbidity Index
• nonlinear risk modeling
• nHSIC
• clinical outcome prediction
• EHR