AdaptHetero: Machine Learning Interpretation-Driven Subgroup Adaptation for EHR-Based Clinical Prediction

However, the in t rinsic complexity and heterogeneity of EHR data limit its effectiveness in guiding subgroup - specific modelin g . W e propose AdaptHetero, a novel MLI - driven framework that transforms interpretability insights into actionable guidance for tailor ing model training and evaluation across subpopulations within individual hospital systems . E valuated on th ree large - scale EH R datasets -- GOSSIS - 1 - eICU, WiDS, and MIMIC - IV -- AdaptHetero consistently identif ies heterogeneous model behaviors in predicting ICU mortality, in - hospital death, and hidden hypoxemia. By integrating SHAP - based interpretation and unsupervised clustering, the framework enhances the identification of clinicall y meaningful subgroup - specific characteristics, leading to improved predictive performance and optimized clinical deployment . Introduction Machine learning interpretation (MLI) techniques are increasingly leveraged in the analysis of electronic health records (EHRs) to reveal latent clinical patterns and to support trustworthy, actionable decision - making in high - stakes healthcare settings .