Explainable Admission-Level Predictive Modeling for Prolonged Hospital Stay in Elderly Populations: Challenges in Low- and Middle-Income Countries
This study addresses the challenge of predicting prolonged hospital stays (>7 days) among older adults in resource-constrained settings of low- and middle-income countries, with the goal of reducing adverse in-hospital events. Leveraging patient and hospital administrative data available at admission, the authors propose a novel feature selection method that integrates information value analysis with graph-theoretic clique structures to identify nine non-redundant, highly interpretable variables. An interpretable logistic regression model built on these features achieves an AUC-ROC of 0.82, accuracy of 0.76, specificity of 0.83, and sensitivity of 0.64 on the validation set. The approach maintains strong predictive performance while significantly enhancing clinical transparency and practical utility for deployment in low-resource healthcare environments.