AI-Assisted Polypharmacy Risk Assessment Model for Safer Geriatric Medication Management

Abstract

Polypharmacy (use of five or more medications) affects more than 40% of older adults in the community and up to 90% of residents in residential aged care, and is a major source of global preventable hospitalisations, adverse drug reactions (ADRs), and drug-drug interactions (DDIs). We introduce the Polypharmacy Risk Assessment Model (PRAM), a new AI-aided clinical decision support system that combines Random Forest (RF) classifiers, XGBoost (XGB) models for risk scoring and Long Short-Term Memory (LSTM) temporal networks within a Bayesian ensemble fusion layer. PRAM analyses multi-source patient information - electronic health records (EHRs), medication history, renal and hepatic function, frailty index, and multimorbidity - to simultaneously identify high-risk medication combinations, assess ADR risk, and provide patient-specific recommendations for dose optimisation. PRAM has been validated on 4,284 geriatric encounters from three tertiary-care facilities with an AUROC of 0.944 and F1-score of 0.921 for DDI risk classification, exceeding that of all five benchmark methods. A six-month prospective trial in geriatric wards further shows an average 36.4% reduction in ADRs, and integrates with the HL7-FHIR EHR standard, creating a deployable framework for algorithmically supported safe geriatric medication use.