A PRIVACY-PRESERVING IOMT DIGITAL TWIN: INTEGRATING WEARABLE MULTIMODAL SENSING AND EDGE-DRL FOR PRECISION GERIATRIC CARDIOLOGY

Abstract

Physiological frailty and the ubiquitous clinical burden of polypharmacy significantly exacerbate cardiovascular disease in the geriatric population. Current guidelines for prescribing and static predictive models do not take into account the non-linear pharmacokinetics of the elderly that leads to adverse drug events, renal toxicity, and acute hemodynamic decompensation. To address the limitations of reactive clinical care, this paper presents the Cardio-Geriatric Digital Twin, an end-to-end privacy-preserving computational framework for continuous cardiovascular trajectory simulation and autonomous polypharmacy optimization. The proposed architecture fuses high-frequency IoT wearable telemetry with unstructured Electronic Health Records (EHRs) via a novel Cross-Modal Transformer fusion core, yielding a highly dynamic, context-aware patient replica. The medication titration process in this simulation environment is formulated as a Partially Observable Markov Decision Process (POMDP) and solved by a clinically constrained Proximal Policy Optimization (PPO) agent. To ensure strict data privacy and regulatory compliance, the agent is trained in a decentralized Federated Learning (FedPPO) protocol over distributed edge nodes. Crucially, the reinforcement learning policy is guided by a strict multi-objective reward function that independently penalizes renal degradation and hyperpolypharmacy, while ensuring the stability of vital hemodynamics. Extensive empirical evaluation on 65,420 simulated longitudinal profiles shows the decisive superiority of the framework over state-of-the-art predictive and heuristic baselines. The proposed model produced an unprecedented Trajectory RMSE of 4.20% and increased the Medication Adherence F1-Score to 0.89, generating disproportionate compliance gains within the highly vulnerable “Frail” patient stratum. And the framework achieved a robust 65.0% Hospitalization Aversion Rate in a 90-day simulation. We develop a highly scalable, empirically validated and interpretable at the level of features, computational framework for proactive, precision-driven geriatric cardiology using Shapley Additive Explanations (SHAP).

Keywords: Wearable MEMS Sensors, Digital Twin, Edge AI, Multimodal Sensing, Federated Learning, Geriatric Cardiovascular Care, and Deep Reinforcement Learning.