Digital Health Integration in Drug-Device Systems

Abstract

Drug-device combination systems increasingly rely on digital health integration to achieve their therapeutic potential -- a continuous glucose monitor paired with an insulin pump without a connected app and decision support algorithm is a less effective system than its closed-loop counterpart. The digital layer is not ancillary; it is often the component that determines whether a drug-device combination delivers its designed therapeutic benefit to real patients in real conditions. Despite this centrality, the integration of digital health technologies into drug-device systems has been largely evaluated on a system-by-system basis, without a comparative framework that examines which digital integration approach best serves which drug-device system context. This study presents the Digital Health Drug-Device Integration Framework (DHDIF), evaluating five digital integration approaches -- mobile health application workflow coordination (mHealth-WC), AI-assisted therapy optimisation dashboard (AI-TOD), real-time remote patient monitoring with device-drug adjustment (RPM-DDA), digital twin-based personalised optimisation (DT-PO), and federated learning for population-level drug-device optimisation (FL-PDO) -- across four drug-device system contexts: insulin pump and CGM diabetes systems, smart inhaler and spirometry systems for respiratory disease, infusion pump remote monitoring for oncology, and cardiac device with medication management systems. Performance was scored using the Digital Integration Effectiveness Score (DIES), a weighted composite of clinical outcome improvement (0.30), user engagement (0.20), system interoperability (0.20), data security compliance (0.15), and scalability (0.15). Real-time remote patient monitoring achieved the highest DIES (0.884) through the best clinical outcome improvement (0.920) and interoperability (0.900). Federated learning for population optimisation ranked second (0.883) with the best interoperability (0.920) and strong scalability (0.880). AI-assisted dashboards (0.876) provided the strongest clinical outcome gains in the oncology infusion context. Digital twin personalisation (0.868) achieved the best clinical accuracy (0.940) but was limited by user engagement (0.820) and scalability (0.820).