FLARE: Federated Learning for Adherence, Risk, and Engagement in clinical trials

FLARE (Federated Learning for Adherence, Risk, and Engagement) will help healthcare teams support patients and research participants sooner and monitor safety more consistently—across both clinical trials and post-market programs—without pooling or moving sensitive health records between organizations. Today, important signals are spread across apps, nurse workflows, patient reports, and care-team documentation. As a result, adherence issues, drop-out risk, and potential adverse events can be missed or identified late, and patients may struggle to find the right guidance and resources at the right time.

Building on an existing clinical platform, FLARE will deliver an end-to-end solution (front-end experience, back-end services, and AI workflows) that enables three practical capabilities:

1. Adherence and engagement risk prediction to identify people who may be at risk of missed doses, reduced engagement, or stopping therapy or participation, so care teams (nurses, pharmacists, and other HCPs) can provide timely, appropriate support based on agreed policies.

2. Safety (AE/SAE) detection support to help teams find and triage potential adverse events earlier by analyzing local structured data and text (for example, check-ins, diaries, messages, and nurse notes), with human review and clear audit trails.

3. Patient/participant engagement through an intelligent, privacy-first conversational interface designed for regulated environments. The interface will understand a user’s intent and needs, help with navigation and next steps, and recommend approved  resources—such as relevant content and text from proprietary program materials (e.g., trial protocols or post-market support content), while staying within governance and compliance rules.

FLARE uses federated learning, meaning models are trained where data already live (within each organization’s secure environment). Patient records do not move; instead, only privacy-protected model updates are shared. By the end of the project, we aim to demonstrate a working multi-partner prototype that reduces manual workload, improves consistency of safety triage, and strengthens support for patients and care teams from trials through long-term treatment.