PREDICARE — Territorial predictive medicine program

Origin and context

PREDICARE originated from the Sentinelle IA Program — an initial exploration of the conditions for deploying predictive AI in primary care for metabolic syndrome management. This upstream program identified the structural constraints of deployment: not technical limitations, but obstacles of architecture, governance and funding.

PREDICARE structures the scale-up of this initial exploration into an operational territorial program. It is no longer about demonstrating that a predictive model can work — it is about demonstrating that a predictive AI system can be deployed, governed and sustained over time in a regulated healthcare environment.

PREDICARE is not an AI research project. It is a predictive medicine industrialization program in a real healthcare territory.

Program scope

PREDICARE is built on three interdependent components:

1. Individual medical digital twin

Each enrolled patient has an individual medical digital twin — a computational representation of their health status, built from multimodal data (biological, clinical, behavioral, environmental) and continuously updated. This component is provided by TweenMe (Qualees).

2. Territorial predictive infrastructure

Individual digital twins aggregate into a territorial predictive infrastructure — enabling early detection of medical abandonment trajectories and risks of progression toward costly chronic pathologies. The unit of analysis is the territory, not the isolated patient.

3. Governance and funding framework

PREDICARE integrates governance constraints (who decides, who validates, who is responsible) and sustainable funding (an economically viable model within the French healthcare system) from inception. This third component is as structurally important as the first two.

RAISE Framework application

PREDICARE is designed according to the RAISE Framework — the five pillars are not a compliance constraint added at the end. They structure architectural decisions from the outset.

Pilot phase and validation

The PREDICARE pilot phase covers a cohort of ~100 patients followed in primary care within the partner GHT perimeter. It targets three distinct objectives:

Technical validation — digital twin quality, predictive model reliability, FHIR interface performance.

Clinical validation — predictive alert relevance, true positive rate, practitioner acceptability.

Organizational validation — integration into existing workflows, cognitive load on physicians, decision circuit functionality.

The pilot phase is designed to be publishable — results will be submitted to a peer-reviewed journal.

Related publications

The PREDICARE corpus structures four distinct audience angles — cabinet, public decision-maker, clinician, financier — around a single thesis: medical drift is not the failure of insufficiently attentive actors, it is the expected effect of an intermittent vigilance regime.

The corpus derives from the PREDICARE v3 memoir (≈ 46,000 words, 128 references).