PREDICARE

Origin & context

PREDICARE derives from the Programme Sentinelle IA — an initial exploration of the conditions for deploying predictive AI in primary care for the management of metabolic syndrome. This upstream programme made it possible to identify the structural constraints of deployment: not technical limitations, but architectural, governance and funding obstacles.

PREDICARE structures the scale-up of this initial exploration into an operational territorial programme. The question is no longer whether a predictive model can function — it is whether a predictive AI system can be deployed, governed and sustained over time in a regulated healthcare environment.

PREDICARE is not a research project on AI. It is a programme for the industrialisation of predictive medicine in a real health territory.

Programme scope

PREDICARE rests on an architecture of three interdependent components:

1. The individual medical digital twin

Each patient included in the programme has an individual medical digital twin — a computational representation of their health state, built from multimodal data (biological, clinical, behavioural, environmental) and continuously updated. This layer is provided by TweenMe (Qualees).

2. The territorial predictive infrastructure

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

3. The governance and funding framework

PREDICARE integrates governance constraints (who decides, who validates, who is accountable) and sustainable funding (a viable economic model within the French healthcare system) from the outset of design. This third component is as structuring 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 of the project; they structure architectural decisions from the outset.

Pilot phase & validation

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

• Technical validation — quality of digital twins, reliability of predictive models, FHIR interface performance
• Clinical validation — relevance of predictive alerts, true positive rate, acceptability by practitioners
• Organisational validation — integration into existing workflows, cognitive load for physicians, functioning of decision-making circuits

The pilot phase is designed to be publishable — results will be the subject of a publication in a peer-reviewed journal.

Funding structure

PREDICARE mobilises a mixed funding architecture, consistent with the nature of the programme — both a territorial research project and an industrial demonstrator.

• BPI France — pre-validation obtained for the innovation component (finalisation in progress)
• Regional funding — committed through territorial health bodies
• Partner GHT — institutional pre-validation for pilot deployment

Associated publication

The PREDICARE programme is the subject of a 110-page research memoir documenting the structural diagnosis that led to the programme, the system architecture and the conditions for scale-up.