NHS England · AI Risk Intelligence
Every day, patients quietly stop seeing their GP — until a health crisis brings them back. This tool spots who is drifting away before it happens, so NHS teams can act early. An AI model trained on GP consultation patterns, prescription history, and deprivation data predicts disengagement risk across England, visible on a live map.
The Problem
In England, around 1 in 8 patients gradually stops engaging with their GP surgery — missing check-ups, skipping repeat prescriptions, going years without a consultation. By the time they return, they often present with preventable conditions that cost far more to treat. NHS systems currently have no early-warning signal for this. The risk is invisible until it becomes a crisis.
The Approach
The model is trained on GP consultation history, prescription patterns, repeat visit frequency, and area-level deprivation data. It learns what disengagement looks like before it fully happens — falling consultation rates, longer gaps between visits, dropping off repeat prescriptions. An XGBoost classifier then assigns each patient a risk score (0–100%), displayed on a choropleth map so commissioners, analysts, and frontline teams can see risk by neighbourhood at a glance.
Use Cases
The map, risk scores, and synthetic patient export are designed to slot into existing NHS and health analytics workflows. Filter the cohort you care about, export it, and connect it to whatever system your team already uses.
Problem: You manage health budgets across a whole region but can't tell which neighbourhoods are quietly losing touch with GP care.
How to use it: Filter the map by your ICB boundary. Export the at-risk patient cohort and route community health teams to the highest-risk Lower Super Output Areas before the next quarter planning cycle.
Problem: Your practice list has thousands of patients. Proactive outreach calls are expensive — you need to know who to call first.
How to use it: Pull the disengaged-only cohort filtered by your practice code. Sort by risk score. Your care coordinators now have a ranked call list, not a random one.
Problem: You're building a clinical decision tool and need a validated UK population risk layer without building the model yourself.
How to use it: Download the JSON export as a structured dataset. The XGBoost risk scores, deprivation quintiles, and SNOMED-coded conditions plug directly into your feature pipeline or patient management API.
Problem: Leadership wants to know whether the most deprived communities are also the least engaged — but your data is siloed across systems.
How to use it: Use the IMD quintile and ICB filters together to surface equity gaps. The built-in causal analysis shows deprivation increases disengagement risk by 5.1 percentage points independent of age and illness burden.
Problem: Your enrolled members who disengage from primary care become your most expensive claims. You need to intervene before the A&E visit.
How to use it: Segment enrolled members by risk score band and ICB. High-risk, high-deprivation members get routed to your wellness programme. Low-touch for the rest.
Synthetic data only. No real patient records are stored or displayed. Every output requires clinical review before any action is taken. GDPR Art.22 compliant.