Ethics & methodology
Transparent by design.
The Missing Million is built for trust. Every score is explainable. Every signal is traceable. Every outreach decision is made by a person who can see, question and override the model.
Not a diagnostic tool
No clinical inference. No replacement for any care pathway.
Not surveillance
Aggregated, anonymised signals already held within the PCN. No new collection.
Not replacing judgement
AI-assisted prioritisation only. Practitioners decide everything.
Not a black box
Every score breaks down into named, weighted, traceable factors.
How scoring works
The composite score is a weighted sum of independently sourced signals, each with documented confidence and provenance. Weights are reviewed quarterly with the clinical lead.
| Factor | Weight | Data source | Confidence |
|---|---|---|---|
| Recently bereaved | 22% | PCN bereavement register | High |
| Missed GP appointments | 18% | GP appointment system | High |
| Living alone | 14% | Council records | High |
| Reduced prescription pickup | 13% | Pharmacy records | Medium |
| Social withdrawal | 12% | Community attendance log | Medium |
| No recent community activity | 11% | VCS partner network | Medium |
| Mobility decline | 6% | GP notes | Medium |
| Winter isolation pattern | 4% | The Missing Million pattern model | Low |
Operating principles
Human review required
No automated outreach. Every flag is reviewed by a social prescriber before any contact is made.
Explainability is a feature, not a footnote
Every score is decomposable to named factors with sources and confidence. If it can't be explained, it isn't shown.
Opt-out by default for sensitive groups
Residents on safeguarding registers, in care settings, or who have opted out are excluded from prioritisation.
Anti-bias safeguards
Demographic parity audits on every model release. Score distributions reviewed by ward, age band and ethnicity. No protected attributes used as input features.
Data governance
Data stays within the PCN's information governance perimeter. No external sharing. No third-party model training. Full audit log of every read and action.
Practitioner override is signal
Dismissals and overrides are tracked, reviewed quarterly, and used to refine factor weights — not ignored.
The promise we make to residents
- We will only ever use information already held within your existing care relationship.
- We will never make a clinical decision about you using this system.
- You can ask to be excluded from prioritisation at any time, with no impact on your care.
- You can ask to see what factors contributed to any contact you receive.
- A real person will always be the one who decides to reach out.