Risk Predictions (AI)¶
Why predictions¶
MedInsight analyzes patient data (documents, diagnoses, medications) and estimates:
- readmission risk — probability of repeat hospitalization;
- complication risk — probability of adverse outcome.
This is decision support for clinicians, not a replacement for clinical judgment.
Who can run predictions¶
Doctors, heads of department, administrators — roles with write permission (WRITE_ROLES).
Running a prediction¶
- Open the patient record.
- Ensure at least one parsed document (
parsed). - Click Run prediction.
- Wait for completion (usually 10–60 seconds).
Screenshot: patient record with Run prediction button and “Processing…” indicator.
Result¶
On the record you see:
- probability of readmission and complications (%);
- risk level: low / medium / high;
- explanation from GPT (if ProxyAPI is available);
- prediction date.
On the dashboard, high-risk patients appear in a separate table.
Plan limits¶
Monthly analysis count is limited by plan:
| Plan | Limit (example) |
|---|---|
| Freemium | 5 |
| Pro | 100 |
| Enterprise | unlimited |
Current usage — on the Subscription page (/subscription).
When the limit is exceeded, the API returns 402 Payment Required.
Notifications¶
When a prediction is ready:
- WebSocket — browser update (if open);
- Email — if SMTP is configured;
- Telegram — if the account is linked.
Validating predictions¶
A doctor can confirm or reject a prediction (via API validate-prediction) — data is used to improve the system.
If GPT is unavailable¶
Without a ProxyAPI key or on network failure, the system uses rule-based fallback — simplified rule-based calculation without GPT explanation.
Security¶
- Patient data does not leave your server except for ProxyAPI requests (if configured).
- Predictions are tied to tenant_id — other clinics cannot see them.