Immunizations Service — AI Integration
Status: populated Owner: TBD Last updated: 2026-04-18 Companion: Service Template
1. Current AI Use
Tier: None (S0/S1)
No AI or ML components are active in the current implementation. All forecast calculations are deterministic, rule-based interval computations driven by the EPI national schedule configuration.
2. Planned AI Capabilities (S3)
The following AI features are scoped for a future enhancement slice:
2.1 Defaulter Outreach Prioritisation (Advisory — Tier B)
| Aspect | Detail |
|---|---|
| Purpose | Rank defaulters by risk score to prioritise outreach effort |
| Input | Patient demographics, prior appointment adherence, distance from facility, dose count missed, seasonal patterns |
| Output | Risk score (0–100) and recommended outreach channel |
| Human gate | Vaccination officer reviews the ranked list; AI does not send outreach directly |
| Fallback | If model unavailable, fall back to FIFO by daysOverdue DESC |
| Model host | ai-gateway-service; immunizations-service calls /ai/v1/risk-score |
2.2 Coverage Anomaly Detection (Advisory — Tier B)
| Aspect | Detail |
|---|---|
| Purpose | Detect unusual drops in vaccination coverage at facility or antigen level |
| Input | Historical coverage timeseries from materialized views |
| Output | Alert if coverage drops > 2 standard deviations below rolling 90-day mean |
| Human gate | Dashboard alert; requires admin acknowledgement |
| Fallback | Static threshold alert (< 70% coverage) if model unavailable |
3. Hard Constraints
- AI must never automatically record, amend, or correct an immunization record.
- AI must never override a contraindication determination.
- AI recommendations must be clearly labelled as advisory in all UI surfaces.
- All AI calls route through ai-gateway-service; immunizations-service does not call external LLM APIs directly.
- AI model inputs and outputs must be logged for auditability (GDPR Art. 22 compliance for automated decision assistance).