Documentation Index
Fetch the complete documentation index at: https://docs.encoreos.io/llms.txt
Use this file to discover all available pages before exploring further.
Spec: PM-50
Created: 2026-04-07
Status: 📝 Planned
Overview
PM-50 provides ML-based claim denial prediction as a pre-submission advisory layer. It integrates into the PM-08 claim submission queue, complements PM-18 rule-based scrubbing, reads documentation completeness metadata from CL-36 via Platform Integration Layer, and uses PF-96 jurisdiction profiles for payer/state-specific context.
Integration Points
| PF Feature | Integration Type | Usage |
|---|
| PF-01 (Organizations & Sites) | Direct dependency | Tenant isolation, site-scoped predictions |
| PF-02 (RBAC) | Direct dependency | Permission checks for billing specialist and revenue cycle director roles |
| PF-10 (Notifications) | Platform Integration Layer | Send documentation request notifications to clinical team |
| PF-96 (Jurisdiction Profiles) | Platform Integration Layer | Read payer/state-specific parameters (filing deadlines, modifier conventions) |
Same-Core Dependencies
| PM Feature | Integration Type | Usage |
|---|
| PM-08 (Claims Management) | Shared UI surface | Risk scores displayed in claim submission queue; scoring triggered on claim queue entry |
| PM-18 (Claim Scrubbing) | Sequential pipeline | PM-18 scrubbing runs first (rule-based), PM-50 adds ML-based prediction after |
| PM-29 (Denial Management) | Data feedback | PM-29 denial outcomes feed PM-50 retraining pipeline (actual_outcome field) |
Cross-Core Dependencies
| Feature | Integration Type | Usage |
|---|
| CL-36 (Clinical Documentation) | Platform Integration Layer (@/platform/clinical) | Read documentation completeness metadata (note section count, word count, required section presence, diagnosis specificity); no PHI content crosses into PM-50 |
Events Published
| Event Name | Publisher | Subscribers | Status |
|---|
pm.denial_prediction.created | PM-50 | None (available for GR dashboards) | 📝 Planned |
pm.denial_prediction.outcome_recorded | PM-50 | None (available for analytics) | 📝 Planned |
Event Payload Schemas
pm.denial_prediction.created:
{
prediction_id: string;
claim_id: string;
risk_level: 'low' | 'medium' | 'high';
organization_id: string;
}
pm.denial_prediction.outcome_recorded:
{
prediction_id: string;
claim_id: string;
actual_outcome: 'paid' | 'denied';
organization_id: string;
}
Edge Functions
| Function | Purpose | Auth |
|---|
predict-denial-risk | Receive claim features, return risk score + reasons + actions | verifyOrgAccess() |
retrain-denial-model | Cron-triggered monthly retraining using historical outcomes | Service role (cron) |
denial-prediction-metrics | Return aggregated model performance metrics for dashboard | verifyOrgAccess() |
@/platform/notifications — Send documentation request to clinical team (FR-3.2)@/platform/clinical — Read documentation completeness metadata from CL-36 (FR-3.1); PM-50 implements stub/adapter returning null signals when CL-36 unavailable
Data Flow
PM-08 (Claim enters queue) → PM-18 (Rule-based scrub) → PM-50 (ML prediction)
↓
Risk Score + Actions
↓
Biller Review (UI)
├─ Acknowledge → Submit
├─ Hold → Investigate
├─ Correct → Re-queue
└─ Override → Submit with reason
Security Boundaries
- No PHI crosses into PM-50 — only categorical/aggregate features and completeness metadata
- All prediction data scoped to organization via RLS with SECURITY DEFINER helper
- Model training data never crosses organization boundaries
- Edge functions validate org access via
verifyOrgAccess() from _shared/auth.ts
References
- Spec:
specs/pm/specs/PM-50-ai-denial-prediction-prevention.md
- Related: PM-08, PM-18, PM-29, PM-49, CL-36, PF-96
- Events:
docs/architecture/integrations/EVENT_CONTRACTS.md
- Platform layers:
docs/architecture/integrations/PLATFORM_INTEGRATION_LAYERS.md
