Predictive denial scrubber

Predictive denial scrubber

Pre-submit denial probability + gap list — runs on every claim before it leaves Pollen8. Find scrubbed-failed claims at /admin/scrub-queue.

Four signals

The scrubber consults four data sources:

1. CMS NCCI bundle — Medicare PTP edits loaded by an admin quarterly. Catches code-pair bundling violations.
2. Prior-auth on file — does the patient have an active auth for this CPT × payer? Checked against legal_auths_on_file rows synced via X12 278.
3. Required-Dx rules — payer-specific dx requirements per CPT. E.g., 99213 requires an associated Dx; 90834 (psychotherapy) requires a mental-health Dx.
4. Tenant-historical risk — claims with the same payer + CPT + place-of-service that were previously denied. Probability model trained on the tenant’s own ERAs (X12 835) — improves over time without external training.

Per-claim verdict

{
  "predicted_denial_probability": 0.31,
  "gaps": [
    {
      "kind": "ncci",
      "issue": "99214 bundled with 96365 — modifier 25 required",
      "fix": "Append modifier 25 to E/M code 99214",
      "severity": "block"
    },
    {
      "kind": "prior_auth_missing",
      "issue": "CPT 70551 requires PA for Aetna",
      "fix": "Obtain prior auth before resubmit",
      "severity": "block"
    },
    {
      "kind": "historical",
      "issue": "13% of prior 99213 claims to Cigna denied in last 90 days",
      "fix": "Verify medical necessity documentation",
      "severity": "warn"
    }
  ]
}

severity: "block" claims are quarantined; "warn" ones submit but log the warning. Configurable per tenant.

UI

/admin/scrub-queue — open issues sorted by severity. Each row shows the claim, the gap list, and a one-click action (apply fix, override, write off). Bulk-fix UI for the common case of “add modifier 25 to every claim flagged with this rule today”.

What it isn’t

Not a coding compliance tool — it doesn’t tell you whether your documentation supports a 99214. It tells you whether this specific claim shape is going to bounce off the payer’s adjudication engine.

Pair it with the AI charge capture for the full pre-submit story: capture caught misses, scrubber caught likely denials.