Prior Authorization

Utilization ManagementAsync jobs required

Chart summarization, criteria-tree clinical review, and policy → questionnaire from a 265-policy corpus.

Prior authMedical necessityChart summaryQuestionnaireCriteria tree

About

A prior-authorization toolkit built on a shipped corpus of 265 payer questionnaires (criteria trees). It summarizes a chart into a structured longitudinal review, answers a questionnaire against the chart to an APPROVE/DENY with per-criterion verdicts, maps which provider documents satisfy which criteria, and converts a raw payer policy PDF into a structured questionnaire.

Questionnaires are browsable and filterable by CPT, ICD, and line of business; each resolves to a recursive criteria tree with an evaluatable-leaf count and a downloadable source policy PDF.

How it works

  1. 1Browse: facets + questionnaire search (CPT / ICD / LOB) → criteria tree by policy_id
  2. 2Chart-summary: OCR → map-reduce extraction → longitudinal summary (problems, meds, labs, gaps)
  3. 3Clinical-review: flatten the criteria tree → evaluate each leaf vs the chart → boolean rollup → APPROVE/DENY
  4. 4Policy-to-questionnaire: a payer policy PDF → recursive criteria-tree draft

Intended use

  • •PA reviewer worklists (chart summary, medical-necessity determination)
  • •Provider-side packet assembly — which documents satisfy which criteria
  • •Policy ingestion — turn a payer PDF into an evaluatable questionnaire

Key outputs

  • ▸clinical-review → decision (APPROVE | DENY) + criteria_verdicts[] with evidence
  • ▸chart-summary → reviewer_summary, active_problems, medications, prior_and_failed_therapies, timeline, documentation_gaps
  • ▸questionnaires → policy_id, guideline_name, cpt/icd codes, evaluatable_count

Model comparison

F1 on Gwen's healthcare benchmark for this task — the Gwen pipeline vs the prompt-optimized model alone, with the uplift the pipeline adds, per model.

#ModelGwen pipelineModel onlyUplift
1
GPT-5.5Best
0.954
0.918+0.036
2
Gemini 3.5 Flash
0.929
0.903+0.026
3
Claude Opus 4.8
0.926
0.833+0.093

Endpoints

Try each endpoint with your signed-in session — usage counts toward your monthly budget.

Use synthetic data only. Do not submit real patient records or PHI when testing endpoints.

Limitations & caveats

  • –Determinations are decision support for human reviewers, not autonomous approvals
  • –Clinical-review / provider-doc-review require a policy_id (preloaded) or an inline criteria tree
  • –Document endpoints run 1–3 minutes; the async /jobs flow is mandatory