CDI — Pre-Bill Reconciliation

Documentation QualityAsync jobs required

Pre-bill reconciliation of intended codes against chart evidence — integrity risks, capture, and queries.

CDIPre-billReconciliationICD-10Physician queries

About

A pre-bill reconciliation engine: the provider supplies the ICD-10 codes they intend to bill, and a map-reduce LLM pipeline compares each against the chart's documented evidence — surfacing over-coding (integrity risk), under-coding (capture opportunity), and specificity gaps.

Each finding carries its track, finding type, the supportable code, evidence quotes, an impact category (HCC/RAF, CC/MCC, DRG), and — where a query is warranted — a non-leading physician query that must clear a non-leading + evidence-supported audit.

How it works

  1. 1Document upload → OCR (or paste pre-OCR'd text)
  2. 2Normalize the provider's intended ICD-10 set
  3. 3Map-reduce reconciliation: confirm / capture / specificity-upgrade / integrity-risk per condition
  4. 4Audit pass validates each drafted query is non-leading and evidence-supported

Intended use

  • •Pre-bill coder / CDI review before claim submission
  • •Integrity programs flagging unsupported intended codes
  • •Specificity and capture worklists with HCC/RAF and CC/MCC impact

Key outputs

  • ▸findings[] — finding_type, intended_code, supportable_code, recommended_code, evidence, impact
  • ▸suggested_query (+ audit: non_leading, evidence_supported) when a query is warranted
  • ▸summary — counts by finding type, codeable_now, requires_query, estimated_impact

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
0.955
0.868+0.087
2
Claude Opus 4.8
0.955
0.885+0.070
3
Gemini 3.5 Flash
0.953
0.923+0.029

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

  • –Integrity-risk (over-coding) detection requires the intended_codes input
  • –Drafts queries — human CDI review before sending remains the expected workflow
  • –Multi-pass LLM (1–3 minutes); the async /jobs flow is mandatory for production uploads