Ortho Coding

Coding & Risk AdjustmentAsync jobs required

Orthopedic CPT + ICD coding, LCD coverage verdicts, and critique-mode chart audit.

OrthoCPTNCCILCD coverageOperative notes

About

Specialty coding for orthopedic and MSK operative reports. The CPT coder reads the operation title and technique narrative to emit procedure codes with modifiers, NCCI-bundled for the chosen setting (practitioner vs facility selects the NCCI/MUE edition), each citing the technique text that supports it. The ICD coder walks the ICD-10-CM tabular tree (section → parent → leaf → 7th character → companion rules) for diagnosis codes.

Two higher-order endpoints build on the coders: coverage reconciles the ICD and CPT sets into a claim and checks, per CPT line, whether an emitted diagnosis justifies medical necessity under the payer/state LCD; audit runs a coder in critique mode against codes a human already assigned, applying the deterministic CMS-edit floor (NCCI, MUE, MPFS, LCD) plus LLM grounding judges, and answers a free-text question from the note.

Uniquely, these endpoints accept multiple files as one job: an op-note split across several PDFs is OCR'd and coded as a single concatenated note with continuous line IDs, preserving cross-file context.

How it works

  1. 1File(s) or pasted text → OCR (text input skips OCR and runs synchronously)
  2. 2/cpt: procedure extraction → modifier assignment → NCCI bundling (setting-aware) → supporting ICD diagnoses included
  3. 3/icd: tabular-tree walk for cited diagnoses
  4. 4/coverage: both coders → per-CPT-line LCD medical-necessity check (payer/state → MAC)
  5. 5/audit: deterministic CMS-edit floor + LLM judges → free-text critique { audit_text, findings }

Intended use

  • •Orthopedic / MSK coding apps from operative reports — routed to for orthopedic coding, surgical bundling, and coverage-check requests
  • •Multi-file surgical cases (op note + addenda) coded as one note via uploadManyAndProcess
  • •Auditor tooling: "is 29881 supported? what's missing?" answered from the chart

Key outputs

  • ▸/cpt → result.cpt_codes[] (with modifiers + citations) and result.icd_codes[] in one call
  • ▸/coverage → per-CPT-line medical-necessity verdicts under the selected payer/state LCD
  • ▸/audit → audit_text (free-text critique) + structured findings

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
Claude Opus 4.8Best
0.891
0.844+0.047
2
GPT-5.5
0.871
0.780+0.091
3
Gemini 3.5 Flash
0.857
0.795+0.061

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

  • –Provide a file OR text per endpoint — text input must stay short (runs synchronously, ≤300s); real multi-file cases must use the async /jobs flow
  • –/audit requires the original chart file(s) or text — calls without a note return 422
  • –payer/state affect coverage verdicts only, never code selection
  • –Specialty scope is orthopedic/MSK; general diagnosis coding belongs to ICD Coding