AI Medical Billing Severity Drift & Upcoding Audit Workbench

Audit Scenario

Study Benchmark Cohort Pre-calibrated against clinical documentation improvement (CDI) severity shifts.

Annual Incurred Claims 10,000 Total audited patient encounters in the payer or health system sample.

AI Scribe / CDI Adoption 68% Proportion of clinical notes generated or prompted by generative EHR tools.

Acuity / Severity Uplift +18.5% Empirical jump in higher-weight MS-DRGs or Level 5 billing codes.

Baseline Avg Allowed Reimbursement $8,450 Standard baseline payment per encounter prior to AI upcoding shift.

Post-Audit Disallowance / Overturn Rate 62% Share of flagged claims where human chart reviews fail to corroborate AI severity.

2-Year Excess Payer Cost $21.2M $10.6M / yr unadjusted

Audited Severity Creep +18.5% 6,800 AI claims

Recoverable Recoupment $13.1M 62% disallowance yield

False Claims / FCA Risk High Risk Triggered: >15% outlier

Severity Tier Distribution: Baseline vs. AI-Assisted Scribe

MS-DRG / E&M coding shifts across audited clinical encounters

Pre-AI Baseline

Observed AI Billing

Audited Encounter Sample (Select row to inspect AI chart prompt)

Claim ID	Patient / Service	Pre-AI Code	AI-Billed Code	Cost Delta	Clinical Audit Finding	Action

Claim Audit Detail: CLM-849201 (Inpatient Sepsis) AI Hallucinated MCC

Original Physician Clinical Impression

"Patient presented with fever (100.8F), mild dysuria, and WBC 11.2k. Urinalysis positive for leukocytes. Initiated oral levofloxacin. Hemodynamically stable, BP 122/78."

AI Ambient Scribe Auto-Generated Addendum

"Systemic Inflammatory Response Syndrome (SIRS) secondary to urinary tract pathogen. Severe Sepsis with Acute Organ Dysfunction suspected based on transient tachycardia. Upgraded MS-DRG 872 to MS-DRG 871 (Major Complication/Comorbidity)."

Baseline: DRG 872 ($7,200) → AI Billed: DRG 871 ($14,800) | Payer Overpayment: +$7,600

Audit parameters applied. 8 representative cohort claims loaded.

Understanding the local billing scenario

Your browser does not support video.

Read the explanation

This page is a synthetic billing scenario, not a finding about real patients or coding compliance. The first multiplication selects the modeled assisted claims: ten thousand encounters times sixty eight percent equals six thousand eight hundred. Changing adoption changes that count. These assumptions do not establish that documentation caused a payment error. The calculation multiplies assisted claims by the baseline payment and the entered severity drift. At the defaults, annual modeled excess is ten million six hundred thirty thousand one hundred dollars. The two year card doubles it. Drift is an input, not a measured causal estimate. This arithmetic is a scenario, not an estimate validated against real reimbursement records. Recovery uses the annual excess times the entered disallowance rate, sixty two percent. That gives six million five hundred ninety thousand six hundred sixty two dollars. It does not multiply the two year total. The page assigns scrutiny labels at fixed drift thresholds; those labels do not prove that a government audit is legally required or that a claim is fraudulent. The selected sample begins with a billed amount of fourteen thousand eight hundred dollars and a baseline of seven thousand two hundred. The local downcode action replaces the billed amount with that baseline, removing a seven thousand six hundred dollar difference in the sample row. This is a browser state transition, not a submitted claim adjustment, clinical judgment, or recovered payment. The cohort cards still follow their slider assumptions.