Medicaid Spending Anomaly Report

Interactive analysis of 227 million Medicaid claims records from the HHS open data release (Feb 14, 2025), identifying providers with statistically anomalous billing patterns.

What this is

On February 14, 2025, the HHS DOGE team open-sourced the largest Medicaid provider spending dataset in department history. This project analyzes that data across three dimensions:

1. Year-over-year spending spikes — Providers whose billing tripled or more in a single year, starting from a baseline above $1M. Flags like CARES Inc (70x spike, $77K to $486M in three years) and Tradition Choice (52x spike in Brooklyn, NY).

2. Minnesota autism pattern — Searches nationally for the same billing code surges (H2015, T1019) used in the Minnesota autism diagnosis fraud. Finds providers with 157x and 106x year-over-year increases in community habilitation billing.

3. Claims-per-beneficiary outliers — Providers billing thousands of claims per patient, suggesting services not rendered or systematic upcoding. LA County DMH bills 4,807 claims per beneficiary for community habilitation; Alabama DMH averages $4.3M per beneficiary for residential care.

Dataset

• Source: HHS Medicaid Provider Spending (open data release, Feb 2025)
• Size: 227,083,361 rows (11GB CSV, 2.7GB Parquet)
• Span: January 2018 through December 2024
• Scope: 617,503 billing providers, 1,627,362 servicing providers, 10,881 HCPCS codes
• Total spending: $1.09 trillion

Each row is a unique combination of billing provider NPI, servicing provider NPI, HCPCS code, and claim month, with fields for total unique beneficiaries, total claims, and total amount paid.

Analysis pipeline

1. Convert 11GB CSV to Parquet using DuckDB with ZSTD compression
2. Run anomaly detection queries (spending spikes, code-specific surges, ratio outliers)
3. Resolve provider identities via the NPI Registry API (NPPES)
4. Present findings with interactive charts (Recharts) and detailed narratives

Running locally

pnpm install
pnpm dev

Open http://localhost:3000.

Caveats

• Anomaly is not fraud. Statistical outliers may reflect legitimate operational changes, new contracts, Medicaid policy changes, or data reporting adjustments.
• Fiscal intermediaries like Public Partnerships LLC and Tempus Unlimited aggregate billing for thousands of individual caregivers. High volumes may reflect their role.
• Government agencies like LA County DMH and Alabama DMH may bill differently due to bundled service agreements and statewide contracts.
• Beneficiary counts represent peak monthly counts, not unique lifetime beneficiaries. Claims-per-beneficiary ratios are upper-bound estimates.

Tech stack

• Next.js 16 (App Router)
• Recharts (interactive charts)
• Framer Motion (scroll animations)
• Tailwind CSS
• DuckDB (data processing)