Fraud, Waste & Abuse (FWA) Pattern Hunter

The Problem

Healthcare fraud, waste, and abuse (FWA) cost the industry an estimated $68 billion annually (CMS/DOJ estimates). Common schemes include upcoding (billing higher-severity diagnosis codes than documented), phantom billing (billing for services never rendered), billing during clinically implausible timeframes (a podiatrist billing 26 hours of surgery in a single day), and phantom clinics (billing from addresses that don't exist or aren't licensed). Traditional claim-by-claim manual audits catch only the most obvious cases; sophisticated schemes exploit temporal and network patterns invisible to single-claim analysis.

Data Requirements

Data Sources:

• Historical claims data (3 to 5 years minimum) , 837/835 claim detail with full procedure/diagnosis codes, dates, provider/member IDs
• Provider master file , NPI, TIN, specialty, state license number, credentials
• State licensing databases (via API or bulk extract) , active license status, credential verification, known address
• Payer member master file , demographics, enrollment history, coverage periods
• CMS PECOS (Provider Enrollment, Chain, and Ownership System) , excluded providers list
• OIG Exclusion List , sanctioned providers, suppliers, clinicians
• Optum/Cotiviti historical fraud data (if available) , known scheme signatures

Data Classification:

• Public: State licensing databases (public query), OIG Exclusion List, CMS PECOS (public), provider addresses (property records)
• Internal Structured: Claims data (837/835 detail), member master file, provider master file, payer billing rules
• Internal Unstructured: Prior investigation case files, SIU notes, provider correspondence
• External Third-Party: State licensing API queries, CMS/OIG bulk downloads
• Sensitive/Regulated: All claims data is PHI; fraud investigations are highly sensitive and subject to HIPAA minimum necessary rules, False Claims Act, and state AG oversight

Data Quality Requirements:

• Timeliness: Claims data must be complete and finalized (no pending adjustments) before analysis; lag of 10 to 30 days acceptable for batch retrospective analysis.
• Completeness: At least 95% of claims must have complete provider NPI, member ID, service date, and diagnosis/procedure codes; missing data impairs pattern detection.
• Accuracy: Diagnosis/procedure codes must be current and valid (ICD-10, CPT with modifiers); provider NPI must match NPPES; service dates must be within member's enrollment period.

Integration Complexity: High , Requires bulk extract of 3 to 5 years of claims data (typically 100+ million claim records), integration with multiple external databases (state licensing, OIG, CMS), and ML pipeline for anomaly detection. Batch processing may take 6 to 12 hours; requires data warehouse or data lake for efficient querying.

Regulatory Alignment

• CMS Fraud Prevention Rule & Anti-Kickback Statute: Payers must have documented FWA detection programs; agent provides audit trail.
• False Claims Act (31 USC § 3729): Knowing submission of false claims can trigger qui tam suits; agent provides evidence trail for investigation.
• HIPAA Minimum Necessary: Investigation data should be limited to needed personnel; agent must support access controls and audit logging.
• OIG Compliance Programs: OIG expects payers to have risk assessment and monitoring; agent demonstrates proactive approach.

Comparable Implementations

• Optum/Cotiviti FWA Detection Solutions , Industry leader in claims analytics and fraud detection; publicly available ROI data shows 5-10x return on fraud investigation investment
• CMS OIG Compliance Program Guidance , Regulatory expectations for payer FWA programs
• Guidehouse: Generative to Agentic AI in Healthcare , Case study on ML-driven anomaly detection in claims