How US Healthcare Works, Part 15: Payment Integrity

Part 13 showed that adjudication runs fast and mostly without people, and part 14 showed how examiners handle what the machine can't. Some wrong payments still get through: a claim coded higher than the record supports, a service that wasn't necessary, a bill that another plan should have paid, and sometimes a bill for care that never happened. Payment integrity is the set of programs that find these, before payment if possible and after it if not.

1. How big the problem is

Medicare measures it carefully. Each year a contractor samples traditional Medicare claims, requests the medical records and checks whether each payment was correct. For fiscal year 2025, the estimated improper payment rate was 6.55%, or $28.83 billion, down from 7.66% the year before.

An improper payment isn't the same as fraud. Most of these are documentation errors: the record didn't support what was billed, or a required form was missing. Some care was probably appropriate but not properly documented. CMS says so itself. But the number shows the scale: even a careful payer pays roughly one claim dollar in fifteen wrongly.

2. Fraud, waste and abuse

The industry splits wrong payments into three kinds, and the response differs:

The main federal laws are the False Claims Act, which lets the government, and whistleblowers on its behalf, recover up to three times the loss from anyone who knowingly submits false claims, and the Anti-Kickback Statute, which forbids paying for referrals of federal program business. Each payer has a special investigations unit (SIU) for suspected fraud.

The cases can be large. In June 2025, the Justice Department and HHS's Inspector General announced their largest national health care fraud takedown: 324 defendants charged in schemes involving more than $14.6 billion in alleged intended losses. "Intended loss" means the amount billed, not the amount actually paid.

3. Before payment: prepayment programs

Stopping a wrong payment is far cheaper than recovering it later, so payers push as many checks as they can before payment:

• Coding and clinical edits, the NCCI and payer edits from part 13, plus licensed edit sets from payment integrity vendors.
• Prepayment review of high-risk claims: the claim is held and records requested before payment. Common for very expensive claims, unusual patterns and providers under investigation.
• Predictive models that score each claim for risk of being wrong or fraudulent, so the riskiest go to review. Medicare has run a Fraud Prevention System that screens claims this way since 2011.

The trade-off is with providers' cash and goodwill. Holding a legitimate claim for weeks to request records is costly for the provider, so payers try to target prepayment review narrowly.

4. After payment: recovery programs

Traditional Medicare uses Recovery Audit Contractors (RACs), paid a share of what they recover, for post-payment review, and Unified Program Integrity Contractors (UPICs) for fraud investigation. Commercial payers use their own teams plus vendors, often paid on contingency.

Recovery usually happens by offset: the payer subtracts the overpayment from the provider's next payments, with a letter explaining which claim was adjusted and why. Providers can dispute it, and contracts limit how far back the payer can look.

5. Risk adjustment audits

Medicare Advantage adds a payment integrity problem that runs the other way, from the government to the plan. As part 11 mentioned, CMS pays MA plans more for members with more documented conditions. If a plan reports diagnoses its members' records don't support, it's overpaid.

CMS checks this with Risk Adjustment Data Validation (RADV) audits, which sample members, request their records and recalculate payment. In May 2025, CMS said it would audit every eligible MA contract every year, a large expansion from the small samples it had audited before. Some of the largest False Claims Act cases of recent years have been about MA risk adjustment.

6. Where the data and AI fit

Payment integrity is pattern-finding over very large datasets, so it was one of the first places payers used statistical models:

• Outlier detection: a provider who bills a code far more often than peers in the same specialty and region.
• Network analysis: links between providers, patients, addresses and bank accounts that reveal fraud rings.
• Record review: language models that read medical records and highlight whether they support the billed codes, so a reviewer spends minutes, not an hour.

The risks are the same as in part 10: a model's flag isn't proof, and legitimate providers who are unusual, a specialist with sicker patients, for example, can look like outliers. Good programs use models to decide what a person looks at, not what gets recovered.

7. Limits

• Recovery is slow and incomplete. Money paid out is hard to get back, especially from providers who close or are under investigation.
• The provider pays for false positives. Every legitimate claim held or audited costs the provider staff time and cash.
• Incentives can distort. Contingency-fee vendors earn more by recovering more, which pushes toward aggressive audits that providers then appeal, often successfully.
• Documentation decides. Most improper payments are about what the record says, so documentation quality matters as much as billing honesty.

Remember

Next, in part 16: how providers get paid, fee schedules and DRGs in detail, capitation, value-based contracts and how CMS pays Medicare Advantage plans.

Sources

• CMS, Fiscal Year 2025 improper payments fact sheet, January 2026
• CMS, Comprehensive Error Rate Testing (CERT)
• HHS Office of Inspector General, 2025 National Health Care Fraud Takedown
• US Department of Justice, The False Claims Act
• CMS, Medicare Fee-for-Service Recovery Audit Program
• CMS, Risk Adjustment Data Validation