A Medicare overpayment review used to start with a paper letter, a request for a sample of charts, and a human reviewer moving through them at a reasonable pace. A billing pattern had to look strange enough that someone actually noticed it. Now the noticing is done by software, and it happens before the check clears. CMS has run predictive analytics against every fee-for-service Medicare claim, nationwide, on a streaming basis since 2011, and the models have only gotten sharper since.
For a clinic, that changes the stakes of a routine billing quirk. A modifier used a little too often, a code that trends higher than peers, a documentation habit that looks fine in the chart and odd in the data — any of it can put a practice on a queue that ends somewhere far more serious than a repayment request.
A single 99215 trend line can start the whole thing
Picture a mid-sized primary care group whose share of level-five office visits (99215) creeps up over eighteen months. Nothing about that number is illegal. Sicker patients, an aging panel, a new provider who documents more thoroughly — there are honest reasons for the drift. The clinicians don’t see a problem because, chart by chart, there isn’t one.
The algorithm doesn’t read charts. It reads the clinic’s billing distribution against every other primary care group in the region and flags the outlier. That flag is the seed. Everything that follows — the contractor letter, the extrapolated overpayment estimate, and, in some cases, the referral to investigators — grows out of a pattern the practice never noticed it was creating.
The contractor reading the flag gets paid to find money
The 99215 trend line lands with a Recovery Audit Contractor or a similar program-integrity vendor. These contractors are paid on contingency — a percentage of what they claw back — and that shapes how aggressively they read a chart and how they characterize a coding judgment call.
A practice pulled into that review doesn’t get to argue nuance on the front end. The contractor requests records, applies its own reading of medical necessity, and issues a determination. If a healthcare fraud defense attorney isn’t already involved by the time the extrapolation shows up, the number on the demand letter can look nothing like the actual dollars at issue in the sampled charts.
Extrapolation turns a small sample into a very large number
The 99215 story gets ugly at the extrapolation step. Auditors sample a few dozen visits, calculate an error rate, and project that rate across every comparable claim in the lookback period.
A modest error rate on a small sample can translate into a repayment demand many times larger than the dollars actually at issue in the reviewed charts. That is where the fight lives. Practices that push back successfully do so by attacking the sample — its size, its selection method, its statistical validity — rather than relitigating each individual chart. Waiting to raise those arguments until the third level of appeal is a common, expensive mistake.
The same flag can be routed somewhere much worse
The 99215 pattern doesn’t only travel to a RAC. The same claims data feeds cross-payer analytics and gets shared with investigators when the pattern looks less like sloppiness and more like a scheme. Payers are also under pressure to hand off suspicious patterns rather than absorb them quietly. A 2025 HHS-OIG report found that 10% of Medicaid managed care plans made zero provider fraud referrals in 2022, and regulators are pushing plans to refer more, not fewer.
Once a pattern crosses from “overpayment” to “potential fraud,” the file changes hands. The clinic may still think it’s in a billing dispute while, on the other end, someone at a Medicaid Fraud Control Unit or a U.S. Attorney’s office is opening a matter. The lag between those two realities is the dangerous part.
What a clinic should actually do the day the letter arrives
The single most useful move is to handle the first audit letter as the beginning of a legal matter, not a billing chore. In practice, that means a short list of things:
- Freeze the records. Preserve the exact charts, audit trails, and billing edits as they exist the day the letter arrives. Any later “cleanup” in the EHR is discoverable and looks terrible.
- Pull the data behind the flag. Run your own utilization report against peer benchmarks so you know what pattern the contractor is looking at before you respond.
- Get counsel involved before the response. The response letter, the sample-selection challenge, and the statistician question all shape what an appeal looks like months later.
- Assume the file travels. Write every response as if an investigator will read it, because one might.
- Don’t self-refund on a hunch. Voluntary repayments made without legal advice can be used as an admission and can widen, not narrow, the scope of what’s reviewed.
The danger isn’t that an algorithm can prove fraud. It’s that it can decide which clinic gets looked at next. By the time a routine billing pattern becomes an investigation, understanding what the data appears to say — and responding before that pattern hardens into a narrative — can matter as much as the individual charts themselves.
This content is provided for informational purposes only and is not a substitute for professional advice. AFP editorial staff were not involved in the creation of this content.
Share
