The Probe Sample With the Auto-Populated Review of Systems
You are the newly hired compliance auditor for a six-provider family medicine practice that grosses about $4.2 million a year and has never been audited by a payer. You pull a random sample of 30 established-patient visits billed 99213 or 99214 from last quarter. Nineteen records fully support the level billed; eight billed 99214 have documentation supporting only 99213; three billed 99213 have documentation supporting 99214. The payer mix is 40% Medicare, 45% one commercial plan, 15% self-pay. While reviewing, you notice the EHR default template inserts a 10-system review of systems and a complete past medical history into every note, and the physicians say they rarely ask those questions. The medical director proposes repaying only the eight overbilled visits and treating the three underbilled visits as an offset. He also wants the template left alone because 'the level is based on medical decision-making now anyway.'
1. Calculate the error rates and explain why the three undercoded visits cannot be netted against the eight overpaid ones.
Read the worked answer
Overcoded: 8 of 30 = 26.7%. Undercoded: 3 of 30 = 10.0%. Any mismatch: 11 of 30 = 36.7%. With a sample of 30, the 95% confidence interval around a 26.7% rate is wide - roughly plus or minus 16 percentage points - so the true overcode rate could plausibly be anywhere from about 11% to 43%. Even the low end is well above the single-digit error rates a working compliance program would expect, and that matters for how you frame the finding to leadership: this looks systemic, not like two or three sloppy visits.
Netting is not available. Each claim is a separate request for payment. The eight 99214 claims were paid more than the documentation supports, and that difference is an overpayment the practice must return within 60 days of identification. The three 99213 claims were paid less than the documentation supports; that is an underpayment, and the practice's remedy is to submit corrected claims within the payer's timely filing limit, not to reduce what it owes on other claims. Self-help offsetting across claims and across payers is not permitted and, if discovered, reads as a refusal to refund. Say this plainly in the report so the medical director's proposal does not survive the meeting.
2. Assess the auto-populated review of systems. Does it change your view of the overpayment, and what is the compliance risk if the template stays?
Read the worked answer
Under the current office E/M guidelines, level of service is selected on medical decision-making or total time; history and examination are not key components. So the fabricated ROS does not by itself change the correct level, and the eight overpaid claims stand or fall on their MDM documentation. That is the medical director's one valid point - and it is a point about the level, not about the record.
The separate problem is documentation integrity. A template that asserts a 10-system review the physician did not perform creates a record that is false on its face. If any visit is billed on time, the record is now unreliable in a second way, and a reviewer cannot tell what actually happened. Once a payer or a relator can show the practice knew the template auto-filled unsupported history, every claim carrying the template is arguable as a false claim regardless of whether the level happened to be correct, because the unsupported content is part of what was certified. It also destroys the practice's ability to defend the sample.
Recommendation: disable auto-population, require the clinician to affirmatively document what was obtained (or document that the history was unavailable), add an attestation, and re-audit after the change. Consider expanding the audit, because a template defect this broad means your 30-chart sample probably understates the exposure rather than overstates it.
3. What is the defensible scope of repayment here, and how do you decide whether to extrapolate?
Read the worked answer
You must return the identified overpayment on the eight claims within 60 days of identification, using the actual difference between the 99214 and 99213 allowed amounts for each payer - not the charge difference. That is not optional and does not wait for a full-population review.
For the remainder, a 30-chart probe is a screening tool, not a basis for extrapolation. A defensible extrapolation needs a statistically valid random sample drawn from a defined population, stratified by provider and payer, with a documented methodology a statistician would sign off on and a confidence level stated up front. If the practice chooses to quantify the full exposure rather than refund only identified claims, it should draw a larger sample (or review the population if volume permits) and apply the same methodology the payer would. What the practice must not do is assume the payer will not extrapolate - Medicare contractors routinely do - or disclose a number it cannot defend.
Practical path: refund the eight, disclose the sampling finding and the template remediation to the Medicare contractor and the commercial payer through their voluntary refund processes, conduct the larger review, and complete provider education with a post-education audit within 90 days. Document the methodology and the decisions in the compliance file, including why the medical director's offset proposal was rejected.