Modifier 25 Findings and the Limits of a Ten-Chart Sample
A twelve-provider group submits roughly 9,000 office E/M claims annually. The internal audit plan requires ten randomly selected charts per provider per year, drawn from a claims extract by the compliance analyst. For the current cycle, 120 charts were reviewed. Forty-one (34%) billed a same-day E/M with modifier 25 alongside a minor procedure, and the reviewer concluded the E/M was not significant and separately identifiable because the note documented only the procedure and a one-line diagnosis. Results by provider range from 0 of 10 (Provider B) to 9 of 10 (Provider A). The compliance committee wants to know whether it can extrapolate a repayment across all 9,000 E/M claims and whether Provider A should be referred for disciplinary action.
1. Can the committee rely on this audit to extrapolate an overpayment across the 9,000 claims and to single out Provider A?
Read the worked answer
Group-level: directionally reliable, not extrapolatable.
- 41 of 120 is 34%, with a 95% confidence interval of roughly 26% to 43% (Wilson). The lower bound alone is a serious problem, so the finding justifies immediate action.
- But a 17-point spread is not a number you can defend as a repayment calculation. Extrapolation for recovery requires a statistically valid random sample sized in advance for a stated precision, with the universe, sampling unit, and error definition documented before the pull.
Provider-level: the sample is far too small to rank people.
- 0 of 10 is statistically compatible with a true error rate as high as about 31%. Provider B is not exonerated.
- 9 of 10 is compatible with a true rate anywhere from roughly 56% to 99%. Provider A is an outlier worth investigating, but ten charts will not survive a peer-review challenge or an employment action.
- Ranking twelve providers on ten charts each invites grievance claims and undermines the audit program's credibility.
Also fix the error definition before you quantify. An unsupported modifier 25 does not always mean the whole E/M is an overpayment. If the visit is still payable at a lower level, the recoverable amount is the level differential, not the full claim. And if the review question is really whether the E/M was separately identifiable, the reviewer needs to state what separate work was absent.
Recommendation: treat this cycle as a risk-assessment exercise, not a recovery exercise. Expand the sample for any provider you intend to act on, using a methodology defensible on its face.
2. The committee asks for a corrective action plan. What do you build, and in what order?
Read the worked answer
Diagnose the root cause before prescribing education. Stratify the 41 errors by:
- Provider and specialty - is the pattern concentrated?
- Procedure paired with the E/M (for example, joint injection, lesion removal, nebulizer) - some pairings are genuinely separable, others almost never are.
- Documentation source - is the E/M text generated by a template that auto-inserts a level-3 note whenever a procedure is scheduled?
- Reviewer - did more than one clinician review, and did they agree?
Then act, in this order:
- Targeted education on the specific requirement: a significant, separately identifiable service, above and beyond the usual pre-, intra-, and post-procedure work, with its own documented history, exam, and medical decision making. Teach it with the group's own de-identified claims, not generic slides.
- 100% pre-bill review of the affected provider's modifier-25 claims for 90 days. Pre-bill is where you stop the loss; post-payment review only measures it.
- System control: an EHR soft stop that requires the coder to affirm a separate-and-identifiable service, or an edit that holds the pair for review.
- Re-audit at 90 days using the same tool, the same sampling method, and the same reviewers, so the result is comparable. Report the change, not just the rate.
- Feedback loop: give each provider their own results with the specific charts, and keep it confidential between the auditor and the provider until the pattern is confirmed.
Do not skip straight to a group-wide memo. If the errors are concentrated in two providers and one template, a group-wide memo dilutes the message and wastes the credibility you need for the targeted conversation.
3. Should the ten-charts-per-provider rule in the audit plan be changed, and how would you justify the change?
Read the worked answer
- A flat ten is a policy, not a sample design. Sample size should be a function of claim volume, coding risk, prior audit history, and the precision you need to act on the result - none of which are in the current rule.
- Risk stratification is the practical fix. For example: high risk (new providers, high modifier-25 or level-5 utilization, prior error rate above threshold) gets 30-40 charts; moderate risk gets 20; low risk with clean history gets 15. A twelve-provider group can absorb that volume.
- Volume matters. A provider with 1,500 annual E/M claims and a provider with 300 should not get the same sample size.
- Document the methodology in the audit plan - universe, sampling unit, stratification, error definition, reviewer qualifications, and the confidence level you are working to. When a reviewer challenges the finding, the methodology is your defense.
- Separate the two purposes. Compliance monitoring needs enough data to detect patterns. Recovery quantification needs a statistically valid sample run with a formal tool. Do not try to make one pull do both jobs.
- Get the change approved by the compliance committee and record why: the committee owning the methodology is what makes the results usable.