Missing Records and Extrapolation in an E/M Audit
A compliance auditor for a 40-provider multispecialty group is auditing outpatient E/M claims. The population is 12,000 claims for 99213–99215 from the prior year. Using a random number generator, the auditor selects 60 claims. After three written requests, 5 records are still missing; two belong to a provider whose records are in a legacy system being migrated. Of the 55 records reviewed, 8 have E/M level errors: 6 were upcoded by one level and 2 were downcoded. The CFO wants to extrapolate the 8 errors to all 12,000 claims, demand repayment from the group's own funds, and report a 13.3% error rate to the board. The auditor is asked to validate the methodology before the report is finalized.
1. How should the five missing records be handled in the error-rate calculation?
Read the worked answer
The auditor should not silently drop the five records. Missing documentation generally means the billed service is unsupported, because the claim cannot be validated. However, the auditor must first document exhaustive retrieval efforts: EHR search, legacy-system vendor, paper files, and provider office. The two legacy-system records deserve extra follow-up before being treated as missing.
The report should present at least three figures:
- Reviewed-only rate: 8/55 = 14.5%.
- Missing-as-unsupported rate: 13/60 = 21.7%.
- Sensitivity range: show both, and explain that the true rate lies between them until the missing records are resolved.
If the missing records are later found and are compliant, the rate falls. If they are noncompliant, it rises. Treating them as a separate 'unable to validate' category is acceptable for internal management, but the auditor should state that unsupported claims create repayment exposure. The key is transparency: no exclusion without a documented rationale.
2. Calculate the error rates and assess whether extrapolation to all 12,000 claims is defensible.
Read the worked answer
The reviewed error rate is 8/55 = 14.5% (6 upcoded = 10.9%; 2 downcoded = 3.6%). The net upcoding rate is 10.9% minus 3.6%, but netting is not appropriate for repayment because upcoded and downcoded claims affect different services.
Extrapolation is not yet defensible as a precise point estimate. A random sample of 55 reviewed claims is small, and the 95% confidence interval around 14.5% is wide-roughly 5% to 24%. The CFO's 13.3% figure is also methodologically inconsistent: 8/60 = 13.3%, but that treats missing records as compliant, which is not supported.
Before extrapolating, the auditor should:
- Confirm the sampling frame is complete and unduplicated.
- Stratify by provider and code, because errors may cluster.
- Retrieve the missing records.
- Calculate overpayments by allowed amount, not charge amount.
- Consider expanding the sample if the error rate is high.
Extrapolation may support a corrective-action estimate, but a demand for repayment should be based on validated claims or a statistically defensible projection reviewed by legal or a statistician.
3. What should the auditor recommend to the CFO before any repayment or board report?
Read the worked answer
The auditor should recommend a hold on repayment and final board reporting until four steps are complete:
- Resolve missing records: make a final retrieval attempt and document the outcome.
- Validate the sample: confirm random selection, inclusion of all providers, and correct claim status (paid vs. denied).
- Stratify and expand: analyze errors by provider, code, and date; if one provider drives most errors, expand that stratum rather than extrapolating globally.
- Calculate financial exposure: use allowed amounts and separate upcoded from downcoded claims; do not net them into a single recovery figure.
The board report should state the reviewed error rate, the missing-record range, the confidence interval, and the limitations. It should avoid accusing any provider of fraud. The CFO can be told that a voluntary refund based on validated overpayments is appropriate, and that extrapolation may be used for internal risk assessment but not as a definitive recovery without statistical and legal review.