Use of Statistical Sampling by the OIG and Others

Early last year, HHS OIG released a Podcast explaining its statistical sampling methodologies. The OIG depends on statistical sampling to cover thousands or even millions of claims in a fair and objective fashion. Without using this method, the OIG’s reviews would make it much more difficult for providers to gather the necessary supporting documentation and appeal contested claims. Moreover, reviews would not be efficient or cost-effective for either the provider or the OIG. The sampling method used in each review varies depending on risk factors and applying the appropriate statistical formulas to calculate any estimated overpayment. As the OIG notes, the courts have held that the methodology need not be precise or optimal as long as it is statistically valid. The OIG ensures that its work meets this standard by evaluating each sample using appropriate laws or regulations, but all methods used must meet four requirements. The methods must be: (a) statistically valid; (b) efficient; (c) representative of the larger group; and (d) producing a valid estimate of any overpayment. The process the OIG follows includes the following steps:

1. When looking at a sample of claims selected from a larger group, the OIG makes an estimate, but only applies the estimate to that specific larger group of claims from which the sample was drawn.
2. The OIG reduces the overpayment estimate in order to properly account for claims that are canceled, refunded to the Medicare program, or are otherwise not in error.
3. The OIG makes clear that if done properly, the process employed creates an accurate and efficient way to look at a high volume of data.
4. The OIG maintains and keeps all documents and data related to every sample so that the data can be reproduced.

Notably, numerous administrative appeal decisions and federal court cases have concluded that statistical sampling is an appropriate way to calculate any overpayment. Courts have also upheld the provider’s right to appeal when the OIG uses a statistical sampling to question costs or claims, or to appeal an individual determination of an overpayment through the normal Medicare appeals process.

The OIG is not the only organization that uses statistical sampling to calculate overpayment demands. State and federal prosecutors rely upon on employing statistically valid sampling methods to make cases. CMS contractors (MACs, ZPICs, RACs, etc.) and State Medicaid Fraud Control Units also depend on sampling.

Tips When Confronted by Demands Based Upon Sampling

It is a mistake to assume government agencies and their contractors always follow statistical sampling methodology properly and focus the defense on challenging individual claims made up in the sample. Realistically, those doing the sampling take short cuts, or do not use the best qualified parties to draw a statistically valid sample. The important thing to remember is that whoever charges inappropriate claims being submitted and acts upon that information must maintain documentation and data related to every sample so that it can be reproduced. Therefore, the first line of defense when confronted by assertions of wrongful claims based upon sample projections is to challenge the claims sampled using real experts. It is far better to request documentation to see if the sampling methodology is valid and reliable. If it is not, the whole case may be just a “card palace.” If there is fault in the method used, then it is advisable to conduct your own sample. The best practice is to take a RAT-STAT sample (the method adopted by the OIG) and, from that cell sample, take a smaller “probe sample.” The probe sample will give you a clear idea of error rates, but will not be statistically valid for projection to the universe. This is important, as the representative sample of the universe is disclosable. Knowing what the real risk exposure is likely to be and the ability to shoot down the demanding parties’ sample methodology will put an organization in the best possible position to move to proper resolution of the issue.