While the specific numbers can vary depending on different situations, significant overpayments can be determined with relatively little effort on the part of auditors. However, there are some very real concerns with this process, and healthcare providers and suppliers that become enmeshed in it must fully understand the vulnerable points of the extrapolation process.

Are the RACs allowed to use this process? If so, will the RACs use this process?

The answer to this question can be found in the transcripts for the Special Open Door Forum held by CMS on Nov. 13, 2008. From page 20 of the transcript:

(Lisa Johnson): "And then during the time when the RAC is reviewing the facility and they review hypothetically 10 patient charts, will they then be extrapolating and estimating based on total amount of Medicare patients seen by that facility? Or is it per case per charts that they review?"

Melanie Combs-Dyer: "This is Melanie again. The Recovery Audit Contractors, both in the demonstration program and in the permanent program, have the ability to perform extrapolation. However, during the demonstration program, none of our demonstration RACs chose to follow the extrapolation process. We don't know yet whether any of our permanent RACs will choose to do extrapolation. But if they do, they will have to follow all of the same instructions that our regular carriers and FIs and (MACs) have to follow about choosing a statistically valid random sample using a statistician and all the other requirements for using extrapolation."

Clearly, the extrapolation process is available to the RACs. The forum held on Nov. 13, 2008 was for Part B, with a similar forum held the day before for Part A. Typically, the information addressing the extrapolation process has been on the Part B side. But is this same process available on the Part A side? In preparing for and monitoring RAC audits, be prepared for the possibility of the extrapolation process being used in any setting.

Note also that the demonstration RACs did not elect to use this process. Based on the significant overpayment amounts found in the demonstration project, the extrapolation process may not have been necessary. However, with the RAC program going nationwide on a long-term basis, the extrapolation process most likely will be used for certain types of cases.

Is the extrapolation process complicated?

Yes: the extrapolation process is a technical, statistical mechanism that demands careful preparation and analysis on the part of the auditors. For a given type of study that is to be replicated across a number of clinics or hospitals, the sample size for a given series must be determined in order to assure that the sampling is statistically valid. This requires a formal analysis on the part of a statistician or mathematician. This also plunges us into an extended discussion of issues such as error rate, precision and confidence intervals - yes, all of those wonderful mechanisms that you encountered in your statistic studies!

The good news is that this process can be streamlined by the use of statistical software such as the OIG's RAT-STAT: this is free software than can be downloaded from the OIG's website. There also is extensive documentation that runs into the hundreds of pages. In studying the RAT-STATS documentation and features, everything that is needed to perform extrapolation on a sound statistical basis is available. The OIG prefers a 90 percent confidence interval with 25 percent precision, as stated in connection with audits meeting Corporate Integrity Agreements (CIAs). The most difficult variable is the estimated error rate that often is addressed with a preliminary probe audit of 30 or more cases. From the probe audit we can determine a mean overpayment and a standard deviation that meets what the RAT-STATS program requires.

Let us consider a small case study.

Case Study 1 - A RAC has determined that there appear to be medically unnecessary hospital admissions in connection with cardiac catheterizations. Of particular interest are admissions of less than 48 hours. The RAC wants to identify the series as various hospitals (i.e., cases over a one- to three-year time period that meet the criteria), perform an audit on a statistically valid sample and then extrapolate the overpayments identified to the whole series at each hospital.

A statistician is retained to establish a valid process. In order to determine the error rate, a probe audit of 30 cases must be performed. Based on the results of the probe audit, the actual number of cases to be selected randomly from the series is determined within specified parameters by RAT-STATS. When the total extrapolated overpayment is calculated, the RAC is to demand the lower limit of a one-sided 90 percent confidence interval.

In Case Study 1, there is a single issue at stake, so random sampling is appropriate both for the probe audit and the final audit. Also note that the fully extrapolated overpayment amount is not to be demanded. A slightly lower amount is to be demanded based upon the statistical process used.

In some cases there may be a need to stratify the sampling. Let us consider a slightly different case, this time for E/M code levels.

Case Study 2 - The RAC is concerned that physicians may not be coding the correct E/M levels based on the 1995 and/or 1997 E/M coding guidelines. The RAC wants to establish an audit program that will allow for a statistically valid sample that can be extrapolated to an entire series.

There are several different types of E/M codes. Assuming a fairly simple case, there are five established patient visit codes, namely 99211-99215. In this case, the series may well consist of thousands of E/M levels, but there is a natural way to stratify or divide it, namely by the five different code levels. Thus, on a percentage basis, we may have the following distribution:

99211 - 10%

99212 - 18%

99213 - 36%

99214 - 23%

99215 - 13%

After the proper sample size has been determined, the sampling will be random within the different strata. For instance, if in our above distribution the sample size is 150, the cases will be randomly selected, with the different strata featuring the following numbers:

99211 - 10% * 150 = 15

99212 - 18% * 150 = 27

99213 - 36% * 150 = 54

99214 - 23% * 150 = 25 (Rounded Up)

99215 - 13% * 150 = 20 (Rounded Up)

These two fairly simple case studies are really just the beginning. Clearly, the extrapolation process can become quite complicated.

Where can you obtain additional information about the extrapolation process?

A fairly detailed discussion of this process can be found in Chapter 3 of CMS publication 100-08, the "Program Integrity Manual," in section 3.10. There is also a very good completion in Program Memorandum B-03-022 dated March 21, 2003. Much of the language in 3.10 was taken from the program memorandum.

What actions can providers and suppliers take relative to the types of audits that use extrapolation?

The very first thing that should be done is to fully understand the process. Do not be intimated by the language and what appears to be an irrefutable mathematical process. Go back to the fundamental process that is being used.

For healthcare claims and associated errors, there tend to be two different categories of errors:

• Episodic

• Systematic

Episodic errors and incidental errors include an incorrect keystroke, or when someone checks off the wrong box on a form. These errors are quite random and generally are not uniformly distributed through a given sample or series.

Systematic errors occur when there exists a mistake that occurs over and over. An example might be clinical personnel failing to understand how to enter charges for injections and infusions. Such misunderstandings will cause the same error to be repeated on a systemic basis.

Extrapolation works quite well if there are systemic errors that are replicated across the sample or series. Extrapolation does not work well with episodic errors, which are unpredictable. Also, if the sampling series covers a significant period of time, such as a year or two, there may be changes in the coding and billing system used by the provider, which will further change the nature of systemic errors.

While we will have to wait to see if extrapolation will be used by the RACs, and if so, to what extent, be prepared to challenge this process if necessary.

Significant overpayment amounts may be claimed based upon the extrapolation process using a relatively small audit sample. If or when you do have to challenge such a process, be prepared to retain your own statistical experts to study it. The manipulation of variables in choosing sample sizes can be critical within the overall context of the issue being addressed.

###

**About the Author**

Duane C. Abbey, Ph.D., CFP, is an educator, author and management consultant working in the healthcare area. He is president of Abbey & Abbey, Consultants, Inc. that specializes in healthcare consulting and related areas. His firm is based in Ames, Iowa. Dr. Abbey earned his graduate degrees at the University of Notre Dame and Iowa State University.

About the Author | EMail the Author