While we can be concerned about the impact of such studies, you also can use this process before the RACs get a chance! If you find a potential problem, you can have such a study performed, identify overpayments (and possibly underpayments), repay overpayments and have your MAC remove the claims in the universe from consideration by your RAC. In other words, you can perform the extrapolation before the RAC gets a chance. While the results of your extrapolation and that of the RAC should be the same, the RACs tend to go overboard in their assessments and sometimes apply minimal guidance during the audit process. By using this approach, you can maintain a little more quality control, which, in the end, may be less expensive than going through an extensive appeals process.
In order to use the extrapolation process, a formal set of steps must be followed carefully. The problem or situation being addressed must represent a systematic error; that is, an error that repeats itself. Episodic errors, or, those that randomly occur, don’t lend themselves to this approach.
While there are different parameters that may be used, the process itself and the qualifications of personnel involved in this process are important. Here are the basic steps:
1. Identify the specific systematic problem for consideration.
2. Identify the universe of claims to be considered.
3. Engage a statistical expert to establish the statistical process.
4. Engage an Independent Review Organization (IRO) to conduct the audit.
5. Develop audit guidelines in cooperation with the IRO.
6. Perform a probe audit to determine the estimated error rate.
7. Calculate the sample size.
8. Randomly select the cases for the sample.
9. Conduct the audit on the sample.
10. Extrapolate the results of the audit to the overall universe.
11. Determine the amount to be repaid.
The statistician or mathematician will determine the appropriate parameters that will guide the overall process. For the case study we will consider, we will use the OIG’s suggested statistical standards of a 90 percent confidence interval with a 25 percent precision level. As indicated in step 6, a probe audit of 30 cases will be used to determine the estimated error rate. This error rate is important in order to use the formulas to determine a statistically valid sample size so that the results can be extended to the entire universe safely.
This may seem like black magic, but actually the process is conceptually straightforward. What is needed is a statistical program that can perform all of these calculations to the desired degree of precision. The OIG has developed such a program, called RAT-STATS. The program can be downloaded free of charge from the OIG’s Web site at http://www.oig.hhs.gov/; also, be sure to download the two manuals that come with it. Yes, the manuals are big, but you only will need certain parts of the program in order to perform extrapolation.
We will be using three different parts of RAT-STATS:
- Variable Appraisals – To determine the error rate for the sample size determination.
- Sample Size Determination – To determine the size of the sample for review.
- Random Numbers – To determine the actual cases for review.
Because there are many statistical terms and concepts included with it, be certain to experiment with this program to see how it works and identify all the things you can do with it.
Case Study – “-59” Modifier
Assume that during a preliminary RAC analysis at your hospital you have discovered that the “-59” modifier automatically is being attached by billing system personnel whenever a CCI edit is encountered. A review of a few claims indicates that documentation justifying the use of the “-59” is not present in all cases, and there also have been concerns that departmental personnel may be attaching the “-59” modifier arbitrarily. Also, some questions have been raised as to the chargemaster generating multiple codes that hit a CCI edit.
Note: The “-59” modifier is the ‘Separate Procedure’ modifier used by both physicians and hospitals whenever a CCI (Correct Coding Initiative) edit is encountered. By using the “-59” modifier, additional payment is gained and the physician or hospital attests to the fact that documentation that justifies the use of the modifier is on file.
Based upon this situation, you may be able to preempt what could be a major RAC investigation by conducting your own extrapolation study.
Now that the potential problem has been identified, you must identify the universe. In this case, assume that the universe encompasses 3,500 cases over a two-year time period. In actual practice, the number probably will be higher, but this is a reasonably tractable number for our case study.
Next, we must engage a statistical expert to study the issue and determine the proper statistical process to be used. While there definitely can be some variation, depending upon circumstances, we will use the OIG approach that is used when an organization is showing compliance within a Corporate Integrity Agreement (CIA). The process and parameters are as follows:
• 30-Case Probe Audit to Determine Error Rate
• 90% Confidence Interval
• 25% Precision
Again, in actual practice the statistician may require something more stringent, such as 10 percent precision or a 95 percent confidence interval. Making the parameters more stringent will cause the necessary sample size to increase, sometimes dramatically.
Next, an IRO, or Independent Review Organization, will need to be retained to conduct the audit and auditing guidelines will need to be developed. In cases of this type there will be some degree of subjectivity that must be left to the discretion of the auditors.
Using the Random Numbers feature of RAT-STATS, we will choose the 30 cases for the probe audit. Let us assume the results of the probe audit indicate that six cases were found in which the “-59” modifier was not sufficiently documented, with the average overpayment being $62.00.
Using the Variable Appraisals for the data from the probe audit, we have:
- Mean – 12.40
- Standard Deviation – 25.53
We now may use this information as input into the Sample Size Determination feature of RAT-STATS, along with the 90 percent confidence interval and 25 percent precision level. The sample size yielded is 174 cases: this is less than 10 percent of the universe, so the statistical expert may want to adjust the sample size upward. For our purposes we simply will use what RAT-STATS has generated. Here is a matrix of the various sample sizes: you can see how the sample sizes vary within the choices of precisions level and confidence level.
Continuing with our case study, the results of the audit of these 174 cases actually have turned out reasonably well. There were 40 cases in which there were overpayments, and the average overpayment was $65.00. We now take the results from the audit and run them through the Variable Appraisals feature of RAT-STATS; we need to know the upper and lower 90 percent confidence levels.
$64,197.00 < Upper 90% Confidence Level
$52,299.00 < Extrapolated Overpayment
$40,401.00 < Lower 90% Confidence Level
The overpayment amount is the lower of a one-sided 90% confidence interval using the OIG’s approach. In this case we arrive at $40,401.00.
When the overpayment is refunded through your FI or MAC, then this universe of claims can be excluded from RAC review.
Our intent was to provide a strategy and a simple case study to illustrate the process of using extrapolation, hopefully, to your benefit. There are many additional concerns we have not addressed. For instance, what if you identify underpayments as well as overpayments when using this process? Can you offset overpayments with underpayments? Also, how closely do you need to work with your FI or MAC? Certainly when you make repayments, you will need to consider interest. Thus, there are some operational challenges that are yet to be defined fully.
Also, if healthcare providers are to use this process, what confidence levels and precision requirements will be made? This whole process depends very heavily on a sound statistical basis for performing the extrapolation. Although the case study we have reviewed is fairly straightforward, there certainly exist advanced statistical variations including stratified sampling. For instance, you may be auditing for proper E/M levels within the different code levels; thus you will want to select a representative sampling within each code level.
Nonetheless, this process may offer a means of addressing certain types of situations preemptively. The RACs sometimes tend to be overaggressive in their assertions of overpayments, making going through the appeals process quite burdensome — thus, you may want to use this approach to avoid that process.
Good luck, and may your overpayments be minimal!
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.