May 3, 2011

Random Sampling, Extrapolation and Appeals: Core Components of Defending a RAC Audit

By

fcohen100In our recent survey of medical practices that have undergone Recovery Audit Contractor (RAC) audits, respondents reported that in more than 50 percent of cases random sampling was claimed to have been used by a RAC auditor.

Normally, there is only one reason to do this: to establish that there is a high and sustained rate of error, which authorizes the auditor to use extrapolation to determine overpayments. Yet only 21 percent of these same respondents indicated that they were aware that extrapolation had taken place. So either there is a communication problem that is occurring during the audit process or a there is a great deal of confusion over what constitutes extrapolation.

I have read many RAC audit letters and reports, so I can testify that contractors are not always clear regarding what was done and how. This is a very important component of the audit to understand, because if random sampling drives extrapolation, practices need to have an awareness of which (if either) are being applied to their specific audit.

In the past I have discussed how a practice can perform an initial randomness assessment by comparing the average (or median) paid amount per claim for the sample to the universe. Without getting into the details of how to conduct a statistically significant two-sample test, the purpose is to eye the figures and get help if the variance seems too large.

Lately, however, I have found some situations in which, when comparing these figures, the sample appeared random at first - but when I dug deeper and looked at the distribution of the codes being audited, I discovered a great deal of disparity.

Case in Point

In a recent case, a particular procedure code was ranked No. 50 (out of 220) in terms of frequency in the universe, yet this same code ranked No. 1 in the sample. It also had the highest average overpayment estimate. Of the 72 codes in the 30-claim sample, this code appeared 26 times, whereas if it was distributed based on the universe we should have seen it appear only four times.  So by having a representation order of magnitude greater than it should have been, the overpayment amount contributed more to the total than it should have.  In this case, the code contributed 65 percent more to the total overpayment estimate than it would have if it had been represented properly in the sample.

Another Example

In a similar case, we saw a procedure code ranked No. 1 in the universe that ranked last in the sample - and it also reported the lowest overpayment estimate. The same situation as that of the above case was present, only this time the overpayment estimate should have been considerably less because more of the distribution would have been absorbed by a lower overpayment per unit. In this case, had it been represented properly it would have reduced the overpayment estimate by 40 percent. So be aware, there are sneaky ways to make an otherwise random-looking sample turn up with an unexpected bias.

Stratification of Sample Frame

The other issue that practices deal with regularly during audits has to do with the stratification of the sample frame. "Strata" is short for stratification, and this is a method used to divide the claims universe into sections based on some specific criteria. Usually this ratio is provider-paid amount per claim.

Strata: An Example

In a recent case, strata were designated as under $100, $100 to $250, $250 to $500 and over $500. This means that those claims for which the total paid amount was less than $100 in essence were treated as a separate analysis. The same went for those claims falling in the other three strata. In this case, when I analyzed the full claims database I found that pretty much everything over $500 was a statistical outlier, meaning that these claims should not have been included in any type of extrapolation.

In another recent case, we found that the average overpayment amount per claim for an "outlier" stratum actually exceeded the average paid amount per claim, meaning that the practice would have been required to pay back more than they were paid.

The Stratified Sample

When a stratified sample is recommended or used, it is important to take a look at the universe first. One way to do this is to use a histogram-type graph to see what the distribution looks like. If you see a long right tail on the graph, many times this means there are a bunch of particularly high paid-amount-per-claim values on the right; these may be outliers and should be excluded from extrapolation. The only appropriate way to handle outlier audits is on a single-claim basis, with overpayment determinations limited only to individual claims.


Appealing Overpayments

The final issue has to do with appeals. Depending on what study you read, anywhere between 35 and 65 percent of all overpayment findings are overturned on appeal - meaning that if a RAC finds that 20 of your claims have been overpaid, you will get about 10 of them back on appeal. The downside is that only around one-third of practices are appealing, creating a huge shortfall in collectible data.

An Appeal that Worked

One of my clients recently underwent an audit of 30 claims, with the RAC finding 24 of them to have been paid improperly. The practice appealed each of the denied claims, and the result was that 20 were reversed back in favor of the practice. Unfortunately, though, the cost of appealing a claim sometimes can outweigh the value of the overpayment demand. Whether or not a practice chooses to do this has to be an individual business decision. But remember, every time a RAC gets away with improperly tagging a claim as being paid in error, it only motivates them to do this more in the future.

Best Practices

In summary, here are my suggestions:

1.   Make sure that your auditor provides you with enough information to replicate their study and findings. This includes what filters were used to scan from the universe of all claims to the sample frame.  Exclusions may include items like zero-paid claims, secondary payers, etc.

2.   Review the sampling methodology from different angles, including paid per claim, distribution of codes and claim types. Make sure the sample is random from all perspectives.

3.   Determine whether the auditor is using extrapolation. Ask specifically, and if they are not and instead claim to be pulling a random sample, find out why.

4.   Appeal every denial with which you disagree on the auditor's determination of overpayment. Pay particularly close attention when overpayment determinations are due to medical necessity issues.

About the Author

Frank Cohen is the senior analyst for The Frank Cohen Group, LLC. He is a healthcare consultant who specializes in data mining, applied statistics, practice analytics, decision support and process improvement.

Contact the Author

frank@frankcohengroup.com

To comment on this article please go to editor@racmonitor.com

Left in the Dark: The Blackout in Region A

Frank D. Cohen, MPA, MBB

Frank Cohen is the director of analytics and business Intelligence for DoctorsManagement, a Knoxville, Tenn. consulting firm. He specializes in data mining, applied statistics, practice analytics, decision support, and process improvement. He is a member of the RACmonitor editorial board and a popular contributor on Monitor Monday.

This email address is being protected from spambots. You need JavaScript enabled to view it.