I was giving a presentation recently on the Comprehensive Error Rate Testing Program, or CERT. This is a study conducted annually by the Centers for Medicare & Medicaid Services (CMS) that, in effect, measures the error rate found for claims that have been previously approved and paid by Medicare.
For 2015, the CERT study reported on the findings of reviews for 49,603 claims that encompassed Part A, Part B, and DMEPOS services. Of these, they found that 12.1 percent were paid in error, of which 11.7 percent were flagged as overpaid and 0.4 percent, or $1.3 billion, were flagged as underpaid.
As in the past when giving this presentation, I didn’t even talk about the underpaid metric because it seemed mostly irrelevant based on both the rate and my own experience at adjusting underpayment. And that is how I continued to view this until quite recently.
I was working on a pre-audit risk analysis with a client, and in addition to reporting line-item overpayment risk, our application also reports line-item underpayment risk. It’s really not that complicated since predicting underpayment, at least using my model, is the same as predicting overpayment. The only difference is the sign; overpayments are negative values and underpayments are positive values. In this case, the underpayment potential was quite significant; in fact, over $2 million, and when I showed it to the client, their focus on this made it difficult to ignore.
In the course of the analysis, the client asked me a very simple question: could I recoup from Medicare the underpayments if I could show that the claims were, in fact, underpaid? His thesis was that if you were to audit a chart and found that the documentation and medical necessity supported a higher level of service, say a 99214 instead of a 99212, Medicare would be obligated to pay him the difference just as he would be obligated to pay back the difference if he found an overpayment. It was a good question, and not one into which I had put a lot of thought. But it made logical sense and got me to thinking about all of the audits for which I was engaged as an expert statistician, particularly extrapolation audits. In extrapolation audits, the auditor reviews some small sample of claims; say 30, and then, based on the overpayment findings of those 30 claims, extrapolates the results to all of the claims within the universe of claims from which the sample was drawn. For example, of the 30 claims, let’s say that the amount the auditor found to have been overpaid was 50 percent of the actual paid amount. In an extrapolation audit, the auditor could then demand repayment for 50 percent of all amounts paid to the provider during the period from which the sample was drawn.
CMS Ruling 86-1 states, in part; “the use of statistical sampling creates a presumption of validity as to the amount of an overpayment which may be used as a basis for recoupment. The burden then shifts to the provider.” This means that, while the process may be considered valid, it is up to the provider to propose objections that would mitigate or exclude the overpayment estimates. This is an interesting concept and at first seems pretty basic and clear. But when it comes to a provider conducting an internal audit for the purpose of voluntary disclosure or maybe payback under the 60-day rule, this concept becomes a bit more opaque.
So, the most basic question is this: Is extrapolation a valid statistical technique? And the answer is, at least from this statistician’s perspective and experience, heck yes. Extrapolation is an excellent way to infer the findings of a study from a small sample to a larger universe of homogenous data. The key to a valid extrapolation, however, is a valid sample, and as long as the sample is both statistically valid and representative of the universe, then an extrapolation may very well be appropriate. And I’m not alone in this. In Chaves County Home Health Service, Inc. v. Louis W. Sullivan, M.D., Secretary, Department of Health and Human Services, the judge for the D.C. Court of Appeals generally upheld the Secretary’s right to extrapolation as long as the sample was statistically valid and representative. The court stated the following: “Absent an explicit provision in the statute that requires individualized claims adjudications for overpayment assessments against providers, the private interest at stake is easily outweighed by the government interest in minimizing administrative burdens (emphasis added); in light of the fairly low risk of error so long as the extrapolation is made from a representative sample and is statistically significant (emphasis added), the government interest predominates.” Wouldn’t this be true in the case of an extrapolation for underpayments? I can’t imagine that the process to audit a claim for underpayment is that much more burdensome than auditing a claim for overpayment. In the end, at least from what coders tell me, and audit is an audit is an audit.
In Illinois Physicians Union v. Miller, the 7th Circuit Court of Appeals upheld the use of sampling audits to recoup Medicaid overpayments from participating physicians, squarely rejecting the contention that “any formula for sampling and extrapolation is improper per se.” In United States ex rel. Loughre v. Unamprovident Corp, et al., the court concluded extrapolation is a reasonable method for determining the number of false claims so long as the statistical methodology is appropriate. And there are many more such decisions and even though many extrapolation audits are thrown out because of statistical errors committed by the auditor, I have never seen a recoupment demand discharged based solely on the fact that it was the result of an extrapolation.
Okay. So we have established that extrapolation is a widely used and accepted technique for conducting government audits and there is substantial precedent to support this. And billions of dollars a year are recouped from providers using this technique. Then why, if extrapolation can be used to recoup overpayments without a claim-by-claim review, can’t a provider recoup underpayments from the government using the same technique? I think this is a very legitimate question, and I had an incredibly difficult time finding a legitimate answer. I spoke with several attorneys, a former AUSA, a former OIG special investigator and a number of compliance experts and they pretty much all said the same thing: To recoup the underpayment you would have to refile each claim individually and that you could only go back six months from the date the claim was paid. Yet, in State of Georgia v. Califano, the United States District Court concluded that the use of statistical samples was not improper and that, in fact, has been recognized and approved by federal courts in a number of cases. And this goes back to 1977! The court goes on to say that, “The weight which must be given to such statistical evidence is necessarily one which must be considered by the fact finder in light of the practical difficulties in obtaining a claim-by-claim review” and an individual review of every claim would be a practical impossibility as well as unnecessary. A practical impossibility because of the time and resources required to review, revise, and resubmit each claim and unnecessary because of this very well-established and widely accepted statistical technique called extrapolation. This is a conundrum that demands a solution!
The 2016 draft Statement of Work (SOW) for the Recovery Audit Program, under Section I. (Purpose), begins as follows: “The Recovery Audit Program’s mission is to reduce Medicare improper payments through the efficient detection and collection of overpayments, the identification of underpayments and the implementation of actions that will prevent future improper payments.” (emphasis added) In fact, it goes on to say that the contract includes looking for underpayments through a review of all claims that have a “high propensity for error based on the CERT program and other CMS analysis.” And the RAC can look for these underpayments for a period of up to three years retrospectively. This is amazingly interesting because, according to the 2015 CERT report, while procedure code 99233 is estimated to have been overpaid 26 percent of the time, procedure code 99231 is estimated to have been underpaid 6.2 percent of the time. And while 99214 is estimated to have been overpaid 6.1 percent of the time, code 99212 is estimated to have been underpaid 13 percent of the time. And amazingly enough, procedure code 99213 is reported to have been overpaid at a rate of 3.6 percent while also being underpaid at a rate of 2.3 percent.
What happens when the RAC finds an underpayment? Does the provider need to resubmit the claim in order to recoup the difference? Again, according to the SOW, “Neither the Recovery Auditor nor the AC may ask the provider to correct and resubmit the claim, although the Recovery Auditor shall issue an Underpayment Notification Letter including the claim(s) and beneficiary detail.” While I couldn’t find anything in the SOW that stated, specifically, that underpayments are used to offset overpayments, under Task 7C. (Contingency Fees), it states that “The contingency fee will be determined by the overpayments collected without consideration given to the underpayments identified (i.e. without netting out the underpayments against the overpayments).” So it sounds to me like that is exactly the case. MGMA published an FAQ page about the RACs and they begin by saying that “RACs will identify overpayments and underpayments. If a RAC finds both from a provider, it offsets the underpayment with the overpayment.” It also says that if a RAC finds an underpayment, it notifies the MAC, which will then “proceed with the claim adjustment and payment to the provider.” And this apparently happens without the provider having to resubmit the claim and for a period of up to three years.
In speaking to several experts, it appeared that there were different opinions about whether, in internal audit, such as for self-disclosure or voluntary payback, an underpayment could offset an overpayment and if so, for what period of time? If an audit can go back six years to extrapolate overpayments, does it make sense that for underpayments, there would be a limit of only six months to a year? The answer is NO, it does make any sense and in fact, I wonder whether this would stand up to a legal challenge. The importance of getting to the bottom of this is becoming more important with the recent rash of Comparative Billing Reports (CBRs) dispatched by eGlobalTech, a Medicare contractor. In the CBR, the provider is advised of both under- and over-utilization of specific services and while some state that the CBR does “not indicate the identification of overpayments,” others are a bit more specific. If the CBR indicates potential over-utilization, the practice might have a responsibility to self-audit and repay if necessary. But if the CBR indicates potential under-utilization and a self-audit reveals underpayment, will Medicare respond in kind? Obviously, we know the answer to be no. And it’s not just the feds. One of my clients recently received a CBR from a private payer and they stated that if the practice’s E&M coding patterns continued to deviate from their peer group, the payer will request a meeting at their office, a request for medical records, and a subsequent medical record audit. ‘Nuff said!
So, here we are, not too far from where we started this. And the question remains: If the government is so confident that extrapolation can fairly represent overpayment, such that they will collect millions of dollars from a provider by reviewing only a handful of claims, then why can’t that same technique be used for underpayments? Looking at the data, one might ask why the overpayment amount, then, is more important than the underpayment amount? If the true mission is to reduce errors, which is accounted for in both underpayments as well as overpayments, then wouldn’t applying the same extrapolation technique to both underpayments and overpayments achieve that mission in a more efficient and expeditious manner? That is, unless reducing error is not the real mission, and I suspect this is more in line with the truth. If the courts have overwhelmingly held that extrapolation is legal for recoupment against the provider, then wouldn’t they likely hold that the same is true for recouping underpayments from the government? I am far from being an attorney, but as a lay person, it makes a lot of sense to me. In all of my research, I could not find a single legal case where a provider sued the government to support the use of extrapolation for underpayments.
Well, maybe it’s time that happened. Perhaps one could use the “what’s good for the goose is good for the gander” defense.
And that’s the world according to Frank.
About the Author
Frank Cohen is the director of analytics and business intelligence for DoctorsManagement, a Knoxville, Tenn.-based consulting firm. Mr. Cohen specializes in data mining, applied statistics, practice analytics, decision support, and process improvement.
Contact the Author
Comment on this Article