After receiving an enormous demand for reimbursement based on a statistical extrapolation, it may be possible to get the extrapolation thrown out by the Administrative Law Judge (ALJ) for a Medicare claim reversal.
If this occurs, you won’t have to pay the large extrapolated amount, but may need to pay only for any individual claims that have been ruled to be invalid, an amount that usually is much smaller.
How do you get this situation to be reversed in your favor? Barraclough Health does it by using an accurate statistical methodologyrather than let the results of the inaccurate methodology used by RACs stand.
In this Barraclough Health Medicare Claims Reversal Case Study, we show how using our statistical methodology saved the client $1,297,700 dollars.
Medicare ZPIC Claims Demand
Dr. X received a Medicare reimbursement demand for approximately $1,300,000 dollars.
But the auditor (the ZPIC) had examined only 35 of the thousands of Dr. X’s claims.
Of these 35 claims, they had rejected 17 of them.
The value of those claims was approximately $2,300 dollars.
Dr. X then contacted Barraclough LLC.
Barraclough’s VALID Statistical Methodology
Barraclough’s expert team completed an extensive analysis of the 4 statistical methodology used by the ZPIC, a formidable task.
The Barraclough team checked:
all of the calculations that were made
the details of how the sample was taken
the formulas that were used in picking the sample size
By doing this extrapolation, the general pattern of decisions was revealed.
The Barraclough team found that the contractor had made many errors in their work, including:
Manufactured data was an essential part of the calculations used in determining the needed sample size.
Picking sample sizes is not an arbitrary act – there should be a method behind it.
One of the formulas that is needed to determine sample size requires an input representing the underlying variation in the variable being estimated.
For an over payment analysis, this would mean that it is necessary to understand the underlying variation in the over payments.
The only way to get this information is to analyze a number of claims to make a measurement.
But the contractor had skipped this step entirely, and had simply manufactured this number, plugged it into the formulas, and decided to use a sample size of 35.
When the Barraclough Team double-checked the calculations, we put the correct number into the same formulas and found that the required sample size was more than 100 times larger.
During the hearing the contractor admitted that they had failed to make the required measurement.
After finding many other problems with the statistical work, it was possible to conclude that the contractor had failed to use an acceptable statistical methodology.
ZPIC Work Falls Short of Standard
Since the MPIM (Medicare Program Integrity Manual) requires that a valid statistical methodology must be used, we were able to show that the work of the ZPIC had fallen short of the standard.
Result: Medicare Claim Reduced Significantly
The extrapolation was thrown out by the ALJ.
Instead of having to pay the extrapolated amount of approximately $1,300,000 dollars, Dr. X. ended up paying approximately $2,300 dollars, a difference of $1,297,700.
So we thought we would reproduce the ruling here, but with only the portions that deal specifically with statistical sampling and extrapolation.
Hospital Insurance and Supplementary Medical Insurance Benefits (Parts A and B) Use of Statistical Sampling to Project Overpayments to Providers and Suppliers
Purpose: HCFA and its Medicare contractors may use statistical sampling to project overpayments to providers and suppliers when claims are voluminous and reflect a pattern of erroneous billing or overutilization and when a case-by-case review is not administratively feasible.
. . . As result of a subsequent audit of the provider’s Medicare claims, the intermediary discovered a large number of bills for medically unnecessary services. . . . The cost of identifying and calculating each individual overpayment itself would constitute a substantial portion of the amount the intermediary might reasonably be expected to recover. . . .
The intermediary notified the provider that, because of the volume of records and the costs of retrieving and reviewing all records for the period as discussed above, it intended to project the overpayment by reviewing a statistically valid sample of beneficiary records and that if it were determined that the provider had been overpaid for the sample cases, it would project the results (again using statistically valid methods) to the entire population of cases from which the sample had been drawn. This would result in a statistically accurate estimate of the total amount the provider had been overpaid for services to these beneficiaries.
[Next are the complaint items filed by the health care provider.]
The provider objected to the intermediary’s use of sampling to project the overpayment on the following grounds:
1. There is no legal authority in the Medicare statute or regulations for HCFA or its intermediaries to determine overpayments by projecting the findings of a sample of specific claims onto a universe of unspecified beneficiaries and claims.
2. Section 1879 of the Social Security Act, 42 U.S.C. 1395pp, contemplates that medical necessity and custodial care coverage determinations will be made only by means of a case-by-case review.
3. When sampling is used, providers are not able to bill individual beneficiaries not in the sample group for the services determined to be noncovered.
4. Use of a sampling procedure violates the rights of providers to appeal adverse determinations.
5. The use of sampling and extrapolation to determine overpayments deprives the provider of due process.
(The succeeding presentation of our decision and supporting facts is applicable also to the use of sampling to project overpayments to suppliers (including physicians) whose claims are processed by Medicare carriers when 100 percent readjudication would be excessively costly or impractical.)
The Supreme Court has long recognized that the Federal Government possesses an inherent right to recover monies illegally or erroneously paid out. . . . The Government’s common law right of recoupment, and its corollary power of recovery by offset, are based on strong considerations of public policy. . . . The common law right to recover Federal funds has been specifically recognized as being fully applicable to the Medicare program. . . . Congress has affirmed the Government’s right to recover Medicare Trust Funds by reasonable means from those who have no right to retain them. . . .
[Next, the ruling makes the “administrative burden” argument.]
Since HCFA’s contractors process vast numbers of Medicare claims . . . A case-by-case review could require a significant diversion of staff from the ongoing claims process, and the cost of determining the amount of an overpayment would be prohibitively high unless a sampling method were used. . . .
We also do not believe that the statutory provisions limiting provider or beneficiary liability preclude the use of sampling.. . .
The use of sampling to determine overpayments for medically unnecessary services or custodial care does not deprive a provider of its right to bill those beneficiaries who knew or should have known that they were receiving these services. . . .
[There are also “public policy” used to justify sampling.]
As between the provider and the Government, strong considerations of public policy favor recovery.. . .
[Next, the famous shifting of the burden of proof is explained.]
Sampling does not deprive a provider of its rights to challenge the sample, nor of its rights to procedural due process. Sampling only creates a presumption of validity as to the amount of an overpayment which may be used as the basis for recoupment. The burden then shifts to the provider to take the next step. The provider could attack the statistical validity of the sample, or it could challenge the correctness of the determination in specific cases identified by the sample (including waiver of liability where medical necessity or custodial care is at issue). . . . If certain individual cases within the sample are determined to be decided erroneously, the amount of overpayment projected to the universe of claims can be modified. If the statistical basis upon which the projection was based is successfully challenged, the overpayment determination can be corrected.
The provisions of the statutes and regulations provide a constitutionally sufficient means by which the provider may challenge an overpayment determination. In cases of denials made through sampling which are based on medical necessity or custodial care, section 1879 of the Act, 42 U.S.C. 1395pp, permits the provider to assert the same appeal rights that an individual has under the statute when the individual does not exercise his rights to appeal. Under Part A, these rights include an opportunity for reconsideration (42 CFR 405.710- 405.716), an oral evidentiary hearing by an administrative law judge (42 CFR 405.720-405.722), Appeals Council review (42 CFR 405.701(c) and 405.724), and finally judicial review if the amount in controversy is $1,000 or more (42 CFR 405.730; 42 U.S.C. 139 5ff (b)(2)). In cases that do not involve medical necessity or custodial care, 42 CFR 405.370, et seq. sets out the applicable procedures through which current payments may be suspended (offset) to recover an overpayment under the Medicare program. . . .
In summary, the use of sampling is a reasonable and cost effective method of projecting overpayments under Medicare. It is not unfair to a provider or supplier to hold it accountable for the receipt of Medicare funds to which it is not entitled under the statute. . . .
Ruling: Accordingly, it is held that the use of statistical sampling to project an overpayment is consistent with the Government’s common law right to recover overpayments, the Medicare statute, and the Department’s regulations, and does not deny a provider or supplier due process. Neither the statute nor regulations require that a case-by-case review be conducted in order to determine that a provider or supplier has been overpaid and to determine the amount of overpayment.
Effective date: This Ruling is effective February 20, 1986.
So the question that arises is this: Are contractors free to employ any accuracy they wish in their work, or are there standards that have been suggested or published by the Federal Government?
As it turns out, there appears to be some guidance from two sources.
In the May 5, 2010, report by the Acting Administrator and Chief Operating Officer of the Centers for Medicare & Medicaid Services (CMS) On page 3 of that report, the section titled “Precision-level requirements” states:
“[Office of Management and Budget] OMBCircular A-123, Appendix C, states that Federal agencies must produce a statistically valid error estimate that meets precision levels of plus or minus 2.5 percentage points with a 90-percent confidence interval or plus or minus 3 percentage points with a 95-percent confidence interval.”
There is a note in the document: Under these assumptions, the minimum sample size needed to meet the precision requirements can be approximated by the following formula, which is used in the examples:
Where n is the required minimum sample size and P is the estimated percentage of improper payments (Note: This sample size formula is derived from Sampling of Populations: Methods and Applications (3rd edition); Levy, P. S. & Lemeshow, S. (1999); New York: John Wiley & Sons; at page 74. The constant 2.706 is 1.645 squared.
In the CMS-issued Federal Register, 72 Fed. Reg. 50490, 50495 (Aug. 31, 2007), the error estimate should meet precision levels of plus or minus 2.5 percentage points with a 90-percent confidence interval, and the State error estimates should meet precision levels of plus or minus 3 percentage points with a 95-percent confidence interval.”
So it appears that these standards, which are fairly good, have been twice promulgated by the Federal Government.
Source: Memorandum Opinion re: Parties’ Cross-Motions for Summary Judgment, files 12/28/2012, and Barraclough analysis.
A number of arguments were made that established clearly that the statistical work was faulty, and from a scientific point of view was completely invalid.
Arthur J. Schwab the United States District Judge wrote in his opinion “Balko is not entitled to the best possible statistical sample of claims that it submitted . . . Instead, Balko is only entitled to a statistically valid random sample.” (Memorandum Opinion, p. 23.)
Question: Is a “statistical valid random sample” one that is so poor that it lacks any scientific credibility?
What has happened in this case does not bode well for health care providers. Here, statistical work that is demonstrably faulty and inferior and definitely not scientifically valid has been signed off on by the Medicare Appeals Council (MAC), and by the Federal Court that reviewed the case.
This type of sloppy scientific work never would be accepted in any other type of case before a Federal Court in which scientific evidence is evaluated in conformity with Rule 702 “Testimony by Expert Witnesses” of the Federal Rules of Evidence. The question is why is this type of poor and inadequate scientific work OK for audits of health care providers but not OK anywhere else?
Barraclough NY LLC supplies experts for litigation support in Medicare and Medicaid appeals cases.
John Balko & Associates d/b/a Senior Healthcare Associates, Plaintiff, v. Kathleen Sebelius Secretary U.S. Department of Health and Human Services, defendant. United States District Court for the Western District of Pennsylvania. Case 2:12-cv-00572-AJS.
This is a case in which the extrapolation was thrown out.
Here, the Medicare Appeals Council agreed with the “determination that the sampling was sufficiently flawed to preclude calculation of an overpayment by extrapolation”.
Although the appellant had made a number of arguments attacking the statistical extrapolation, the MAC relied on two errors in throwing out the extrapolation:
“The errors are: 1) the PSC provided the independent statistical expert with sample data which assigned some claims to the wrong stratum; and 2) the PSC provided the independent expert with a second CD containing an Excel set of sample data with significant discrepancies from the first set of data, and the PSC was unable to clarify the discrepancies, to identify which set of data was applicable, or to explain the significance of the second set of data.”
This provides at least three check points when providing litigation support to a health care provider:
First, always have the statistical expert carefully check that all claims in any strata strictly fit the definition of the strata;
Second, look for any instance in which inconsistent records have been handed over by the contractor; and
Third, demand detailed explanations from the contractor for each and every inconsistency found in the data.
Unfortunately, it has been our experience at Barraclough that these arguments do not always result in an extrapolation being thrown out. Rulings are inconsistent with each other – sometimes this argument works, sometimes it does not.
It has been our experience at Barraclough that contractors almost always skip the step of taking a probe sample when calculating the required sample size. Even though they do this, they frequently rely on RAT-STATS to make the sample size calculations. The inputs into RAT-STATS requires the variation in the variable being estimated, that is, RAT-STATS requires as one of its crucial inputs the variation (e.g., the mean and standard deviation) of the overpayments (which is the variable being estimated). Since the contractors skip taking a probe sample, they plug the wrong data into the RAT-STATS program, and then make their calculation of sample size by using the variation of the payments instead of the underpayments. This almost always results in RAT-STATS claiming that a smaller sample size is adequate. In the MPIM, Chapter 3, Section 3.10.2, we see a sketch of what a “properly executed” sample design is. In includes:
(1) defining the universe, (2) [defining] the frame, (3) [specifying] the sampling units, (4) using proper randomization, (5) accurately measuring the variables of interest, and (5) using the correct formulas for estimation
It can be argued that taking a probe sample so as to be able to plug the correct (and required) data into the RAT-STATS program falls under the fifth category “accurately measuring the variables of interest”. It follows that if the probe sample is not taken, then according to the MPIM, proper procedures have not been used. Note: RAT-STATS is a free statistical software package that providers can download to assist in a claims review. The package, created by OIG in the late 1970s, is also the primary statistical tool for OIG‘s Office of Audit Services.
“Failure by the [contractor] to follow one or more of the requirements contained herein does not necessarily affect the validity of the statistical sampling that was conducted or the projection of the overpayment.
An appeal challenging the validity of the sampling methodology must be predicated on the actual statistical validity of the sample as drawn and conducted.
Failure by the [contractor] to follow one or more requirements . . . should not be construed as
necessarily affecting the validity of the statistical
sampling and/or the projection of the overpayment.”
The language quoted from the MPIM seems to indicate that no appeal may be based on the sampling methodology.
No matter how the contractor arrives at their sample, it does not seem to be reviewable.
Statistical extrapolation in Medicare and Medicaid audits can be problematical. They are not always done correctly, and it actually has been our experience that they frequently are not done correctly at all.
Steven E. Skwara identifies three challenges that can be made against statistical sampling in health care fraud cases:
Reproducibility. If the results can not be reproduced, then there is reasonable argument that the results are not scientific. Documentation is crucial.
Sample Size. Larger sample, more accurate results. There is a great deal of leeway given by Medicare courts.
Variability. This is not often looked for, but a high degree of variability in the data may signal problems. Ask your statistician.