We will be undergoing testing for a few days starting 2009-10-22 so do not not consider the page completed until the end of October.
For comments, please contact Edward M. Roche
Medicare fraud is an important problem. Qualified Independent Contractors (QICs) and Recovery Audit Contractors (RACs) perform audits to recover over-payments made to health care providers. The extrapolated refund demand is the result of an analysis of a statistically random sample (SVRS) of provider claims. Coding analysis is used to determine a general error rate for claims submitted. This error rate then is extrapolated against the total universe of paid claims to create the final overpayment demand.
Forensic statistics can be used to expose errors made in the extrapolations leading to the overpayment demand.
RAC; Recovery Audit Contractor; CMS; Medicare Fraud; Audit; Overpayment; Overpayment Demand; Statistics; Forensic Statistics; Defense Strategy; Extrapolation
Refund demands are part of the process through which the Medicare Trust Fund is able to recover claims paid out due to either error or intentional fraud.
By any measure, Medicare Fraud is a significant problem:
According to the American Medical Association:
"Provisions of the Health Insurance Portability and Accountability Act of 1996 (HIPAA), Public Law l04-191, enhance the ability of the federal government to recoup health care spending that may be categorized as inappropriate payments, including fraud. Among other provisions, this law: created the Medicare Integrity Program that allows the Department of Health and Human Services (HHS) to enter into contracts with private entities to review and audit activities where Medicare provides coverage; earmarked funding to virtually double the number of HHS 0ffice of the Inspector General (OIG) auditors and investigators in addition to an expansion of the Federal Bureau of Investigation's (FBI) ability to investigate health care fraud; and established a monetary reward program to encourage Medicare beneficiaries to report virtually anything they believe to be questionable behavior."
This section focuses on the circumstances that gave rise to government action.
This section will outline the governmental authority on the topic, including both statutory and regulatory authority on federal and state levels. Relevant case law and administrative rulings will also be discussed in this section. Any federal register notices and preambles relevant to the topic should also be included in this section.
This section will summarize any guidance provided by federal or state government agencies and departments. Where possible, links to a file containing copies of relevant provisions/chapters/documents or articles should be included. Links to websites are discouraged as websites and links tend to be revised or overwritten.
This section is intended to anticipate future government action on the topic.
The road leading to an overpayment demand starts with 42 CFR 405. There we find specifications for a standard administrative procedure. Here is an overview:
The first step is the identification of which health care providers ("providers") deserve greater scrutiny. This identification can come through a variety of ways, including the reporting of a whistleblower. In general, the investigator is give a wide leeway in this part of the process, which is designed to be inherently flexible.
One of the more common ways to identify a suspect provider is by the application of computer operated algorithms that screen through millions of claims to detect patterns that may look like abuse or fraud.
Using a database containing the average billing for various medical procedures, vast data mining programs sift through the claim records of thousands of health care providers and identify those who present atypical billing patterns.
Although conceptually simple, in practice this is a difficult process. There is an apple and oranges problem. What exactly constitutes a “coherent class” of health care providers? How realistic is it to compare the billing practices? This actually may be one of the first problems with the ensuing audit; the targeted health care provider may not fit well into the class they are said to represent.
When this is the case, it may form part of the basis for an argument on behalf of the provider. However, since this issue relates only to the selection of possible providers to audit, and not to the results of an actual audit, it may not be successful in relieving responsibility for any fraud that has been found.
After the health care provider is picked as a target, a statistical analysis is performed on all of the claims they submitted during a fixed time period, usually 3-4 years. The purpose of this analysis is to identify a Statistically Valid Random Sample (SVRS) of claims, then pull them out for further analysis.
It is impracticable to analyze all claims of a provider. Therefore, a sample is taken of some claims.
Statistical theory says there is no need to examine all of the thousands of claims submitted by the health care provider. Providing the sample is valid, then a careful analysis should give a reliable picture of the overall universe of paid claims.
Once the SVRS has been identified, the Program Integrity Manual, known universally as “the PIM”, gives the auditor authority to request complete information on each claim in the sample.
When confronted with a demand letter, the next step is for the health care provider to pull together the complete records on each of the claims, get them photocopied, and then send them in for analysis.
There is no provision made in either the CFR or PIM for the expenses associated with fulfilling this request. For small practitioners, it may be the doctor themselves who do this work.
It is necessary to provide every record in the file.
After the claims arrive, they are analyzed one by one.
“Medical necessity” is one of the key screening criteria used. If the claim does not meet this hurdle, it then is considered to be an error, that is, an overpayment.
In addition to “medical necessity”, there are many other reasons why claims might be placed in the “overpayment” category.
Since claims are analyzed according to procedure codes, this phase in the process many times is called “coding review”.
After the coding review is completed, an overall error rate is determined. For example, if there were 100 claims in the sample, and 25 were found to have been overpayments, then the analyst would report out an “error rate” of 25 percent.
Once this “error rate” is established, it then is applied against the large “universe” of claims that have been paid to the health care provider. There are statute of limitations rules for how far back in time the auditor can go. Generally this is 4 years.
Here is an example: If Dr. X has been found to have an “error rate” of 25 percent, then 4 years of his claims would define the “universe”. How much was paid out? If in those 4 years, $12 million dollars has been paid out, a relatively simple calculation is taken:
In our example, since $3 million dollars is 25% of the $12 million received, Dr. X receives a letter demanding $3 million be returned.
Auditors are paid by being given a portion of whatever monies are recovered for Medicare. This number is calculated as a percentage, and can be as high as 12.5% of whatever monies are recovered. Here, this would be $500,000 dollars for the auditor, with the remaining $2.5 million being returned to the Medicare Trust Fund.
Theoretically at least, all of the parties benefit from this arrangement. The Medicare Trust Fund gets back monies that were wrongfully paid out. The auditor receives a commission to sustain its operations.
In fact, there is more subtlety to the statistical analysis than reported here. For example, most calculations are given with a number for their confidence interval, usually given as 10-90. But we will skip over these details here for the time being.
The only problem arises if Dr. X does not trust the analysis, or does not believe the process was fair.
For the vast majority of health care providers, the piles of supporting documentation, combined with what appears to be inpenetrable statistical analysis convince them to pay up without a fight. For others, the result may be different.
Many believe the analysis is wrong, but by simply comparing the amount in controversy against what they perceive to be the possible legal costs for filing an appeal, they go ahead and pay anyway, just to get it “off their back”.
Here we are concerned with those health care providers who decide to litigate. For them, the CFR has set out a clear procedure of appeal. The basic steps are as follows:
QIC -- Qualified Independent Contractor
ALJ -- Administrative Law Judge
For all practical purposes, the bulk of overpayment appeals are taken up only to the ALJ level.
Although there are many avenues for appeal, almost all rely upon a robust argument that the coding review of claims was faulty. In effect, the appellant argues that many of the claims classified as overpayments, that is as “errors”, were in fact justified.
ALJ hearings may go on for hours as each claim from the sample is discussed and reevaluated de novo.
Most if not all appellants use a specialist in coding to counter the analysis done by the Qualified Independent Contracter (QIC).
Coding analysis is the process by which an independent third party reviews each and every claim in the sample. Essentially, they re-do the work already done by the QIC auditor.
If the auditor has made any mistakes, these are identified.
There are many ways in which a coding analysis can help improve the prospects for the appellant health care provider. Regulations may have been wrongly interpreted; arbitrary decisions could have been made regarding medical necessity; some claims possibly were rejected on spurious or absurd grounds -- all of these and more can form the basis of the appellant’s argument. (NB: The provider is the appellant.)
The results of the coding analysis are then bundled together with any other basis for an appeal and then sent for redetermination. A review of the statistical analysis may be developed at this stage in the process.
After all of these documents are assembled into a record they are sent up to the next level in the appeal process. A neutral third party then revisits the entire matter. This is called the “redetermination”.
The results are published into three categories:
At this point, many health care providers settle; but what factors determine if an appellant settles or not?
One factor may be whether or not the auditor’s analysis is credible. In order to determine the credibility of the auditor's analysis, further work must be done. The underlying question is "Have they made their case?"
Some cases examined have found substantive and material errors.
Here is a list of a few of the more common problems:
According to the rules, the health care provider is owed a complete explanation of the analysis that is behind the refund demand. Certainly this process is not random, or subject to individual prejudices.
On occasion, the explanation may not be complete. It is not uncommon to see entire parts of the analysis eliminated altogether.
It is sometimes found that the explanations are misleading. For example, the explanation may make extensive reference to complicated statistical formulas, although they are not actually used in the analysis.
In other cases, complex statistical formulas may be listed, but without any explanation of how they actually were used.
In some cases, the auditor has simply provided a photocopy of 6-7 key statistical formulas from a popular textbook.
The implication is that these formulas and their underlying valid statistical methodology have been used in the estimation of how much is to be recovered in the refund. This, however, may not be the case.
It may be that the auditor has put in these formulas as a type of “window dressing” to convince the reader that underlying their refund demand lies a sophisticed statistical methodology.
Each auditor has a somewhat fixed methodology for its work. As a result, many explanations are composed of generic "boilerplate" language. It is though that they may be how some irrelevant information may be placed into the record.
It is rare that the judge can read or clearly understand these mathematical notions, and the same can be said of the judge’s staff, or most if not all of the attorneys involved. So the result is what would be expected: These formulas receive little scrutiny, and as a result are more likely to give the impression of a solid statistical analysis when in fact none may have been performed.
In some cases, the provider may find that the auditor has provided only a partial explanation, but has omitted to explain critical steps in the analysis. For example, the auditor may claim to be using certain formula, but never show how the formula were used.
Claiming to use valid statistical formula is not the same as actually showing that you did it.
The auditor is required to show their work. Merely providing a description of what the auditor claims was done is not sufficient.
Careful analysis by the provider may find instances of the auditor using the formulas improperly. By checking the spreadsheets showing the statistical analysis, it is possible to determine if the inputs used for the formulas are actually what is reqired, and not fake or manufactured data. These potential flaws can be detected only by re-working the statistical analysis underlying the refund demand.
The PIM requires that no matter what approach is taken, the auditor must use a valid statistical procedure. The term "valid statistical procedure" is repeated many times throughout the PIM.
Some cases have found evidence the auditors have skipped entire steps in the procedures specified by the PIM. For example, it may be possible to identify the taking of liberties in the sampling process, particularly as regards the underlying error rates.
In some cases, it has been found that instead of taking a probe sample to determine the underlying error rate in the universe of claims, critical assumptions are made regarding the underlying error rate, even before even a single claim has been subjected to analysis.
Since the error rate is a factor that determines the size of the sample required, this is a critical step in the process.
Some explanations may contain large amount of completely irrelevant information. For example, in one case, as part of their justification for the statistical procedure used the auditor included more than 100 pages of a photocopied software manual.
Technically, providing excessive information is not a violation of any rule. Nevertheless, it does place a burden upon the provider, since all of the informaiton must be analyzed. Under some circumstances, depending upon the amount of excessive information provided, a question of bad faith may arise.
It is difficult to predict the response from the ALJ as they realize this pile of irrelevant information has been inserted into the record, thus requiring a careful but unnecessary reading by the trier of fact.
In our example case, Dr. X was asked to return $3 million dollars. This was because the auditor determined there was a 25% claim error rate.
In reality, receiving a completely “favorable” result from the redetermination phase is rare. So lets assume that based on the redetermination, the underlying error rate was reduced from 25% to 12%. Now, Dr. X is asked to pay back a different amount: ($12 million)x(12%) = $1,440,000 dollars.
Did litigation make sense?
As a result of retaining an attorney to navigate the appeal process, the amount that must paid back has been reduced from $3 million to $1,440,000 dollars, a difference of $1,560,000 dollars.
Lets assume that the price of hiring the coding expert was $15,000 dollars, and the total charges from the attorney are around $28,000 dollars for a total of $43,000 dollars.
In this case, Dr. X has spent $43,000 dollars so as to reduce what he must return by $1,560,000 dollars. This amounts to a “litigation tax” of a little less than 2.9% -- a bargain!
We should note, however, that as the amount in controversy decreases, the litigation tax rises. At some point, litigation becomes such a high percentage of what is saved, that it might not be worth it.
For example, here $1,560,000 was saved and the litigation costs were around 2.9%. As the amount saved decreases, litigation costs become more expensive. If $877,500 is saved, litigation is about 5%. If only $277,646 is saved, then litigation is more than 15%. At $117,132 dollars, litigation is about 35%, about the same level as contingency work. As the amount falls further, then litigation becomes less cost effective. At $65,887 dollars, litigation is 65.3%. As we approach $43,000 dollars, the litigation costs are 100%. If less than that is saved, then litigation costs become greater than the benefits to the appellant. For example, if only $15,635 is saved, then litigation costs are 206%.
Of course, we have assumed that as a result of the “partially favorable” result from the redetermination, the level of errors has dropped from 25% to 12%.
Unfortunately, there is no reliable published information on the average percentage amount of reduction, and even if there were, there are so many variations in cases it likely would be meaningless. Anecdotally, a safer bet would be that the most errors will be reduced is about a third, and only under exceptional circumstances. So going back to our numbers, still the “litigation tax” would amount to less than 5 percent.
It follows that if there is an “unfavorable” result, then nothing is saved, and Dr. X is out by $43,000 dollars and still is obligated to pay back the $3,000,000 dollars.
As the amount in controversy balloons, the appellant can tolerate litigation that achieves a relatively smaller percentage reduction in what must be paid back. Another way to say this is that for smaller cases, the attorney must be more effective.
Some have found three tactics to be useful in reducing or eliminating altogether the overpayment demand:
Due Process: The attorney remains on alert for any failure in administrative procedures. Experience shows that these often come during the discovery process. QICs may appear reluctant to hand over their complete work unless asked to do so.
Coding Analysis: An expert should be brought in to re-examine the coding work done during either the initial or redetermination stage. The analyst is on the lookout for biased interpretations of the reimbursement rules, particularly as regards “medical necessity” and documentation.
Forensic Statistics: A mathematical statistics expert should be brought in to examine the underlying methodology used to find the initial sample of claims used to define the error rate and extrapolate the overpayment demand.
Based on experience in using forensic statistics to defeat overpayment demands, some have developed a check-list of seven things the legal team can do to obtain for their client the best possible outcome.
It is a common error to ignore the statistical side and rely exclusively on coding analysis.
Coding analysis is absolutely essential, but also “incremental” in nature -- it will allow Dr. X in each round to get a reduction in the error rate, and thus he will have to pay less.
Forensic statistics works differently. Practice has shown that many statistical mistakes made are extremely serious, and if exposed can work wonders for the appellant. Instead of getting incremental reductions, it works to eliminate altogether the validity of the extrapolation. When this strategy is successful, Dr. X may end up being responsible for payment of the overpayments uncovered in the sample, but the extrapolation to the overall universe of claims – a much larger number – will be excluded.
The best strategy uses both types of attack and this doubles the chances of delivering a substantially better outcome.
Although the C.F.R. provides for “good faith” in discovery, your actual case may fall short.
Many audit contractors tend to provide “boiler plate” for much of their documentation and may skip on providing details explaining how they used statistical procedures.
As a consequence, your discovery requests must be comprehensive and leave no wiggle room for evasion. Here is a reasonable example of a workable discovery request:
[A]ll statistical information, formulas, applications, spreadsheets, samplings and extrapolations used in every step of the [Dr. X ] case.
[A]ll statistical information, formulas, applications, spreadsheets, samplings and extrapolations used in every step of the [Dr. X ] case.
By ensuring that all relevant information is requested up front, any failure by the QIC to provide such information raises a “good faith” question.
It is common for audit contractors to respond to discovery requests by providing paper print-outs of their work.
Should this occur, it may be extraordinarily burdensome on the provider:
Paper copies may burden your client with thousands of unnecessary dollars in expenses because it complicates the statistical forensics process.
If any electronic (spreadsheet) data is received during the discovery process, confirm that it is not password protected. An additional consideration is to check that the “hide column” spreadsheet software feature is not covering up critical columns in the spreadsheet. This problem may occur inadvertently.
The attorney always holds the threat of filing for sanctions if discovery requests are not met, but in practice this rarely is allowed by the ALJ.
The attorney may find it very useful to weed out all “boilerplate” information before handing it out to the experts to analyze. This is done as soon as materials arrive from the audit contractor. Here is how one expert put it:
“Many times more than 95% or more of the materials received from the audit contractor are boilerplate. After a while, you start to recognize the same stuff over and over. In once case, they even supplied a copy of Chapter 3 of the RAT-STATS manual as ‘evidence’ of the quality of their work.”
By identifying these materials ahead of time, you can save your client significant money before the case is handed over to the forensic statistics team.
In some instances portions of the PIM may be quoted selectively. For example, a popular quotation points out that “the statistical sample is not a basis for an appeal”.
This convenient quotation leaves out important other accompanying language that when examined may lead to a completely opposite rule.
Due diligence requires that each and every instance of a PIM quotation be carefully analyzed to ensure the audit contractor has not shaded the meaning or quoted without full context.
Getting a clear understanding of the mathematical statistics behind the QIC’s work is the most crucial element to defeat an extrapolated refund demand. Doing this is challenging and can be entrusted only to a trained expert possessing a great deal of experience.
Always keep in mind the following: Usually very few parties involved have a sufficient knowledge of statistics. This includes your client, yourself as the attorney (unless you are an exception), the QIC, and almost certainly the ALJ.
When faced with the dozens of pages of statistical analysis and boilerplate, the most common reaction is to wave it through and not question it.
Only a trained eye can see the cracks in the façade.
The key to discrediting a extrapolated overpayment demand is to show that the audit contractor failed to use a “Statistically Valid Random Sample” when it picked out the claims that later would be subjected to the auditor’s coding analysis.
Here are only a few of the many anomalies that might be detected in analysis of audit contractors’ work:
No Documentation: In some cases, even though as many as 100 pages may be supplied, in fact the statistical procedures used are not even documented. The forensic statistician is left completely in the dark about what has transpired.
“Bait & Switch” Variables: Many times the audit contractor will describe a process in which one variable is used as a so-called “proxy” for another. Although this usually is explained in what at first appears to be a convincing way, switching variables always raises a red flag and usually invalidates the analysis.
Biased Assumptions: Always look for any statistical or numerical assumption made in the analysis. Behind every number is an assumption that has an effect on the provider's case. Understanding all assumptions and how they are made is a key to unlocking any material errors in the analysis. For example, it is sometimes possible to find an assumption of a 100% error rate built into an analysis before even a single claim has been analyzed.
Those are only a few of the points to look for. There are many others that can be discovered by a trained expert.
It can be argued that much more can be saved if a strong statistical challenge is made to that initial high number.
The bottom line is this: Rather than focusing merely on reducing the value of the initial extrapolated demand number, instead, focus on getting it invalidated altogether.
The only way to do this is through a statistical attack.
When you are successful, your client will become responsible only for those claims that actually have been examined, not for an entire universe of unexamined claims.