Background and Fundamentals of the Use of Statistical Sampling in the FCA and Medicare Appeals Arenas
Origins of Sampling and Extrapolation
CMS Ruling 86-1
Statistical sampling and extrapolation originated in the administrative arena, and in 1986, the Health Care Financing Administration (HCFA), the predecessor to the Centers for Medicare & Medicaid Services (CMS), issued CMS Ruling 86-1, which was the first ruling that allowed a fiscal intermediary to use sampling and extrapolation in place of a claim-by-claim review. CMS has relied on this decision to justify the use of statistics to support a demand for repayment of claims billed to federal healthcare programs, which set off an evolution of how data analytics would be used across the legal spectrum of administrative, civil, and criminal law.
The acting administrator for HCFA determined in CMS Ruling 86-1 that a contract auditor was permitted to use sampling and extrapolation as opposed to individual claim-by-claim review because the government had a significant interest in the cost-effective recovery of improper payments and, even though there was no express authorization for extrapolation, there was also no express prohibition. Despite citing no statutory or regulatory authority within the ruling, CMS Ruling 86-1 asserted that the use of sampling and extrapolation grew out of the government’s “federal common law right” to recover property.
The ruling also set forth that providers were not denied due process because of their ability to appeal extrapolated findings through the administrative appeals process. CMS emphasized that providers should and would have a fair opportunity to contest adverse determinations based on statistical sampling: “Sampling does not deprive a provider of its rights to challenge the sample, nor of its rights to procedural due process.”
Medicare Integrity Program
In 1996, Congress created the Medicare Integrity Program to strengthen the Secretary of the Department of Health and Human Services’ ability to deter fraud and abuse. As part of this “Medicare Integrity Program,” Congress authorized the Secretary to “use extrapolation to determine overpayment amounts to be recovered,” but, in approving the Secretary’s use of extrapolation in determining Medicare overpayments, Congress provided limited guidance on how extrapolation was to be performed. However, there has always been a requirement set forth in 42 U.S.C. § 1395ddd(h) for contractors to identify underpayments as well as overpayments.
Medicare Program Integrity Manual
In 2000, to carry out the congressional objective on using extrapolation, CMS set forth its Medicare guidelines for statistical sampling and overpayment estimation in the Medicare Program Integrity Manual (MPIM). The MPIM, Chapter 8, Section 4, provides detailed requirements for CMS contractors in developing an audit plan and executing the sampling and extrapolation process. As explained further below, the MPIM includes guidelines regarding preservation and documentation of the universe as well as the evaluation and inclusion of zero-paid claims.
Statistical Sampling in the FCA Arena
In the FCA arena, data analytics are used to find fraud, and over the last decade, statistical sampling has transitioned from a mechanism primarily used in calculating damages to a primary tool used to support and validate the core elements of FCA liability. In FCA litigation, the burden is on the government to prove that the statistical sampling plan utilized was factually sound.
Government-Initiated FCAs Based on Data Analysis
Recently, the federal government has dramatically increased its use of data analytics methods to identify outlier providers, support damages, and allege liability for FCA violations. Although data analytics were more frequently used in the civil context, the DOJ’s Criminal Division’s Health Care Fraud Unit launched a new data analytics team in 2017 with the primary goal of using data analytics to identify and prosecute fraud under the FCA. The greater degree of government sophistication with data analytics has led to more government-initiated (i.e., non-relator) FCA cases.
For example, in 2021, the U.S. Attorney’s Office for the Eastern District of Pennsylvania secured three FCA settlements with providers regarding P-Stim electro-acupuncture device usage after identifying these providers through the U.S. Attorney’s Office’s own data analytics mechanisms. These settlements totaled nearly $2,000,000. In collaboration with other government agencies, the U.S. Attorney’s Office was able to identify inconsistent and improper billing methods to support their demands and as evidence of liability.
Government Collaborations for Claims Data Analysis
The most centralized entity in the healthcare fraud analysis arena throughout the U.S. healthcare system is the CMS Center for Program Integrity (CMS-CPI). CMS-CPI is a specific division of CMS that oversees all CMS interactions and collaborations with stakeholders relating to program integrity, including the DOJ, U.S. Department of Health and Human Services Office of Inspector General (HHS-OIG), state law enforcement agencies, and other federal entities. The purpose of CMS-CPI’s collaboration with these other entities is detecting, deterring, monitoring, and combating fraud and abuse, as well as taking action against those that commit or participate in fraud. CMS-CPI has also developed initiatives such as the Health Care Fraud Prevention Partnership, which is a group of over 240 private and public partners focused on data and information sharing, as well as the Major Case Coordination program, which is a collaboration between CMS-CPI, HHS-OIG, and DOJ that has led to large-scale enforcement actions.
Additionally, the Medicare Fraud Strike Force, established under HHS-OIG, and DOJ’s Health Care Fraud Strike Force, were created to harness data analytics through federal, state, and local resources. Most recently, the COVID-19 Fraud Enforcement Task Force, comprising the civil and criminal divisions of the DOJ, the Executive Office of U.S. Attorneys, and the Federal Bureau of Investigation, has been using data analytics to identify public health emergency-related fraud through data mining.
Limitations of Statistical Sampling
Statistical sampling has historically been used in FCA litigation to determine damages, and in United States v. Cabrera-Diaz, statistical sampling was permitted for supporting computation of damages for default. However, in United States ex rel. Loughren v. UnumProvident Corp., a bellwether jury trial was required before statistical sampling was permitted to extrapolate the total number of false claims for the purpose of determining damages.
More recently, the government has started making a concerted effort towards using statistical sampling methods to prove liability. In United States v. Life Care Centers of America, the court explicitly held that statistical sampling could be used to prove liability in Medicare overpayment claims brought under the FCA when claim-by-claim review is “impracticable.” However, it would be up to each fact-finder to determine how much weight to give to the use of statistical sampling and extrapolation.
Additionally, some circuit courts have held that statistical sampling cannot be used to prove liability in instances where payment hinges on a patient’s medical necessity, as statistical sampling cannot substitute expert medical judgment; however, circuit courts remain split on this issue. In Life Care Centers of America, the court determined that when all patients’ medical charts were “intact and available for review by either party,” the use of statistical sampling was not appropriate, even though the available medical records were voluminous. Moreover, in United States ex rel. Michaels v. Agape Senior Cmty., Inc., the court held that the use of statistical sampling must be limited when “a thorough review of the detailed medical chart of each individual patient” is required, such as when a physician must use “subjective clinical judgment” to determine a patient’s life expectancy for the purpose of determining hospice eligibility.
Until the courts adopt a more uniform approach on how statistical sampling and extrapolation are utilized in FCA claims, defendants must be prepared for the possibility of courts permitting multiple types of statistical methods.
Statistical Sampling in the Medicare Appeals Arena
While CMS is the agency responsible for administering the Medicare program, CMS has contracted with various private entities to assist CMS in processing and auditing claims for reimbursement and in the appeals process itself. Unified Program Integrity Contractors (UPICs) have become the primary vehicle for CMS to investigate and data-mine for fraud in Medicare and Medicaid claims processing. UPICs perform integrity work with Medicare Parts A and B, durable medical equipment, home health and hospice, Medicaid, and the Medicare-Medicaid data match program.
UPICs are authorized to: (1) prevent fraud by identifying program vulnerabilities; (2) proactively identify incidents of potential fraud, waste, and abuse that exist within its service area and take appropriate action; (3) investigate allegations of fraud; (4) explore available sources of fraud leads in its jurisdiction; (5) initiate appropriate administrative actions where there is reliable evidence of fraud, including, but not limited to, payment suspensions and revocations; and (6) refer any necessary provider or supplier to the provider outreach and education staff at the Medicare Administrative Contractor (MAC). As explained further below, UPICs are tasked with identifying both overpayments and underpayments. One of the primary tools that UPICs and other Medicare contracting entities use in performing their duties is statistical sampling and extrapolation, as described in the MPIM.
If a Medicare provider wishes to appeal claims denials, they are subject to the lengthy appeals process set forth in 42 U.S.C. § 1395ff. The Medicare appeals regulations allow for five levels of appeal: (1) redetermination, (2) reconsideration, (3) administrative law judge (ALJ) Office of Medicare Appeals (OMHA) hearing, (4) Medicare Appeals Council (Council) review, and (5) federal district court review. Partially favorable decisions on the individual sample claims at any level of the appeal require recalculation of the extrapolated overpayment demand. Statistical sampling and extrapolation can be invalidated at any appeal level, but appellants must assert the reasons they disagree with how the statistical sampling and/or extrapolation was conducted at each appeal level to preserve the arguments.
Due Process Challenges to Statistical Sampling and Extrapolation
Two particular due process challenges to statistical sampling and extrapolation warrant attention for potential success at the federal level. Each of these due process challenges has shown recent success at the ALJ and/or Council levels of the Medicare claims appeal process. These due process challenges are: (1) failure to produce the universe of claims, and (2) failure to include zero-paid claims in the sampling frame. Before delving into these challenges, we must briefly orient ourselves to the elements of a due process claim in the administrative context.
The U.S. Supreme Court has indicated that there are two steps to a viable claim for violation of procedural due process in the administrative context. Step one is “to identify a property or liberty interest entitled to due process protections[.]” In the Medicare claims appeal context, courts have expressly recognized that Medicare “beneficiaries have a protected due process ‘property interest’ in ‘receiving the medical insurance benefits for which they paid a monthly premium.’” While beneficiaries are the primary parties in interest to the Medicare program, providers and suppliers are likewise parties in interest “as assignees of the beneficiaries” and, therefore, have a valid property interest in receiving Medicare payments for services rendered.
Once step one is completed and it is determined that due process protection applies, the second step is to determine what process is due that has not been met. It is well-settled that “[t]he fundamental requirement of due process is the opportunity to be heard at a meaningful time and in a meaningful manner.” The requirement that a party must be heard in a meaningful manner is embodied in the “failure to produce the universe” due process challenge, which operates as a failure of notice for an improper taking of property.
Failure to Produce the Universe
As discussed further above, CMS Ruling 86-1 requires that CMS and its contractors produce to a provider sufficient documentation to recreate the sampling frame and thereby challenge the statistical validity of the sample. CMS has enshrined this requirement in the form of sub-regulatory guidance across the MPIM. Documentation needed to recreate the sampling frame includes the universe of claims, which consist of all claims submitted by the provider within the chosen time period for CMS’s review. For example, MPIM Chapter 8, Section 8.4.4.4.1 requires “[a]n explicit statement of how the universe is defined and elements included shall be made and maintained in writing … [and] [d]ocumentation shall be kept in sufficient detail so that the sample frame can be re-created should the methodology be challenged.”
In short, where CMS and its contractors have failed to comply with the MPIM’s mandate to carefully document, preserve, and produce a statistical sampling and extrapolation from start to finish, a provider is denied its due process right to recreate or replicate the process to determine if it was performed correctly and to determine if a valid challenge should be raised. By extension, where a provider is denied its due process right to dispute and contest an extrapolated overpayment, the provider is not extended appeal rights in accordance with 42 U.S.C. § 1395ff(b) and the extrapolation should be invalidated. This due process challenge has shown success at both the ALJ level and at the Council level. While there is a dearth of caselaw involving this due process challenge at the federal district court level, there have been indications of viability as well.
Failure to Include Zero-Paid Claims
The U.S. Supreme Court has long held that a party generally must show prejudice to prevail on a due process claim. The second due process challenge against statistical sampling and extrapolation discussed herein—“failure to include zero-paid claims”—is well-poised to meet this prejudice requirement, as that failure itself creates bias in the extrapolation.
Auditors have historically misused vague and conflicting terms regarding zero-paid claims in the MPIM that are biased against providers. These vague and conflicting terms are clarified by the hierarchy of authority in the Medicare claims appeal context. Where a conflict exists between a statute and a regulation, or a statute and the MPIM, the statute will govern. Likewise, a regulation will govern over conflicting language in the MPIM. In addition, the statistical sampling and extrapolation guidelines in the MPIM are entitled only to Skidmore deference given their role as sub-regulatory guidance. As will be discussed below, that hierarchy is central to this due process challenge because there is statutory authority mandating identification of overpayments and underpayments.
For the statistical sampling and extrapolation process to function properly and not be biased against the provider, underpayments, including unpaid or zero-paid claims, must be present in the universe, sampling frame, and sample to ensure the actual net overpayment is calculated. Auditing only paid claims is biased toward only the potential errors made by the provider. This eliminates from consideration those claims that most likely would have been underpaid and certainly would have no chance of being overpaid. By doing this, the audit sample is biased towards only those claims that have a higher likelihood of being overpaid rather than underpaid. This bias renders the sample ineligible for use in any inferential statistical calculation, including extrapolation. Additionally, removal of all the unpaid claims mathematically increases the extrapolation estimate because it is working off the presumption that all the claims in the universe were paid.
Congress has recognized this risk of bias by including provisions in the Social Security Act and associated regulations that require contractors to identify underpayments as well as overpayments in the recovery audit process. Likewise, through sub-regulatory guidance, CMS has included and reiterated the requirement that the sampling frame include zero-paid claims in at least ten separate provisions of the MPIM.
To date, this due process challenge for failure to include zero-paid claims has been somewhat less successful than the due process challenge for failure to produce the universe. While a number of ALJs have agreed with the assertion that a failure to include zero-paid claims biases and invalidates the statistical sample and extrapolation, others have found exclusion of zero-paid claims to be permissible. Similarly, this due process challenge has yet to see significant success before a federal district court.
Conclusion and Developing Areas
As statistical sampling and extrapolation based on data analysis has dramatically evolved over the past 30 years in arenas such as FCA litigation and Medicare appeals, it has become imperative for defendants to understand how to challenge the processes and presumptions of such applications to support allegations of fraud and arm themselves with the tools to ward off potential liability.
As mentioned above, each of the due process challenges to statistical sampling and extrapolation discussed in this article—failure to produce the universe and failure to include zero-paid claims—has not frequently come before federal courts. However, there are several cases pending at the federal district court level that raise these due process challenges. Providers should monitor these and other potential cases raising those due process challenges and other developments in the Medicare claims appeal arena, as the legal framework for statistical sampling and extrapolation continues to rapidly evolve.