Nobody argues the need for scientific statistical sampling and extrapolation in cases of overpayment determination. However, the government often falls short in making its case, says Michael Intriligator, PhD, economics professor at the University of California, Los Angeles. "There is no question you need scientific statistical sampling," asserts Intriligator, who has successfully challenged the sampling methodology used by the Centers for Medicare & Medicaid Services (CMS). "But it has to be done properly."
If it is not done properly, Intriligator says, it can be thrown out either at the fair-hearing level or more likely at the administrative law judge (ALJ) level. In other instances, this can occur at the Medicare Appeals Board level.
"Usually, you do lose at the fair-hearing level," says Intriligator. "But I have had a couple of cases where we actually won at the fair-hearing level." He says that can be accomplished through successful legal or statistical challenges.
According to Intriligator, whether to use a statistical challenge depends on the amount of money involved as well as some of the statistical results. He also notes that carriers and the ALJ increasingly are bringing in their own statistical experts.
Intriligator says there are three primary issues, which he calls "the big three," that health care attorneys should consider when deciding whether to challenge carriers and intermediaries:
- Sample size. According to Intriligator, sample size is the No. 1 issue. Frequently, he says the sample size of the claims selected is too small and inconsistent not only with generally accepted statistical principles but with CMS’ own guidelines for a "basic sample size." In some instances, the number of claims selected can be less than a tenth of the number required, he says.
- Documentation. The second key area is documentation. Intriligator says the government must provide enough documentation to create an audit trail so the study can be replicated. He says all the authorities, statistical textbooks, and guides are very clear on this point.
"I find it very ironic that many times they will claim an overpayment based on the fact that there was not adequate documentation of the medical procedures," he argues. "Yet, when the carrier does the statistical exercise, they do not provide enough documentation."
- Randomness of the sample. To be statistically valid, the sample must be selected at random, with no biases or other distortions that could make it not "representative," says Intriligator. Sampling that specifically omits low-charge claims, for example, would not yield a reasonable sample, he says.
The coefficient of variation is a measure of the variability in the sample, Intriligator explains. If the variability is too high, it means the estimates are very imprecise and there is inaccuracy, he says. Likewise, if the coefficient is too high, the study is unacceptable.
Intriligator says one package that is widely used is the Health and Human Services’ Office of the Inspector General’s Office of Audit Services software package RAT-STATS. "This is actually a library of different computer routines, some of which I find perfectly acceptable," he says. "For example, their random number generator I find perfectly acceptable."
"The rest of it, I have some doubts about," he adds. Intriligator says the manual offers no information about what goes on inside the program. "It only tells you how to operate the program," he explains. "There is no information about what is inside that program."
Another area of possible error is improper stratification. "Stratification and the choice of strata is a very important issue," says Intriligator. This process divides the population into different subpopulations that are relatively homogeneous. For example, hospital services might be stratified into inpatient/outpatient or other categories, while physician services might be stratified by diagnostic categories, he explains.
Nevertheless, this is not always done. "If they did not use stratification with a heterogeneous population, the results are nonsense," he argues. "It is like adding apples and oranges." Conversely, some cases use stratification when it should not be used.
Outliers in the sample are another issue that should be looked at, says Intriligator. Sometimes the sample includes unrepresentative outliers that can bias the results, he explains.
"Any one of these issue areas or some combination of them could represent a basis for challenging the sampling/extrapolation," Intriligator concludes. It also is possible to challenge the qualifications and capabilities of those performing the study, he adds.
Note: Intriligator and Lester Perling, a health care attorney with Broad & Cassel in Tampa, FL, have co-authored a monograph on this subject published by the American Health Lawyers Association called Statistical Sampling in the Medicare Program: Challenging Its Uses.