Abstract

When a healthcare provider is suspected of billing abuse, Medicare integrity investigations sample a payment population to estimate the total overpayment using a (nominal) 90% lower confidence bound for purposes of recovery. The classical method Mean-per-Unit (MPU) method based on the finite-population Central Limited Theorem tends to be conservative for right-skewed overpayment populations. For a left-skewed population, this method often produces coverage probability well below 90% under high error rates, which are common in Medicare investigations. The Empirical Likelihood (EL) method has been shown to perform well at low error rates in the accounting/auditing literature, but has not been tested at high error rates. In this talk, we discuss our testing of the EL method on Medicare payment populations at all error rates. Preliminary results are very promising: At typical sample sizes (30-90), the EL method is viable (controls the confidence level at or above 90%) more often than the MPU method, and is more efficient (i.e. tends to have higher overpayment recovery) when both methods are viable.