Medicare began contracting with risk-based managed care plans in 1985 with the primary goal of containing costs in the Medicare program. The Health Care Financing Agency developed a risk adjustment method known as the adjusted average per capita cost (AAPCC) to pay risk- based HMOs for medical services used by beneficiaries. It turned out that the AAPCC does not adequately adjust for underlying risk, resulting in overpayments to HMOs rather than savings to Medicare. The 1997 Balanced Budget Act mandated the implementation of a new, health- based capitation payment methodology to better adjust HMNO premiums. This new methodology-like the AAPCC-was developed using data from the Medicare Fee-for-Service (FFS) program The FFS-based models may be biased not only because they do not capture all the underlying risk of HMO enrollees, but because of fundamental differences in the way HMOs practice medicine. The purpose of this research is to investigate the bias in HCFA's FFS- based payment methodologies in predicting HMO enrollees' resource use. It uses a unique large data set of inpatient admissions in California Medicare HMOs in the early 1990s. Most previous assessments of HCFA's payment models rely solely on FFS data because HMO enrollees' medical care utilization and expenditures are generally not available. Using this data, it is possible to evaluate how HCFA's FFS- based methodologies actually perform on real Medicare HMO samples. First, sample selection models of individual inpatient utilization in the HMO and FFS sectors will be developed, and consistently estimated. Based on that, the prediction bias in HCFA's FFS-based methodologies attributable to FFS and HMO selection bias and HMO practice style will be quantified. Finally, the extent to which the payment methodologies result in overestimation-or perhaps underestimation-of HMO enrollees' resource use will be simulated. The final results will inform policymakers how Medicare risk adjustment methodologies may be improved.
