Risk-adjusted capitation payments: how well do principal inpatient diagnosis-based models work in the German situation? Results from a large data set

Five models of risk adjusters were tested as a (proxy) measure for health status with data from a large German sickness fund. The first two models use standard demographic and socio-demographic variables. One model incorporates a simple binary indicator for hospitalization and the last two are based on the hierarchical coexisting conditions (HCCs: DxCG® Risk Adjustment Software Release 6.1) using in-patient diagnoses. Special investigations were done on the subgroups of insurees who left, joined or stayed with the fund over the observation period. Age and gender grouping accounted for 3.2% of the variation in total expenditure for concurrent as well as prospective models. The current German risk adjusters age, sex, and invalidity status account for 5.1 and 4.5% of the variance in the concurrent and prospective models, respectively. Age, gender, invalidity status and in-patient HCC covariates explain about 37% of the variations of the total expenditures in a concurrent model and roughly 12% of the variations of total expenditures in a prospective model. Only modest improvement can be achieved with the long-term-care (LTC) indicator. For high-risk (cost) groups, substantial under-prediction remains; conversely, for the low-risk group, represented by enrolees who did not show any health care expense in the base year, all of the models over-predict expenditure. Special investigations were done on the subgroups of insurees who left, joined or stayed with the fund over the observation period.