Attributing inpatient medicare costs to diabetes among the Texas elderly

This study compares alternative methods for attributing hospital utilization and costs to diabetes. Findings from five "numerator" methods, found in the literature and based on presence of certain diagnoses or combinations of diagnoses in the billing records, were compared to benchmark findings derived from attributable risk calculations.

Estimates of non-HMO, short-term, nonspecialized hospital stays, hospital days, and costs attributable to diabetes in Texas were derived from the 1995 Medicare inpatient database (MEDPAR) for persons aged at least 65 years at the end of 1994. Attributable risk calculations applied age-, sex-, and ethnicity-specific estimates of diabetes prevalence, based on the combined 1987-1994 National Health Interview Surveys, to 1995 Medicare non-HMO, Part A (hospital insurance) enrollment among the Texas elderly. Alternative prevalence estimates were based on the 1994-1996 Texas Behavioral Risk Factor Surveillance System.

The five numerator methods yielded cost estimates that were 10, 10, 75, 144, and 172% of the benchmark estimate.

This study documents great variation in diabetes cost estimates that might result from alternative methods for selecting diagnoses or combinations of diagnoses as criteria for attributing costs to diabetes. Whereas no method that ignores population prevalence yielded an accurate cost estimate, I suggest that further empirical study may be helpful in selecting those combinations of diagnoses that might, on average, reasonably estimate diabetes costs in situations where population denominators are unavailable or prevalence is unknown.