Small area variation analysis. Methods for comparing several diagnosis-related groups.

In small-area variation analysis, the variation of health care utilization rates, e.g., admission rates, among small areas is calculated. Frequently, the variation of one diagnosis, diagnosis-related group (DRG), or procedure is compared with the variation of another. Unfortunately, the methods generally used to make these comparisons are not consistent. They differ on whether they 1) adjust for the prevalence of the DRGs, 2) distinguish between variation among areas and variation within areas, 3) weight all areas equally, and 4) adjust for multiple admissions per person. None has an associated confidence interval. These discrepancies occur in part because there is no statistical model of small area variation. Without such a model, it is not known how to measure variation, and thus, it is not known how to compare different DRGs. Here, the authors use data on 473 DRGs from 28 counties in Washington state to study the nature of variability. The variation was higher for the more prevalent DRGs, suggesting that adjusting for prevalence may be reasonable. The true coefficient of variation appears to be a "natural" measure of variation, but the usual small area variation statistics do not provide good estimates of the true coefficient of variation. A new estimate is proposed that can be used to compare and test the variability of several DRGs.