Measuring small area variation in hospital use: site-of-care versus patient origin data

There has been increasing attention paid to small area variation in hospital discharge rates. While there is general agreement about the importance of correcting for the migration of patients to hospitals outside their geographic area when constructing population-based hospital use rates for these small areas, there have been no studies of the sensitivity of simple correlations or multiple regression results to these adjustments. Given the paucity of patient origin data, which is needed to adjust hospital discharge rates for patient crossovers, the problems of measurement error present in the more readily available site-of-care data need to be addressed. This paper analyzes the variation in hospital discharge rates, both an unadjusted site-of-care rate and an adjusted patient origin rate, across the 68 counties in the lower peninsula of Michigan in 1980. The results indicate that both simple correlations and multiple regression results of these rates with socio-economic and health care resource characteristics of the counties are very sensitive to the specification of the discharge rate, with the analysis of the unadjusted rate potentially leading to incorrect policy recommendations. The explanatory power of the socio-economic characteristics is underestimated and that of health care resource measures most likely overestimated when the discharge rate is not adjusted for patient crossovers.