Variations in practice admission rates: the policy relevance of regression standardisation

Abstract

Objectives: To explore variations in general practice admission rates, comparing standardisation by regression with direct standardisation of the data to identify explained and unexplained variation.

Methods: Data from hospital episode statistics and the attribution dataset on 8048 cataract admissions from 109 practices in an English health district (North Yorkshire) between 1995 and 1998. Multiple regression was used to estimate the effect of practice characteristics, socio-economic factors, waiting times and distance on practice admission rates. Rankings of practices by the residuals from the regression were compared with rankings by directly standardised admission rates.

Results: The regression model yielded intuitively plausible results and explained 35% of the cross-practice variation in directly standardised admission rates. Standardisation by regression, compared with direct standardisation, made as least as much difference to the ranking of practices as direct standardisation compared with crude admission rates. Regression standardisation suggested that 10 practices not identified as 'unusual' by comparison of their rates to the district mean were in fact 'unusual', and that six practices identified as unusual by comparison with the district mean were not unusual once allowing for the explanatory factors used in the regression model.

Conclusions: Given the increasing importance of systematic performance assessment to support quality improvement, care must be taken when interpreting variations in health care activity even after conventional standardisation of the data. If significant variations are detected, regression analysis can assist in explaining some of it, which is the starting point in informing discussions about whether variations are justified or unjustified.