[Hospital standardised mortality ratio (HSMR): adjustment for severity of primary diagnosis can be improved]

To study the impact of casemix variations within primary diagnostic groups on hospital standardised mortality ratios and to improve the current HSMR model.

Descriptive, model analysis.

We calculated the HSMRs and the standardised mortality ratios at diagnostic group level (SMRs) of 84 Dutch hospitals over a 5-year period using 2 HSMR models. Both models adjusted for severity of diagnosis at admission (primary diagnostic level according to the International Classification of Diseases (ICD-9)) using a severity classification. The severity classification in the current model (model 1) was based on WHO mortality data, whereas our adapted 'Dutch' model (model 2) was based on Dutch hospital mortality data. The HSMRs and SMRs and the corresponding c-statistics of both models were compared in order to test whether the models adequately adjusted for high-risk versus low-risk ICD-9 codes.

Model 1 did not adequately adjust for casemix differences on the ICD-9 diagnostic level for 40 out of 48 diagnostic groups analysed; model 2 did not adequately adjust for 25 diagnostic groups. Model 2 showed better c-statistics than model 1. Relative differences between models 1 and 2 in SMR outcomes varied from -63% to 202% when comparing model 2 to model 1. At the HSMR level these differences varied from -6.7% to 4.3%.

The HSMR model currently used in the Netherlands does not adequately adjust for casemix differences at the level of ICD-9 primary diagnoses. A model using a severity classification of ICD-9 codes based on actual Dutch hospital mortality allows for a better, albeit not perfect, adjustment. We recommend implementation of the 'Dutch' severity classification in order to improve the HSMR model.