Variations in hospital standardised mortality ratios (HSMR) as a result of frequent readmissions

BACKGROUND: We investigated the impact that variations in the frequency of readmissions had upon a hospital's standardised mortality ratio (HSMR). An adapted HSMR model was used in the study. Our calculations were based on the admissions of 70 hospitals in the Netherlands during the years 2005 to 2009. METHODS: Through a retrospective analysis of routinely collected hospital data, we calculated standardised in-hospital mortality ratios both by hospital and by diagnostic group (H/SMRs) using two different models. The first was the Dutch 2010 model while the second was the same model but with an additional adjustment for the readmission frequency. We compared H/SMR outcomes and the corresponding quality metrics in order to test discrimination (c-statistics), calibration (Hosmer-Lemeshow) and explanatory power (pseudo-R(2 )statistic) for both models. RESULTS: The SMR outcomes for model 2 compared to model 1, varied between -39% and +110%. On the HSMR level these variations ranged from -12% to +11%. There was a substantial disagreement between the models with respect to significant death on the SMR level as well as the HSMR level (~ 20%). All quality metrics comparing both models were in favour of model 2. The susceptibility to adjustment for readmission increased for longer review periods. CONCLUSIONS: The 2010 HSMR model for the Netherlands was sensitive to adjustment for the frequency of readmissions. A model without this adjustment, as opposed to a model with the adjustment, produced substantially different HSMR outcomes. The uncertainty introduced by these differences exceeded the uncertainty indicated by the 95% confidence intervals. Therefore an adjustment for the frequency of readmissions should be considered in the Netherlands, since such a model showed more favourable quality metric characteristics compared to a model without such an adjustment. Other countries could well benefit from a similar adjustment to their models. A review period of the data collected over the last three years, at least, is advisable.