Risk-adjusted hospital mortality rates for stroke: evidence from the Australian Stroke Clinical Registry (AuSCR)

Cadilhac and colleagues1 explore an important issue in the measurement and reporting of stroke outcomes. We agree that appropriate risk-adjustment methods are essential to compare hospital outcomes. We also agree that stroke severity is an important predictor of mortality for individual patients. However, we do not agree that determining stroke severity is essential for robust risk-adjustment approaches.

The Bureau of Health Information has recently published its second report on 30-day mortality.2 Our approach includes adjustment for comorbidity, and we developed separate models for ischaemic and haemorrhagic stroke, given the significant differences in outcomes and risk factors. The approach of Cadilhac and colleagues does neither, which means that we do not know the impact of including comorbidity adjustment and severity in the same model. Further, our method is applicable to small hospitals with as few as 50 patients in a 3-year period. In settings such as Australia, where many patients reside outside major cities, it is important to assess outcomes in both small and large hospitals.

The results of Cadhilac et al show that adjusting for stroke severity affected hospital rankings, and across the two models, hospital rates changed on average by 0.01 (range, 0.001–0.026). The impact on outlier status is not described. In our view, it is the outlier status that is the most salient element of public reporting, and indirect standardisation should not be used to rank hospitals.3

Measurement approaches in the United States and Canada do not adjust for severity, and it is not currently possible to do so using administrative records in Australia. While Cadhilac et al have shown that including severity information can affect rankings, such rankings are not appropriate to assess hospital performance when they are based on indirect standardisation and do not take account of hospital size.