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

We thank Sutherland and colleagues for their comments on our article, and for identifying discrepant data in Box 2 and Appendix 4, now corrected.1 International guidelines explicitly refer to including stroke severity in risk adjustment.2 We maintain that stroke severity is required for reliable mortality rate comparisons. Stroke severity may modify individual hospital rankings by up to 25%.2 We have subsequently conducted new analyses using linked administrative and stroke registry data to permit inclusion of ICD-10 coded comorbidities, including calculating an Elixhauser Comorbidity Index.3 Consistent with our original findings, models with the stroke severity variable still provided the best fit for standardising mortality among the hospitals. Therefore, the registry data offer an important adjunct to work undertaken by different organisations focused on performance monitoring using administrative data, and are likely to be more acceptable to clinicians.

Our aim was to illustrate a worked example of new statistical methods for calculating risk-adjusted mortality from our stroke registry. Outlier status was not compared in detail. However, identification of outliers must be based on appropriate risk adjustment. Because the mix of stroke type differs among hospitals and we assessed overall survival, stroke type was appropriately used as an independent variable. We also accounted for hospital size and year. We agree that hospitals from metropolitan and regional and rural areas should be included, as was the case in our study (> 50% regional and rural). Because the precision of model estimates is reduced when hospitals have small numbers,2 we conservatively set a minimum sample size of 200 episodes per hospital. As identified by others,4 and in our experience, interpretation of these data is complex because patients may be cherry-picked to stay or go depending on transfer protocols and available resources. Within the ever-evolving field of acute stroke, regional and statewide protocols to direct patients to hospitals with stroke units and other specialist resources (eg, thrombectomy intervention) are common. Therefore, accounting for differences in stroke severity is essential because of this selection bias. Generally, to justify having a stroke unit, annual patient numbers should be over 100.5 Performance monitoring is an evolving discipline, so we welcome the involvement of the Bureau of Health Information in our efforts.