Assessing the outcome of stroke in Australia

Appropriate risk adjustment of stroke outcome data is needed for assessing and ensuring quality of care

Australia prides itself on providing high quality health care. But how is it measured? A common benchmark in hospitals is the outcome for patients as measured by routinely collected mortality data, with hospitals ranked according to their performance on this measure. However, “league tables” that rank hospitals by crude (unadjusted) mortality rates may not accurately reflect their processes and quality of care if the rates are not adjusted for other factors that can influence outcomes, such as casemix (Box).1-3

In this issue of the Journal, Cadilhac and colleagues report for the first time Australian mortality rates for stroke (30 days after hospitalisation) that are adjusted for important prognostic factors (covariates) not routinely recorded in hospital admission databases.4 The Australian Stroke Clinical Registry (AuSCR) prospectively collected clinical data on 15 951 patients who were admitted with acute stroke to 28 participating Australian hospitals (18 metropolitan and 10 rural or regional), each of which provided at least 200 episodes of stroke care between 2009 and 2014.4,5 Baseline AuSCR data included information routinely collected by hospitals (age, sex, country of birth, Indigenous status, socio-economic status of postcode, year of stroke, stroke type) as well as additional prognostic covariates (a history of previous stroke; ability to walk on admission). The AuSCR data were linked to 2372 national death registrations, realising an overall crude 30-day mortality rate of 14.6%. The crude 30-day mortality rate ranged between 5.2% and 19.6%, despite similar adherence to evidence-based processes of care in the 28 hospitals (such as treating patients in stroke units).4 Patients who died as the result of their stroke within 30 days of hospitalisation were, on average, older than 30-day survivors, and were more frequently women, unable to walk on admission, and hospitalised for a haemorrhagic or recurrent stroke. After adjusting for prognostic covariates recorded in hospital admission data, the 30-day risk-adjusted mortality rate (RAMR) ranged from 8% to 20% across the 28 hospitals. After adjusting for prognostic covariates recorded in the clinical registry, the 30-day RAMR ranged between 9% and 21%. The ranking of the 28 hospitals according to their risk-adjusted 30-day stroke mortality rates varied according to which covariates were included in analyses, particularly for hospitals with high crude mortality rates. The models with the best fit were those that included stroke severity, as indicated by ability to walk on admission, as a covariate; data for this factor are currently recorded only in the AuSCR.4

The authors of the AuSCR study are to be congratulated for overcoming challenges to obtaining data held by the states (hospital data) and by the National Death Index for linkage to a non-governmental national clinical registry, the AuSCR. Their study illustrates the importance of adjusting analyses for key baseline variables (such as stroke severity) when comparing mortality rates for patients hospitalised for stroke. It also highlights the capacity of registries of clinical quality data to inform and complement hospital and national outcome data in the quest to measure, monitor and benchmark patient outcomes. Further, the AuSCR study provides an insight into the potential of clinical registries that systematically collect standardised data about processes of care to identify variations in clinical practice and to assess the appropriateness of care in the context of evidence-based standards and guidelines.6 These data may facilitate the evaluation of the effects of compliance with standards and of variations in care on patient outcomes, and assist in the design of interventions to reduce variation and to improve outcomes.7

In recognition of the potential for clinical quality registries to fill information gaps in the measurement and monitoring of the appropriateness and effectiveness of health care, a national framework has been developed that describes a mechanism for the secure disclosure, collection, analysis, and reporting of individual patient record-level data for high burden clinical conditions, such as stroke.8 Remaining challenges for clinical stroke registries such as AuSCR include the evolving definition9 and coding10 of stroke, ascertaining a high proportion of the eligible patient population, accurate and complete recording of prognostic data by clinical units, measuring outcomes that are important to patients (such as disability and return to usual activities), and providing clinicians with timely feedback that encourages adherence to evidence-based care.