Does linking the australian stroke clinical registry with admissions data provide a better explanation of variability in stroke risk-adjusted mortality rates?

Abstract

Background and Aims: Stroke risk-adjusted mortality rates (RAMR) are influenced by the methods used to adjust for potential confounders. To determine whether linking data from Admissions and Registry datasets provides a better explanation of variability in 30-day RAMR than using admission data alone. Method: Cohort design linking Australian Stroke Clinical Registry (AuSCR) data for patients from 2009 to 2013 in Queensland with hospital admissions and national death registrations. The Elixhauser Index, a validated method for measuring patient comorbidity, was derived from hospital admissions ICD-10 codes. Model A contained data including variables available in hospital admissions datasets (i.e. demographics, stroke type and the Elixhauser index). Model B included the variables contained in model A plus additional information from the AuSCR such as stroke severity and recurrent stroke. Generalised linear latent and mixed models were used to calculate 30-day RAMR. Models were compared using Bayesian information criterion (BIC) and the C-statistic: 95% confidence intervals (CI). Results: Of 2986 episodes of care, 363 patients (12%) died within 30 days of admission. RAMRs for hospitals ranged from 6% to 16%. According to the model fit statistics, Model B (BIC: 1581; C-statistic: 0.842; 95%CI: 0.82, 0.86) provided better explanation than Model A (BIC: 1900; C-statistic: 0.790; 95%CI: 0.77, 0.81). Both the magnitude of difference in BIC and statistically significantly different c-statistics indicate that Model B was strongly superior in explaining variability in 30-day RAMR. Conclusion: The addition of severity and recurrent stroke to mortality models provides a better explanation of variability in RAMR than risk adjustment available from administrative data alone.L616982741 <br />