Impact of stroke units on mortality: a Bayesian analysis

Abstract

Background: Observational studies of the impact of care delivered in stroke units on mortality after acute stroke have been difficult to interpret because of bias.

Methods: This study used Bayesian methodology to (i) sequentially update current knowledge based on a systematic review of observational studies, (ii) adjust for the likely effect of observational bias in a conservative, scenario-based approach and (iii) evaluate the likely impact of further studies. The data were interpreted both unmodified and adjusted for bias by explicitly correcting for point estimate and precision bias.

Results: As of 2007, the total evidence base was almost 47 000 patients. The unadjusted data yielded a current 95% posterior credible interval (Cr.I.) for the odds ratio of death within 1 year after stroke unit versus alternative models of stroke care of (0.77, 0.85). Given the adjusted data, the current 95% Cr. I. is (0.79, 0.93).

Conclusions: The current estimates are robust, with demonstrable stability despite the addition of recent large studies. Further studies are very unlikely to alter current knowledge but may have a role in ensuring regional stability of outcome.