ObjectivesTo determine whether stroke patients’ acute hospital length of stay (AHLOS) varies between hospitals, over and above case mix differences and to investigate the hospital-level explanatory factors.DesignA multicentre prospective cohort study.SettingEight National Health Service acute hospital trusts within the Anglia Stroke & Heart Clinical Network in the East of England, UK.ParticipantsThe study sample was systematically selected to include all consecutive patients admitted within a month to any of the eight hospitals, diagnosed with stroke by an accredited stroke physician every third month between October 2009 and September 2011.Primary and secondary outcome measuresAHLOS was defined as the number of days between date of hospital admission and discharge or death, whichever came first. We used a multiple linear regression model to investigate the association between hospital (as a fixed-effect) and AHLOS, adjusting for several important patient covariates, such as age, sex, stroke type, modified Rankin Scale score (mRS), comorbidities and inpatient complications. Exploratory data analysis was used to examine the hospital-level characteristics which may contribute to variance between hospitals. These included hospital type, stroke monthly case volume, service provisions (ie, onsite rehabilitation) and staffing levels.ResultsA total of 2233 stroke admissions (52% female, median age (IQR) 79 (70 to 86) years, 83% ischaemic stroke) were included. The overall median AHLOS (IQR) was 9 (4 to 21) days. After adjusting for patient covariates, AHLOS still differed significantly between hospitals (p<0.001). Furthermore, hospitals with the longest adjusted AHLOS’s had predominantly smaller stroke volumes.ConclusionsWe have clearly demonstrated that AHLOS varies between different hospitals, and that the most important patient-level explanatory variables are discharge mRS, dementia and inpatient complications. We highlight the potential importance of stroke volume in influencing these differences but cannot discount the potential effect of unmeasured confounders.