Development and external validation of a prognostic model for ischaemic stroke after surgery.

Abstract Background There is an under-recognised patient cohort at elevated risk of postoperative ischaemic stroke. We aimed to develop and validate a prognostic model for the identification of such patients at high risk of ischaemic stroke within 1 yr after noncardiac surgery. Methods This was a hospital registry study of adult patients undergoing noncardiac surgery between 2005 and 2017 at two independent healthcare networks in Massachusetts, USA without a preoperative indication for therapeutic anticoagulation. Logistic regression was used to fit a model from a priori defined candidate predictors for the outcome 1 yr postoperative ischaemic stroke. To enhance clinical applicability, the model was simplified to a scoring system and externally validated. Results In the development (n=107 756) and validation (n=141 724) cohorts, 1.4% and 0.5% of patients had an ischaemic stroke up to 1 yr postoperatively. The final model included 13 variables (patient characteristics, comorbidities, procedural factors), considering sub-models conditional on a previous history of ischaemic stroke. Areas under the curve were 0.89 (95% confidence interval 0.89–0.90) and 0.88 (95% confidence interval 0.86–0.89) in the development and validation cohorts. Decision curve analysis indicated positive net benefits superior to other prediction instruments. Conclusions Stroke after surgery (STRAS) screening can reliably identify patients with a high risk for ischaemic stroke during the first year after surgery. A STRAS-guided risk stratification may inform the recruitment to future randomised trials testing the efficacy of treatments for the prevention of postoperative ischaemic stroke.