Abstract A109: Novel clinical risk prediction tool for dementia after transient ischemic attack and stroke

Background: People with acute stroke and transient ischemic attack (TIA) are at high risk of subsequent dementia, but prediction tools for dementia are lacking. Methods: We included adult participants without prior dementia, with first TIA, acute ischemic stroke (AIS) or intracerebral hemorrhage (ICH) in the Ontario Stroke Registry, between 2002-2013 who survived to hospital discharge. We identified incident dementia with a validated definition using linked administrative data until March 2024. We used Fine-Gray subdistribution hazard models to account for competing risk of mortality and backwards selection to identify the most important variables associated with incident dementia. Dementia risk scores were calculated from beta coefficients of the models, separately for TIA and stroke. We derived models for 1-, 5-, and 10-year dementia risk in the OSR and externally validated them in the Ontario Stroke Audit cohort (not included in the derivation cohort). Results: We included 7554 people with TIA with a mean follow-up of 9 years, and 13833 with AIS and 2340 with ICH with a mean follow-up of 7.5 years. The top factors associated with increased dementia risk for people with TIA were older age, prior dependence, diabetes, depression, cognitive symptom at presentation, and any disability at discharge, and for people with stroke were older age, female sex, diabetes, prior stroke or TIA, depression, ICH (vs. AIS), visual field deficit, cognitive symptom at presentation, or greater modified Rankin score at discharge. The area under the ROC curve for 1-, 5-, and 10-year dementia risk was good for all models in the derivation cohort (0.8, 0.78, 0.76, respectively, for TIA and 0.77, 0.75, 0.71, respectively, for stroke) and decreased negligibly in the validation cohort. Calibration was excellent, with highly comparable rates for predicted and observed dementia (Maximum 50% at 10 years; Figures 1 and 2). Model characteristics remained favorable when evaluating AIS and ICH separately. Conclusion: We developed and validated clinical risk prediction tools for dementia after TIA and stroke from a large cohort with long-term follow-up. The calibration of the risk scores was excellent, suggesting that these novel scores may assist with selecting a population of people with TIA and acute stroke who could be targeted in clinical trials of interventions to lower dementia risk. The risk scores will be made available through an online calculator after publication of the results.