Abstract W MP50: Risk Prediction of Incident Acute Ischemic Stroke (AIS): Pooled Data Analysis of Cohort Studies

Abstract

Introduction: Risk assessment and estimation can be used to support the choice of behavioral and therapeutic interventions. Objective: To derive and validate a race and sex sensitive risk prediction algorithm for incident AIS. Methods: Risk prediction functions for incident AIS were developed and validated using ARIC, CHS, and MESA research databases obtained from the NHLBI BioLINCC. ARIC and CHS cohort members with prior stroke/TIA or cancer at baseline examination and CHS cohort members above age 84 years were excluded. A random 60% sample of the ARIC and CHS cohorts, separately, were pulled and combined to form the derivation cohort. The remaining 40% of ARIC and CHS data were pooled and used as the validation cohort. MESA was used as an external validation cohort. Subjects with missing data on established cardiovascular risk factors were excluded. Cox proportional hazards regression was used to develop best fitting prediction model in the derivation cohort. Performance of the model was assessed at 5 and 10 years. Measures of calibration and discrimination were undertaken in both validation cohorts. Results: The derivation cohort included 9,762 subjects (54.4% women, 23.8% blacks) with 708 cases of AIS with a median follow-up of 18.3 years. The ARIC-CHS validation cohort included 6,538 subjects (55.6% women, 22.8% blacks) with 475 AIS and a median follow-up of 18.3 years. The MESA validation cohort included 6,566 subjects (52.6% women, 27.9% blacks) with 106 AIS and a median follow-up of 7.6 years. The final risk equation demonstrated very good discrimination (concordance statistics: Harrell’s C = 0. 80, Somer’s D =.59). The accuracy of probabilistic predictions for 10-year stroke risk in the derivation cohort was 0.820 (p<.001) for ROC area, and 0.039 for Brier score; corresponding values in ARIC-CHS validation cohort were 0.807(p<.001) and 0.040. Similar measures for 5-year stroke risk in the derivation cohort were 0.828 (p<.001) for ROC area, and 0.016 for Brier score; and corresponding values in MESA validation cohort were of 0.810(p<.001) and 0.011. Conclusion: We have derived and validated a new race and sex sensitive prediction equation based on readily available predictors in clinical practice that quantifies absolute risk of first AIS at 5 and 10 years.