Development of a Point-based Prediction Model for the Incidence of Total Stroke

Abstract

Background and Purpose— An individualized risk score for the development of stroke may be a useful tool to motivate patients to modify their risk behaviors. We developed and validated a point-based prediction model (risk score) for stroke incidence using a Japanese cohort of general men and women. Methods— The Japan Public Health Center-based prospective study cohort II (age range, 40–69 years at baseline in 1993–1994; n=15 672) was used to derive the point-based model according to Cox regression results. The model was externally validated using the Japan Public Health Center study cohort I and also by bootstrap methods within cohort II. The model discrimination was evaluated by the area under the receiver operating characteristic curve, model calibration, by the Grønnesby–Borgan χ 2 statistic. Vascular age was also calculated. Results— During 14 years of follow-up, 790 incident stroke cases occurred. Variables selected for the model were age, sex, current smoking, body mass index, blood pressure, antihypertensive medication use, and diabetes mellitus. Interactions of sex with current smoking and of antihypertensive medication use with systolic blood pressure were statistically significant. The point-based model discriminated reasonably well (area under the receiver operating characteristic curve, 0.73). The area under the receiver operating characteristic curve of the point-based model applied externally to cohort I was reasonably good: 0.69. A 50-year-old man with diabetes mellitus and hypertension has an estimated vascular age of 69 years. High normal blood pressure and grade 1 hypertension accounted for one third of the stroke incidence. Conclusions— We developed score to predict 10-year stroke risk using variables that are easily available in the community setting.