Layperson-oriented vs. clinical-based models for prediction of incidence of ischemic stroke: National FINRISK Study

A simplified model not requiring a clinic visit is important for cardiovascular prevention. We compared such a model, with one requiring clinical measurements for prediction of ischemic stroke.

Five population-based Finnish cohorts comprising 14 296 men and 16 065 women aged 25-64 years were randomly recruited from 1982, and followed up using the national registers until the end of the 2007. The final Cox model included age, prior history of diabetes and hypertension, happy marriage, capability to walk 500 m (self-estimate), regular exercise, vegetable/fruit intake, smoking, body mass index, and systolic blood pressure; the layperson-oriented model was developed by taking blood pressure away from the final model.

Four hundred sixty-nine men and 371 women developed ischemic stroke events. The area under the receiver operating characteristic curve (95% confidence interval) for 10 years incidence of ischemic stroke was 0·817 (0·791-0·843) and 0·813 (0·787-0·839) for the model with and without systolic blood pressure in men and 0·815 (0·782-0·848) and 0·812 (0·779-0·844), respectively, in women (P > 0·10). The predicted 10-year events rate matched well with the observed one across deciles of the predicted risk in men (χ(2)  = 11·57, 9df, P = 0·239) and in women (χ(2)  = 11·18, P = 0·263). The overall net reclassification improvement after adding blood pressure was 8·8% (P = 0·016) in men and 3·2% (P = 0·234) in women. The predicted 10-year risk of the ischemic stroke based on global vascular event models that includes coronary heart diseases did not match well with the observed stroke risk.

The layperson model performed as well as the clinical-based one.