To examine whether modeling a set of readily available information could add to predictive performance of blood pressure measures.
Method and results
From participants of the ongoing Tehran Lipid And Glucose Study, which has been followed for 6 years, 4656 (2695 women) nondiabetic adults, free from hypertension at baseline with no history of cardiovascular disease (mean age 42 years), were selected for the current analyses. We ascertained 805 cases of incident hypertension. Incident rate (per 1000 person-years) was 29.3 [95% confidence interval (CI) 26.7–32.1] for women and 30.9 (95% CI 27.8–34.3) for men (P = 0.457). In multivariable analyses, age, SBP, and DBP significantly contributed to the risk of incident hypertension in both sexes, waist circumference and family history of premature cardiovascular disease among women, and smoking among men. A point-score system for predicting incident hypertension was developed by converting Weibull regression coefficients of predictors to integer values and its performances were compared with those of SBP and DBP. The C-statistic for the prediction model was 0.730 among women and 0.741 among men, and calibration was good in both sexes. Relative integrated discrimination improvement and cut-point-free net reclassification improvement for the point-score system vs. SBP ranged between 22.0 and 69.7%. The corresponding figures for DBP ranged between 29.6 and 51.3%.
In this large, community-based study, the simplified utilization of the survival regression models was superior to blood pressure measures alone for predicting incident hypertension.