RESUMO
This study aimed to develop and validate the hypertension risk prediction models of the CArdioVascular disease Association Study (CAVAS). Overall, 6,186 participants without hypertension at baseline were randomly divided into derivation and internal validation sets in a 6:4 ratio. We derived two prediction models: the first used the Framingham hypertension risk prediction factors (F-CAVAS-HTN); the second considered additional risk factors identified using stepwise Weibull regression analysis (CAVAS-HTN). These models were externally evaluated among Ansan and Ansung (A&A) participants, and the external validity of the Framingham and A&A prediction models (F-HTN and A&A-HTN) were assessed using the internal validation set of CAVAS. The discrimination, calibration, and net reclassification were determined. During the 4-year follow-up, 777 new cases of hypertension were diagnosed. All four models showed good discrimination (C-statistic ≥ 0.7). Internal calibrations were good for both the coefficient-based and the risk score-based F-CAVAS-HTN models, respectively (Hosmer-Lemeshow chi-square, H-L χ2 < 20, P ≥ 0.05). However, the two CAVAS models (H-L χ2 ≥ 20, P < 0.05, both) as well as the F-HTN and the A&A-HTN prediction models (H-L χ2 = 155.39, P < 0.0001; H-L χ2 = 209.72, P < 0.0001, respectively) were not externally calibrated. The F-CAVAS-HTN may be better than models with additional risk factors or derived for another population in the view of the findings of the internal validation in the present study, although future studies to improve the external validity of the F-CAVAS-HTN are needed.