Cardiovascular Disease Risk Prediction Model

All the statistical data analyses were performed in R with version 4.0.3 (2020‐10‐10). Kolmogorov–Smirnoff test was used to perform normality test for continuous variables. For the normally distributed variables, we used Student's t‐test and one‐way ANOVA to compare among different groups, otherwise nonparametric method Wilcoxon test was used for the variables that do not follow normal distribution. For categorical variables, the χ² test was used to compare the difference among groups. p Value < 0.05 indicates there were significant differences among the groups.
Sex‐specific Cox proportional hazard regression³⁵, ³⁶, ³⁷ was used to construct a risk prediction model to predict the probability of developing CVD over a 9‐year follow up. In order to maximize performance of the calibration and discrimination for the prediction models and to minimize the impact of extreme observations, natural log–transformation for all continuous covariates in the model was performed. Risk factors in our models were selected by using the traditional cardiovascular risk factors including age, current smoking state (yes/no), CHOL, HDLC, and diabetes.³⁸ In addition to these factors, we also include the SBP and physical activity in our model. Hence, the selected factors were the same as China‐PAR¹⁴ in addition to physical activity. Besides, the interaction terms between age and other risk factors were also included in the model. Participants were randomly divided into training cohort and validation cohort based on the R function “createFolds” in the R package caret.³⁹ Ten‐fold cross validation was performed for the internal validation. C statistic, calibration χ² were calculated to evaluate the performance of the equations.⁴⁰, ⁴¹, ⁴² PA equations were also compared with the China‐PAR equation¹⁴ by the ROC curves. We also calculated the AUC for quantitative comparison; the log‐rank test was used to calculate p‐value for checking whether the difference was significant.⁴³