Validation and generalizability of the predictive model of ischemic cardiovascular diseases in Chinese / 中华心血管病杂志

ABSTRACT

<p><b>OBJECTIVE</b>To evaluate the applicability of developed prediction models of ischemic cardiovascular diseases (ICVD) in Chinese to other Chinese populations.</p><p><b>METHOD</b>We used the independent prospective cohort established in early 1990's from China Multi-Center Collaborative Study of Cardiovascular Epidemiology (MUCA), as the validation cohort, to test the hypothesis. The area under ROC curve (AUC) based on the application of the Cox optimal model and the simplified model to the validation cohort were calculated and to test the ability of the prediction models to discriminate events from nonevents. Applicability was evaluated by comparing the mean probability estimates in each decile of probability in the validation cohort with the observed incidence with the Hosmer-Lemeshow test.</p><p><b>RESULTS</b>The validation cohort enrolled a total of 17 329 men and women aged 35 to 59 years baseline 1992 - 1994. In this paper, we used data from the remaining 15 100 participants after excluding 2229 subjects for at missing value of risk factors. During 11-year follow up of the cohort, there were 347 ICVD events (206 for men and 141 for women), including 83 coronary heart disease events (56 men and 27 women) and 268 ischemic strokes (154 men and 114 women). ROC curves for men and women showed good and almost identical discrimination for optimal model (the AUCs (95% CI) were 0.796 (0.762 - 0.829) for men and 0.791 (0.755 - 0.828) for women), simplified model (the AUCs (95% CI) were 0.792 (0.758 - 0.825) for men and 0.783 (0.746 - 0.821) for women) and score system (the AUCs (95% CI) were 0.791 (0.757 - 0.825) for men and 0.779 (0.741 - 0.817) for women) in the validation cohort. The predicted 10-year risk of ICVD by optimal models and observed incidence of ICVD in the validation cohort in each decile were compared. Hosmer-Lemeshow chi2 was 3.7 for men (P = 0.879) and 27.7 for women (P < 0.001). Whereas the largest difference between the observed rate and the predicted rate was only 1%.</p><p><b>CONCLUSIONS</b>The prediction models for estimating 10-year risk of ICVD had satisfied predictive capability when they were applied to the validation cohort and are applicable to other Chinese populations.</p>