Evaluation of risk prediction model for stroke risk based on Cox's and Weibull model in Korea

Kim, Youn Nam; Cho, Ur Rin; Nam, Byung Ho; Park, Il Soo; Jee, Sun Ha

Abstract

OBJECTIVE
The objective was to compare Cox proportional hazards model and Weibull model for predicting long-term probabilities for stroke risk in the Korean Cancer Prevention Study(KCPS).
METHODS
The subjects comprised of 385,279 Korean aged 55 to 64 years who received health insurance from the National Health Insurance Corporation and who had medical examinations in 1992 and 1995. 70% of the subjects were used for model building and the rest for model evaluation. The final prediction model for stroke includes age, systolic blood pressure, diabetes, total cholesterol and smoking. Subjects were follow-up for identification of incident stroke cases between 1993 and 2005. Comparisons included predicted coefficients of stroke risk factors, incidence probabilities over 10 years, and the area under a receiver operating characteristics (ROC) curve for both Cox's proportional hazard model and Weibull model.
RESULTS
The average age of study population was 55.5 years in men and 56.3 years in women, respectively. Percentage of men and women in study population were 58.0% and 42.0%, respectively. The study findings satisfied proportionality according to the two models. There was no significant difference in coefficients between the two models of prediction models in men and in women. Moreover, there was no difference in incidence probabilities of stroke and c-statistics. C-statistics were 0.68 for men as same as for women.
CONCLUSION
There was no difference for the prediction of the stroke risk in the Korean population using Cox's proportional hazard model and Weibull model, thus the two models were found to be efficient for this purpose.