| Cardia-cerebrovascular disease(CVD)was one of the most common complication in patients with type 2 diabetes.Diabetes was a major public health problem in China,but there were no prediction tools having been proved suiting for Chinese diabetic patients.In China,the relevant departments had been promoting of chronic disease management based on communities.However,personality health intervention for diabetics was still tough.This study was based on "the procject for prevention and control of hypertension and diabetes" in Jiaonan district,Qingdao.We built a type 2 diabetes health management cohort by data fusion,data cleaning and normalized finishing.We explored the potential risk factors and built a CVD risk model for type 2 diabetes in three years using proportional sub-distribution hazards model.Internal validation using Bootstrap and external validation based on a diabetes check-up subqueue from Shandong Multi-center Health Management Large Database was used to evaluate effectiveness and performance of the model.Finally,we built a simple scoring model by Framingham study risk score functions,which would provide a simple assessment tool for community clinic.The results of the study are as follows:1.Diabetic patients’ average age was 65.49±10.63 years old in this cohort.Gender differences were observed when comparing several risk factors in baseline,which hint at the need to build prediction models by gender stratification.2.The incidence density of CVD events in patients with type 2 diabetes was 39.26‰during the follow-up period,in which the incidence density was 33.65‰ for male and 43.09‰ for female.Female diabetes’ CVD incidence density was higher than male diabetes’(P<0.05).3.The variables of the final prediction model included age,body mass index,total cholesterol,high density lipoprotein cholesterol,history of hypertension,family history of CVD,current smoking and diabetes duration.4.The following is the results of proportional sub-distribution hazards models considering competition events:the AUC of the prediction model for male was 0.720(95%CI:0.683～0.758),while the AUC of the prediction model for female was 0.706(95%CI:0.679～0.732).The results of Cox hazards models are as follow:the AUC of the prediction model for male was 0.718(95%CI:0.698～0.734)and 0.706(95%CI:0.690～0.732)for female.Compared with Cox hazards model,proportional sub-distribution hazards model showed a better discriminant ability.5.The internal validation using Bootstrap of repeated sampling 1000 times showed that the AUCs of FG model and Cox model were 0.707(95%CI:0.671～0.744)and 0.703(95%CI:0.603～0.804)for male,respectively,and 0.705(95%Cl:0.691～0.71)and 0.679(95%Cl:0.588～0.769)for female,respectively.The external validation showed that the AUC of the model was 0.689(95%CI:0.639-0.735)for male,and 0.713(95%CI:0.615～0.799)for female.The prediction models had good discriminant ability and test efficiency,and also with excellent stability and extrapolation effect.6.The simple scoring tool had a good performance:the AUC of male score model was 0.698(95%CI:0.677～0.718)and 0.703(95%CI:0.686～0.719)in female’s.Conclusion:1.In type 2 diabetes patients,there were a number of indicators existing gender differences,and the female diabetes had a higher CVD risk.2.Considering the competition risk and use Fine and Gray model with a adjust risk set,our models had a more stable estimation.3.Risk score model can be easily and intuitively used in diabetic community management institutions. |
PDF Full Text Request