Establishment And Evaluation Of Risk Model Of Type 2diabetes Mellitus In Urumqi Adults

Objective To Establish and evaluate the risk model of type 2 diabetes in residents of Urumqi.Methods We used random cluster sampling to recruit examiners in the A and the B community health service center in Midong and Xinshi district in Urumqi.From July 2018 to January 2019,we did questionnaires,physical measurements,body fat testing,and laboratory testing among residents at the age between 35 and 74 years old.Data from the A community was set to be the training set,and single factor logistic regression model and random forests algorithm were used to select variables in it.After that,multivariate logistic regression was used to establish a diabetes risk model,ten-fold cross-validation was used to make internal verification.Data from the B community was set to be the test set to evaluate the diabetes risk model,and discrimination and calibration of the model was evaluated by area under the receiver operating characteristic(ROC)curve(AUC)and statistical test the difference between the predicted value of the model and the actual observed value.Results The study included 1221 subjects from the A community,and1004 subjects from the B community.The risk model of diabetes mellitus was assessed by using multivariate logistic regression after screening variables with univariate logistic regression(method 1)based on the data of community B residents,the sensitivity was70.1%,the specificity was 64.6%,the predictive consistency rate was 65.4%,and the AUC was 0.717,95%CI(0.672,0.761).The risk model of diabetes mellitus was assessed by using multivariate logistic regression after screening variables with random forests algorithm(method 2)based on the data of community B residents,the sensitivity was70.1%,the specificity was 60.9%,the predictive consistency rate was 62.4%,and the AUC was 0.708,95%CI(0.664,0.749).There was no statistically significant difference between the two AUC.The difference between the predicted and actual observed values of diabetes in the two prediction methods was not statistically significant by Hosmer-Lemeshow(H-L)test.That is to say,both prediction methods have goodcalibration.It can also be seen from the diagram of the relationship between the predicted value of the model and the observed value that the calibration of the two prediction methods is better.Conclusion The risk model of diabetes established by method 1 and method 2 both had high predictive power.The two methods have high discrimination and good calibration.