Prediction Model Of Atherosclerotic Cardiovascular Disease In Uygur And Kazak People Based On Survival Analysis

Objective:1.To understand the incidence of atherosclerotic cardiovascular disease(ASCVD)among Uygur and Kazak people in rural Xinjiang through a prospective cohort study;2.Using Cox regression,Lasso-Cox,and Random Survival Forest algorithms,prediction models were constructed based on gender in this population,and the performance of the China Atherosclerotic Cardiovascular Disease Risk Prediction Study(China-PAR)and Framingham Prediction Model(FRS)in this population was compared;3.Rank the importance of the variables screened in different models to determine the possible risk factors of ASCVD in this population.Method:1.Through multi-stage stratified cluster sampling,Jiashi County,Kashgar Prefecture,southern Xinjiang ,the 51st Regiment of the 3rd Division of the XPCC,and Narati Town,Xinyuan County,Yili Prefecture,northern Xinjiang were selected for field epidemiological investigation;Completed the baseline survey of Jiashi County and Xinyuan County in 2010-2012,and completed the baseline survey of the 51st Regiment of the Third Division in 2016;The population of Jiashi County and Xinyuan County were followed up in 2013,2016 and 2017,and the 51 regiments of the Third Division were followed up in2019,2020,2021 and 2022.The baseline and outcome information of the respondents were collected through questionnaires.The inpatient medical records and social security information of local hospitals were continuously collected to ensure the integrity of the information.2.Excluding subjects with ASCVD at baseline,loss of follow-up,and missing blood samples during follow-up,a total of 11902 subjects were included,with a median follow-up time of 5.79 years.Randomly,the data was split into training and test data,with a 7:3 ratio for each gender.Parameter optimization and model construction were carried out in the training data,and model performance evaluation was carried out in the test data.3.Cox regression,Lasso-Cox and RSF were used for variable screening to construct different gender ASCVD prediction models.The performance of the models was compared with that of the classical China-PAR and FRS models.The discrimination of the model was evaluated by the consistency index,the net reclassification index,and the comprehensive discrimination improvement index;The calibration was evaluated by Greenwood-Nam-D’Agostino(GND)goodness of fit test(P>0.05 represents good calibration of the model)and Blair score;Clinical effectiveness was evaluated through decision curve(DCA)analysis.4.Evaluate the importance of variables on the subset of variables screened by Cox regression,Lasso-Cox,and RSF models,and evaluate the impact of different variables on the incidence risk of ASCVD.Results:1.The cumulative incidence of ASCVD in Xinjiang Uygur and Kazak population was 10.48%(1247/11902);The cumulative incidence rate of ASCVD in the Kazakh population was 13.51%(242/1791),of which the incidence rate of males was 11.25%(82/729),and the incidence rate of the female was17.74%(160/902);The cumulative incidence rate of ASCVD in Uighur population was 9.94%(1005/10111),of which the incidence rate of males was 7.19%(370/5147),and the incidence rate of females was 12.79%(635/4964).2.According to the randomization principle,the data of different genders were divided into training data and test data according to 7:3.There was no difference between the training data and test data for the same variable.A total of 57 variables including general demographic information,disease history and family history,anthropometric indicators,and blood biochemical indicators were included in the study.Cox regression,Lasso-Cox and RSF were used to screen the self-variable quantum set to establish a prediction model for different sexes.8,9 and 20 variables were included in the male population,and 9,11 and 25variables were included in the female population.3.The comparison of model discrimination showed that the prediction performance of Cox regression and Lasso-Cox regression in the test set of the male population was similar(P>0.05),superior to RSF,China-PAR and FRS(P<0.05).The C statistics of each model were Lasso-Cox 0.752 95%CI(0.711,0.793),Cox regression 0.745 95%CI(0.701,0.788),China-PAR 0.720 95%CI(0.674,0.766),RSF 0.713 95%CI(0.665,0.760)and FRS 0.711 95%CI(0.665,0.757).The prediction performance of Cox regression,Lasso-Cox,RSF and China-PAR models in the test the set of female population was similar(P>0.05),which was superior to FRS.The C statistics of each model were Lasso-Cox 0.751 95%CI(0.722,0.780),Cox regression 0.750 95%CI(0.720,0.779),RSF 0.744 95%CI(0.713,0.775),China-PAR 0.739 95%CI(0.707,0.771)and FRS 0.725 95%CI(0.694,0.756).4.The calibration of the model:in the male population,the number of patients predicted by Cox regression,Lasso-Cox,RSF,China-PAR,and FRS was 114.87,114.71,203.44,90.6,7 and 90.82 respectively;The GNDx~2values were 9.25(P=0.32),12.47(P=0.13),43.41(P<0.01),72.42(P<0.01)and 38.04(P<0.01)respectively;Among the female population,the number of patients predicted by Cox regression,Lasso-Cox,RSF and China-PAR was 204.38,202.63,326.94 and 138.06 respectively;The GNDx~2values were 14.70(P=0.11),11.66(P=0.25),33.22(P<0.01)and 142.49(P<0.01),respectively.5.Clinical application value:The results of the decision curve(DCA)showed that Cox regression and Lasso-Cox had clinical application value.When the risk threshold was 0.00-0.20,the net benefit of Cox regression in the male population was greater than Lasso-Cox;When the risk threshold was 0.00-0.25,the net benefit of Cox regression in female population was greater than that of Lasso-Cox regression.6.The importance of variables in Cox regression,Lasso-Cox,and RSF model variable subsets showed that in addition to traditional ASCVD risk factors such as age,blood pressure,and ethnic group,metabolic indicators such as LDL-C,AIP,and waist to hip ratio(WHR),which reflect the degree of obesity,were also important predictors of ASCVD in men.In the female population,in addition to traditional risk factors such as age and systolic blood pressure,hip circumference(HC)and apolipoprotein B(APOB),which reflect obesity,were also important predictors of ASCVD in the female population.Conclusion:1.The cumulative incidence rate of ASCVD in Xinjiang Uygur and Kazak population was 10.48%,and the incidence density was 18.55/1000 person years,higher than the national average;The cumulative incidence rate of ASCVD in Kazak population was 13.51%,and the cumulative incidence rate of ASCVD in Uygur population was 9.94%;The incidence rate of women in both ethnic groups was higher than that of men.2.Cox regression had higher performance than other prediction models,and the China-PAR model and FRS model had a certain underestimation of the risk prediction of ASCVD in this population.3.The results of variable importance of different subsets of variables showed that traditional risk factors such as age and systolic blood pressure were more important influencing factors for the incidence of ASCVD in this population,and metabolism-related LDL-C,APOB,AIP and WHR reflecting the degree of obesity were also important predictors of ASCVD in this population.