Establishment And Internal Validation Of Risk Assessment Model Of Metabolic Syndrome In Oil Workers

Objectives Logistic regression,random forest,convolutional neural network and support vector machine risk assessment models for metabolic syndrome of oil workers were established.The prediction performance of the models was evaluated from two aspects of differentiation and calibration,so as to obtain the optimal prediction model for metabolic syndrome of oil workers.Methods A cross-sectional study was used in this study.The subjects of this study were workers in an oil company who underwent occupational health examination in North China Downhole Oilfield Hospital from April 2017 to October 2018.After uniform and strict training,investigators conduct questionnaires,physical and blood biochemical tests on oil workers to collect relevant data and develop uniform diagnostic criteria.The subjects were divided into two groups according to whether they had Met S or not.IBM SPSS19.0 was used for univariate analysis and multivariate unconditional Logistic regression analysis,and relevant influencing factors of Met S of oil workers were screened out.A large number of literatures were reviewed and clinical experts were consulted,and input variables for the models established in this study were finally determined.Using Met S as the dependent variable of the models,the risk assessment models of Logistic regression,random forest,convolutional neural network and support vector machine were established by Python,respectively.The performance of the models was compared through the evaluation indexes such as accuracy,sensitivity,specificity,F1 score and Integrated Calibration Index,so as to obtain the optimal prediction model.Results 1 A total of 1468 oil workers were included in this study,and a total of 597 oil workers were diagnosed as Met S,with a prevalence rate of 40.67%.Among them,the central obesity rate was 56.81%,the abnormal blood pressure rate was 55.99%,the abnormal blood glucose rate was 49.39%,the abnormal triglyceride rate was 32.90%,and the abnormal high density lipoprotein rate was 19.28%.2 The single factor analysis of the present study,according to the results of two groups in age,BMI,gender,marital status,educational level,family average per capita monthly income,family history of high blood pressure,diabetes,family history,frequency,frequency of meat,dairy,edible salt intake frequency of carbonated drinks,physical exercise,smoking,drinking,shift work,occupational heat,noise contact,red blood cell count,blood platelet count,hemoglobin,uric acid,alanine aminotransferase,these differences were statistically significant(P<0.05).3 Multivariate analysis of the results of this study showed that age,family monthly income per capita,BMI,family history of diabetes,salt intake,frequency of dairy consumption,frequency of soda consumption,physical exercise,smoking,shift work,occupational heat,uric acid and alanine aminotransferase were predictors of Met S.Combined with relevant literature results and expert opinions,the above 13 variables were finally determined as the input variables of the models.4 The accuracy of Logistic regression,random forest,convolutional neural network and support vector machine risk prediction models was 82.49%,95.98%,92.03%,86.10%,Sensitivity was 87.94%,95.52%,90.59%,86.22%,specificity was 74.54%,96.65%,94.14%,85.93%,and Youden’s index was 0.62,0.92,0.85,0.72,respectively,F1 score was 0.86,0.97,0.93,0.89,AUC was 0.81,0.96,0.92,0.86,Integrated Calibration Index was 0.075,0.073,0.074,0.075,Brier score was 0.15,0.08,0.12,0.13,observed/expected ratio was 0.83,0.97,1.13,0.85,Calibrationin-the-large was 0.109,0.099,0.098,0.103,respectively.The accuracy,sensitivity,specificity,F1 score and calibration effect of the random forest risk prediction model are better than those of the other three models.Conclusions 1 Age,family monthly income per capita,BMI,family history of diabetes,salt intake,frequency of dairy consumption,frequency of soda consumption,physical exercise,smoking,shift work,occupational heat,uric acid and alanine aminotransferase affect the occurrence of Met S in oil workers.2 This study found that the random forest risk assessment model was the optimal model to evaluate the occurrence of Met S in oil workers,which can provide a basis for the related health management of this occupational population.Figure 7;Table 15;Reference 154...