Study On Influencing Factors And Prediction Models Of Fatty Liver Disease(FLD)in Population Of A Guangxi County

Background and Objective: Fatty liver disease(FLD)is a serious public health problem that is rapidly increasing and trending younger.In ethnic minority areas in China,the incidence of FLD reveals different characteristics because of environmental and lifestyle discrepancy,furthermore,FLD is influenced by both environmental and genetic factors,transcription factor EB(TFEB)is encoded by TFEB gene,and involving in lysosome biosynthesis,lipid metabolism and other biological processes.Evidences from animal studies indicate that TFEB is closely related to FLD,however epidemiological references from population are limited.Therefore,in this study,we analyzed the environment,gene and other related factors of FLD,and constructed prediction models for FLD based on machine learning arithmetic,so as to provide reference for the prevention and treatment of FLD in this area.Methods: Based on a healthy population cohort in Gongcheng County,Guangxi,in the chapter 1,402 patients with FLD and 624 health individuals,which were matched by sex and age(with variation of ±3 years)in a proportion of 1:1.5,were selected as the cases group and control group,respectively and the behavior,diet and other factors related to FLD were analyzed.According to the same method,228 patients in the case group and 342 individuals in the control group were selected from the totality and were used to explore the relationship between SNP loci polymorphisms of TFEB gene and FLD,and the gene-environment interaction was analyzed by multiplication and addition interaction models.In the chapter 2,the total data set was divided into training set and test set according to the ratio of 7:3,the former was used for model training and the latter was used for testing the performance of prediction models.After that three common machine learning models(Random forest,RF;Support vector machine,SVM;and Artificial neural network,ANN)were used to construct the prediction models of FLD,next their prediction performance was evaluated.Results: results of the influencing factors of FLD and its association with TFEB gene polymorphisms were as the follow: 1)between control group and case group,the history of diabetes and hypertension,smoking,smoking amount,and daily meditation time existed differences,in the case group except the waist circumference,BMI,SBP,DBP,Hb A1 C class index were significantly higher and the HDL-L was significantly lower than control group,other indicators did not have any difference.2)Multivariate Logistic regression analysis indicated that history of hypertension and diabetes,smoking amount,LDL-C and HDL-C were significantly negatively correlated with FLD,while daily sitting time,waist circumference,BMI,TC,ALT and UA were significantly positively correlated with FLD.3)The distribution of all SNP sites was uniform between the case group and control group,and there was no significant difference.4)Logistic regression analysis showed that there was no significant association between SNP locus and FLD.5)rs1015149,rs1062966,rs11754668 and rs2273068 had significant multiplicative interactions with smoking amount,and rs1062966 and BMI,and rs11754668 and alcohol intake also had significant multiplicative interactions.There was no significant additive interaction between each SNP locus and environmental factors in the incidence of FLD.The results of prediction models were as follows: 1)Through variable screening,13 indicators,including gender,age,history of diabetes,history of hypertension,smoking amount,daily sitting time,waist circumference,BMI,LDL-C,HDL-C,TC,ALT and UA,were utilized to construct the prediction models of FLD.2)Testing on test set,the FLD prediction accuracy of RF,SVM and ANN model were 80.52%,57.14% and 74.03%,respectively.Through comprehensive consideration of sensitivity and specific degrees,RF was the best model with the highest comprehensive performance,but the worst model was SVM.Conclusions: 1)For the population in this country,hypertension,history of diabetes,smoking amount,waist circumference,BMI,LDL-C,HDL-C and other factors may be related to FLD,and there were not directly associated between the polymorphism of rs1015149,rs1062966,rs11754668,rs14063,rs2273068 and rs73733015 of the TFEB gene with the susceptibility to FLD,but the risk of FLD may be changed via gene-environment interaction.2)Based on gender,age,history of hypertension,history of diabetes mellitus,smoking amount,ALT and UA,etc.13 indicators,the FLD prediction comprehensive performance of RF was better than that of ANN and SVM in test set,but it is still necessary to select the appropriate prediction model according to the actual work.