| Aims ①To analysis the risk factors and develop prediction models based on H2O AutoML for the incidence of NAFLD in normal population.②To analysis the risk factors and develop prediction models based on monogram for the incidence of NAFLD in lean population.③ To analysis the correlation of intestinal mucosal barrier function and the inflammatory factors in patients with NAFLD.Methods ①A total of 7,458 subjects was recruited in the study.The data were loaded using H2O AutoML to develop various machine learning models to predict normal NAFLD.The model was evaluated by accuracy,sensitivity,specificity,and visualized the important variables.②A total of 5,037 individuals BMI<23 kg/m2 were included in this study,and the data were separated for training and validation.The logistic regression method was used.And a nomogram,a type of prediction model,was constructed according to the logistic regression analysis and the significant clinical factors.The performance of this model was evaluated based on its discrimination,calibration,and clinical utility.③ From March 2021 to November 2022 in the gastrointestinal department of first affiliated hospital of Soochow university,twenty patients who performed gastroscopic surgery,colonoscopy was normal,and abdominal ultrasound indicated fatty liver were selected as the experimental group(NAFLD group).At the same time,twenty patients who performed gastroscopic surgery,colonoscopy was normal,and abdominal ultrasound indicated normal liver were selected as the control group(Non-NAFLD group).Blood samples of NAFLD patients and non-NAFLD individuals were collected and the levels of DAO,CRP,TNF-α,IL-6 were detected.R software(4.1.0)and SPSS(27.0)was used for statistical analysis of the research data.Results ①The individuals were divided into the training(n=5966)or validation(n=1492)cohorts at a ratio of 8 to 2.Ten machine learning models were fitted.The best model was a GBM model.BMI.TG.ALT.GGT.and HDL were the important variables.The accuracy was 0.799.sensitivity was 0.777 and specificity was 0.809.②The individuals were divided into the training(n=4,068)or validation(n=969)cohorts at a ratio of 8 to 2.Sex.age.RBC.platelets.ALT,TC,TG,LDL,creatinine,uric acid,glycosylated hemoglobin are independent risk factors for NAFLD in lean people.The nomogram was constructed based on seven predictors:ALT,TC,TG,LDL,creatinine,uric acid,and hemoglobin A1C.The model based on these factors showed good predictive ability in the training set and in the internal validation set,with AUCs of 0.870 and 0.887,respectively.The calibration curves and decision curve analysis(DCA)displayed good clinical utility.③In NAFLD group,DAO level[19.80(6.06,33.54)]ng/ml was significantly higher than that in non-NAFLD group[10.47(2.19,18.75)]ng/ml and the difference was statistically significant(P=0.013).The level of lymphocyte in NAFLD group[2.20(1.52,2.88)]*109/L was higher than that in non-NAFLD group[1.61(1.35,1.87)]*109/L,and there was a significant difference between the two groups(P=0.001).The level of TNF-α in NAFLD group[76.33(19.61,133.05)]pg/ml was higher than that in non-NAFLD group[46.39(18.41,74.37)]pg/ml,and there was a significant difference between the two groups(P=0.041).Lymphocyte and TNF-α was higher than that in non-NAFLD group and there was a significant difference between the two groups.Conclusions①The AutoML model based on BMI,TG,ALT,GGT,HDL has a good performance in predicting NAFLD in normal population.②The nomogram model based on ALT,TC,TG,LDL,creatinine,uric acid and glycosylated hemoglobin has a good performance in predicting NAFLD in lean population.③The levels of DAO and inflammatory related factors in NAFLD patients were higher than those in the control group,indicating that NAFLD was associated with increased intestinal permeability and inflammatory status. |
