Regression Models For Predicting Visceral Fat Mass Based On Fatty Liver Disease And BMI Or Weight In Adults

Part Ⅰ Regression models for predicting visceral fat mass based on fatty liver disease diagnosed by two-dimensional ultrasound in adultsBackgroundObesity is the presence of excessive adipose tissue deposition in the body.Obesity is a heterogeneous disorder that obese individuals may have substantial differences in body fat deposition,metabolic status,and health risk levels despite similar body weight.The regional distribution,especially visceral adipose tissue(VAT),rather than the total amount of adipose tissue is more important for the morbidity and mortality of metabolic diseases.Visceral obesity is characterized with increased visceral and ectopic adipose tissue deposition(EAT),adipocyte dysfunction,chronic inflammation,adipokine dysregulation,and insulin resistance(IR).IR is strongly associated with type 2 diabetes,hypertension,dyslipidemia,atherosclerosis,and other metabolic diseases.Therefore,VAT is a strong indicator for metabolic syndrome,dyslipidemia,hypertension and cardiovascular disease.VAT can be assessed by CT,MRI or Dual energy X-Ray absorptiometry(DXA).They have been shown to evaluate body composition in total and local regions,allowing the measurement of body fat and muscles.However,the radiation exposure,high cost and technical requirements of CT or MRI have limited their use in large population.The DXA with corescan software can automatically measure VAT,which is has been proven to be strongly correlated with MRI and CT.DXA has lower radiation exposure than CT,however,it might be limited in healthy screening due to the feasibility issues in some remote areas.Recently,several studies have established prediction equations for VAT using simple anthropometric parameters,but the application of these equations is limited due to racial differences in the distribution of visceral fat in China.ObjectiveTo assess the correlation between fatty liver and VAT and construct prediction models for VAT by simple anthropometric measurements and validate the models in validation group.MethodsA total of 515 subjects were enrolled.The age,gender,height,weight,blood examination,ultrasound images,and DXA data were collected from the subjects’medical records,ultrasound database,and DXA database.Ultrasound images were retrospectively analyzed by three experienced sonographers.All subjects were signed to four groups,that is normal(Group 1),mild(Group 2),moderate(Group 3)and severe(Group 4).Statistics were performed using SPSS 24.0(IBM SPSS,Chicago,IL)and MedCalc 19.0.4(MedCalc soft,Ostend).P<0.05 was considered statistically significant.All subjects were randomly divided into a derivation group(70%)and a validation group(30%).The Kolmogorov-Smimov test was used evaluate the normal distribution of the data.Differences in categorical variables were tested using chi-square tests.Observer agreements were tested by the intraclass correlation coefficient(ICC)test.The Mann-Whitney U test were used to assess the differences in anthropometric parameters,lipid profile and VAT mass between the derivation and validation groups.The Kruskal-Wallis test was used to compare the VAT mass among different groups of hepatic steatosis.The correlation coefficients between the VAT mass and included parameters were evaluated with Spearman’s correlation analysis.Multiple linear stepwise regression analysis was used to develop prediction models with DXA-VAT mass as a dependent variable for males and females both together and separately.The correlations between DXA-VAT mass and the predicted VAT mass were assessed using Spearman’s correlation.Bland-Altman plots were plotted to illustrate the agreement of the prediction models.Separate receiver operating characteristic(ROC)curves constructed for DXA-VAT masses less than and greater than1280g.The sensitivity,specificity and area under the curve(AUC)of the prediction models calculated to determine the performance of the prediction models.The didfference in AUCs between the two models were compared using the DeLong test.The Hosmer-Lemeshow test was used to assess the calibration of the prediction models.ResultsA total of 515 subjects(262 males and 253 females)were included in this study.The median age was 58 years(IQR=50-65 years)for all participants,57 years(IQR=49-64 years)for males and 60 years(IQR=52-66 years)for females.There were no statistically significant differences between male and female in heart rate,blood pressure,degree of hepatic steatosis,and triglycerides(TG).There were statistically significant differences in age,height,weight,BMI,VAT,and high-density lipoprotein(HDL).Subjects were randomly assigned to the derivation group(n=366,70%)or validation group(n=149,30%).There were no significant differences between two groups in age,gender,weight,height,BMI,alanine aminotransferase(ALT),total cholesterol(TC),TG,HDL,low-density lipoprotein(LDL),VAT,prevalence of diabetes mellitus(DM%)or prevalence of hypertension(HBP%).There were significant differences in fasting blood glucose(FBG)and aspartate aminotransferase(AST)between two groups.The inter-observer correlation coefficient for three sonographers was 0.905(95%CI:0.891-0.918).The intra-observer correlation coefficient for one sonographer was 0.921(95%CI:0.891-0.943).The DXA-VAT mass and the grade of hepatic steatosis were moderately correlated(r=0.527,P<0.001).In the derivation group,age,gender,BMI,HDL,TG and the grade of hepatic steatosis were included in the final prediction model 1(F=143.074,P<0.001).The prediction model 2 included the age,gender,height,weight,HDL,TG and the grade of hepatic steatosis(F=156.734,P<0.001).Subgroup prediction models were also developed for males and females.The variances of subgroup models were slightly lower than those for both gender.In the validation group,the predicted VATs of model 1(r=0.875,P<0.001)and model 2(r=0.870,P<0.001)were strongly correlated with the DXA-VAT mass.The mean deviations between the predicted VAT and DXA-VAT mass in model 1 and model were 12.8(95%IC:-55.266-80.876)and 12.1(95%IC:-54.479-78.646),respectively.The proportional deviations in tweo models were not statistically significant(r=0.005,P=0.953;r=0.004,P=0.962,respectively).The validation group was divided into two groups of DXA-VAT<1280g(n=70)and VAT≥1280g(n=79).ROC curves were drawn and the AUC,sensitivity and specificity were cauculated.There was no statistical significance between the AUCs of the two prediction models(P=0.933).The Hosmer-Lemeshow test showed that the P values for model 1 and model 2 were 0.696 and 0.683,respectively.Conclusion1.The prediction model 1 including age,gender,BMI,degree of liver steatosis,TG and HDL and the prediction model 2 including age,sex,height,weight,degree of liver steatosis,TG and HDL had good agreements and perfromance.The variance of the gender subgroup prediction models was slightly lower than those of model 1 and model 2.2.VAT mass was correlated with weight,BMI,TG and HDL.Weight and BMI were the strong predictors for VAT mass,and combining age,gender,degree of liver steatosis,TG and HDL could increase the variance of the models.3.The diagonsis of fatty liver using ultrasound had good observer agreements.The association between fatty liver and VAT mass could be established through regression models.Part Ⅱ Correlation of ultrasound-diagnosed fatty liver disease with DXA-measured body composition parametersBackgroundFatty liver disease(FLD)is a global public health problem with the prevalence has been rising globally,usually paralleling the prevalence of obesity.Obesity is strongly and independently associated with FLD.In overweight and obesity,the free fatty acids(FFA)are increased and flow to the liver.In the liver,FFAs can be metabolized by esterification for the production of triglycerides that are stored within hepatocytes,which subsequently leads to multi-organ insulin resistance(IR).In the overweight or obesity,the adipokines,such as leptin and adiponectin,are imbalanced and the cytokines are increased due to the chronic low-grade inflammation in the liver which leads to the interactions between the adipokine and cytokine affecting the fat deposition in the liver.The prevalence is significantly higher in people with overweight or obesity than in those with normal weight or underweight,with a prevalence of up to 80%in overweight or obesity peoples compared to 16%in those with normal weight without metabolic risk factors.BMI is one of the most commonly used methods to assess obesity,and the increased BMI have shown to be associated with the development of FLD,but BMI cannot distinguish the body fat mass and lean tissue mass.Dual-energy X-ray absorptiometry(DXA)can automatically measure the whole-body or local soft tissue and bone mineral content(BMC),such as body fat mass(FM),fat mass percentage(FM%),lean mass(LM),lean mass percentage(LM%),fat mass index(FMI),and visceral adipose tissue(VAT).Among that,FMI is a method assessing the whole-body fat mass,and FM%a method assessing the whole-body fat percentage.Some studies found that FM%,visceral fat area(VFA)and FMI are associated with the FLD and metabolic syndrome.Recently,the muscle tissue loss and or muscle strength loss have received increasing attention in its association with FLD.ObjectiveTo compare the differences in body composition parameters between FLD groups and non-FLD groups and to investigate the association between body composition parameters with FLD.Draw the ROC curves to calculate the best threshold values for predicting FLD.Methods515 subjects were collected in the study.The clinical data,ultrasound images and DXA data were collected from subjects’ medical records,ultrasound database and DXA database,including age,gender,height,weight,blood biochemical profile,and so on.All ultrasound images were divided into non-FLD group and FLD group(including mild,moderate,and severe)based on the ultrasound performance.We divided the subjects in males group and females group due to the differences in body composition between males and females and compared the differences in body composition parameters between non-FLD groups and FLD groups in gender subgroup.Statistical analyses were performed using SPSS 24.0(IBM SPSS,Chicago,IL)software and MedCalc.P<0.05 was considered statistically significant.The Kolmogorov-Smirnov test was conducted to assess the normal distribution of the data.Categorical variables were expressed by percentages and continuous variables were expressed by median ± interquartile ranges(IQRs).The chi-square test was used to compare categorical variables.The differences in general information,blood biochemical profiles and body composition parameters between the two groups in gender subgroup were performed with the Mann-Whitney U test.Binary logistic regression was used to calculate the odds ratio(OR)and 95%confidence interval(95%CI)for body composition parameters such as VAT,BMI,and VAT/ASMI.Multiple logistic regression analysis was also used to adjust for confounders,with model 1 adjusting for age;model 2 adjusting for factors including model 1+systolic blood pressure(SBP),diastolic blood pressure(DBP),presence of hypertension(1 for yes,0 for no),and presence of diabetes(1 for yes,0 for no);and model 3 adjusting for factors including model 2+ALT,AST,FPG,TG,Chol,HDL and GGT.ROC curves were plotted and the AUC,Youden index,sensitivity,specificity and cut-off value were calculated.Results1.The weight,BMI,A/G,ALT,GGT,FPG,TG and TG/HDL were significantly higher in the FLD group than those in the non-FLD group in male and female group(all P<0.01);while HDL was significantly lower in FLD group than that in the non-FLD group(P=0.001).In addition,the differences between the non-FLD and FLD groups were statistically significant in age,height,AST,and Chol(all P<0.05)in male.VAT,FMI,ASMI and VAT/ASMI were significantly lower in the non-FLD group than in the FLD group both in male and female group(all P<0.01),while LM%was significantly higher in the non-group than the FLD group(P<0.001).2.The correlations between BMI,VAT,VAT/ASMI and FLD were analyzed for each subgroup.A crude logistic regression was constructed for the BMI,VAT and VAT/ASMI as the continuous variables.The results showed that BMI,VAT and VAT/ASMI increased the risk of FLD in both male and female,and the relationship still existed after adjusting for the confounders according to model 1,model 2,and model 3.With an increase of one unit in BMI,VAT and VAT/ASMI,the risk of hepatic steatosis was increased in both male and female.Then,logistic regression based on quartiles of BMI,VAT and VAT/ASMI of Q1,Q2,Q3,and Q4 with adjustment for confounders revealed that the Q2-Q4 group of BMI,VAT and VAT/ASMI was linearly associated with an increased risk of FLD compared with Q1.The relationship still existed even after adjusting for age in model 1,demographic characteristics in model 2,and blood biochemicals profiles in model 3.3.Receiver operating characteristic(ROC)curves were constructed for male and female based on BMI,VAT and VAT/ASMI,and the area under the curve(AUC),sensitivity,specificity,Youden index and cutoff values were calculated.The ROC curves for BMI,VAT and VAT/ASMI were statistically significant(all P<0.001).In male,the AUCs for BMI,VAT and VAT/ASMI were 0.829,0818 and 0.778,respectively.In female,the AUCs for BMI,VAT and VAT/ASMI were 0.741,0.805 and 0.785,respectively.The cut-off values for BMI and VAT were 24.2 and 1557g for male and the cut-off value for VAT was 954g for female.Conclusion1.The FLD group had higher BMI,VAT,FMI,FM%,ASMI and VAT/ASMI than the non-FLD group,while the FLD group had significantly lower LM%than the non-FLD group.2.BMI,VAT and VAT/ASMI were risk factors for FLD.There was an increased risk for FLD if the BMI,VAT and VAT/ASMI were higher.3.ROC curves showed that VAT was a predictor of FLD in both male and female,with the optimal cut-off values of 1557g and 954g for male and female,respectively.