The Establishment Of NAFLD Prediction Model Based On Tongue Diagnosis And The Correlation Study Of MRI Liver Fat Content

BackgroundNonalcoholic fatty liver disease(NAFLD)was the leading cause of chronic liver diseases in the world.It can progress to cirrhosis and even liver cancer,increasing the risk of death from cardiovascular diseases.So far,there has been no recognized effective drug approved for NAFLD treatment.Early effective lifestyle modification can significantly improve the poor prognosis.Traditional Chinese medicine(TCM)tongue diagnosis has great potential in disease screening.However,there was no research based on TCM tongue diagnosis for NAFLD screening.At the same time,magnetic resonance imaging-proton density fat fraction(MRI-PDFF)has become the most accurate non-invasive method for quantitative assessment of hepatic fat content,but the association between the amount of hepatic fat based on MRI-PDFF measurement and tongue features remained unclear.Objectives1.To establish a NAFLD prediction model on the basis of finding the characteristics of the tongue difference between NAFLD patients and the control group;2.To explore the correlation between hepatic fat content based on advanced magnetic resonance technology of hepatic fat quantification and tongue features;3.To investigate the influence of dietary habits and lifestyle factors on NAFLD for revealing the potential high-risk factors that affect the presence of NAFLDMethods1.708 subjects were enrolled consecutively from a health check-up center at a tertiary hospital for this observational study.Demographics,anthropometrics and laboratory data were collected by the unified information collection table.At the same time,the smart health mirror combined with the standardized 24-ColorChecker were used to collect the tongue image of each subject.After automatic color correction,the tongue image annotation platform was introduced to extract the tongue features,mainly including tongue color,fur color,tongue body fat or thin,fur thickness,greasy or curdy tongue coating,moist or dry tongue coating,teeth mark,tongue fissure,tongue spots,tongue petechia,ecchymosis,and peeled fur.All subjects were divided into NAFLD group and the control group according to the medical history and ultrasound diagnosis.The chi-square test or Fisher exact test was used to determine the difference of tongue features between the NAFLD group and the control group.2.The tongue features and clinical parameters(age,gender,abdominal obesity and BMI)are assigned as covariates,and NAFLD diagnosis is used as the dependent variable.The independent predictors of NAFLD are determined by univariate and multivariate logistic regression analysis.The NAFLD prediction scoring model was constructed according to the coefficient(? value)of corresponding covariate,including a simple score model(SSM)with only tongue features and a complex score model(CSM)composed of tongue features and clinical parameters.Using Hosmer-Lemshow goodness-of-fit test and area under the receiver operating characteristic curve(AUC)to evaluate the calibration and discrimination of each model in turn,and find the optimal threshold value(also called cut-off value)of each model based on the Youden index.In addition,the novel established NAFLD prediction scoring model was conducted subgroup analyses and internal validation,and compared the difference of AUC for predicting NAFLD by Delong test with existing serum noninvasive diagnostic models of liver steatosis such as FLI and HSI.3.131 of total subjects were included to undergo MRI-PDFF measurement for assessing hepatic fat content after tongue image collection.According to MRI-PDFF,these subjects were divided into non-NAFLD control groups(MRI-PDFF<5%),Mild fatty liver group(5%?MRI-PDFF<11%)and moderate-severe fatty liver group(MRI-PDFF?11%).The difference of tongue features between three groups was assessed by Mantel-Haenszel chi-square linear trend test.The association between MRI-PDFF and clinical parameters as well as existing eight serum noninvasive liver steatosis models was evaluated by Spearman's correlation.4.To carry out lifestyle assessment on 327 subjects who finally completed the diet questionnaire in the overall subjects.The dietary questionnaire was used to acquire the average intake frequency and average intake servings of each time of 17 common food types in the past one year,and then convert them into the average weekly intake frequency(times/week)of each food and average cumulative intake amount(serv./day).Then,using lifestyle factors such as diet,sleep,exercise,smoking,and drinking as covariates,multivariate logistic regression analysis was used to adjust other confounders to screen out the independent risk factors of NAFLD.Moreover,the frequency and cumulative intake amount of spicy food were further divided into four groups of Q1,Q2,Q3,Q4 and four groups of G1,G2,G3,and G4,respectively,according to their corresponding quartiles.Logistic regression analysis was used to calculate odds ration(OR)and 95%confidence interval(CI)of the presence of NAFLD in subjects with different levels frequency and intake amount of spicy food.A linear trend test for OR value was conducted to explore whether there was a dose-response relationship between the frequency and intake amount of spicy food and risk of NAFLD.Results1.The overall median age was 37(IQR:31-46)years old,ranging from 20 to 65 years old,with a male to female ratio of 2.4:1,including 430 cases(60.7%)in the NAFLD group and 278 cases(39.3%)in the control group.In terms of baseline characteristics,the levels of age,BMI,waist circumference,waist-to-hip ratio,waist-to-height ratio,ALT,AST,GGT,blood pressure,blood lipid,and the proportion of males,smoking history,and metabolic syndrome in the NAFLD group were significantly higher than those in the control group(all P values<0.05).In terms of tongue features,dark red tongue(72.6%vs 45.7%),red tongue(15.6%vs 4.05),yellow fur(72.1%vs 30.2%),thick fur(81.6%vs 49.3),greasy fur(80.2%vs 21.9%),enlarged tongue(36.7%vs 8.3%),teeth mask(52.1%vs 31.7%),tongue fissure(44.7%vs 24.5%),tongue spots(39.3%vs 29.1%)and tongue petechia(10.9%vs 2.9%)in NAFLD group have a higher proportion compared to the control group(all P values<0.05).2.Multivariate logistic regression analysis was conducted based on those tongue features with statistically significant differences after univariate analysis.The results showed that the tongue color,fur color,greasy tongue coating,tongue body fat or thin,teeth mark,tongue fissure,tongue spots,and tongue petechia were independent predictors of the presence of NAFLD.Taking the integer according to the coefficient(? value)to determine the score of corresponding tongue feature.The simple score model(SSM)was the sum of the scores,ranging from 0 to 24.The complex score model(CSM)was composed of these tongue features scores combined with the corresponding scores of three clinical parameters of age,BMI and abdominal obesity,ranging from 0 to 28.3.The AUC of SSM for predicting the presence of NAFLD was 0.907(95%CI:0.884-0.929,P<0.001),the cut-off value was ?12 points,and the corresponding sensitivity,specificity,positive predictive value(PPV),negative predictive value(NPV),positive likelihood ratio(PLR),the negative likelihood ratio(NLR)and accuracy was 79.1%,87.8%,90.9%,73.1%,6.47,0.24 and 82.5%,respectively.In total and different subgroups,the percentage of NAFLD in the group of SSM ?12 was significantly higher than that in the group of SSM<12(all P values<0.001).The AUC of CSM for predicting the presence of NAFLD was 0.937(95%CI:0.919-0.955,P<0.001),the cut-off value was ?11 points,and the corresponding sensitivity,specificity,PPV,NPV,PLR,NLR and accuracy was 90.2%,83.5%,89.4%,84.7%,5.47,0.12 and 87.6%,respectively.In total and different subgroups,the percentage of NAFLD in the group with CSM ?11 was significantly higher than that in the group with CSM<11(all P values<0.001).Hosmer-Lemeshow test showed that both SSM and CSM had good calibration capabilities(both P>0.05).In total and three subgroups of female,age<45 years,and non-smoking,the AUC of CSM for predicting NAFLD was significantly greater than SSM(all P values<0.05),while there was no significant difference in other subgroups(all P values>0.05).In addition,the AUC of SSM for predicting NAFLD was significantly higher than that of existing several noninvasive liver steatosis models(AUC range from 0.783 to 0.837),BMI(0.784),waist circumference(0.790),ALT(0.752),AST(0.695),TG(0.775),metabolic syndrome(0.687),as well as single differential tongue features(AUC range from 0.511 to 0.791)(all P values<0.05).4.Among the 131 subjects who underwent MRI-PDFF detection,35 cases(26.7%)in the control group,48 cases(36.6%)in the mild fatty liver group,and 48 cases(36.6%)in the moderate-severe fatty liver group.With the increase of liver fat content,the proportion of dark red tongue gradually increased(45.7%vs 68.8%vs 70.8%,P trend=0.007),the proportion of white fur gradually decreased(57.1%vs 39.6%vs 31.3%,P trend=0.048),the thickness of the tongue coating gradually increased,especially the thicker coating gradually increased(8.6%vs 20.8%vs 31.3%,P trend=0.001),the proportion of greasy coating gradually increased(31.4%vs 66.7%vs 77.1%,P trend<0.001),the proportion of enlarged tongue gradually increased(14.3%vs 37.5%vs 56.3%,P trend=0.005),the proportion of teeth mask gradually increased(37.1%vs 58.3%vs 68.8%,Ptrend=0.005),and the proportion of tongue spots also gradually increased(31.4%vs 70.8%vs 62.5%,P trend=0.010),but there was no significant difference in the appearance rates of tongue fur moistening,tongue fissure,tongue petechia,ecchymosis and peeled fur(all P trend>0.05).At the same time,MRI-PDFF was positively correlated with BMI,WC,WHR,WHtR,SBP,DBP,ALT,AST,UA,FBG,TG,TC,LDL-C and existing noninvasive models of liver steatosis(all P values<0.05),which was inversely related to HDL-C(P=0.001).5.Among the 327 subjects who completed the diet questionnaire,128 cases(39.1%)were in the control group,and 199 cases(60.9%)were in the NAFLD group.After adjusting for various confounders,multivariate logistic regression analysis showed that barbecue food intake amount(OR=3.66,95%CI:1.40-9.55),nut intake frequency(OR=1.18,95%CI:1.04-1.34),the frequency of spicy food intake(OR=1.10,95%CI:1.01-1.21)and adequate exercise(OR=0.32,95%CI:0.15-0.71)were independently related to NAFLD(P values<0.05).Compared with the spicy food intake frequency Q1 group,the OR of having NAFLD in the Q2,Q3,and Q4 groups were 1.60(95%CI:0.88-2.89,P=0.122),1.95(95%CI:1.01-3.75,P=0.047)and 2.61(95%CI:1.40-4.86,P=0.003),respectively,which obviously showed an increasing trend(P trend=0.002).Compared with the spicy food intake amount G1 group,the OR of having NAFLD in the G2,G3,and G4 groups were 2.09(95%CI:1.11-3.93,P=0.022),2.09(95%CI:1.11-3.93,P=0.022)and 2.74(95%Cl:1.46-5.18,P=0.002),respectively,which also showed a clear increasing trend(P trend=0.003).After further adjusted various confounding factors,the increasing trend still existedConclusions1.The distinctive tongue features between patients with NAFLD and control group can be found through tongue elements such as tongue color,tongue shape,fur color,fur thickness,greasy tongue coating,teeth mask,tongue spots,tongue fissure and tongue petechia,which were also related to the degree of liver steatosis based on MRI-PDFF.2.The NAFLD prediction model based on tongue features had high accuracy and could be used as a novel method for NAFLD screening3.Less exercise and excessive intake of nuts,spicy food and grilled food were independent risk factors for NAFLD,and the frequency and intake amount of spicy food were dose-responsive to NAFLD risk.