Establishment and Evaluation of a Noninvasive Metabolism-Related Fatty Liver Screening and Dynamic Monitoring Model: Cross-Sectional Study

Introduction

Nonalcoholic fatty liver disease (NAFLD) is regarded as an important cause of liver disease, affecting more than 25% of the general population worldwide; more than 50% of patients with NAFLD also have dysmetabolism [1,2]. In 2020, experts redefined NAFLD as metabolically associated fatty liver disease (MAFLD), and much emphasis was placed on the presence of metabolic-related diseases or dysfunction [3-5]. Researchers have found that MAFLD is a multisystem disease, and liver steatosis is associated with type 2 diabetes, chronic kidney disease, cardiovascular disease, and other diseases that interact and form a vicious cycle [6-14]. China has the highest incidence of NAFLD or MAFLD morbidity in Asia [3,15,16]. Therefore, much attention should be given to MAFLD by enhancing awareness of MAFLD and optimizing its management.

To date, guidelines have suggested that liver biopsy could serve as the gold standard to diagnose histological liver damage, but noninvasive, quantitative assessment of liver fibrosis may also have prognostic implications. Ratziu et al [17] collected liver biopsy samples from 51 patients and found that 41% of the patients were at different stages of liver fibrosis or had nonalcoholic steatohepatitis. The uneven distribution of histological lesions inevitably led to sampling error when performing biopsy. Abdominal imaging, such as B-ultrasound imaging and the controlled attenuation parameter (CAP) technique, can be used to diagnosis liver disease; the former is less sensitive to mild steatosis, while the latter can detect steatosis of more than 5% and is one of the most common noninvasive methods for quantifying hepatic steatosis and fibrosis clinically [18,19]. The European Association for the Study of the Liver, European Association for the Study of Diabetes, and European Association for the Study of Obesity updated the clinical practice guidelines that propose that the nonalcoholic fatty liver disease fibrosis score (NFS) and fibrosis-4 (FIB-4) index can be used as prognostic markers for the progression of liver disease [20]. The NFS has higher specificity in the older adult population (individuals aged >65 years old) [21,22]. The predictive performance of the NFS, FIB-4 index, and aspartate aminotransferase-to-platelet ratio index (APRI) has been consistent in relation to rates of liver-related disease and mortality but is less valuable for the prediction of liver fibrosis [23]. One study found that the combination of the NFS, FIB-4 index, and liver stiffness measurement greatly improved the diagnostic accuracy, and the performance was similar to that of liver biopsy [24]. A cross-sectional study found that the triglyceride glucose (TyG) index was positively correlated with the likelihood and severity of NAFLD. The TyG index is generally considered a biomarker of steatosis, while its causal role in the judgement of fibrosis progression remains unclear [25,26]. In addition, the hepatic steatosis index (HSI) is more accurate in discriminating between MAFLD and nonfatty liver disease (non-FLD) than ultrasound. The predictive ability of the CAP for steatosis is superior to that of the HSI, and the HSI is more effective at discriminating patients with moderate-to-severe disease [18,27].

Studies have shown that numerous anthropometric indicators, such as BMI, waist-height ratio (WHtR), waist-hip ratio, and body adiposity index, are applicable for the quantification of visceral steatosis [28-32]. Body fat scales, a new popular domestic tool for health analysis, can be used to analyze basic parameters of body conditions such as the basal metabolic rate (BMR), body water distribution, and fat distribution. Reputable experts in the field have conducted extensive long-term studies on NAFLD and MAFLD, yet few noninvasive scoring models that accurately reflect disease activity or progression have been identified [33,34].

Therefore, there is an urgent need to identify more accurate predictive indicators and develop new screening methods for early MAFLD screening. The aim of this study was to construct a noninvasive prediction system for MAFLD, explore this new system for early screening in public, and establish a home-based tool for regular self-assessment and monitoring of MAFLD.

Methods

Study Population

The participants came from Hangzhou, Shaoxing, and Quzhou from March 2021 to November 2021, and a total of 2097 participants were enrolled (Figure 1). All participants signed the informed consent form and completed the examination as required. There were 1758 eligible participants in the training set who truthfully and completely answered the questionnaire, which contained items regarding height, weight, drinking history, past medical history, and other basic information.

To validate the results of the training set, there were 200 eligible participants grouped into the testing set.

All participants were diagnosed using the liver stiffness measurement and classified according to the CAP. CAP values <238 was considered to indicate a healthy liver, ≥238 and <259 was considered to indicate mildly fatty liver, ≥259 and <292 was considered to indicate moderately fatty liver, and ≥292 was considered to indicate severely fatty liver [35,36].

‎

Exclusion Criteria

Patients who met one or more of the following criteria could not participate in this study: (1) <18 years old; (2) long-term use of various health products and drugs; (3) presence of cirrhosis or liver cancer; (4) previous organ transplantation; and (5) patients for whom B-ultrasound imaging indicated FLD but who could not be diagnosed with MAFLD.

Diagnostic Criteria

The researchers in this study entered and organized the data, and the following diagnostic criteria were used to distinguish MAFLD patients [3]: (1) overweight or obese (BMI ≥23 kg/m² in Asians), (2) presence of type 2 diabetes mellitus, and (3) at least 2 of the following metabolic risk abnormalities: waist circumference ≥90 cm in Asian men and ≥80 cm in Asian women; blood pressure ≥130/80 mm Hg or specific drug treatment; triglyceride (TG) level ≥1.7 mmol/L or specific drug treatment; high-density lipoprotein cholesterol (HDL-c) level <1.0 mmol/L for men and <1.3 mmol/L for women or specific drug treatment; fasting plasma glucose (FPG) level of 5.6 mmol/L to 6.9 mmol/L, 2-hour postload glucose level of 7.8 mmol/L to 11.0 mmol/L, or glycosylated hemoglobin level of 5.7% to 6.4%; and homeostasis model assessment of insulin resistance score ≥2.5.

Data Collection and Model Selection

All items were completed under the guidance of the researchers. The participants underwent fasting blood tests. Waist circumference and hip circumference were measured with the participants wearing thin clothes. Body composition analysis was performed with bare feet. The patients were in a supine position during the FibroScan exam, and the right upper limb was held high and flat close to the ear. The probe was moved a small distance from the anchor point so that the most suitable detection point could be determined.

We collected basic information, including sex, age, height, weight, BMI, waist circumference, hip circumference, blood pressure, heart rate, and alcohol consumption history. The following laboratory results were included: alanine aminotransferase (ALT), aspartate aminotransferase (AST), glutamyl transpeptidase (GGT), alkaline phosphatase, hemoglobin, total cholesterol (TC), TG, HDL-c, low-density lipoprotein cholesterol (LDL-c), uric acid, and FPG levels, as well as white blood cell, red blood cell, and platelet (PLT) counts.

A body composition analyzer (InBody770, Biospace) was used to measure body composition and determine total body water (TBW), intracellular water, skeletal muscle mass, protein, and body fat mass (BFM). A body fat scale (3 Pro, Huawei) was used to determine the BMR, fat%, and visceral fat area (VFA).

The models or formulas involved in this study, including BMI, FIB-4 index, Forns index score, APRI, glutamyl transpeptidase-to-platelet ratio index (GPR), HSI, and TyG index, were developed using the following standard equations:

BMI=weight/height²

FIB-4=age×AST/(PLT×√ALT)

Forns index score=7.811-3.131×In PLT(109/L)+0.781×In GGT+3.467×In age-0.014×TC

APRI=(AST/upper limit of normal)/PLT×100

glutamyl transpeptidase-to-platelet ratio index=(GGT/upper limit of normal)/PLT×100

HSI=8×(ALT/AST)+BMI (female+2, diabetes+2)

TyG=ln (TG×FPG/2)

Statistical Analysis Methods

Participants were divided into the non-FLD group, which was the healthy group; MAFLD group; and NAFLD group.

All data obtained in this study were analyzed using SPSS version 26.0 (IBM Corp). The continuous variables were tested for normality and homogeneity of variance. A t test was performed for measurement data that followed a normal distribution, and the results are expressed as the mean (SD). Nonnormally distributed data were analyzed using nonparametric tests, and the results are represented by quartiles. The chi-square test or Fisher precision probability test was used for quantitative data such as sex. ANOVA was followed by post hoc analysis tests to compare numerical data among the 3 groups (MAFLD, NAFLD, and non-FLD). A P<.01 indicated that the difference was statistically significant.

Binary logistic regression analysis was performed to explore the potential anthropometric indicators with predictive ability to screen for MAFLD. A receiver operating characteristic (ROC) curve was drawn based on the selected indicators, and the area under the ROC curve (AUC) was calculated correspondingly. The indicator with the highest AUC was considered the most valuable indicator. The maximum Youden index (using the formula sensitivity + specificity - 1) was used to define the optimal cutoff value. Potential confounding variables were added into the logistic regression equation step by step, including age; blood pressure; and FPG, TC, TG, HDL-c, and LDL-c levels. Calibration Model I (age, blood pressure, and FPG level were added to the logistic regression equation) and Model II (age; blood pressure; and FPG, TC, TG, HDL-c, and LDL-c levels were added to the logistic regression equation) were established, and the predictive ability was evaluated before and after calibration. All significant indicators were included for the combination of diagnostic tests, and ROC curves were drawn. A new prediction model, the MAFLD Screening Index (MFSI), was constructed using logistic regression analysis, and the model was validated with the testing set. All tests were 2-tailed, and P<.01 was considered statistically significant.

Ethical Considerations

Every participant signed a written informed consent form and participated in the study anonymously. We ensured it was not possible to identify individual participants in any images used in manuscripts or other materials.

Every participant was given an allowance of ¥300 (US $0.14) upon completion of the research project. The study protocol was approved by the Ethics Committee of Shulan Hangzhou Hospital (approval number KY2021001).

The study did not involve additional invasive procedures, and there were no associated adverse reactions.

Results

Comparing Numerical Data Among the 3 Groups (MAFLD, NAFLD, and Non-FLD)

Using ANOVA to compare the basic information and anthropometric indicators among the 3 groups, the results showed that all parameters were significantly different among the MAFLD, NAFLD, and non-FLD groups. After the post hoc analysis, PLT count (P=.10) was not significantly different between the MAFLD and non-FLD groups, and white blood cell count (P=.26), TC (P=.35), LDL-c (P=.11), VFA (P=.07), and Fat% (P=.38) were not significantly different between the MAFLD and NAFLD groups (Table 1).

Predictive Performance of Different Anthropometric Indicators

The variables in the previous section with a P<.01 were further included in the logistic regression analysis. The ROC curves and optimal cutoff points for the selected indicators are shown in Table 2 and Figure 2. The AUCs of the WHtR, the Forns index, the HSI, the TyG index, TBW, BFM, and BMR were 0.866, 0.684, 0.873, 0.835, 0.760, 0.842, and 0.778, respectively (P<.001; Table 2).

‎

According to the ROC curve and AUC (Figure 2), HSI had the strongest predictive performance for MAFLD in the training set, and the performance ranking was as follows: HSI, WHtR, BFM, TyG index, TBW, and Forns index. The confounding factors were further corrected for in the logistic regression analysis (Model I: age, blood pressure, and FPG level were added to the logistic regression equation; Model II: age; blood pressure; and FPG, TC, TG, HDL-c, and LDL-c levels were added to the logistic regression equation). After correction for confounding factors, the odds ratio (OR) of the Forns index in Model I was 1.043 (95% CI 0.851-1.277), and that in Model II was 1.050 (95% CI 0.854-1.293; Table 3). The results showed that the performance of the Forns index for MAFLD screening was unstable and the performance of other anthropometric indicators was not easily influenced by confounders.

The HSI and WHtR showed better predictive performance than the other indicators. The sensitivity of the HSI was higher than that of the other anthropometric indicators (sensitivity=81.4%), and the specificity of the WHtR was higher than that of the other anthropometric indicators (specificity=79.8%). The combination of WHtR, HSI, and BFM increased the predictive ability for MAFLD, and the AUC was 0.900 (specificity=81.8%, sensitivity=85.6%; P<.001; Table 2).

Development of a New MAFLD Screening Model

The HSI, WHtR, and BFM displayed strong power in screening for MAFLD. The HSI was calculated based on BMI, ALT levels, and AST levels and was not suitable for early screening for MAFLD. The purpose of establishing a new model was to reduce the need for invasive procedures and reduce the frequency of medical visits, as well as to screen for MAFLD in high-risk populations. The predictive ability of TBW was stable after correcting for confounders (Model I: 95% CI 1.055-1.118; Model II: 95% CI 1.042-1.108; Table 3). Therefore, TBW was included in the new model. Logistic regression analysis was used to establish the MAFLD early screening model, which was named the MFSI. The formula was as follows: MFSI=–13.968+0.120×TBW+0.254×BFM+10.793×WHtR (Figure 3). The AUC of the MFSI was 0.896 (specificity: 83.8%, sensitivity: 82.1%; P<.001; Table 4). Collectively, the performance of the MFSI and the WHtR/HSI/BFM combination models was similar.

‎

Performance of the MFSI in the Testing Set

There were a further 200 participants enrolled in the testing set, including 51 non-FLD patients and 149 MAFLD patients. To evaluate the predictive ability of the MFSI for screening for MAFLD in a high-risk population, the MFSI was used with the testing set, and ROC curves were drawn based on the MFSI; BFM; WHtR; HSI; TBW; and the combined model with WHtR, HSI, and BFM (Figure 4). The AUC (testing set) of the MFSI was 0.917, the specificity was 89.8%, and the sensitivity was 84.4%. The AUC (testing set) of the combined model with WHtR, HSI, and BFM was 0.920, the specificity was 89.8%, and the sensitivity was 81.6% (Table 5). The performance of the MFSI was similar to that of the combined model with WHtR, HSI, and BFM in the testing set.

‎

MAFLD Rating Table for Prediction of MAFLD

The scoring system based on the MFSI and the application program was more practical for patient self-assessment. The MAFLD Rating Table (MRT) also included TBW, BFM, and WHtR. An MRT score ranging from 0 to 2 indicated a healthy individual, and a score ≥3 indicated MAFLD (Table 6). The AUC of the MRT for MAFLD prediction was 0.876 (P<.001; Figure 5).

‎

Discussion

WHtR, BFM, and TBW were predictors of MAFLD. The AUC of the WHtR was 0.866 (specificity=79.8%, sensitivity=80.8%), the AUC of BFM was 0.842 (specificity=76.9%, sensitivity=76.2%), and the AUC of TBW was 0.760 (specificity=75.3%, sensitivity=65.3%). The novel MFSI model, derived through logistic regression analysis, included the WHtR, BFM, and TBW. Notably, the MFSI demonstrated independence from laboratory findings. Upon validation, the MFSI exhibited stability while offering advantages in terms of sensitivity and specificity for MAFLD screening (training set: AUC=0.896, specificity=83.8%, sensitivity=82.1%; testing set: AUC=0.917, specificity=89.8%, sensitivity=84.4%).

Researchers found that the measurement of visceral fat can predict the occurrence of chronic diseases, such as diabetes, hyperuricemia, and metabolic syndrome [37,38]. NAFLD and MAFLD affect more than 25% of the global population and are considered different stages of the disease course. Because of long-term subtle inflammation and unobvious clinical manifestations, some patients gradually develop liver fibrosis and cirrhosis [1]. It is important to raise awareness within the population and optimize the management of this disease.

In recent decades, researchers have considered that the APRI, FIB-4 index, BMI, HSI, and TyG index have high accuracy for the diagnosis of liver fibrosis. Nonfibrosis scores were higher in patients with MAFLD than in those with NAFLD [11,39-42]. Similar conclusions were drawn in this work. To distinguish patients with MAFLD in our study population, we compared traditional indicators and body composition between the MAFLD and non-FLD groups and found there was a significant difference between the 2 groups. Although traditional indicators had efficient performance for the prediction of liver fibrosis, it was doubtful that these indicators were robust for the screening of MAFLD before being confirmed by histological liver examination. Previously published work mainly focused on the predictive ability of indicators for the detection of liver fibrosis, while much work omitted the performance of these indicators for the early screening of MAFLD.

Lee et al [27] suggested a new indicator named the HSI, and they found that NAFLD cannot be diagnosed when the HSI was <30.0, with a sensitivity of 92.5% (95% CI 91.4-93.5) The HSI showed similar performance for MAFLD diagnosis in this study. Italian researchers proposed another new indicator, the fatty liver index (FLI), which is calculated based on waist circumference, BMI, TG levels, and GGT levels. When the FLI is <30, a diagnosis of FLD can be ruled out, and when the FLI is ≥60, patients can be diagnosed with FLD. Waist circumference and BMI are the most robust predictors for the screening of FLD [43]. In contrast to the HSI, the FLI was established by incorporating waist circumference. However, BMI and waist circumference are totally different for people with varied dietary habits, and the study did not take this into account. Zheng et al [44] found that the WHtR had great performance for MAFLD screening, with a sensitivity of 96% and specificity of 64%. In our study, the WHtR showed a sensitivity of 80.8% and specificity of 79.8% for MAFLD screening.

TG and FPG levels are considered 2 pivotal inducers of metabolic syndrome. TGs are produced excessively in the process of fat accumulation, and insulin resistance accelerates hepatic steatosis. The TyG index can be used as a simple alternative marker for the detection of insulin resistance in the diagnostic test combining TG and FPG levels. The prevalence and severity of MAFLD are positively correlated with the TyG index [25,45-47]. The AUC of the TyG index for predicting MAFLD was 0.835 (95% CI 4.560-9.427), which might be valuable for clinical practice.

A meta-analysis revealed that the visceral adiposity index was an independent predictor of MAFLD, which could be used to predict potential morbidity [48]. However, the predictive ability of the visceral adiposity index has not been verified. Wang et al [49] found that nonobese MAFLD patients had higher BFM and VFA values than the healthy population, and most of them had abnormal lipid metabolism. In addition, BFM and VFA were valuable for distinguishing MAFLD patients from nonobese people [49,50]. This conclusion was also confirmed in this study (BFM for the prediction of MAFLD: AUC=0.842, sensitivity=76.2%, specificity=76.9%).

This study aimed to establish a home-based model for early screening of MAFLD to promote disease self-assessment and management. Compared with previously published models that rely heavily on laboratory indicators, our model combined body composition and the WHtR to screen for MAFLD, and the body parameters that were used to build the screening model can be easily obtained using a body fat scale at home. The mobile device software can record specific values and perform calculations.

There were 2 significant advantages of our model: (1) The need for an invasive examination and medical expenditures were reduced; (2) early screening models can provide early warning signs of disease, prompting people to modify diet and exercise or seek medical treatment if necessary; (3) patient-physician interactions were enhanced.

There were also some limitations of our work. First, this study was limited by geographical factors, and regional bias existed. Second, due to ethical considerations, the results in this study cannot be confirmed by histological liver examination. Third, in some villages we went to for recruitment, we were unable to obtain a radiological diagnosis due to manpower, transportation, and other constraints. In addition, it was difficult to follow participants who underwent physical examination in different areas, and reexamination data could not be compared with previous data.

Although our study found that the new MFSI model and MRT were valuable for MAFLD prediction, disease diagnosis still requires experienced clinicians, and those with the disease or at high risk should seek timely medical attention.