A nomogram for assessing the risk of MAFLD based on the albumin-to-gamma-glutamyl transferase ratio Authors: Shengnan Meng, Luonan Li, Yaning Liu, Shijie Yin, Yuxuan Zhao, Yuxi Zhang, Xiaolin Zhang, Fengxue Yu Published: 2025 ABSTRACT AimThe prevalence of metabolic-associated fatty liver disease (MAFLD) is rapidly increasing, posing a pressing challenge to global health. The purpose of this study is to establish a predictive model based on the immunoinflammatory marker, the albumin–gamma-glutamyl transferase ratio (AGTR), to facilitate the early recognition and risk assessment of MAFLD.MethodsThis study used data from 2331 participants in the NHANES database from 2017 to 2018. LASSO and logistic regression analyses were employed to identify risk factors and to establish a nomogram prediction model for MAFLD. External validation was conducted using data from the NHANES 2021–2023. Sensitivity analysis was employed to investigate the independent predictive value of AGTR for MAFLD.ResultsThe results indicated that among 67 variables, BMI, waist-to-hip ratio, waist circumference, AGTR and the triglyceride–glucose index (TyG) were all independent influencing factors for MAFLD. These risk factors were used to create a nomogram prediction risk model, which had AUCs for the training set, internal validation and external validation sets of 0.847 (95% CI 0.830–0.866), 0.834 (95% CI 0.806–0.863) and 0.851 (95% CI 0.834–0.868), respectively. The Hosmer–Lemeshow test p values were all greater than 0.05. Calibration and DCA indicate that the predictive model possesses good consistency and clinical validity. The sensitivity analysis revealed that AGTR remained an independent predictor following adjustment for demographic and lifestyle confounders.ConclusionsAs an independent immune-inflammatory predictor of MAFLD, early monitoring of AGTR may be crucial for predicting the emergence of MAFLD. Meanwhile, the nomogram model established in this study can identify high-risk patients for MAFLD. Introduction Metabolic-associated fatty liver disease (MAFLD), formerly known as non-alcoholic fatty liver disease (NAFLD), is a chronic metabolic disorder characterized by multisystem involvement, triggered by a confluence of genetic, environmental, metabolic, and stress factors [1]. In recent years, MAFLD has become more commonplace worldwide. In addition to being directly linked to a higher incidence of liver problems, including fibrosis and hepatocellular carcinoma [2], it also causes various extrahepatic diseases, including insulin resistance-related malignancies [3], which imposes a significant socioeconomic burden. Numerous studies have established inflammation as a significant trigger for hepatic steatosis [4]. This complex pathophysiological process is co-regulated by multiple intrahepatic and extrahepatic factors [5]. Albumin, synthesized by the liver, possesses antioxidant and anti-inflammatory properties. Hypoalbuminemia is associated with conditions, such as malnutrition, inflammatory responses, and worsening liver function [6]. Serum albumin has been identified as a novel indicator of early hepatic dysfunction, with hypoalbuminemia strongly linked to adverse outcomes in patients with cirrhosis and MAFLD [7]. Evidence indicates that γ-glutamyl transferase (GGT) is modulated by inflammatory cytokines. Elevated GGT levels not only signify impaired liver function but also serve as an independent predictor of cirrhosis in MAFLD, potentially involved in the immunoinflammatory regulation of liver tumors [8, 9]. Consequently, integrating both markers comprehensively reflects the liver’s synthetic capacity and anti-inflammatory function represented by albumin, as well as the inflammatory response and oxidative stress injury indicated by GGT, thereby providing a holistic assessment of hepatic functional status and inflammatory activity. Furthermore, compared to using albumin or GGT alone, the AGTR demonstrates enhanced sensitivity in detecting subtle hepatic injury and offers a more stable and reliable basis for evaluating the overall hepatic metabolic-inflammatory milieu. It is commonly established that a liver biopsy is the most reliable method of identifying MAFLD. However, liver biopsy is rarely recommended for routine screening because of its invasive nature [10]. It is noteworthy that although NAFLD and MAFLD have similar pathological features, there are significant differences in their diagnostic criteria, which requires us to re-evaluate and validate the predictors of NAFLD in the context of MAFLD research. This study endeavors to construct a risk nomogram for MAFLD using the novel immunoinflammatory marker AGTR. The aim is to provide clinical practitioners with a non-invasive, efficient, and accurate method to identify those who are at high risk for MAFLD, thereby offering a solid scientific basis for disease intervention and treatment. Methods Database The National Health and Nutrition Examination Survey (NHANES) is a comprehensive cross-sectional research program designed to assess the health and nutritional status of the United States population. All participants in this study signed informed consent forms, ensuring the ethical compliance and transparency of the study [11]. Study design and participants This study utilized data from the 2017–2018 NHANES as the training and internal validation set, comprising 9254 participants surveyed during this period. Simultaneously, data from 2021 to 2023 were employed as the external validation set, which included 11,933 participants. We excluded participants who were younger than 18 years, those without essential biochemical data and critical biochemical values, those with incomplete controlled attenuation parameter (CAP) data from transient elastography, and those missing data necessary for the diagnosis of MAFLD. After screening, 2331 and 1845 eligible participants from the 2017–2018 and 2021–2023 cycles, respectively, were ultimately included in the study (Fig. 1).Fig. 1Screening process diagram in this study Screening process diagram in this study Data collection and definitions Based on clinical experience and previous research, this study selected factors that may influence the occurrence of MAFLD. These factors included demographic characteristics (age, sex, race, education level, income level), anthropometric measurements [body mass index (BMI), waist circumference (WC), arm circumference (AC), waist-to-hip ratio (WHR)], lifestyle habits (smoking, alcohol consumption), comorbidities (hypertension, diabetes), and laboratory tests [neutrophils (NEU), Neutrophil Percentage (NEU%), lymphocytes (LYM), Lymphocyte Percentage (LYM%), white blood cells (WBC), red blood cells (RBC), platelets (PLT), monocytes (MONO), Monocyte Percentage (MONO%), eosinophils (Eos), Eosinophil Percentage (Eos%), basophils (BASO), Basophil Percentage (Baso%), hematocrit (HCT), mean platelet volume (MPV), mean corpuscular volume (MCV), hemoglobin (HB), mean corpuscular hemoglobin (MCH), mean corpuscular hemoglobin concentration (MCHC), red cell distribution width (RDW), alanine aminotransferase (ALT), aspartate aminotransferase (AST), alkaline phosphatase (ALP), gamma-glutamyl transferase (GGT), plasma glucose (PG), low-density lipoprotein cholesterol (LDL-C), high-density lipoprotein cholesterol (HDL-C), uric acid (UA), total cholesterol (TC), triglycerides (TG), total protein (TP), globulin (GLO), albumin (ALB), total bilirubin (TBIL), blood urea nitrogen (BUN), creatinine (Crea), high-sensitivity C-reactive protein (hsCRP), fasting plasma glucose (FPG), glycated hemoglobin (HbA1c)]. In this study, race was categorized into four groups: non-Hispanic whites, non-Hispanic blacks, Mexican Americans, and other races. There were two categories for education levels: below and above high school, or high school. The categories of smoking status were non-smokers, former smokers, and current smokers. Drinking status is categorized as: never drinking (having never consumed alcohol or fewer than 12 drinks in a lifetime), moderate drinking (≤2 drinks/day for men and ≤1 drink/day for women), and heavy drinking (>2 drinks/day for men and >1 drink/day for women). Three categories were used to categorize the poverty income ratio: low (PIR < 1.30), medium (1.30 ≤ PIR < 3.50), and high (PIR ≥ 3.50). Three categories for BMI were established: normal (BMI < 25), overweight (25 ≤ BMI < 30), and obese (BMI ≥ 30). Diabetes can be detected through the serum biomarkers HbA1c ≥ 6.5%, FPG ≥ 126 mg/dL, or by inquiring whether individuals have been diagnosed with diabetes by a healthcare professional [12]. A diagnosis of hypertension can be made if the systolic blood pressure is ≥140 mmHg or the diastolic blood pressure is ≥90 mmHg, or by questionnaires that inquire whether participants have been advised by a physician of a hypertension diagnosis [13]. Composite indices included in this study are as follows, WHR = WC/hip * 10, NLR = NEU/LYM, dNLR = NEU/(WBC–NEU), PLR = PLT/LYM, MLR = Mono/LYM, SII = PLT × NEU/LYM, NPAR = NEU/ALB (g/L), CAR = hsCRP (mg/L)/ALB (g/L), SIRI = NEU × Mono/LYM, UHR = UA (umol/L)/HDL-C (mmol/L), AGTR = ALB (g/L)/GGT (IU/L), TyG = ln [(TG (mg/dL) * FPG (mg/dL))/2]. Definition of MAFLD The diagnostic criteria for MAFLD are hepatic steatosis along with one of the three disorders listed below: overweight/obesity, type 2 diabetes, or metabolic dysregulation. The CAP threshold of 274 dB/m was chosen for this study’s assessment of hepatic steatosis [14]. Metabolic dysregulation is defined as satisfying a minimum of two of the subsequent criteria: (1) WC ≥ 102 cm for men and ≥88 cm for women of Caucasian ethnicity; (2) blood pressure ≥ 130/85 mmHg; (3) YG ≥ 1.70 mmol/L; (4) HDL-C < 1.0 mmol/L in men and <1.3 mmol/L in women; (5) prediabetes, such as FPG levels ranging from 5.6 to 6.9 mmol/L, or HbA1c ranging from 5.7 to 6.4%; (6) HOMA–IR score ≥ 2.5; (7) hsCRP levels > 2 mg/L [15]. Statistical analysis The R software (version 4.3.1) was used for all statistical analyses. This study employed a phased strategy to handle missing data. For variables with a missing rate of ≤20%, we applied multiple imputation, generating five imputed data sets, and pooled the results according to Rubin’s rules [16]. Furthermore, given the critical importance of key variables, such as fasting blood glucose for the diagnosis of MAFLD, observation samples with a missing rate exceeding 20% for these variables were excluded to ensure the reliability of the core data. Following this process, the final missing rate for all variables was below 20%. In addition, the proportions of missing data for each variable were visualized using a bar chart (Fig. 2). If the continuous data had a regular distribution, they were presented as mean ± standard deviation (SD) or median (P25,P75), and the data between groups were compared using the t test or Mann–Whitney U test. For comparing rates or proportions, the χ2 test was employed. To mitigate the risk of model overfitting, this study initially employed Least Absolute Shrinkage and Selection Operator (LASSO) regression for preliminary variable screening. The optimal regularization parameter λ (λ−1se) was determined via tenfold cross-validation to identify significant predictor variables. Subsequently, the selected variables were incorporated into a multivariate logistic regression analysis to further identify independent risk factors for MAFLD, based on which the final nomogram prediction model was constructed. The performance of the nomogram is evaluated by constructing the receiver operating characteristic (ROC) curve, calibration curves, and decision curve analysis (DCA). After categorizing AGTR into quartiles, a linear trend test was employed to assess the presence of linear trends. Subgroup analyses were conducted to determine whether the association between AGTR and MAFLD varied across different subgroups and to evaluate potential interaction effects within these subgroups.Fig. 2Bar plot of missing values Bar plot of missing values Results Baseline characteristics of participants Participants were assigned at random to a training set and a validation set in a 7:3 ratio. The training set comprised 701 MAFLD patients and 930 non-MAFLD controls; the internal validation set contained 288 patients and 412 controls; and the external validation set included 710 patients and 1135 controls. Table 1 displays the clinical features and demographic data of the participants. Table 1Baseline characteristics of participantsVariablesTraining set(N = 1631)Internal validation(N = 700)PAge51.00 (34.00, 64.00)52.00 (33.00, 64.00)0.988aSex0.789b Male797 (48.9)347 (49.6) Female834 (51.1)353 (50.4)Race, n (%)0.622b Mexican America247 (15.1)95 (13.6) Non-Hispanic white534 (32.7)246 (35.1) Non-Hispanic black374 (22.9)160 (22.9) Other476 (29.2)199 (28.4)Education, n (%)0.736b High school and below727 (44.6)306 (43.7) Above high school904 (55.4)394 (56.3)PIR, n (%)0.840b <1.30458 (28.1)189 (27.0) 1.30 ≤ PIR < 3.50681 (41.8)300 (42.9) PIR ≤ 3.50492 (30.2)211 (30.1)BMI, n (%)0.355b Normal weight (<25)444 (27.2)204 (29.1) Over weight (25–30)509 (31.2)227 (32.4) Obesity (≥30)678 (41.6)269 (38.4)Drinking, n (%)0.940b No388 (23.8)169 (24.1) Moderate623 (38.2)262 (37.4) Heave620 (38.0)269 (38.4)Smoking, n (%)0.343b Never943 (57.8)419 (59.9) Previous386 (23.7)169 (24.1) Current302 (18.5)112 (16.0)WC (cm)99.10 (87.70, 111.30)97.50 (86.88, 109.23)0.066aAC (cm)33.00 (29.50, 36.80)32.70 (29.67, 36.50)0.397aWHR0.94 (0.88, 0.99)9.39 (8.79, 9.90)<0.001aHbA1c (%)5.60 (5.30, 6.00)5.60 (5.30, 5.90)0.323aFPG (mg/dL)104.00 (96.00, 116.00)103.00 (96.00, 115.00)0.522aDiabetes, n (%)0.439b No1256 (77.0)550 (78.6) Yes375 (23.0)150 (21.4)Hypertension, n (%)0.118b No673 (41.3)314 (44.9) Yes958 (58.7)386 (55.1)ALT (U/L)18.00 (13.00, 26.00)17.00 (13.00, 25.00)0.456aAST (U/L)19.00 (16.00, 24.00)19.00 (16.00, 23.00)0.243aALB (g/L)40.00 (38.00, 42.00)40.00 (38.00, 42.00)0.727aALP (IU/L)75.00 (62.00, 92.00)75.50 (63.00, 90.00)0.974aGLO (g/dL)31.00 (28.00, 34.00)31.00 (28.00, 34.00)0.654aBUN (mmol/L)5.00 (3.93, 6.07)5.00 (3.93, 6.07)0.640aCrea (umol/L)74.26 (61.88, 87.52)75.14 (61.88, 89.28)0.406aGGT (IU/L)21.00 (15.00, 33.00)20.00 (14.00, 31.00)0.074aTBIL (umol/L)6.84 (5.13, 10.26)6.84 (5.13, 10.26)0.584aTC (mmol/L)4.71 (4.06, 5.43)4.68 (4.09, 5.51)0.649aTP (g/L)71.00 (68.00, 74.00)71.00 (69.00, 74.00)0.212aTG (mmol/L)1.23 (0.86, 1.73)1.15 (0.82, 1.67)0.090aUA (umol/L)321.20 (264.70, 386.60)321.20 (272.12, 380.70)0.876aPG (mmol/L)5.38 (5.00, 6.00)5.33 (4.94, 5.94)0.358aWBC(×109/L)6.50 (5.40, 7.90)6.40 (5.30, 7.90)0.322aLYM% (%)31.20 (24.80, 37.00)30.65 (25.65, 36.90)0.932aMomo% (%)8.00 (6.90, 9.40)8.10 (6.80, 9.50)0.634aNEU% (%)56.90 (50.55, 63.60)57.05 (50.10, 62.90)0.783aEos% (%)2.40 (1.60, 3.70)2.60 (1.60, 3.90)0.310aBaso% (%)0.70 (0.60, 1.00)0.80 (0.60, 1.00)0.405aLYM (×109/L)2.00 (1.60, 2.40)2.00 (1.60, 2.40)0.260aMomo (×109/L)0.50 (0.40, 0.60)0.50 (0.40, 0.60)0.757aNEU (×109/L)3.60 (2.80, 4.80)3.60 (2.80, 4.70)0.391aEos (×109/L)0.20 (0.10, 0.20)0.20 (0.10, 0.30)0.155aBaso (×109/L)0.00 (0.00, 0.10)0.00 (0.00, 0.10)0.886aRBC (× 1012/L)4.78 (4.44, 5.14)4.77 (4.43, 5.10)0.263aHB (g/dL)14.20 (13.20, 15.20)14.05 (13.20, 15.10)0.131aHCT (%)42.30 (39.60, 45.20)42.10 (39.48, 44.90)0.237aMCV (fl)88.80 (85.40, 92.20)88.80 (85.40, 92.20)0.909aMCHC (g/dL)33.50 (32.90, 34.00)33.40 (32.90, 34.00)0.047aMCH (pg)29.90 (28.50, 31.10)29.80 (28.50, 31.10)0.506aRDW (%)13.60 (13.10, 14.20)13.60 (13.10, 14.30)0.713aPLT (×109/L)231.00 (195.00, 273.00)230.50 (195.00, 274.25)0.979aPLT (×109/L)8.20 (7.70, 8.80)8.20 (7.60, 8.80)0.255aHDL-C (mmol/L)1.29 (1.09, 1.58)1.32 (1.09, 1.58)0.701aLDL-C (mmol/L)2.79 (2.22, 3.44)2.82 (2.28, 3.49)0.550ahsCRP (mg/L)1.92 (0.89, 4.26)1.69 (0.76, 4.22)0.114aSII431.05 (302.33, 610.69)424.05 (305.67, 615.15)0.830aNLR1.82 (1.38, 2.52)1.84 (1.38, 2.44)0.914adNLR1.32 (1.03, 1.75)1.32 (1.00, 1.71)0.795aPLR117.14 (92.70, 147.63)121.28 (94.69, 146.72)0.383aCAR0.05 (0.02, 0.11)0.04 (0.02, 0.11)0.105aLMR3.80 (3.00, 5.00)3.80 (2.88, 4.76)0.333aSIRI0.94 (0.63, 1.49)0.97 (0.64, 1.40)0.943aUHR241.17 (176.41, 329.52)239.77 (181.68, 322.16)0.751aAGTR1.86 (1.23, 2.71)2.00 (1.29, 2.84)0.064aTYG8.51 (8.05, 8.99)8.45 (8.01, 8.92)0.056aNPAR0.09 (0.07, 0.12)0.09 (0.07, 0.12)0.261aMAFLD0.437b No930 (57.0)412 (58.9) Yes701 (43.0)288 (41.1)aMann–Whitney U testbChi-square test Table 1Baseline characteristics of participantsVariablesTraining set(N = 1631)Internal validation(N = 700)PAge51.00 (34.00, 64.00)52.00 (33.00, 64.00)0.988aSex0.789b Male797 (48.9)347 (49.6) Female834 (51.1)353 (50.4)Race, n (%)0.622b Mexican America247 (15.1)95 (13.6) Non-Hispanic white534 (32.7)246 (35.1) Non-Hispanic black374 (22.9)160 (22.9) Other476 (29.2)199 (28.4)Education, n (%)0.736b High school and below727 (44.6)306 (43.7) Above high school904 (55.4)394 (56.3)PIR, n (%)0.840b <1.30458 (28.1)189 (27.0) 1.30 ≤ PIR < 3.50681 (41.8)300 (42.9) PIR ≤ 3.50492 (30.2)211 (30.1)BMI, n (%)0.355b Normal weight (<25)444 (27.2)204 (29.1) Over weight (25–30)509 (31.2)227 (32.4) Obesity (≥30)678 (41.6)269 (38.4)Drinking, n (%)0.940b No388 (23.8)169 (24.1) Moderate623 (38.2)262 (37.4) Heave620 (38.0)269 (38.4)Smoking, n (%)0.343b Never943 (57.8)419 (59.9) Previous386 (23.7)169 (24.1) Current302 (18.5)112 (16.0)WC (cm)99.10 (87.70, 111.30)97.50 (86.88, 109.23)0.066aAC (cm)33.00 (29.50, 36.80)32.70 (29.67, 36.50)0.397aWHR0.94 (0.88, 0.99)9.39 (8.79, 9.90)<0.001aHbA1c (%)5.60 (5.30, 6.00)5.60 (5.30, 5.90)0.323aFPG (mg/dL)104.00 (96.00, 116.00)103.00 (96.00, 115.00)0.522aDiabetes, n (%)0.439b No1256 (77.0)550 (78.6) Yes375 (23.0)150 (21.4)Hypertension, n (%)0.118b No673 (41.3)314 (44.9) Yes958 (58.7)386 (55.1)ALT (U/L)18.00 (13.00, 26.00)17.00 (13.00, 25.00)0.456aAST (U/L)19.00 (16.00, 24.00)19.00 (16.00, 23.00)0.243aALB (g/L)40.00 (38.00, 42.00)40.00 (38.00, 42.00)0.727aALP (IU/L)75.00 (62.00, 92.00)75.50 (63.00, 90.00)0.974aGLO (g/dL)31.00 (28.00, 34.00)31.00 (28.00, 34.00)0.654aBUN (mmol/L)5.00 (3.93, 6.07)5.00 (3.93, 6.07)0.640aCrea (umol/L)74.26 (61.88, 87.52)75.14 (61.88, 89.28)0.406aGGT (IU/L)21.00 (15.00, 33.00)20.00 (14.00, 31.00)0.074aTBIL (umol/L)6.84 (5.13, 10.26)6.84 (5.13, 10.26)0.584aTC (mmol/L)4.71 (4.06, 5.43)4.68 (4.09, 5.51)0.649aTP (g/L)71.00 (68.00, 74.00)71.00 (69.00, 74.00)0.212aTG (mmol/L)1.23 (0.86, 1.73)1.15 (0.82, 1.67)0.090aUA (umol/L)321.20 (264.70, 386.60)321.20 (272.12, 380.70)0.876aPG (mmol/L)5.38 (5.00, 6.00)5.33 (4.94, 5.94)0.358aWBC(×109/L)6.50 (5.40, 7.90)6.40 (5.30, 7.90)0.322aLYM% (%)31.20 (24.80, 37.00)30.65 (25.65, 36.90)0.932aMomo% (%)8.00 (6.90, 9.40)8.10 (6.80, 9.50)0.634aNEU% (%)56.90 (50.55, 63.60)57.05 (50.10, 62.90)0.783aEos% (%)2.40 (1.60, 3.70)2.60 (1.60, 3.90)0.310aBaso% (%)0.70 (0.60, 1.00)0.80 (0.60, 1.00)0.405aLYM (×109/L)2.00 (1.60, 2.40)2.00 (1.60, 2.40)0.260aMomo (×109/L)0.50 (0.40, 0.60)0.50 (0.40, 0.60)0.757aNEU (×109/L)3.60 (2.80, 4.80)3.60 (2.80, 4.70)0.391aEos (×109/L)0.20 (0.10, 0.20)0.20 (0.10, 0.30)0.155aBaso (×109/L)0.00 (0.00, 0.10)0.00 (0.00, 0.10)0.886aRBC (× 1012/L)4.78 (4.44, 5.14)4.77 (4.43, 5.10)0.263aHB (g/dL)14.20 (13.20, 15.20)14.05 (13.20, 15.10)0.131aHCT (%)42.30 (39.60, 45.20)42.10 (39.48, 44.90)0.237aMCV (fl)88.80 (85.40, 92.20)88.80 (85.40, 92.20)0.909aMCHC (g/dL)33.50 (32.90, 34.00)33.40 (32.90, 34.00)0.047aMCH (pg)29.90 (28.50, 31.10)29.80 (28.50, 31.10)0.506aRDW (%)13.60 (13.10, 14.20)13.60 (13.10, 14.30)0.713aPLT (×109/L)231.00 (195.00, 273.00)230.50 (195.00, 274.25)0.979aPLT (×109/L)8.20 (7.70, 8.80)8.20 (7.60, 8.80)0.255aHDL-C (mmol/L)1.29 (1.09, 1.58)1.32 (1.09, 1.58)0.701aLDL-C (mmol/L)2.79 (2.22, 3.44)2.82 (2.28, 3.49)0.550ahsCRP (mg/L)1.92 (0.89, 4.26)1.69 (0.76, 4.22)0.114aSII431.05 (302.33, 610.69)424.05 (305.67, 615.15)0.830aNLR1.82 (1.38, 2.52)1.84 (1.38, 2.44)0.914adNLR1.32 (1.03, 1.75)1.32 (1.00, 1.71)0.795aPLR117.14 (92.70, 147.63)121.28 (94.69, 146.72)0.383aCAR0.05 (0.02, 0.11)0.04 (0.02, 0.11)0.105aLMR3.80 (3.00, 5.00)3.80 (2.88, 4.76)0.333aSIRI0.94 (0.63, 1.49)0.97 (0.64, 1.40)0.943aUHR241.17 (176.41, 329.52)239.77 (181.68, 322.16)0.751aAGTR1.86 (1.23, 2.71)2.00 (1.29, 2.84)0.064aTYG8.51 (8.05, 8.99)8.45 (8.01, 8.92)0.056aNPAR0.09 (0.07, 0.12)0.09 (0.07, 0.12)0.261aMAFLD0.437b No930 (57.0)412 (58.9) Yes701 (43.0)288 (41.1)aMann–Whitney U testbChi-square test Baseline characteristics of participants aMann–Whitney U test bChi-square test Selection of MAFLD prediction factors With the occurrence of MAFLD as the dependent variable, a LASSO regression model was established by incorporating all 63 variables from the data set. The LASSO solution path and cross-validation plot were constructed (Fig. 3). The optimal regularization parameter λ (λ−1se) was determined via tenfold cross-validation to prevent overfitting, identifying seven variables: BMI, WC, WHR, hypertension, diabetes, AGTR, and TYG. These seven variables were then included in a multivariate logistic regression model for further analysis, from which five independent risk factors were ultimately identified: BMI, WC, WHR, AGTR, and TYG (Table 2). The VIF values for these five predictors were 2.32, 2.82, 1.45, 1.09, and 1.10, respectively. As all VIF values were below 5, no significant multicollinearity was detected among the variables.Fig. 3Utilizing LASSO regression for the selection of demographic and clinical characteristics. a Coefficient trajectory plot of LASSO generated based on a sequence of log(λ), encompassing the coefficient trajectories for 63 features. b Tenfold cross-validation for the tuning of the hyperparameter selection within the LASSO modelTable 2Multivariate logistic regression analysisΒSEWald χ2OR95% CIP valueBMI Over weight0.9490.20820.8162.5821.728–3.909<0.001 Obesity1.2820.27122.3793.6052.124–6.154<0.001Hypertension0.2210.1372.6021.2480.954–1.6300.105Diabetes0.2370.1602.1941.2680.927–1.7350.138WC0.0310.00815.0161.0381.023–1.055<0.001WHR0.3190.1118.2591.3691.102–1.7020.005AGTR−0.2790.06617.8700.7570.663–0.860<0.001TYG0.8020.11548.6352.2291.785–2.798<0.001 Utilizing LASSO regression for the selection of demographic and clinical characteristics. a Coefficient trajectory plot of LASSO generated based on a sequence of log(λ), encompassing the coefficient trajectories for 63 features. b Tenfold cross-validation for the tuning of the hyperparameter selection within the LASSO model Table 2Multivariate logistic regression analysisΒSEWald χ2OR95% CIP valueBMI Over weight0.9490.20820.8162.5821.728–3.909<0.001 Obesity1.2820.27122.3793.6052.124–6.154<0.001Hypertension0.2210.1372.6021.2480.954–1.6300.105Diabetes0.2370.1602.1941.2680.927–1.7350.138WC0.0310.00815.0161.0381.023–1.055<0.001WHR0.3190.1118.2591.3691.102–1.7020.005AGTR−0.2790.06617.8700.7570.663–0.860<0.001TYG0.8020.11548.6352.2291.785–2.798<0.001 Multivariate logistic regression analysis Establishment and evaluation of a nomogram for MAFLD Based on the aforementioned five independent risk factors, this study developed a nomogram model to predict the risk of MAFLD occurrence (Fig. 4). The nomogram transforms complex regression equations into an intuitive graphical tool, enabling clinicians to quickly integrate these readily available indicators during routine clinical practice and calculate an individual’s probability of developing MAFLD. Furthermore, it facilitates the efficient and low-cost identification of undiagnosed high-risk asymptomatic individuals from the general patient population, allowing for recommendations for further liver imaging examinations to achieve early detection and intervention.Fig. 4A Nomogram for predicting the risk of MAFLD A Nomogram for predicting the risk of MAFLD ROC curves of the nomogram prediction model were plotted for the training set, internal validation set, and external validation set, respectively (Fig. 5). The model demonstrated AUC values of 0.847 (95% CI 0.829–0.865), 0.834 (95% CI 0.806–0.863), and 0.851 (95% CI 0.834–0.868) in the training, internal validation, and external validation sets, indicating good predictive accuracy. The Hosmer–Lemeshow test results indicated good calibration of the model across all data sets, with p values greater than 0.05 in the training set (χ2 = 8.708, P = 0.368), internal validation set (χ2 = 10.65, P = 0.222), and external validation set (χ2 = 9.218, P = 0.324). In addition, the calibration curves indicate that the estimated risks were close to the actual proportions, demonstrating good model calibration (Fig. 6).The DCA demonstrates that using the AGTR-based model yields a higher net benefit than treating all or none across clinically relevant thresholds (Fig. 7). In summary, the predictive model demonstrated good consistency between the training and validation data sets.Fig. 5a AUROC of the model in the training set. b AUROC of the model in the internal validation set. c AUROC of the model in the external validation set. The area under the curve (AUC) is a critical metric for assessing the performance of a model. A higher AUC value indicates superior model performance, signifying the model’s enhanced ability to accurately classify samplesFig. 6a Calibration plot of the training set. b Calibration plot for the internal validation set. c Calibration plot for the external validation set. The x-axis represents the model-predicted prevalence of MAFLD, while the y-axis denotes the actual prevalence of MAFLD. A 45-degree reference line signifies perfect calibrationFig. 7a DCA of the model in the training set. b DCA of the model in the internal validation set. c DCA of the model in the external validation set. The x-axis represents the threshold probability (Pt), and the y-axis denotes the net benefit (NB) a AUROC of the model in the training set. b AUROC of the model in the internal validation set. c AUROC of the model in the external validation set. The area under the curve (AUC) is a critical metric for assessing the performance of a model. A higher AUC value indicates superior model performance, signifying the model’s enhanced ability to accurately classify samples a Calibration plot of the training set. b Calibration plot for the internal validation set. c Calibration plot for the external validation set. The x-axis represents the model-predicted prevalence of MAFLD, while the y-axis denotes the actual prevalence of MAFLD. A 45-degree reference line signifies perfect calibration a DCA of the model in the training set. b DCA of the model in the internal validation set. c DCA of the model in the external validation set. The x-axis represents the threshold probability (Pt), and the y-axis denotes the net benefit (NB) Comprehensive evaluation of the predictive model A comprehensive evaluation of the predictive model demonstrated favorable performance across all metrics: sensitivity ranged from 82.0 to 86.1%, specificity from 66.5 to 72.8%, and accuracy from 74.6 to 76.3%; positive predictive value was between 64.3 and 68.1%, and negative predictive value between 85.0 and 87.3%. In addition, the model achieved Brier scores ranging from 0.154 to 0.164. These results collectively indicate that the predictive model possesses robust discriminative ability and good calibration (Table 3). Table 3Model efficacy for training and validation setsGroupsSENSPEACCPPVNPVBrierTraining set0.8350.7050.7610.6810.8500.158Internal validation set0.8610.6650.7460.6430.8730.164External validation set0.8200.7280.7630.6530.8650.154SEN sensitivity, SPE specificity, ACC accuracy, PPV positive predictive value, NPV negative predictive value Table 3Model efficacy for training and validation setsGroupsSENSPEACCPPVNPVBrierTraining set0.8350.7050.7610.6810.8500.158Internal validation set0.8610.6650.7460.6430.8730.164External validation set0.8200.7280.7630.6530.8650.154SEN sensitivity, SPE specificity, ACC accuracy, PPV positive predictive value, NPV negative predictive value Model efficacy for training and validation sets SEN sensitivity, SPE specificity, ACC accuracy, PPV positive predictive value, NPV negative predictive value Comparison of MAFLD prediction models We calculated three commonly used non-invasive MAFLD scoring indices: Fatty Liver Index (FLI), Hepatic Steatosis Index (HSI), and US Fatty Liver Index (USFLI). The diagnostic performance of these three classical models was compared with our newly developed nomogram. The results showed that the AUC values for FLI, HSI, and USFLI were 0.503 (95% CI 0.474–0.531), 0.507 (95% CI 0.479–0.535), and 0.503 (95% CI 0.475–0.532), respectively. While the traditional scoring models showed AUC values close to 0.50, the AUC of our model significantly increased to 0.847, demonstrating its superior discriminative ability for the disease. Sensitivity analysis Correlation between MAFLD and AGTR Although LASSO regression and multivariate logistic regression statistically adjusted for the inter-relationships among the selected predictors, traditional confounding factors were not included in the final model. Therefore, we performed a fully adjusted logistic regression by incorporating these variables to validate the independent predictive value of AGTR. Two models were constructed for analysis: Model I included only the five variables retained in the final model. Model II further adjusted for age, sex, ethnicity, hypertension, diabetes, smoking, and alcohol consumption based on Model I. In both multivariate logistic regression models, AGTR showed a highly significant association with the probability of MAFLD onset, with higher AGTR levels associated with a reduced risk of MAFLD. We further categorized AGTR into quartiles, and the results indicated that higher AGTR values corresponded to a lower probability of MAFLD. Moreover, a significant linear trend of AGTR was observed in both models (P for trend < 0.001) (Table 4). Table 4Association of AGTR with MAFLDModel 1Model 2OR (95%CI)POR (95%CI)PAGTR0.759 (0.658, 0.851)<0.0010.737 (0.642, 0.843)<0.001Quartiles Q11 (ref)1 (ref) Q20.790 (0.565, 1.103)0.1670.786 (0.559, 1.103)0.165 Q30.606 (0.428, 0.857)<0.0010.588 (0.410, 0.842)0.003 Q40.480 (0.328, 0.701)<0.0010.467 (0.313, 0.694)<0.001Pfor trend<0.001<0.001Model 1: Adjusted only for the variables selected by LASSO regressionModel 2: Further adjusted for age, sex, ethnicity, hypertension, diabetes, smoking, and alcohol consumption based on Model 1 Table 4Association of AGTR with MAFLDModel 1Model 2OR (95%CI)POR (95%CI)PAGTR0.759 (0.658, 0.851)<0.0010.737 (0.642, 0.843)<0.001Quartiles Q11 (ref)1 (ref) Q20.790 (0.565, 1.103)0.1670.786 (0.559, 1.103)0.165 Q30.606 (0.428, 0.857)<0.0010.588 (0.410, 0.842)0.003 Q40.480 (0.328, 0.701)<0.0010.467 (0.313, 0.694)<0.001Pfor trend<0.001<0.001Model 1: Adjusted only for the variables selected by LASSO regressionModel 2: Further adjusted for age, sex, ethnicity, hypertension, diabetes, smoking, and alcohol consumption based on Model 1 Association of AGTR with MAFLD Model 1: Adjusted only for the variables selected by LASSO regression Model 2: Further adjusted for age, sex, ethnicity, hypertension, diabetes, smoking, and alcohol consumption based on Model 1 Subgroup analysis Subgroup analyses revealed no significant interaction effects across subgroups defined by sex, race, smoking status, alcohol consumption, education level, poverty–income ratio, BMI category, hypertension, or diabetes (P for interaction > 0.05). This indicates that the inverse association between AGTR and MAFLD remained consistent across all subgroup populations (Table 5). Table 5Stratified analysis of the association between AGTR and MAFLDVariablesOR (95% CI)PP for interactionSex0.82 Male0.51 (0.43, 0.61)<0.001 Female0.50 (0.43, 0.58)<0.001Race<0.0010.929 Mexican American0.48 (0.37, 0.63)<0.001 Non-Hispanic White0.48 (0.40, 0.58)<0.001 Non-Hispanic Black0.51 (0.39, 0.66)<0.001 Other Race0.52 (0.43, 0.63)<0.001Smoking<0.0010.319 Non-smokers0.48 (0.42, 0.56)<0.001 Former smokers0.49 (0.39, 0.63)<0.001 Current smokers0.60 (0.47, 0.77)<0.001Drinking<0.0010.914 Non-drinkers0.52 (0.42, 0.65)<0.001 Moderate drinkers0.51 (0.42, 0.60)<0.001 Heavy drinkers0.49 (0.41, 0.58)<0.001Education<0.0010.152 High school and below0.55 (0.47, 0.64)<0.001 Above high school0.47 (0.41, 0.55)<0.001PIR<0.0010.289 <1.300.48 (0.39,0.59)<0.001 ≥1.30, <3.500.56 (0.47, 0.65)<0.001 ≥3.500.46 (0.38, 0.56)<0.001BMI<0.0010.27 <250.55 (0.41, 0.73)<0.001 25–300.64 (0.53, 0.77)<0.001 ≥300.52 (0.43, 0.62)<0.001Diabetes<0.0010.26 No0.56 (0.49, 0.63)<0.001 Yes0.47 (0.36, 0.61)<0.001Hypertension<0.0010.39 No0.52 (0.43, 0.62)<0.001 Yes0.57 (0.50, 0.66)<0.001In the subgroup analysis, the covariates of age, sex, race, BMI, poverty income, education level, hypertension, diabetes, smoke, drink, WC, WHR, and TYG were adjusted Table 5Stratified analysis of the association between AGTR and MAFLDVariablesOR (95% CI)PP for interactionSex0.82 Male0.51 (0.43, 0.61)<0.001 Female0.50 (0.43, 0.58)<0.001Race<0.0010.929 Mexican American0.48 (0.37, 0.63)<0.001 Non-Hispanic White0.48 (0.40, 0.58)<0.001 Non-Hispanic Black0.51 (0.39, 0.66)<0.001 Other Race0.52 (0.43, 0.63)<0.001Smoking<0.0010.319 Non-smokers0.48 (0.42, 0.56)<0.001 Former smokers0.49 (0.39, 0.63)<0.001 Current smokers0.60 (0.47, 0.77)<0.001Drinking<0.0010.914 Non-drinkers0.52 (0.42, 0.65)<0.001 Moderate drinkers0.51 (0.42, 0.60)<0.001 Heavy drinkers0.49 (0.41, 0.58)<0.001Education<0.0010.152 High school and below0.55 (0.47, 0.64)<0.001 Above high school0.47 (0.41, 0.55)<0.001PIR<0.0010.289 <1.300.48 (0.39,0.59)<0.001 ≥1.30, <3.500.56 (0.47, 0.65)<0.001 ≥3.500.46 (0.38, 0.56)<0.001BMI<0.0010.27 <250.55 (0.41, 0.73)<0.001 25–300.64 (0.53, 0.77)<0.001 ≥300.52 (0.43, 0.62)<0.001Diabetes<0.0010.26 No0.56 (0.49, 0.63)<0.001 Yes0.47 (0.36, 0.61)<0.001Hypertension<0.0010.39 No0.52 (0.43, 0.62)<0.001 Yes0.57 (0.50, 0.66)<0.001In the subgroup analysis, the covariates of age, sex, race, BMI, poverty income, education level, hypertension, diabetes, smoke, drink, WC, WHR, and TYG were adjusted Stratified analysis of the association between AGTR and MAFLD In the subgroup analysis, the covariates of age, sex, race, BMI, poverty income, education level, hypertension, diabetes, smoke, drink, WC, WHR, and TYG were adjusted Discussion Based on the NHANES database, this study developed and externally validated a clinical prediction model for estimating the risk of MAFLD in a physical examination population. The model incorporates traditional indicators, including waist circumference, TyG index, BMI, and waist-to-hip ratio, as well as the novel indicator AGTR. It demonstrated favorable predictive performance across the training, internal validation, and external validation data sets, achieving an AUC of 0.847. In sensitivity analyses, AGTR remained an independent predictor after adjusting for demographic and lifestyle confounders. Subgroup analysis further confirmed AGTR as a protective factor against MAFLD consistently across various subgroups, with no significant interaction effects observed between subgroups. These findings substantiate AGTR as an independent protective factor for MAFLD. Compared to establish non-invasive MAFLD scoring indices—namely, the FLI, HSI, and USFLI—the nomogram model developed in this study demonstrated superior predictive performance. In recent years, multiple prediction models based on either machine learning or traditional regression methods have been reported. For instance, Huang et al. developed and validated six machine learning models in a health screening population [17]. Their CatBoost model achieved the highest predictive performance in the training set, with an AUC of 0.810 (95% CI 0.768–0.852), while the logistic regression model performed best in the internal and external validation sets, with AUCs of 0.778 (95% CI 0.759–0.794) and 0.806 (95% CI 0.788–0.821), respectively. Another study focusing on non-obese type 2 diabetic patients with MAFLD reported a nomogram with AUCs of 0.815 (95% CI 0.772–0.858) in the training set and 0.787 (95% CI 0.754–0.820) in the internal validation set [18]. In contrast, the nomogram constructed in this study exhibited improved discriminative ability across the training, internal validation, and external validation sets. Anthropometric markers are essential for predicting MAFLD. Elevated BMI leads to an increase in free fatty acid (FFA) levels, which is the main factor for the buildup of fat in the liver [19]. In the context of insulin resistance, the metabolism of carbohydrates shifts from muscle glycogen storage to the liver, increasing lipid buildup and FFA production [20]. These inter-related mechanisms collectively promote the development of MAFLD. BMI is currently utilized widely in the evaluation of MAFLD and as a proxy indicator for measuring visceral fat [21]. The correlation between WC and the prevalence of MAFLD has been extensively reported, and a rise in waist size is thought to be a key factor contributing to the increased incidence of MAFLD [22]. The use of WC to measure central obesity has become standard procedure in epidemiological investigations. In addition, research suggests that the WHR is a more accurate indicator for assessing abdominal and visceral obesity compared to WC [23]. As a result, the inclusion of BMI, WC, and WHR in this research enables a thorough evaluation of how obesity affects the risk of MAFLD. It is now recognized that insulin resistance (IR) has a significant role in the development of MAFLD [24]. The gold standard for determining IR is the hyperinsulinemic–euglycemic clamp method [25]. However, its utility in clinical practice is restricted due to its complexity and high cost. Because insulin levels are not routinely measured in the clinical context, the Homeostasis Model Assessment of Insulin Resistance (HOMA–IR) is limited in its ability to assess IR [26]. In contrast to the HOMA–IR index, the TyG index provides a quick, dependable, and affordable way to evaluate IR [27]. As essential elements of the metabolic syndrome, TG and FPG play a critical role in the development of fatty liver. Elevated levels of triglycerides can accelerate the transfer of FFA to the liver and increase hepatic glucose production, thereby promoting the development of organ steatosis [28]. Many studies have discovered a relationship between the TyG index and the incidence of MAFLD. To forecast the occurrence and severity of MAFLD, the TyG index can be a useful risk assessment tool [29]. Albumin, a crucial protein generated and metabolized in the liver, is associated with specific inflammatory mediators [30]. Its physiological functions include controlling blood colloidal osmotic pressure in addition to maintaining the exchange of intracellular, extracellular, and tissue fluids [31]. Moreover, albumin has a variety of non-colloidal uses, such as molecular transport, immunological regulation, antioxidant and anti-inflammatory properties, and endothelium stability [32]. Albumin has the capability to bind various inflammatory mediators and modulate immune responses in systemic inflammation and sepsis through toll-like receptor signaling pathways [33]. Albumin in plasma is the primary transporter of circulating FFA, facilitating the transfer of FFA from adipose tissue to non-adipose tissues [34]. Studies have indicated that decreased albumin levels might cause negative consequences by disrupting the body’s fluid distribution [35]. Nevertheless, the predictive value of albumin in reflecting inflammation or its independent role requires further clarification. Elevated serum GGT levels are considered an independent marker of systemic inflammatory activation [24]. GGT, as a surface enzyme, is capable of cleaving extracellular glutathione (GSH). Further evidence suggests that elevated blood GGT activity may react to oxidative stress by encouraging the uptake of glutathione into cells, hence supporting the involvement of GGT in oxidative stress and inflammation [36]. Cruz et al. [37] have revealed that GGT levels have a stronger correlation with the severity of fatty liver disease than traditional inflammatory markers, such as ALT. In addition, there is also an association between serum GGT and MetS. Studies have indicated that GGT may enhance the likelihood of insulin resistance, and elevated serum GGT levels significantly increase the risk of MetS and its components, which are considered major factors in the pathogenesis of MAFLD [38]. Our findings demonstrate that the albumin-to-gamma-glutamyl transferase ratio (AGTR) serves as an independent protective factor against MAFLD. Albumin exhibits inherent antioxidant and anti-inflammatory properties, while gamma-glutamyl transferase (GGT) reflects levels of oxidative stress and glutathione metabolism [39]. Albumin scavenges reactive oxygen species and modulates inflammatory signaling, whereas elevated GGT indicates intensified oxidative stress and compromised antioxidant defense [40]. Previous research suggests that decreased albumin levels correlate with increased free fatty acids, which in turn stimulate GGT synthesis and release [25]. Consequently, AGTR represents a composite biomarker reflecting the balance between endogenous antioxidant capacity and systemic oxidative stress [41]. Furthermore, AGTR not only indicates hepatic synthetic function but also captures systemic redox balance and the body’s overall antioxidative and anti-inflammatory status. This interpretation aligns with accumulating evidence emphasizing the crucial role of oxidative stress in MAFLD pathogenesis and its progression to fibrosis and cardiometabolic complications. However, current research on AGTR’s predictive value in MAFLD remains limited, and the synergistic relationship between albumin and GGT in MAFLD development has not been fully elucidated. AGTR may reflect systemic oxidative stress beyond mere hepatic dysfunction. Future studies should further clarify the mechanistic connections among AGTR, redox homeostasis, and hepatic lipid metabolism. This study has a number of notable strengths in the construction of a prediction model for MAFLD. First, this is the first study to utilize AGTR to construct a prediction model for MAFLD, offering a novel perspective for disease risk assessment. In addition, the study incorporated various newly proposed inflammatory biomarkers associated with MAFLD as variables, such as SII and SIRI. This study has a number of notable strengths in the construction of a prediction model for MAFLD. First, this is the first study to utilize AGTR to construct a prediction model for MAFLD, offering a novel perspective for disease risk assessment. In addition, the study incorporated various newly proposed inflammatory biomarkers associated with MAFLD as variables, such as SII and SIRI. However, upon rigorous screening, these biomarkers failed to demonstrate significant clinical relevance. Second, the variables selected in this study are easily obtainable indicators in clinical settings, enhancing the model’s simplicity and applicability. This study has several limitations. First, as NHANES is a cross-sectional survey reflecting only U.S. population characteristics, the model may not capture ethnic or pathophysiological differences in AGTR, nor establish causality or temporal sequence between AGTR and MAFLD. Second, variable collection relied on self-reported questionnaires, which may introduce recall or reporting bias. Third, the use of transient elastography for diagnosing hepatic steatosis, compared to liver biopsy, may lead to an underestimation of MAFLD prevalence. Finally, although internal and external validation were performed, larger multi-center clinical cohort studies are still needed to verify the nomogram’s generalizability and external applicability. Conclusion This study utilized comorbidities, anthropometric characteristics, and serological indicators to establish a predictive model based on the novel immune-inflammatory marker AGTR for assessing the risk of MAFLD. The model holds significant clinical value and helps medical practitioners identify patients with MAFLD in a timely manner. It can also serve as an effective tool for the general population to self-monitor their risk of MAFLD, thereby facilitating early detection, diagnosis, and treatment.