Development and validation of an online dynamic nomogram for nonobese metabolic dysfunction-associated steatotic liver disease based on body composition analysis

Abstract

Background

The factors influencing nonobese Metabolic dysfunction-associated steatotic liver disease are discussed, and an online dynamic nomogram model is constructed.

Methods

A retrospective analysis was conducted on the clinical data of 216 non-obese patients with MASLD and 322 non-obese healthy controls from the Second Affiliated Hospital of Soochow University. The data were randomly divided into a training set and a validation set in a 7:3 ratio. Potential variables were initially screened through univariate analysis, correlation analysis, and LASSO regression (including cross-validation). Subsequently, binary logistic regression and SHAP methods were employed to identify independent risk factors, and an online dynamic ROC curve was constructed. Model performance was evaluated using the area under the curve (AUC), sensitivity, specificity, calibration curves, and recalibration methods.

Results

LASSO regression initially screened 11 potential variables. Binary logistic regression ultimately identified six independent risk factors which is past history (odds ratio (OR): 2.399, P = 0.0008), triglyceride(TG) (OR: 1.176, P = 0.008), high density lipoprotein cholesterol(HDL-C) (OR: 0.173, P = 0.014), low density lipoprotein cholesterol(LDL-C) (OR: 3.916, P = 0.001), fat mass(Fat) (OR: 4.299, P = 0.0009) and dPhaseAngle(OR: 3.174, P = 0.022). SHAP analysis revealed that TG was the most influential feature, while HDL-C showed a negative contribution. All variance inflation factor (VIF) values were below 1.5, indicating no significant multicollinearity. The AUC values were 0.869 in the training cohort and 0.802 in the validation cohort. The calibration curve demonstrates that the prediction results of the line graph for the outcome closely approximate the ideal curve, aligning with the actual outcomes.

Conclusion

An online dynamic nomogram for nonobese MASLD patients with good identified performance was constructed, which can be used as a practical approach for personalized early screening and auxiliary diagnosis of potential risk factors and can assist physicians in making personalized diagnoses and treatments for patients.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12876-026-04972-6.

Introduction

Metabolic Dysfunction-associated steatotic liver disease(MASLD) is a disease which is closely related to the liver, and its onset is closely linked to metabolic abnormalities.The disease, formerly known as nonalcoholic fatty liver disease (NAFLD), renamed by the international expert group in 2023,is known to be associated with insulin resistance, overnutrition and genetic susceptibility, and is a chronic metabolic stress liver disease. In recent years, with the change of lifestyle and the improvement of life quality, the incidence of fatty liver has been increasing year afer year [1].The global prevalence of MASLD ranges from 25% to 30%, with significant regional variations [2].A recent meta-analysis provides more specifc prevalence rates for diferent regions as follows: South Asia 34% (23–47%), South-East Asia 33% (19–51%), East Asia 30% (26–34%), and the Pacifc region 28% (25 − 32%) [3].In China, the prevalence of MASLD is 32%, and there is an increasing trend [4].Traditionally, MASLD has been conventionally regarded as being associated primarily with obesity. However, approximately one-third of affected individuals possess a normal body mass index (BMI) [5]. Notably, while MASLD prevalence reaches 30% -40% in Western countries and is closely linked to obesity, Asian populations can develop this condition even with lower BMI, indicating higher susceptibility to metabolic disorders.The prevalence of nonobese MASLD within China ranges from 4.2% to 19.33%, constituting 15% to 56% of the total MASLD patient population [6]. Compared to their obese counterparts, nonobese MASLD patients typically exhibit a BMI within the normative range, a standard physique, and a relatively balanced body fat distribution, presenting an ostensibly healthy appearance [7]. Nevertheless, they may manifest similar metabolic dysfunctions, potentially attributable to factors such as visceral adipose tissue accumulation and genetic predispositions, albeit distinct from the pathological profile observed in obese MASLD patients [8]. Consequently, nonobese patients are frequently overlooked, resulting in underdiagnosis. This oversight leads to disease progression and the forfeiture of optimal therapeutic intervention. Therefore, identifying non-invasive, accurate, and straightforward serological diagnostic approaches assumes particular importance. The objective of this study is to identify indicators for nonobese MASLD and construct corresponding diagnostic models, thereby providing a scientific foundation for the early screening of MASLD in individuals possessing a normal body mass index.

Materials and methods

Patient data

This study was conducted in the Physical examination Center of the Second Affiliated Hospital of Soochow University from January 2024 to April 2025. A total of 1720 eligible participants with BMI<24 kg/m² were investigated [9], among which 219 cases of MASLD patients and 1501 cases of non-MASLD patients were found. Of these participants,538 underwent body composition analysis, including 216 with MASLD disease and 322 without.The diagnosis of MASLD was made by two senior attending physicians based on the results of ultrasound and imaging examinations and past history and laboratory test results [1].The subjects included: (1) patients diagnosed with MASLD, (2) patients with complete clinical data, (3) patients who had signed an informed consent form and agreed to collect their medical data.The following patients were excluded from the study: (1) those with other liver diseases, such as viral hepatitis, drug-induced liver disease, autoimmune liver disease, cirrhosis, or liver cancer, (2) patients with other major organic diseases or those requiring medication that could affect the study results, (3)those with any history of alcohol consumption, (4) patients with incomplete general information and medical history.The control group consisted of nonobese individuals without MASLD. Inclusion criteria for controls were: (1) BMI < 24 kg/m²,(2) no evidence of hepatic steatosis on abdominal ultrasound or FibroTouch, (3) complete clinical data. Exclusion criteria were the same as those for the MASLD group.This study has been approved by the Ethics Committee of The Second Affiliated Hospital of Soochow University (Ethics number: LK2024111) and informed consent has been obtained from patients.

Data collection

Data collection covered the following items: age, height, weight, blood pressure, and past history, including metabolic diseases such as hypertension and diabetes, and other history. For statistical analysis, the history of metabolic diseases such as hypertension and diabetes was classified as metabolic disease history, and the rest were classified as other disease history. Laboratory tests were carried out by the staff from 7 a.m. to 10 a.m. and fasting blood samples of participants were collected by vein and sent to the Laboratory Testing Center of the Second Affiliated Hospital of Soochow University for testing. The test items include blood routine, fasting blood glucose (FBG), total cholesterol (TC), triglyceride (TG), low density lipoprotein cholesterol (LDL-C), high density lipoprotein cholesterol (HDL-C), alanine aminotransferase (ALT), aspartate aminotransferase (AST), γ-glutamyl transferase (GGT) and other indicators.The human body composition analysis was analyzed by DBA-610 to measure various indicators, including intracellular fluid (inw), extracellular fluid (exw), total body water (tbw), protein (pro), muscle mass (mus), fat free mass (ffm), fat mass (fat), percent body fat (pbf), visceral fat area (vfa), waist-to-hip ratio (whr), basal metabolic rate (bmr), dphase angle (dPhA), and limb skeletal muscle index (dASMI) and so on.The operation is carried out by two professional technicians according to the manual, and the temperature is maintained at 20 to 25 degrees celsius. The examinee should fast, empty his or her bladder and bowels, take off shoes, hats and heavy clothes, stand on the scale, keep even breathing, and the height and weight are accurately recorded by professionals.The abdominal ultrasound examination was performed using GE-VIVid T9 color ultrasonic diagnostic instrument (probe frequency 5.0 MHz) to observe the liver echo intensity, intrahepatic duct structure and deep attenuation.The examination was performed by a professional physician. The examinee should be fasting and lie in a supine position. The liver area was repeatedly explored in the frontal and lateral positions, and whether fatty liver was diagnosed according to relevant guidelines.The CAP(controlled attenuation parameter) value was measured by FibroTouch-pro5000 to complete the quantitative detection of liver steatosis, which was expressed as CAP value with units of dB/m.The operation is performed by two professional doctors according to the user manual, The detection point is located at the right axillary line to the 7th and 8th intercostal or 8th and 9th intercostal lines of the axillary line. The effective detection should be recorded continuously for 10 times, and the median is taken as the final measurement result. The success rate of the final detection should be greater than 60%, and the interquartile distance should be less than 1/3 of the median of the measured value.

Statistical analysis

The study population was stratified into two cohorts based on the presence or absence of MASLD. A descriptive analysis of baseline characteristics for both cohorts was conducted. Continuous variables underwent assessment for normality utilizing the Kolmogorov-Smirnov test. Variables adhering to a normal distribution were expressed as mean ±standard deviation, with group differences evaluated using Student’s t-tests. Variables deviating from normality were described using the median and interquartile range, and group differences were assessed employing the Mann-Whitney U test. Categorical variables were summarized by frequency and percentage, comparisons between the two groups’ baseline data utilized either the chi-square test or Fisher’s exact test, as appropriate. Screening of identified variables was performed utilizing R software (version 4.2.0) and associated packages, including glmnet, Matrix, shapviz and so on. The collected dataset was randomly partitioned by statistical software into training and validation cohorts at a 7:3 ratio, followed by variable comparison. Variables demonstrating statistically significant differences in univariate analysis were incorporated into Lasso regression analysis. The optimal lambda parameter was determined via 10-fold cross-validation to identify independent risk factors, the importance of indicators was analyzed using SHAP, enabling the construction of an online dynamic nomogram for MASLD diagnosis.The nomogram’s performance was evaluated through receiver operating characteristic (ROC) curve analysis and calibration curve assessment.The area under the ROC curve (AUC), ranging from 0.5 (indicating no discriminatory capacity) to 1.0 (denoting perfect discrimination), quantified recognative recog accuracy. All statistical analyses within this study were executed using SPSS version 25.0 and R version 4.2.0. Two-tailed tests were applied, and a p-value less than 0.05 was deemed statistically significant.

Result

Patient features

The detailed flow diagram is presented in Fig. 1. A total of 216 patients with MASLD and 322 patients without MASLD were included in this study from January 2024 to April 2025, all of whom were diagnosed in the Physical examination Center of the Second Affiliated Hospital of Soochow University. All the patients included in the study met the established inclusion and exclusion criteria.Of these patients, 70% were selected for the training cohort and the remaining 30% for the validation cohort.The clinicopathological characteristics of the patients are summarized in Table 1. The Wilcoxon test and chi-square test were used to compare the indices between the nonobese MASLD and non-MASLD groups. In the training cohort, some significant indicators (P < 0.05, Table 2) were selected, including gender(P < 0.001),history(P < 0.001),age (P < 0.001),waist(P < 0.001),BMI(P < 0.001),ALT(P < 0.001),AST(P < 0.001),GGT(P < 0.001),FBG(P < 0.001),TC(P < 0.001),TG(P < 0.001),HDL-C(P < 0.001),LDL-C(P < 0.001),inw(P < 0.001),exw(P < 0.001),tbw(P < 0.001),pro(P < 0.001),mus(P < 0.001),ffm(P < 0.001),fat(P < 0.001),vfa(P < 0.001),whr(P < 0.001),bmr(P < 0.001),dPhaseAngle(P < 0.001) and dASMI (P < 0.001) were statistically significan.The percentage of pbf (P = 0.601) was not statistically significant.

Fig. 1.

Flow chart of the study process

Table 1.

Patient demographics and clinical characteristics

Characteristics	Training Cohort			Validation Cohort
	Nonobese MASLD	Non- MASLD	P	Nonobese MASLD	Non-MASLD	P
N	147	229		69	93
Gender(%)			< 0.001			< 0.001
male	109 (74)	100 (44)		51(73.91)	39(41.93)
female	38 (26)	129 (56)		18(26.09)	54(58.07)
History (Metabolic disease)			< 0.001			0.005
yes	35(23.81)	19(8.30)		16(23.19)	7(7.53)
no	112(76.19)	210(91.70)		53(76.81)	86(92.47)
Age /year(Median (Q1,Q3))	51(42, 58.5)	42 (36,54)	< 0.001	50.54± 9.95	45.24 ± 12.30	0.003
Waist/cm(Median (Q1,Q3))	85 (79, 88)	76 (70, 83)	< 0.001	8 5(80,88)	74(70,83)	< 0.001
BMI kg/m2(Median (Q1,Q3))	23.1(22.2,23.5)	21.7(20.3,23.1)	< 0.001	23.1(22.1,23.7)	21.7(20.8,22.8)	< 0.001
ALT U/L(Median (Q1,Q3))	23 (17, 31)	14 (11, 21)	< 0.001	21(18,31)	14(12,20)	< 0.001
AST U/L(Median (Q1,Q3))	19(16, 24)	17(15, 20)	< 0.001	20(16,230	18(15,21)	0.061
GGT U/L(Median (Q1,Q3))	27 (19.5, 41)	17 (12, 25)	< 0.001	30(21,49)	16(12,24)	< 0.001
FBG mmol/l(Median (Q1,Q3))	5.13 (4.81, 5.64)	4.87 (4.56, 5.18)	0.011	5.07(4.64,5.83)	4.9(4.61,5.22)	0.005
TC mmol/L(Mean ± SD)	5.244 ± 1.038	4.861± 0.884	< 0.001	5.32(4.52,5.85)	4.72(4.12,5.35)	0.002
TG mmol/L(Median (Q1,Q3))	1.69 (1.26, 2.53)	1.04 (0.77, 1.48)	< 0.001	1.89(1.54,;2.73)	0.92(0.72,1.37)	<`0.001
HDL-C mmol/l(Median (Q1,Q3))	1.18 (1.06, 1.36)	1.47 (1.23, 1.76)	< 0.001	1.22(1.06,1.38)	1.44(1.27,1.68)	< 0.001
LDL-C mmol/L(Mean ± SD)	3.345 ± 0.941	2.965 ± 0.787	< 0.001	3.31 ± 0.92	2.91 ± 0.89	0.006
Inw kg(Median (Q1,Q3))	23.3 (19.6, 25.25)	19 (17.2, 23.4)	< 0.001	23.4(20.3,25.4)	19.6(17.2,23.8)	< 0.001
Exw kg(Median (Q1,Q3))	14(12.1, 15.1)	11.7 (10.6, 14.3)	< 0.001	14(12.3,15.4)	12.2(10.6,14.3)	< 0.001
Tbw %(Median (Q1,Q3))	37.2 (31.75, 40.15)	30.9 (27.8, 37.5)	< 0.001	37.5(32.2,40.7)	31.7(27.9,38.3)	< 0.001
Pro kg(Median (Q1,Q3))	10.1 (8.45, 10.9)	8.2 (7.4, 10.1)	< 0.001	10.1(8.8,11)	8.5(7.4,10.3)	< 0.001
Mus kg(Median (Q1,Q3))	47.6(40.45, 51.45)	39.3(35.3, 47.9)	< 0.001	47.9(41.3,51.9)	40.4(35.5,48.9)	< 0.001
Ffm kg(Median (Q1,Q3))	50.8 (43.2, 54.85)	42.1 (38, 51.1)	< 0.001	51(44.2,55.4)	43.2(38.0,52.2)	< 0.001
Fat %(Median (Q1,Q3))	17(15.2, 18.8)	15.3 (13.1, 17.7)	< 0.001	17.48 ± 3.17	15.09 ± 3.55	< 0.001
Pbf %(Mean ± SD)	26.14± 5.63	25.81± 5.89	0.984	26.16 ± 4.66	25.46 ± 6.01	0.401
Vfa cm2(Mean ± SD)	76.7 ± 18.23	67.69 ± 19.9	< 0.001	78.26 ± 17.38	65.84 ± 19.84	< 0.001
whr(Median (Q1,Q3))	0.9 (0.88, 0.93)	0.87 (0.84, 0.89)	< 0.001	0.91 ± 0.04	0.87 ± 0.04	< 0.001
Bmr kcal/day(Median (Q1,Q3))	1467.2 (1304, 1554.45)	1278.6 (1191.9, 1473.7)	< 0.001	1471.3 (1324.5,1567.2)	1303 (1190.6,1497.7)	< 0.001
dPhaseAngle ° (Mean ± SD)	5.91± 0.67	5.53± 0.66	< 0.001	5.97 ± 0.59	5.54 ± 0.69	< 0.001
dASMI kg/m2(Median (Q1,Q3))	7.37(6.64, 7.78)	6.46 (5.88, 7.35)	< 0.001	7.325(6.7,7.68)	6.52(5.92,7.43)	< 0.001

Table 2.

The results of receiver operating characteristic (ROC) curve

Characteristics	AUC	Cut-off value	Sensitivity	Specificity	Youden index	95%CI	P
history	0.578	1.0	0.236	0.919	0.155	0.555–0.617	0.012
age	0.632	46.0	0.671	0.568	0.24	0.589–0.681	0.000
waist	0.764	78.0	0.838	0.578	0.416	0.722–0.799	0.000
ALT	0.738	17.0	0.796	0.602	0.399	0.696–0.765	0.000
GGT	0.746	20.0	0.759	0.643	0.402	0.713–0.786	0.000
FBG	0.640	5.0	0.606	0.621	0.228	0.597–0.686	0.000
TG	0.793	1.38	0.741	0.72	0.461	0.760–0.828	0.000
HDL-C	0.738	1.33	0.738	0.668	0.376	0.704–0.778	0.000
LDL-C	0.630	3.39	0.514	0.739	0.253	0.575–0.672	0.000
fat	0.659	15.1	0.792	0.463	0.254	0.594–0.711	0.000
dPhaseAngle	0.672	5.589	0.741	0.571	0.312	0.623–0.717	0.000

Factor selection for the model, calibration, and validation of the nomogram

The aforementioned variables were integrated into the baseline model and subsequently narrowed to eleven potential predictors through LASSO regression analysis applied to the training cohort, with cross-validation of the LASSO model (Fig. 2A). A cross-validated error plot of the LASSO regression model is shown in Fig. 2B. This figure demonstrates that the most regularized and parsimonious model, achieving a cross-validated error within one standard error of the minimum, incorporated eleven variables. The standard error of the minimum distance corresponded to λ = 0.031. The variables selected for this model comprised: history, age, waist, ALT, GGT, FBG, TG, HDL-C, LDL-C, fat, and dPhaseAngle.

Fig. 2.

Results of the LASSO regression analysis. A Plot of the LASSO coefficient profiles. B Tuning parameter (λ) selection cross-validation error curve

As demonstrated in Table 2; Fig. 3, the receiver operating characteristic (ROC) analysis of the specified diagnostic indicators reveals that the area under the curve (AUC) values exceed 0.5 for the following parameters, history (AUC = 0.578), age (AUC = 0.632), waist circumference(AUC = 0.764), alanine aminotransferase (ALT) (AUC = 0.738), gamma-glutamyl transferase (GGT) (AUC = 0.746), fasting blood glucose (FBG) (AUC = 0.64), triglycerides (TG) (AUC = 0.793), high-density lipoprotein cholesterol (HDL-C) (AUC = 0.738), low-density lipoprotein cholesterol (LDL-C) (AUC = 0.63), fat content (AUC = 0.659), and dPhaseAngle (AUC = 0.672). The corresponding critical values, sensitivity, and specificity for these parameters are detailed in Table 2, with the ROC curve analysis of candidate diagnostic indicators presented in Fig. 3.

Fig. 3.

ROC curve analysis of candidate diagnostic indicators

In order to determine whether the above eleven variables were independent risk factors for MASLD, a binary logistic regression analysis was used to evaluate the effects of history, age, waist, ALT, GGT, FBG, TG, HDL-C, LDL-C, fat and dPhaseAngle. The AUC of the training Cohort data was 0.85, and the identification effect of the model was good. The analysis results showed in Table 3 and history, TG, HDL-C, LDL-C, fat and dPhaseAngle were statistically significant and were included in the model(Fig. 4A). Similar results were obtained in validation cohorts (Fig. 4B). Results of the correlation analysis showed that the six factors(history, TG, HDL-C, LDL-C, fat and dPhaseAngle) were all linearly correlated with each other, while HDL-C was negatively correlated with other indices (Fig. 5). To further assess multicollinearity among the six selected predictors, we calculated the variance inflation factor (VIF) for each variable. The VIF values were as follows: HDL-C 1.437, dPhaseAngle 1.328, TG 1.279, fat 1.150, LDL-C 1.121, and history 1.041. All VIF values were well below the commonly accepted threshold of 5, indicating that multicollinearity was not a significant concern in our model.

Table 3.

The results of logistic regression analysis of prediction indicators

Predictor	Estimate	SE	OR (95%CI)	Z	p
history	0.875	0.338	2.399(1.247–4.716)	2.587	0.0008
age	0.023	0.011	1.023(1.001–1.046)	2.074	0.279
ALT	0.024	0.009	1.024(1.006–1.044)	2.548	0.130
waist	0.033	0.021	1.034(0.992–1.077)	1.569	0.119
GGT	0.004	0.003	1.004(0.998–1.011)	1.35	0.328
FBG	0.196	0.089	1.216(1.012–1.470)	2.204	0.278
TG	0.162	0.12	1.176(0.951–1.526)	1.352	0.008
HDL-C	-1.757	0.456	0.173(0.069–0.414)	-3.853	0.014
LDL-C	0.552	0.141	1.737(1.323–2.302)	3.916	0.001
fat	0.175	0.041	1.191(1.102–1.292)	4.299	0.0009
dPhaseAngle	0.729	0.23	2.073(1.329–3.278)	3.174	0.022

Fig. 4.

Forest maps of the logistic regression analysis of the training cohort (A), Validation cohort (B),

Fig. 5.

Linear correlation analysis of the eight indicators. The number in the right of the plot was the correlation coefficient

To enhance model interpretability, we employed the SHAP (Shapley Additive exPlanations) method to assess the contribution of each variable to the model recognition. This approach provides an intuitive representation of the contribution of each feature to the outcome, thereby aiding in a deeper understanding of the model’s internal mechanisms. The results indicate that TG is the most influential factor in the model, suggesting a critical role in the model’s decision-making process. The contributions of fat and dPhaseAngle are comparable, indicating that these two variables have similar explanatory power in model. The distributions of HDL-C and LDL-C are relatively symmetrical, which may be related to their synergistic roles in human metabolism. In contrast, past history exhibits the smallest SHAP value, indicating its relatively limited identified contribution in the current model (Fig. 6A, B). These findings not only validate our model construction rationale but also provide important insights for subsequent optimization and clinical application.

Fig. 6.

SHAP Feature Importance Ranking and Effect Direction in MASLD Identification Model. (A Bar plot): Feature importance ranking based on mean absolute SHAP values. (B. Beeswarm plot): Visualization of SHAP value distribution and effect direction for each feature

The final logistic model included six independent predictors(history, TG, HDL-C, LDL-C, fat and dPhaseAngle)and was developed as a simple-to-use nomogram, which is illustrated in Fig. 7A and available online (https://stop15.shinyapps.io/DynNomapp/) and presented in Fig. 7B. For example, one person, whose history is no, TG is 0.79mmol/L, HDL-C is 1.47mmol/L, LDL-C is 2.7mmol/L, fat is 16.3% and dPhaseAngle is 5.153°. The total score of the predicted score was 112, indicating a probability of 9.52% for MASLD (Fig. 7A).

Fig. 7.

Nomogram identification model for nonobese MASLD diagnosis. A Established nomogram in the training cohort by incorporating the following six parameters. B Online dynamic nomogram accessible at https://stop15.shinyapps.io/DynNomapp/

As demonstrated in Fig. 8A-D, the area under the curve values for the model were 0.869(95%CI:0.834–0.905) in the training cohort and 0.802(95%CI:0.734–0.870༉in the validation cohort, indicating robust identified capability. Other performance indicators under the aggregated optimal probability threshold are shown in Table 4. Regarding calibration accuracy, the calibration curve on the training set demonstrates a high consistency between predicted probabilities and actual observed probabilities (calibration intercept is -0.084 and slope is 0.56). However, on the validation set, the original calibration intercept remains-0.084 while the calibration slope drops to only 0.56, indicating that the model’s predicted risk values are too extreme to be used directly as absolute risk estimates. To improve probability calibration on the validation set, a logistic regression-based recalibration method was employed, constructing the calibration model: logit (calibrated probability) = a + b*logit (original probability), with a = -0.084 and b = 0.56. After applying this method, the calibration slope on the validation set increased to 1, the intercept approached 0 (-5.36 × 10^− 16), and the calibration curve perfectly coincided with the ideal straight line.

Fig. 8.

Evaluation of validity and reliability of the model.ROC curves of the nomogram identification model in the training cohort (A), validation cohortt (B),Training set calibration curve(C) and Validation set calibration curve (after post-hoc logical correction) D

Table 4.

Predictive performance of the nomogram in the training and validation sets

Dataset	AUC(95%CI)	cutoff	Sensitivity (95%CI)	Specificity (95%CI)	PPV(95%CI)	NPV(95%CI)	F1 Score (95%CI)
Train	0.869(0.834–0.905)	0.370	0.846(0.777–0.890)	0.780(0.720–0.832)	0.716(0.643–0.781)	0.885(0.833–0.926)	0.775(0.725–0.825)
validation	0.802(0.734–0.870)	0.296	0.881( 0.778–0.947)	0.674(0.570–0.766)	0.656(0.548–0.753)	0.889(0.793–0.951)	0.752(0.671–0.822)

Discussion

Nonobese Metabolic dysfunction-associated steatotic liver disease(MASLD) has become one of the leading causes of chronic liver disease worldwide, with obese MASLD being more prevalent, yet nonobese MASLD should not be overlooked [10]. This disease is not only the main cause of cryptogenic cirrhosis, but is also closely related to insulin resistance, metabolic disorders, and cardiovascular diseases. As the disease progresses, The disease burden of patients will be significantly increased and the quality of life will be significantly decreased [11]. At present, MASLD has become an important public health problem. Nonobese MASLD is asymptomatic in the early stage, with low clinical laboratory indicators and normal BMI value, which makes it difficult to attract people’s attention and easy to be missed [12]. However, liver biopsy, as the gold standard for diagnosing fatty liver [6], is an invasive and invasive procedure that is not readily accepted by the population.Ultrasonic and fibroscan are not sensitive to mild fatty liver, and detect more than 20%~30% of fatty degeneration [13]. In addition, these examinations all depend on the level of operators, which is highly subjective, and it is difficult to popularize in economically backward areas.The new expert consensus affirms the value of blood biomarkers in diagnosing MASLD, but there is still a lack of effective identified models [14]. The incidence of nonobese MASLD in the nonobese physical examination population was 40.15% in our study, which was higher than that in existing studies (4.2%~19.33%) [6]. Therefore, the development of identification tools for nonobese population is very important for disease prevention and management, and it is also one of the hot research topics in recent years.

The body composition mainly discusses the composition data of human body, the detection methods and the influence of internal and external factors on the quantitative relationship of each component, which reflects the response process of the body to different changes in the inside and outside of the body at the molecular, cellular and tissue levels.The measurement of human body can not only accurately reflect the content of body fat, muscle and water [15], It can also reflect the visceral fat area (VFA), waist-to-hip ratio (WHR) and so on, so as to judge whether the body composition is normal, which has certain screening value for metabolic diseases [16]. Changes in body composition are associated with the occurrence and development of many diseases [17]. The body composition analysis is suitable for the diagnosis, management, and risk stratification of MASLD [18]. With the progress of society and the improvement of people’s living standards, nonobese MASLD is attracting more and more attention.

The results of Logistic regression in our study showed that the history of metabolic disease, TG, HDL-C, LDL-C, fat and dphase Angle were the identified indicators for the diagnosis of nonobese MASLD, which were consistent with most of the previous studies [12]. At present, it has been shown that the incidence of MASLD in nonobese people is often associated with nutritional status, metabolic factors and genetic inheritance [12]. MASLD is a hereditary disease, and a large number of studies have shown that the presence of patatin-like phospholipase domain protein 3rs738409 gene is closely associated with it [19]. In addition, abnormal blood lipid is an important factor in the occurrence and development of MASLD. Dyslipidemia is usually caused by increased lipolysis in liver tissue, activation of new fat production, and excessive accumulation of fat due to high-fat diet mechanisms [20]. Nonobese MASLD patients have more visceral fat accumulation, and visceral adipose tissue has stronger adipolysis activity than subcutaneous adipose tissue, leading to the increase of high levels of free fatty acids (FFA) in the liver, which aggravates the TG burden of the liver [21]. High TG, especially the increase of ceramide and diacylglycerol, can aggravate insulin resistance and induce inflammatory response of liver fibrosis, which further aggravates the disease [20]. As our SHAP analysis demonstrates, TG is the most identified feature in the model, exhibiting the highest average absolute SHAP value. The scatter plot reveals that its high values (yellow dots) predominantly correspond to positive SHAP values, indicating that elevated TG levels significantly increase the model’s recognized risk assessment. HDL-C is a protective factor for nonobese MASLD, cardiovascular and cerebrovascular diseases, and type 2 diabetes (T2DM).It delays the progression of nonobese MASLD by activating the adenylate activated protein kinase signaling pathway and regulating lipid metabolism [22]. Fat in body composition is also an independent risk factor for nonobese MASLD. People with high visceral fat were more likely to develop MASLD, which was consistent with previous studies [16]. dPhaseAngle is a common indicator to evaluate the nutritional health level of patients in clinical practice [23], A high value reflects good cell health, while low of it may indicate damage to the cell membrane or reduced cell integrity, often associated with disease or malnutrition [24]. In patients with inflammatory bowel and liver diseases, dPhaseAngle was consistently found to be related to nutritional status and body composition, particularly the depletion of lean body mass and sarcopenia. It has been associated with prognosis, disease stage and severity and found to be helpful in monitoring fluid shifts and response to interventions [25]. Variations in dPhaseAngle values can serve as a sensitive indicator for assessing hepatic metabolic function and disease progression, providing supplementary information for early identification and risk stratification of MASLD.To further assess the multicollinearity among the six selected predictors, we calculated the variance inflation factors (VIF) for each variable. All VIF values were significantly below the commonly used threshold of 5 (ranging from 1.041 to 1.437), indicating no significant multicollinearity in this model and ensuring the stability of the regression coefficients.

In this study, an online dynamic nomogram for MASLD in nonobese patients was constructed for the first time by computer software.The AUC values for the training cohort and validation cohort were 0.869 and 0.802, respectively indicating excellent diagnostic performance. From a clinical practical perspective, the model demonstrated high sensitivity (88.1%) and negative predictive value (88.9%) on the validation set. This ROC curve effectively identifies MASLD patients, with an accuracy of nearly 90% in correctly identifying individuals as truly disease-free when predicted as “asymptomatic.” This characteristic makes the ROC curve particularly suitable as a primary screening tool for MASLD, enabling efficient exclusion of low-risk individuals from large populations and reducing unnecessary further investigations. The moderate specificity (67.4%) and positive predictive value (65.6%) suggest that additional diagnostic tests are warranted for individuals predicted to be at high risk.Regarding calibration accuracy, although the calibration curve on the training set demonstrates a high consistency between predicted probabilities and observed probabilities (calibration intercept: -0.084; slope: 0.56), the validation set initially exhibited an excessively extreme distribution of predicted risks (calibration slope: only 0.56). To address this issue, we employed a recalibration method based on logistic regression: logit (post-calibration probability) = a + b*logit (original probability), where a = -0.084 and b = 0.56. After applying this method, the calibration slope for the validation set increased to 1, with the intercept approaching zero (-5.36 × 10^− 16), and the calibration curve perfectly overlapped with the ideal line. This recalibration step is crucial for ensuring that the model’s predicted probabilities can be reliably interpreted as absolute risk estimates in clinical practice.

However, our study also has some limitations: (1) The diagnosis of MASLD is based on ultrasound and imaging examinations rather than biopsy, which may lead to misclassification bias in the early-stage. It is reported that ultrasound detected mild steatosis (minimum amount 10% on histology) with a cut-off valued ≥ 2 (sensitivity 90.1%, specificity 90%), moderate steatosis (≥ 30%) with a cut-off value ≥ 3 (sensitivity 86.4%, specificity 92.5%) and severe steatosis (> 66%) with a cut-off value ≥ 5 (sensitivity 88.5%, specificity 87%) [26]. CT had a sensitivity and specificity of 72% and 88%, respectively, for steatosis (> 5% fat at biopsy) detection and 82% and 94% for at least moderate steatosis (over 20%-33% fat at biopsy) detection [27]. Future studies should consider liver biopsy as the diagnostic standard whenever possible to ensure accuracy.In addition, future studies may utilize CT or MRI to assess abdominal fat distribution, particularly visceral fat area (VFA) and subcutaneous fat area (SFA). CT and MRI can provide more accurate quantitative and spatial distribution information of fat, facilitating deeper exploration of the mechanisms by which different fat reservoirs influence MASLD.(2) Due to the cross-sectional design, the study may be biased due to participants’ high concern for their health. Future studies could expand to include multi-center designs and long-term follow-up data to further refine the research. (3)Although the line graph performed well in the training set and internal validation set, its generalization ability needs to be evaluated in larger external validation sets and prospective studies.(4)In addition, although we assessed multicollinearity using variance inflation factor and found no significant concerns (all VIF values < 1.5), we acknowledge that LASSO regression does not completely eliminate collinearity among selected variables. However, the low VIF values suggest that the impact of any remaining collinearity on the interpretation of odds ratios is likely minimal.

In summary, this research established an online diagnostic model, based on body composition and biochemical blood indicators, for the detection of nonobese metabolic dysfunction-associated steatotic liver disease (MASLD). This model represents the first instance utilizing metabolic indicators as its core foundation, providing an intuitive visual interface amenable to widespread implementation. It further demonstrates robust discrimination and high calibration accuracy. Given the current absence of specific pharmacological interventions for nonobese MASLD, this model effectively identifies potential patients within the nonobese population, thereby assisting clinicians in the early identification of high-risk individuals. Furthermore, the model facilitates the adoption of preventative measures, such as dietary and lifestyle modifications, among physical examination participants, thereby mitigating disease progression and alleviating the associated socioeconomic and healthcare burdens.

Conclusion

This study was the first to develop and validate online nomograms based on the independent risk factors to dynamically indentify diagnosis in individuals with nonobese MASLD. This model demonstrated superior performance and discriminative capability, which can used as a practical approach for the personalized early screening and auxiliary diagnosis of the potential risk factors.

Abbreviations

MASLD

Metabolic dysfunction-associated steatotic liver

NAFLD

Nonalcoholic fatty liver disease

ROC

Receiver operating characteristic curve

AUC

Area under the curve

PPV

Positive predictive value

NPV

Negative predictive value

SHAP

Shapley Additive exPlanations

OR

Odds ratio

BMI

Body mass index

FBG

Fasting blood glucose

TC

Total cholesterol

TG

Triglyceride

LDL-C

Low density lipoprotein cholesterol

HDL-C

High density lipoprotein cholesterol

ALT

Alanine aminotransferase

AST

Aspartate aminotransferase

GGT

γ-glutamyl transferase

Inw

Intracellular fluid

Exw

Extracellular fluid

Tbw

Total body water

Pro

Protein

Mus

Muscle mass

Ffm

Fat free mass

Fat

Fat mass

Pbf

Percent body fat

Vfa

Visceral fat area

Whr

Waist-to-hip ratio

Bmr

Basal metabolic rate

DPhA

Dphase angle

dASMI

Device-derived appendicular skeletal muscle index

CAP

Controlled attenuation parameter

Authors’ contributions

All the authors worked together to complete the paper. As the corresponding authors, ZW and YZ guided the design of the entire. XS contributed to collecting ultrasound data and body composition data.TZ and SW performed the experiments and contributed to the acquisition and analysis of data. The manuscript was drafted by TZ and HY. All authors read and approved the final manuscript.

Funding

This project was supported by National Natural Science Fund (82371218) and by pre-research fund project of Second Affiliated Hospital of Soochow University(SDFEYJC2330).

Data availability

We are willing to share the raw data. Data has been provided within the manuscript or supplementary information files.

Declarations

Ethics approval and consent to participate

This study has been approved by the Ethics Committee of The Second Affiliated Hospital of Soochow University (Ethics number: LK2024111) and informed consent has been obtained from patients. Research on human data must comply with the Helsinki Declaration.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.