Evaluation of the lipid accumulation product for MASLD risk stratification in type 2 diabetes: establishing sex-specific thresholds and clinical utility

Abstract

Background

While visceral adiposity and lipid dysregulation are established drivers of metabolic dysfunction-associated steatotic liver disease (MASLD), the clinical utility of the lipid accumulation product (LAP) for identifying prevalent MASLD in patients with type 2 diabetes mellitus (T2DM) remains insufficiently characterized. This study aimed to characterize the dose-response relationship between LAP and MASLD in T2DM, establish optimal sex-specific diagnostic thresholds, and evaluate its clinical net benefit to guide non-invasive screening.

Methods

This study included 495 inpatients with T2DM. log₁₀LAP was prioritized using the Boruta algorithm. Diagnostic cut-offs were determined via ROC analysis. The mathematical reliability of these thresholds was evaluated using 1,000-run stratified bootstrapping (internal validation), while the biological generalizability of LAP was further examined in an independent NHANES cohort (external validation). The dose–response relationship was characterized by restricted cubic splines (RCS). Clinical utility and stability were assessed using decision curve analysis (DCA) and subgroup analyses.

Results

Through a systematic feature-selection approach using the Boruta algorithm, log₁₀LAP was objectively identified as the most robust indicator for MASLD.log₁₀LAP was independently associated with MASLD (OR 1.83, 95% CI 1.43–2.35). Optimal sex-specific cut-offs were 20.4 for men and 27.0 for women, yielding a positive predictive value of 85.93%. RCS analysis revealed a significant linear association, with MASLD probability increasing monotonically with log₁₀LAP. DCA demonstrated a consistently higher net benefit for the LAP-based model over the “screen-all” strategy at threshold probabilities > 0.20. Subgroup analyses confirmed robustness across age and BMI strata, with the highest discriminative power in patients aged < 55 years (AUC 0.855) and reliable performance in non-obese individuals (AUC 0.711). External analysis in the NHANES cohort (N = 630) demonstrated consistent independent associations between LAP and MASLD risk (P < 0.05).

Conclusions

log₁₀LAP is a robust linear predictor of MASLD in T2DM. Implementing tailored thresholds provides superior diagnostic precision and clinical net benefit, particularly for younger and non-obese populations, supporting its use as a prioritized non-invasive screening tool.

Graphical abstract

Supplementary Information

The online version contains supplementary material available at 10.1186/s12933-026-03205-0.

Research Insights

What is currently known about this topic?

• Synergistic Risk in T2DM Patients: T2DM is a critical driver of MASLD, with a significantly higher prevalence (over 60–70%) and a more severe metabolic profile compared to the general population.

• Limitations of Traditional Indices: Conventional metrics like Body Mass Index (BMI) fail to capture visceral adiposity and ectopic fat deposition, which are the primary drivers of hepatic steatosis and insulin resistance in diabetic individuals.

• Potential of the LAP: While LAP has emerged as a robust, non-invasive indicator of visceral fat accumulation, its clinical utility and optimal diagnostic thresholds have predominantly been established in healthy cohorts, leaving its effectiveness in the distinct metabolic milieu of T2DM patients insufficiently validated.

What is the key research question?

This study investigates whether log₁₀LAP maintains its priority as a reliable indicator associated with MASLD in patients with type 2 diabetes, how the overall dose-response relationship behaves within this population, and whether implementing sex-specific thresholds enhances the discriminative utility for identifying prevalent MASLD in this high-prevalence inpatient population.

What is new?

• Through a systematic feature-selection approach, log₁₀LAP was objectively identified as a primary indicator for MASLD in patients with T2DM, exhibiting stronger discriminative power than traditional anthropometric and lipid metrics .

• This study defined a steady, linear dose-response relationship between log₁₀LAP and MASLD risk. This continuous metabolic impact highlights that even incremental increases in visceral adiposity significantly elevate MASLD risk in the T2DM population, providing a more intuitive assessment than categorical variables.

• We established and validated sex-specific optimal cut-off values (20.4 for men and 27.0 for women). The stability and generalizability of these thresholds were confirmed through 1,000-run internal bootstrapping and external validation using the NHANES cohort (N = 630). DCA further demonstrated that integrating log₁₀LAP into clinical assessment offers a substantial net benefit for identifying prevalent MASLD in high-prevalence settings.

How might this study influence clinical practice?

• Refined Identification of Prevalent MASLD: The identified sex-specific log₁₀LAP cut-off values (20.4 for men and 27.0 for women) may assist clinicians in detecting MASLD among T2DM patients in hospital settings, providing a preliminary basis for further diagnostic evaluation and patient management.

• Cost-Effective Screening: By utilizing simple anthropometric and biochemical measures (WC and TG), LAP offers a low-cost, non-invasive triage tool prior to specialized imaging, making it ideal for routine metabolic monitoring in resource-limited clinical settings.

• Targeted Metabolic Management: The identified linear dose-response relationship underscores the need for aggressive visceral adiposity control in diabetic patients, suggesting that even moderate reductions in LAP levels are associated with a lower the risk trajectory of MASLD.

Introduction

Metabolic dysfunction-associated steatotic liver disease (MASLD) has emerged as a global public health challenge, affecting approximately 30% of the adult population worldwide [1–9] .In China, the prevalence of MASLD has risen sharply, becoming a primary driver of chronic liver disease, cirrhosis, and hepatocellular carcinoma [10–14]. T2DM is one of the most potent risk factors for MASLD, with over 70% of diabetic patients estimated to have concomitant hepatic steatosis [15–18]. Given this synergy, early identification of MASLD within the T2DM population is critical to preventing advanced fibrosis and cardiovascular complications.

The LAP, calculated from waist circumference (WC) and triglyceride (TG) levels, has been proposed as a robust, non-invasive indicator of visceral adiposity [13–15, 19–21]. In the general population, LAP has demonstrated superior performance in identifying metabolic syndrome and steatotic liver disease compared to traditional anthropometric indices like BMI [22–28]. However, the applicability of LAP in the T2DM population warrants independent investigation due to their fundamentally different metabolic profiles. Unlike the general population, where MASLD prevalence is approximately 30%, the prevalence in T2DM cohorts often exceeds 60–70% [29], potentially altering the predictive power and clinical utility of adiposity markers.

Furthermore, patients with T2DM exhibit a distinct metabolic milieu characterized by profound insulin resistance, “overflow” of free fatty acids, and chronic low-grade inflammation [9, 30–34]. These factors lead to a higher baseline of TG and WC, suggesting that the “standard” LAP reference ranges and cut-offs derived from healthy or general populations may lead to significant misclassification in diabetic patients. Consequently, it is unclear whether LAP remains a sensitive “value-added” biomarker or if its diagnostic efficiency is blunted by the overall metabolic derangement inherent in T2DM.

To address this gap, we utilized a clinical cohort of patients with T2DM to investigate the diagnostic value of LAP specifically within this high-risk metabolic context. This study aims to: (1) Identify the priority of log₁₀LAP among various metabolic variables using the Boruta algorithm; (2) Characterize the dose-response relationship between log₁₀LAP and MASLD risk using RCS analysis; (3) Establish sex-specific diagnostic cut-offs tailored to the T2DM population and evaluate their performance; (4) Validate the mathematical stability of these cut-offs through 1,000-run stratified bootstrapping; (5) Quantify the clinical utility using DCA and evaluate model robustness through rigorous subgroup and sensitivity analyses. (6) Assess the biological generalizability of the findings using an independent external cohort from the NHANES database; By identifying population-specific cut-off values, this study contributes a more refined, non-invasive indicator for identifying prevalent MASLD in patients with type 2 diabetes within a hospital setting.

Methods

Study design and data source

A total of 993 inpatients admitted to the Second Affiliated Hospital and Yuying Children’s Hospital of Wenzhou Medical University between September 1, 2024, and September 30, 2025, were screened for this study. Patients were excluded if they presented with at least one of the following conditions: (1) presence of viral hepatitis, including HBV, HCV, and HEV (N = 75) and Acute infection during hospitalization (N = 31); (2) absence of type 2 diabetes mellitus (N = 103); (3) heavy alcohol consumption (N = 61), defined as daily intake > 25 g for men and > 15 g for women according to the criteria of the National Institute for Nutrition and Health, Chinese Center for Disease Control and Prevention [35–37]; (4) pregnancy (N = 2); or (5) incomplete data for LAP calculation (N = 216) or MASLD diagnosis (N = 10). Following these exclusions, a final cohort of 495 participants with T2DM was enrolled in this retrospective study.

This study design was approved by the Institutional Review Board of the Second Affiliated Hospital and Yuying Children’s Hospital of Wenzhou Medical University (No.2025-K-270-01), and has been performed in accordance with ethical standards laid down in the Declaration of Helsinki. Due to the retrospective nature of the analysis, the requirement for written informed consent was waived by the Institutional Review Board. The data processing and selection procedure is illustrated in Supplementary Fig. 1.

To evaluate the generalizability of the diagnostic association between LAP and MASLD, an independent external validation was conducted using data from the National Health and Nutrition Examination Survey (NHANES) database. We extracted a validation subset (N = 630) from the (2017–2018) cycle, applying inclusion and exclusion criteria consistent with the primary cohort: adults (≥ 18 years) with a documented diagnosis of T2DM and complete data for LAP and hepatic steatosis assessment. Hepatic steatosis in the NHANES cohort was defined by a controlled attenuation parameter (CAP) ≥ 285 dB/m obtained via vibration-controlled transient elastography (VCTE).

Covariates

Medical history and baseline characteristic data, including age, sex, height, weight, WC, blood pressure, duration of diabetes, tobacco and alcohol consumption, medication use (e.g., glucose-lowering and lipid-lowering drugs), and laboratory parameters, were abstracted from individual medical records. Laboratory measurements included TG, total cholesterol (TC), high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), fasting blood glucose (FBG), HbA1c, alanine aminotransferase (ALT), aspartate aminotransferase (AST), gamma-glutamyl transferase (GGT), serum uric acid, and creatinine.

Anthropometric measurements were conducted on the first day of admission by trained medical staff. Body mass index (BMI) was calculated as weight in kilograms divided by the square of height in meters (kg/m²). Waist circumference was measured to the nearest 0.1 cm at the midpoint between the lower costal margin and the iliac crest. The primary outcome, MASLD, was diagnosed based on the 2023 international multi-society consensus criteria, requiring the presence of hepatic steatosis via ultrasonography in combination with type 2 diabetes mellitus [38–40]. The criteria for diagnosing comorbid conditions such as hypertension, hyperlipidemia, and coronary heart disease were adopted from previously established guidelines [41, 42]. Tobacco and alcohol consumption were treated as dichotomous variables, where participants were categorized into two groups: current/past users (Yes) and never users (No). Similarly, medical histories including hypertension and hyperlipidemia were recorded as binary presence. The specific criteria for these variables were as follows: Smoking: Participants were classified into smokers (including current smokers and former smokers) and non-smokers (those who had never smoked). Drinking: Drinking status was categorized into current drinkers (any alcohol consumption within the past year) and non-drinkers (never consumed alcohol). As noted in our exclusion criteria, daily alcohol intake for drinkers did not exceed the heavy drinking thresholds established by the National Institute for Nutrition and Health, Chinese Center for Disease Control and Prevention. Hypertension: Defined as a documented medical history of elevated blood pressure (SBP ≥ 140 mmHg and/or DBP ≥ 90 mmHg) or the current use of antihypertensive medications. Hyperlipidemia: Defined as a documented medical history of elevated lipid levels (tTotal Cholesterol ≥ 6.2 mmol/L or LDL-C ≥ 4.1 mmol/L) or the current use of lipid-lowering medications (e.g., statins or ezetimibe).

For the external validation cohort (NHANES), covariates were harmonized to match the primary cohort as closely as possible. Hepatic steatosis was defined using the Controlled Attenuation Parameter (CAP) obtained via Vibration-Controlled Transient Elastography (VCTE), with a threshold of ≥ 285 dB/m indicative of significant steatosis [43]. In alignment with the 2023 consensus, MASLD in the NHANES subset was diagnosed as the presence of hepatic steatosis among participants with pre-existing T2DM.

Exposures

The LAP was used as the primary exposure and was calculated using the following sex-specific formulas [44, 45]:

  

To stabilize variance and address the skewed distribution of LAP values, a log-transformation was applied, denoted as log₁₀LAP.

Fasting venous blood samples were collected from participants after an overnight fast of at least 8 h by trained medical staff at the Second Affiliated Hospital and Yuying Children’s Hospital of Wenzhou Medical University. Serum TG levels were measured using the GPO-PAP enzymatic method on a Siemens ADVIA 2400 Chemistry System (Siemens Healthineers, Germany). WC was measured to the nearest 0.1 cm at the midpoint between the lower costal margin and the iliac crest at the end of a normal expiration. The coefficients of variation (CV) for the laboratory measurements of TG were consistently below 2.5%, ensuring high analytical precision.

Outcome

MASLD was the primary outcome, defined according to the 2023 international multi-society consensus. Diagnosis required the presence of hepatic steatosis plus at least one of five cardiometabolic risk factors (CMRFs) [46–48]. In this cohort, T2DM served as the requisite CMRF for all participants. Hepatic steatosis was assessed via high-resolution B-mode abdominal ultrasonography (Philips IU22) after an 8-hour fast. Although liver biopsy is the gold standard, ultrasonography is a validated and widely accepted modality for steatosis screening in large-scale clinical studies due to its high specificity and non-invasiveness (cite: Rinella et al., 2023; AASLD Guidelines). Steatosis was identified based on standardized semi-quantitative features: (1) diffuse hyperechogenicity (“bright liver”); (2) ultrasound beam attenuation; and (3) obscured visualization of intrahepatic vessels or the diaphragm [49]. To ensure diagnostic rigor, scans were performed by experienced radiologists blinded to clinical data, with all findings independently reviewed by two senior sonographers and discrepancies resolved by consensus. Diabetes diagnosis followed established clinical guidelines: HbA1c% ≥ 6.5%, random glucose ≥ 11.1 mmol/L, fasting glucose ≥ 7.0 mmol/L, or 2-h post-glucose challenge ≥ 11.1 mmol/L [50].

Statistical analysis

Continuous variables were reported as means ± standard deviations (SD) or medians (interquartile ranges, IQR), and categorical variables as frequencies and percentages. Differences between groups were assessed using the t-test, Mann-Whitney U test, or chi-square test as appropriate. A comparison of baseline characteristics between included (N = 495) and excluded participants (N = 216) revealed no significant differences in major clinical and metabolic parameters, suggesting a minimal risk of selection bias (Supplementary Table 1). To ensure robust variable selection, the Boruta algorithm [51], a wrapper method based on a random forest classifier, was employed to rank the importance of clinical predictors for MASLD. This approach identifies the most significant contributors by comparing real features with “shadow” features, confirming the prioritized role of log₁₀LAP. The incremental discriminative value of LAP relative to its individual components (waist circumference and triglycerides) was assessed using Net Reclassification Improvement (NRI) and Integrated Discrimination Improvement (IDI).

Multivariate logistic regression analysis was conducted to investigate the independent association between log₁₀LAP and MASLD. Four incremental models were constructed: Model 1: Adjusted for age and sex. Model 2: Adjusted for variables in Model 1 plus smoking, alcohol consumption, BMI, hypertension, and hyperlipidemia. Model 3: Adjusted for variables in Model 2 plus WBC, PLT, ALT, AST, Cr, VD, and UA. Model 4: Adjusted for variables in Model 3 plus medication use, including, HbA1C, Duration.of.diabetes, ARB, statins, and ezetimibe.In this study, the logistic regression models were rigorously assessed to ensure compliance with all fundamental assumptions. Model assumptions were rigorously verified, including posterior predictive checks, Cook’s distance for influential observations, and Variance Inflation Factors (VIF < 5) to ensure no significant multicollinearity (Supplementary Fig. 2). To visualize the dose-response relationship between log₁₀LAP and MASLD risk, RCS with four knots were utilized for the total population. To minimize the impact of extreme outliers and ensure model stability, the X-axis was truncated to the 1.5th and 98.5th percentiles of the log₁₀LAP distribution. Diagnostic performance was evaluated using the Area Under the Receiver Operating Characteristic curve (AUC). To facilitate clinical implementation, sex-specific optimal cut-off values were determined based on the maximum Youden index, with sensitivity, specificity, and predictive values (PPV/NPV) reported. The incremental value of log₁₀LAP over the baseline model was quantified using NRI and IDI. The clinical utility of the LAP-based diagnostic model was assessed via DCA, which quantified the net benefit across a range of threshold probabilities compared to “screen-all” and “screen-none” strategies. In our primary analysis, log₁₀LAP was evaluated as a continuous variable and further categorized into quartiles (Q1–Q4) to assess its association with MASLD. To verify the consistency and robustness of our findings, several sensitivity analyses were performed: Multi-granular grouping: log₁₀LAP was re-evaluated as tertiles and quintiles to confirm the stability of the observed trend.Subgroup evaluations: Consistency was tested across categories of age, sex, BMI, smoking status, hyperlipidemia, and vitamin D levels, with interaction analyses conducted to identify potential modifiers.Model refinement and bias control: We conducted sensitivity analyses by excluding BMI from the models to ensure the independent effect of log₁₀LAP. Additionally, data were trimmed at the 0.5th and 99.5th percentiles to minimize outlier influence, and E-values were calculated to quantify potential unmeasured confounding. For comparative analysis, the triglyceride–glucose (TyG) index was calculated as ln [fasting triglycerides (mg/dL) × fasting glucose (mg/dL) / 2 [52], and its incremental discriminative value for prevalent MASLD was evaluated in parallel with log₁₀LAPTo further assess the relationship between LAP and liver fibrosis risk, the fibrosis-4 (FIB-4) index [53] was calculated using established formulas. Correlation analysis between log₁₀LAP and FIB-4 was performed using Spearman’s rank correlation. Additionally, participants were stratified according to clinically relevant FIB-4 thresholds to compare log₁₀LAP levels across fibrosis risk categories.External validation was performed using data from the National Health and Nutrition Examination Survey (NHANES). Participants with type 2 diabetes and complete data required for LAP calculation were included.Baseline characteristics of the NHANES cohort were summarized, and the association between log₁₀LAP and MASLD was evaluated using multivariable logistic regression models consistent with those applied in the primary cohort. To ensure the robustness of the external validation, several sensitivity analyses were performed, including evaluating log₁₀LAP as tertiles and quintiles, as well as re-assessing the models after excluding outliers. Model discrimination, calibration, and clinical utility were further evaluated using AUC, calibration curves, and DCA, respectively.

Statistical analyses were performed using R Statistical Software (v4.2.3; R Foundation for Statistical Computing; http://www.R-project.org) and a Free Statistical Analysis Platform (v2.1.1; Beijing Free Clinical Medical Technology Co., Ltd.). Descriptive statistics were calculated for the entire cohort. A two-tailed test was used to determine statistical significance, which was set at P < 0.05 considered significant.

Results

Feature selection via Boruta algorithm

The Boruta algorithm identified six confirmed important features with importance scores significantly exceeding the shadow variables (shadowMax). Among these, log₁₀LAP exhibited the highest importance score, demonstrating the strongest discriminative association with prevalent MASLD. This was followed by BMI, ALT, VD, PLT, and WBC (Fig. 1) .To further clarify the incremental discriminative value of LAP relative to its individual components, we performed a comprehensive model comparison using ROC curves, NRI, and IDI. As illustrated in Supplementary Fig. 3, the addition of log₁₀LAP to the baseline model achieved the highest area under the curve (AUC: 0.7974), which was notably superior to the models incorporating WC (AUC: 0.7432) or TG (AUC: 0.7549) alone. Consistent with the ROC results, the addition of log₁₀LAP yielded a significantly greater improvement in risk reclassification (NRI: 0.5881, P < 0.001; IDI: 0.1076, P < 0.001) compared to the individual components (Supplementary Table 2).

Fig. 1.

Feature selection via Boruta algorithm in WMU cohorts. Boxplots illustrate the importance scores (Z-scores) of candidate predictors for MASLD. Green boxes represent confirmed important features, with log₁₀LAP, BMI, and ALT showing the highest importance.Blue boxes represent shadow attributes (min, mean, max) used as statistical baselines; yellow boxes are tentative, and red boxes are rejected variables

Fig. 2.

Subgroup analysis of the association between log₁₀LAP and MASLD risk. OR: odds ratio, CI confidence interval, log₁₀LAP, log-transformed lipid accumulation product. The forest plot displays the Odds Ratios (OR) and 95% CIs for the association between log₁₀LAP and MASLD across various subgroups. The overall adjusted OR is 1.83 (1.43–2.35). No significant interaction was found across all pre-specified subgroups (P for interaction > 0.05). The diamond markers represent the overall effect sizes, while the square markers represent subgroup-specific ORs. Models were adjusted for all variables included in Model 4

Characteristics of the population by log₁₀LAP

This study included 495 participants with an average age of 58.9 ± 14.8 years. Detailed missing data analysis is presented in Supplementary Table 3; all clinical variables exhibited high data integrity with missingness ratios below 5%, ranging from 0% to 4.44% (for Vitamin D). Of these, 275 (55.6%) were men, and the mean log₁₀LAP was 5.12 ± 1.15. Table 1 presents the characteristics of the participants across the log₁₀LAP quartiles. Higher log₁₀LAPscores were typically associated with younger age and higher BMI. Additionally, higher LAPlog was associated with hypertension and hyperlipidemia, as well as higher levels of ALT, AST, and uric acid. In contrast, higher log₁₀LAP showed a marginal association with lower vitamin D levels. No significant differences were observed across the log₁₀LAP quartiles regarding sex, smoking status, drinking status, HbA1C, Duration.of.diabetes or the use of statins and ezetimibe.

Table 1.

Baseline characteristics of research participants stratified by quartiles of log₁₀LAP^a (WMU)

Variables	Total	Q1(n = 124)	Q2(n = 124)	Q3(n = 124)	Q4(n = 123)	P
	(N = 495)	≤ 4.3	4.3-5.0	5.0-5.9	≥ 5.9
Age, Mean ± SD(year)	58.9 ± 14.8	58.8 ± 16.0	61.9 ± 12.7	60.5 ± 13.6	54.4 ± 15.7	< 0.001
Sex, n (%)						0.133
Men	275 (55.6)	77 (62.1)	71 (57.7)	59 (47.6)	68 (54.8)
Women	220 (44.4)	47 (37.9)	52 (42.3)	65 (52.4)	56 (45.2)
BMI, Mean ± SD(kg/m2)	24.2 ± 4.9	21.5 ± 4.2	23.5 ± 4.5	24.3 ± 4.4	27.2 ± 4.8	< 0.001
WC, Mean ± SD, cm	86.7 ± 10.0	77.7 ± 7.0	85.5 ± 6.9	89.8 ± 8.3	93.9 ± 9.8	< 0.001
Hypertension, n (%)						< 0.001
No	201 (40.6)	69 (55.6)	46 (37.4)	39 (31.5)	47 (37.9)
Yes	294 (59.4)	55 (44.4)	77 (62.6)	85 (68.5)	77 (62.1)
Hyperlipidemia, n (%)						< 0.001
No	180 (36.4)	63 (50.8)	59 (48)	40 (32.3)	18 (14.5)
Yes	315 (63.6)	61 (49.2)	64 (52)	84 (67.7)	106 (85.5)
Smoking, n (%)						0.195
No	358 (72.8)	87 (70.7)	91 (74)	98 (79)	82 (67.2)
Yes	134 (27.2)	36 (29.3)	32 (26)	26 (21)	40 (32.8)
Drinkc, n (%)						0.377
No	385 (78.3)	100 (81.3)	94 (76.4)	101 (81.5)	90 (73.8)
Yes	107 (21.7)	23 (18.7)	29 (23.6)	23 (18.5)	32 (26.2)
WBC, Mean ± SD (×109/L)	7.2 ± 2.7	6.9 ± 2.6	7.1 ± 3.2	7.3 ± 2.4	7.5 ± 2.5	0.385
PLT, Mean ± SD (×109/L)	223.8 ± 74.6	224.3 ± 90.2	218.9 ± 74.1	222.3 ± 59.1	229.7 ± 72.2	0.716
ALT, Median (IQR) (U/L)	19.0 (14.0, 29.0)	16.0 (12.0, 23.8)	18.0 (14.0, 24.5)	18.0 (13.0, 30.5)	21.5 (17.0, 38.5)	< 0.001
AST, Median(IQR) (U/L)	19.0 (16.0, 25.0)	18.0 (15.0, 23.0)	19.0 (16.0, 22.2)	20.0 (16.0, 26.0)	21.0 (16.0, 29.8)	0.004
Cr, Median (IQR) (umol/L)	68.0 (53.0, 85.0)	67.0 (51.2, 83.0)	69.5 (52.8, 89.5)	69.0 (56.0, 84.0)	66.0 (51.5, 82.5)	0.632
UA, Mean ± SD (umol/L)	347.2 ± 114.6	307.1 ± 107.1	330.8 ± 116.7	365.5 ± 107.4	385.1 ± 112.0	< 0.001
VD, Mean ± SD (ng/mL)	22.1 ± 8.6	23.6 ± 10.4	22.3 ± 8.6	21.9 ± 7.6	20.5 ± 7.4	0.051
TG, Median ( mmol/L)	1.4 (1.0, 2.2)	0.9 (0.7, 1.1)	1.1 (1.0, 1.3)	1.6 (1.3, 1.9)	3.2 (2.4, 4.7)	< 0.001
HbA1c, Mean ± SD (%)	9.2 ± 2.5	9.4 ± 2.7	9.4 ± 2.6	9.0 ± 2.4	9.1 ± 2.2	0.484
Duration.of.diabetes, Median (IQR) (Year)	10.0 (3.0, 15.0)	8.0 (3.8, 10.2)	10.0 (5.0, 16.2)	10.0 (3.0, 16.0)	7.0 (3.0, 16.0)	0.408
Arb, n (%)						0.162
No	335 (67.7)	93 (75)	84 (68.3)	77 (62.1)	81 (65.3)
Yes	160 (32.3)	31 (25)	39 (31.7)	47 (37.9)	43 (34.7)
Statins, n (%)						0.295
No	117 (23.6)	36 (29)	25 (20.3)	25 (20.2)	31 (25)
Yes	378 (76.4)	88 (71)	98 (79.7)	99 (79.8)	93 (75)
Ezetimibe, n (%)						0.104
No	423 (85.5)	106 (85.5)	109 (88.6)	98 (79)	110 (88.7)
Yes	72 (14.5)	18 (14.5)	14 (11.4)	26 (21)	14 (11.3)
MASLD, n (%)						< 0.001
No	170 (34.3)	73 (58.9)	58 (46.8)	26 (21)	13 (10.6)
Yes	325 (65.7)	51 (41.1)	66 (53.2)	98 (79)	110 (89.4)

ALT, alanine aminotransferase; AST, aspartate aminotransferase; BMI, body mass index; WC, waist circumference; Cr, creatinine; IQR, interquartile range; log₁₀LAP, log-transformed lipid accumulation product; PLT, platelet count; SD, standard deviation; UA, uric acid; VD, Vitamin D; WBC, white blood cell count. TG, triglycerides; MASLD, metabolic dysfunction-associated steatotic liver disease;

Data are presented as mean ± SD for normally distributed continuous variables, median (IQR) for skewed continuous variables, or number (percentage) for categorical variables

One-way ANOVA for normally distributed continuous variables; Kruskal-Wallis test for skewed continuous variables; Chi-squared test or Fisher’s exact test for categorical variables

^aCategorized into 4 levels based on log₁₀LAP quartiles: Q1 (< 4.3), Q2 (4.3–5.0), Q3 (5.0–5.9), and Q4 (> 5.9)

Association between log₁₀LAP and the risk of MASLD

Following multivariate adjustments, log₁₀LAP was significantly positively associated with the risk of MASLD. In the overall population, a 1 unit increase in log₁₀LAP resulted in an 83% increase in the risk of MASLD (OR 1.83, 95% CI 1.43–2.35, P < 0.001). Stratification by log₁₀LAP quartile showed that the highest log₁₀LAP subgroup (Q4,≥5.9) experienced a 4.33-fold increase in MASLD risk (OR 5.33, 95% CI 2.38–11.91, P < 0.001) compared with the lowest log₁₀LAP subgroup (Q1,≤4.3). These associations were consistent in separate analyses of men and women (Table 2). In men, an increase in log₁₀LAP was significantly associated with elevated risk of MASLD (OR 1.57, 95% CI 1.09–2.25, P = 0.015), and the Q4 subgroup showed a 2.23-fold higher risk compared to the Q1 subgroup (OR 3.21, 95% CI 1.12–9.14, P = 0.027). Similarly, in women, a 1 unit increase in log₁₀LAP was associated with an increased risk of MASLD (OR 1.82, 95% CI 1.23–2.69, P = 0.003), and the Q4 subgroup experienced a 8.9-fold increase in risk compared with the Q1 subgroup (OR 9.9, 95% CI 2.61–37.54, P < 0.001).

Table 2.

Association between log₁₀LAP and the risk of MASLD in the WMU cohort

	Categories	Unadjust		Model 1		Model 2		Model3		Model 4
		OR (95%CI)	P value	OR (95%CI)	P value	OR (95%CI)	P value	OR (95%CI)	P value	OR (95%CI)	P value
ALL	log10LAP	2.27 (1.85 ~ 2.79)	< 0.001	2.28 (1.86 ~ 2.80)	< 0.001	1.90 (1.52 ~ 2.36)	< 0.001	1.84 (1.46 ~ 2.31)	< 0.001	1.83 (1.43 ~ 2.35)	< 0.001
	log10LAP Q4
	Q1 (≤ 4.3)	1(Ref)		1(Ref)		1(Ref)		1(Ref)		1(Ref)
	Q2 (4.5-5.0)	1.63 (0.99 ~ 2.69)	0.057	1.66 (1.00 ~ 2.75)	0.049	1.33 (0.78 ~ 2.25)	0.295	1.32 (0.77 ~ 2.27)	0.319	1.09 (0.59 ~ 2.01)	0.777
	Q3 (5.0-5.9)	5.40 (3.08 ~ 9.46)	< 0.001	5.55 (3.15 ~ 9.79)	< 0.001	4.03 (2.22 ~ 7.33)	< 0.001	3.90 (2.09 ~ 7.29)	< 0.001	4.59 (2.28 ~ 9.25)	< 0.001
	Q4 (≥ 5.9)	12.11 (6.15 ~ 23.84)	< 0.001	12.07 (6.11 ~ 23.84)	< 0.001	6.33 (3.01 ~ 13.31)	< 0.001	5.84 (2.69 ~ 12.66)	< 0.001	5.33 (2.38 ~ 11.91)	< 0.001
	P for trend		< 0.001		< 0.001		< 0.001		< 0.001		< 0.001
Men	log10LAP	2.33 (1.77 ~ 3.05)	< 0.001	2.27 (1.73 ~ 2.97)	< 0.001	2.19 (1.65 ~ 2.9)	< 0.001	1.61 (1.17 ~ 2.23)	0.004	1.57 (1.09 ~ 2.25)	0.015
	log10LAP Q4
	Q1 (≤ 4.29)	1(Ref)		1(Ref)		1(Ref)		1(Ref)		1(Ref)
	Q2 (4.29–4.98)	2.51 (1.26 ~ 5.01)	0.009	2.54 (1.27 ~ 5.08)	0.008	2.37 (1.18 ~ 4.79)	0.016	2.11 (0.97 ~ 4.55)	0.058	1.01 (0.43 ~ 2.37)	0.987
	Q3 (4.98–5.93)	6.91 (3.22 ~ 14.86)	< 0.001	6.9 (3.20 ~ 14.87)	< 0.001	6.29 (2.88 ~ 13.76)	< 0.001	4.41 (1.82 ~ 10.65)	0.001	5.13 (1.61 ~ 16.37)	0.006
	Q4 (≥ 5.93)	11.73 (4.99 ~ 27.6)	< 0.001	10.39 (4.35 ~ 24.8)	< 0.001	9.04 (3.65 ~ 22.37)	< 0.001	3.46 (1.24 ~ 9.65)	0.018	3.21 (1.12 ~ 9.14)	0.027
	P for trend		< 0.001		< 0.001		< 0.001		0.002		0.025
Women	log10LAP	2.24 (1.64 ~ 3.06)	< 0.001	2.25 (1.64 ~ 3.07)	< 0.001	2.03 (1.47 ~ 2.79)	< 0.001	1.86 (1.29 ~ 2.69)	0.001	1.82 (1.23 ~ 2.69)	0.003
	log10LAP Q4
	Q1 (≤ 4.46)	1(Ref)		1(Ref)		1(Ref)		1(Ref)		1(Ref)
	Q2 (4.46–5.17)	3.07 (1.41 ~ 6.68)	0.005	3.10 (1.38 ~ 6.97)	0.006	3.00 (1.3 ~ 6.96)	0.01	3.02 (1.25 ~ 7.28)	0.014	1.28 (0.43 ~ 3.76)	0.658
	Q3 (5.17–5.99)	6.12 (2.66 ~ 14.1)	< 0.001	6.16 (2.64 ~ 14.35)	< 0.001	5.32 (2.22 ~ 12.75)	< 0.001	5.31 (2.07 ~ 13.62)	0.001	4.96 (1.68 ~ 14.63)	0.004
	Q4 (≥ 5.99)	19.33 (6.61 ~ 56.53)	< 0.001	19.38 (6.61 ~ 56.8)	< 0.001	15.39 (5.13 ~ 46.13)	< 0.001	13.28 (3.88 ~ 45.43)	< 0.001	9.90 (2.61 ~ 37.54)	0.001
	P for trend		< 0.001		< 0.001		< 0.001		< 0.001		< 0.001

OR, odds ratio; CI confidence interval

log₁₀LAP, log-transformed lipid accumulation product. BMI Body-mass index, WBC White blood cell count, PLT Platelet count, ALT Alanine aminotransferase, AST Aspartate aminotransferase, Cr Serum creatinine, UA Uric acid, VD Vitamin D, ARB Angiotensin-receptor blocker, Statins HMG-CoA reductase inhibitors, Ezetimibe Ezetimibe

Unadjust: unadjusted

Model 1: adjusted for age and sex

Model 2: adjusted for variables in Model 1 plus smoking, alcohol consumption, BMI, hypertension, and hyperlipidemia

Model 3: adjusted for variables in Model 2 plus WBC, PLT, ALT, AST, Cr, VD, and UA

Model 4: adjusted for variables in Model 3 plus ARB, statins, and ezetimibe, HbA1C, Duration.of.diabetes

Restricted cubic spline regression model

RCS analysis with four knots was performed to characterize the dose-response relationship between log₁₀LAP and the probability of MASLD. To ensure model stability and minimize the influence of extreme outliers, the X-axis was restricted to the 1.5th and 98.5th percentiles of the log₁₀LAP distribution. In the total population, a significant linear association was observed between log₁₀LAP levels and MASLD risk (P for overall association < 0.001; Supplementary Fig. 4). The test for non-linearity was non-significant (P for non-linearity = 0.085), with the probability of MASLD increasing monotonically and steadily across the spectrum of log₁₀LAP levels.

Subgroup analysis

Subgroup analysis was performed across various subgroups to assess potential differences in the association between log₁₀LAP and MASLD risk. No significant interactions were found in subgroups categorized according to age, sex, smoking status, BMI, hypertension, or vitamin D levels (Figure. 2).

Incremental discriminative value of log₁₀LAP compared to BMI

The discriminative performance of log₁₀LAP and BMI for prevalent MASLD was evaluated using ROC curve analysis (Fig. 3), along with NRI and IDI metrics. Compared to the Baseline Risk Model (model 4 excluding BMI), which yielded an AUC of 0.6943, the addition of BMI increased the AUC to 0.7729. The incorporation of log₁₀LAP into the baseline model resulted in the highest discriminative performance, yielding an AUC of 0.7974. Further quantitative assessment of model improvement showed that the addition of log₁₀LAP to the baseline model resulted in a significant NRI of 0.5881 (95% CI 0.4111–0.7651, P < 0.001) and an IDI of 0.1076 (95% CI 0.0797–0.1356, P < 0.001). For comparison, the addition of BMI yielded an NRI of 0.6004 (95% CI 0.4228–0.7779, P < 0.001) and an IDI of 0.0735 (95% CI 0.0513–0.0957, P < 0.001). (Table 3)

Fig. 3.

Comparison of ROC curves for MASLD prediction baseline tisk model: model 4 excluding BMI. The receiver operating characteristic (ROC) curves illustrate the predictive performance of different models for MASLD. Baseline risk model (red line): Includes age, sex, smoking, drinking, hypertension, and hyperlipidemia (AUC = 0.6943). Adding BMI (blue line): Represents the baseline risk model with the addition of Body Mass Index (AUC = 0.7729). Adding log₁₀LAP (green line): Represents the baseline risk model with the addition of log-transformed lipid accumulation product (AUC = 0.7974). BMI Body-mass index, LAP Lipid accumulation product. log₁₀LAP, log-transformed lipid accumulation product

Table 3.

Reclassification and Discrimination Performance of log₁₀LAP and BMI

Model	NRI (95%Cl)	P	IDI (95%Cl)	P
Baseline tisk model	Ref.	Ref.	Ref.	Ref.
+BMI	0.6004 [0.4228–0.7779]	< 0.001	0.0735 [0.0513–0.0957]	< 0.001
+log10LAP	0.5881 [0.4111–0.7651]	< 0.001	0.1076 [0.0797–0.1356]	< 0.001

Baseline tisk model: model 4 excluding BMI

BMI Body-mass index, LAP Lipid accumulation product, NRI Net reclassification improvement, IDI Integrated discrimination improvement, CI Confidence interval.log₁₀LAP, log-transformed lipid accumulation product

Sensitivity analysis

Several sensitivity analyses were conducted to verify the robustness of our findings. First, the findings remained consistent when participants were categorized according to log₁₀LAP tertiles (Supplementary Table 4) and log₁₀LAP quintiles (Supplementary Table 5). Second, after excluding extreme values (the top 0.5% and bottom 0.5% of log₁₀LAP), the results remained robust, indicating that the association was not driven by outliers (Supplementary Table 6). Third, sensitivity analysis excluding BMI from the multivariable model yielded consistent results, confirming the independent discriminative value of log₁₀LAP (Supplementary Table 7). Fourth to assess the potential impact of unmeasured confounding, we calculated the E-value (Supplementary Fig. 5). The primary results were unlikely to be overturned by unmeasured confounders, further confirming the stability of the observed associations.

Evaluation of clinical utility and subgroup consistency

Based on the multivariable logistic regression model, the optimal probability threshold was established at 0.6611. This corresponds to sex-specific raw LAP cut-off values of 20.4 for men and 27.0 for women. Due to the strictly monotonic nature of the log₁₀ transformation, the diagnostic performance (sensitivity and specificity) of the clinical cut-offs in raw values is mathematically identical to their counterparts in the log-transformed scale (Supplementary Table 8). At these thresholds, the model demonstrated a high Positive Predictive Value (PPV) of 85.93% (Supplementary Table 9). The clinical utility of the model was evaluated via DCA. As illustrated in Supplementary Fig. 6, the LAP-based model (red line) yielded a consistently higher net benefit compared to both the ‘screen-all’ and ‘screen-none’ strategies across a wide threshold probability range of approximately 0.15 to 0.90. Notably, the LAP model also demonstrated increased net benefit over the BMI-augmented model (blue line) across the same range, suggesting that using LAP for MASLD screening provides greater incremental value in clinical decision-making than BMI. While the net benefit of the ‘screen-all’ strategy declined sharply and reached zero at a threshold of approximately 0.65, the LAP model maintained substantial clinical utility even at higher risk thresholds. Subgroup analysis (Supplementary Table 10) further confirmed the diagnostic stability of the model across various clinical strata. The AUC remained robust regardless of sex (men: 0.790; women: 0.774). Notably, the highest discriminative power was observed in the younger subgroup (Age < 55 years; AUC 0.855, 95% CI 0.793–0.917). Additionally, the model sustained reliable performance in patients without obesity (BMI < 24 kg/m²; AUC 0.711) and patients with obesity (BMI ≥ 24 kg/m²; AUC 0.747).

Comparison of LAP with the TyG index

To evaluate the comparative diagnostic performance of LAP, we compared it against the well-established TyG index. The addition of log₁₀LAP to the baseline model significantly improved the AUC to 0.7974, which was higher than that achieved by adding the TyG index (0.7417) (Supplementary Fig. 7). Furthermore, log₁₀LAP demonstrated a significant incremental value over the baseline model, with a NRI of 0.5881 (P < 0.001) and an IDI of 0.1076 (P < 0.001). These improvements were notably greater than those observed for the TyG index (NRI: 0.4505; IDI: 0.0511), (Supplementary Table 11) suggesting that LAP is a more effective marker for identifying MASLD in patients with T2DM.

Association between LAP and FIB-4

To address the clinical concern of liver fibrosis risk in the T2DM population, we further evaluated the relationship between LAP and the FIB-4 index. Correlation analysis demonstrated that log₁₀LAP was weakly and negatively associated with the FIB-4 index (rho = -0.111, P = 0.013), suggesting that LAP provides independent metabolic information distinct from age-driven fibrosis scores (Supplementary Fig. 8A).

Furthermore, we stratified the population by the clinical FIB-4 threshold (1.3). Notably, participants in the low-fibrosis-risk group (FIB-4 < 1.3, n = 263) exhibited significantly higher log₁₀LAP levels compared to those in the intermediate/high-risk group (FIB-4 ≥ 1.3, n = 232) (5.24 ± 1.45 vs. 4.98 ± 1.10, P = 0.028; Supplementary Fig. 8B).

Robustness and external validation of LAP

To assess the mathematical reliability and clinical utility of the identified LAP thresholds, a comprehensive validation framework was implemented. First, internal stability was evaluated via 1,000 bootstrap resamples (Supplementary Table 12). The optimism-corrected AUCs were 0.7574 for men and 0.7626 for women, with near-zero optimism values (− 0.0008 and − 0.0006, respectively), indicating minimal overfitting. The sex-specific cut-offs (20.4 for men and 27.0 for women) remained stable during resampling, with their diagnostic performance (sensitivity and specificity) demonstrating high reliability within the T2DM inpatient population.Second, external generalizability was examined in an independent cohort from the NHANES 2017–2018 dataset (N = 630) (Supplementary Fig. 9). The baseline characteristics of this cohort, stratified by log₁₀LAP quartiles, are detailed in Supplementary Table 13. Multivariate logistic regression analysis (Supplementary Table 14) confirmed that log₁₀LAP was independently and strongly associated with MASLD risk. Notably, participants in the highest quartile (Q4) exhibited a 10.69-fold higher risk of MASLD compared to those in Q1 (OR = 10.69, 95% CI 5.80–19.68; P < 0.001), even after adjusting for comprehensive metabolic and medication-related confounders.Third, the incremental discriminative value of LAP was compared across three nested assessment models (Supplementary Fig. 10). In the ROC analysis, adding log₁₀LAP to the baseline model (Model 3) achieved a significantly higher AUC of 0.8149, outperforming both the BMI-augmented model (Model 2, AUC = 0.7951) and the baseline model (Model 1, AUC = 0.7082) (Supplementary Fig. 10A). Calibration curves (Supplementary Fig. 10B) demonstrated excellent agreement between predicted and observed MASLD prevalence for the LAP-augmented model. Furthermore, DCA (Supplementary Fig. 10C) illustrated that the inclusion of log₁₀LAP provided the highest clinical net benefit across a broad range of threshold probabilities. Finally, a series of sensitivity analyses were performed to ensure the robustness of these findings. The potent association between log₁₀LAP and MASLD remained consistent and statistically significant when log₁₀LAP was analyzed by tertiles (Supplementary Table 15), quintiles (Supplementary Table 16), or after the exclusion of extreme values (0.5th–99.5th percentiles) (Supplementary Table 17), reinforcing the stability of the results across different population contexts.

Discussion

This hospital-based cohort study provides several clinical insights into the management of MASLD in patients with T2DM. First, an elevated log₁₀LAP was independently correlated with a higher risk of MASLD, an association that remained robust across multiple adjustment models, including those accounting for metabolic health and medication use. Importantly, our results demonstrated that LAP is not merely a proxy for WC and TG, but offers a superior synergistic effect, as evidenced by the significantly higher NRI and IDI compared to its individual components. This suggests that the mathematical integration of anatomical and biochemical markers in LAP captures a more comprehensive pathological state than either marker alone. Second, RCS analysis revealed a consistent linear dose-response relationship between log₁₀LAP and MASLD risk in the total population. Notably, despite this linear trend, we identified distinct sex-specific diagnostic thresholds—20.4 for men and 27.0 for women—which yielded high positive predictive values and significantly improved risk reclassification. Although age differed across LAP quartiles, the association between log₁₀LAP and MASLD remained robust after multivariable adjustment and age-stratified analyses, suggesting that the observed relationship was unlikely to be solely driven by age differences. These findings imply that visceral lipid accumulation, as quantified by LAP, serves as a continuous driver of hepatic steatosis, but requires tailored clinical action levels in diabetic populations.

Previous research on MASLD risk factors in T2DM populations has left several critical issues unresolved. First, although existing studies have established associations between static anthropometric measures (e.g., BMI) or individual lipid parameters, they have often failed to capture the synergistic effect of anatomical fat distribution and biochemical dysregulation that LAP uniquely integrates [54, 55]. Second, methodological limitations, such as ignoring extreme outlier influence or lacking clinical utility assessments (DCA), have obscured the practical application of these markers [55, 56] .This study contributes to the advancement of this field through three key methodological strengths: (1) By employing a systematic feature-selection framework (Boruta algorithm), log₁₀LAP was objectively prioritized as a core predictor among a wide array of clinical variables, outperforming traditional metrics. (2) A RCS model was employed to establish a stable linear dose-response relationship, with 1.5th and 98.5th percentile trimming to ensure results were not driven by noise. (3) Beyond statistical significance, DCA was implemented to quantify the clinical net benefit, providing actionable sex-specific thresholds (20.4 for men; 27.0 for women) that are substantially higher than those reported in the general population. While the high prevalence of MASLD (66.7%) in our T2DM cohort might suggest a ‘treat-all’ approach, our DCA results indicate that the LAP-augmented model offers a superior net benefit. This suggests that LAP can serve as an efficient triage instrument to prioritize high-risk patients for advanced evaluations—such as transient elastography or fibrosis scoring—while avoiding unnecessary intensive follow-up for the remaining one-third of the population, thereby optimizing resource allocation in clinical practice.

A critical concern raised in clinical practice is whether identifying MASLD per se adds value in T2DM patients, given their high pre-test probability of steatosis. Our study addresses this by exploring the relationship between LAP and the FIB-4 index, a marker of advanced fibrosis. We observed a weak negative correlation (ρ = -0.111, P = 0.013), indicating that LAP and FIB-4 capture distinct clinical dimensions. Interestingly, patients in the low-fibrosis-risk group (FIB-4 < 1.3) exhibited significantly higher LAP levels than those at higher risk (5.24 ± 1.45 vs. 4.98 ± 1.10, P = 0.028). This ‘metabolic-structural mismatch’ suggests that LAP is a more sensitive marker for the ‘metabolic driving force’ (visceral lipid accumulation) during the early stages of the disease spectrum, whereas FIB-4 reflects cumulative architectural damage often more prominent in older patients. Thus, LAP serves as a ‘metabolic sentinel’ that identifies high-risk individuals during the golden window for intervention—before significant fibrosis manifests.The combination of log₁₀LAP with fibrosis-specific markers, such as FIB-4, may provide a more comprehensive assessment for identifying patients at high risk of prevalent MASLD. This synergy warrants further prospective investigation to evaluate its utility in the clinical management of patients with type 2 diabetes.It is worth noting that while log₁₀LAP exhibited a numerically higher AUC compared to BMI (0.797 vs. 0.773), the absolute difference in discriminative power appears incremental. However, the true clinical value of LAP lies not merely in a marginal increase in AUC, but in its superior ability to reclassify risk, as evidenced by the significant improvements in NRI and IDI. While BMI remains a fundamental and robust marker of general adiposity, LAP integrates biochemical dysregulation (triglycerides) with anatomical fat distribution (waist circumference), capturing the ‘metabolic quality’ of adipose tissue rather than just its quantity. This synergy is particularly relevant for identifying MASLD in patients with a ‘normal’ BMI but high visceral lipid burden—a common clinical phenotype in Asian populations that BMI alone often fails to risk-stratify.

The association between log₁₀LAP and MASLD is potentially elucidated through visceral lipotoxicity and systemic inflammation. Elevated LAP indicates an over-accumulation of visceral adipose tissue, a major source of pro-inflammatory cytokines and free fatty acids (FFAs). These FFAs are delivered directly to the liver via the portal vein, inducing oxidative stress and promoting de novo lipogenesis [57, 58]. In our study, the linear relationship observed across the population suggests that even incremental increases in visceral adiposity contribute to hepatic steatosis without a discernible “safe” lower limit in the T2DM context. The observed sex-specific thresholds (20.4 for men and 27.0 for women) likely reflect fundamental sexual dimorphism in lipid partitioning and hormonal modulation. First, men typically exhibit a ‘centralized’ adiposity pattern, where excess energy is preferentially stored as visceral adipose tissue (VAT). VAT is highly sensitive to lipolytic stimuli, releasing a high flux of FFAs directly into the portal circulation. In contrast, women tend to have a ‘peripheral’ distribution, with greater storage in subcutaneous gluteal-femoral depots, which act as a metabolic sink to protect the liver from lipid overflow. Second, estrogen exerts a multi-faceted protective effect on the liver by upregulating genes involved in mitochondrial β-oxidation and downregulating de novo lipogenesis via estrogen receptor α (ER α). This hormonal shield implies that women may tolerate a higher cumulative lipid burden, as captured by LAP, before manifesting clinical MASLD. Consequently, these distinct physiological frameworks necessitate the use of tailored thresholds to optimize diagnostic precision in both sexes.

Furthermore, the sex-specific thresholds we identified may reflect biological variations in fat distribution and hormonal influences on lipid metabolism between men and women. The observed sex-specific thresholds may be explained by underlying biological differences in fat distribution and lipid metabolism. Compared with women, men tend to accumulate more visceral adipose tissue, which is more metabolically active and strongly associated with hepatic fat deposition. In contrast, women generally have a higher proportion of subcutaneous fat, which may exert a relatively protective metabolic effect. In addition, sex hormones, particularly estrogen, play an important role in regulating lipid metabolism, insulin sensitivity, and hepatic fat accumulation. These physiological differences may contribute to distinct LAP thresholds required to reflect MASLD risk in men and women. In this study, we addressed the reliability of the proposed LAP thresholds through a multi-layered validation strategy. While the internal bootstrap validation confirms the mathematical stability of the 20.4 and 27.0 cut-offs within our clinical population, the NHANES analysis provides crucial external support. Although ethnic and diagnostic differences preclude a direct threshold transfer, this cross-population consistency underscores the underlying robustness of LAP as a biological marker for MASLD. Consequently, log₁₀LAP represents a robust, non-invasive indicator that accounts for both anatomical and biochemical factors, offering enhanced discriminative capacity for identifying prevalent MASLD in high-risk diabetic populations within hospital settings.

Strengths and limitations

Our study possesses several key strengths, particularly its methodological methodological strengths that significantly advance MASLD screening strategies in diabetic populations.

(1) Prioritized Biomarker Identification: By utilizing the Boruta algorithm, log₁₀LAP was identified as a prioritized risk factor among a wide array of clinical variables. By utilizing simple anthropometric and biochemical measures, log₁₀LAP offers a practical approach for identifying prevalent MASLD during routine clinical assessments. This suggests its potential as a feasible alternative to more resource-intensive diagnostic methods in settings where advanced imaging may be limited.First, as a retrospective study, nearly half of the initially evaluated patients were excluded, primarily due to missing waist circumference data required for LAP calculation. While this introduces a potential for selection bias, our rigorous comparison between included (n = 495) and excluded (n = 216) participants revealed no statistically significant differences in age, sex, BMI, or MASLD prevalence (P > 0.05). This suggests that the missingness occurred at random and our analyzed cohort remains highly representative of the T2DM population in a tertiary hospital setting. Nevertheless, the single-center nature of this study and the relatively high prevalence of MASLD (65.7%) in our hospitalized cohort mean that the proposed thresholds (20.4 for men and 27.0 for women) should be interpreted as potential reference values rather than universal standards. (2) Clinically Actionable Thresholds: Although RCS analysis revealed a consistent linear dose-response relationship in the total population, we successfully established sex-specific diagnostic thresholds (20.4 for men and 27.0 for women). These thresholds, validated by high positive predictive values and DCA, provide clinicians with actionable targets to identify high-risk individuals and intervene before advanced liver injury occurs. (3) Robustness and Population Specificity: Focusing on hospitalized patients with T2DM addresses a critical gap, as this high-risk population exhibits complex metabolic profiles requiring more nuanced markers than the general public. The reliability of our findings is further reinforced by rigorous bias mitigation, including trimming of extreme values (0.5th and 99.5th percentiles) and sensitivity analyses across age and BMI strata. The reliability of our findings is supported by an external validation using the NHANES cohort, a geographically and ethnically distinct population. This supplementary analysis suggests that the association between log₁₀LAP and prevalent MASLD remains consistent across different clinical settings and ethnicities, providing preliminary evidence for the robustness of log₁₀LAP as a stable indicator for identifying MASLD in diverse populations.(4) while we evaluated LAP alongside BMI, TyG, and FIB-4, data on other specialized tools such as the NAFLD Fibrosis Score (NFS) or transient elastography were not available in this cohort. Future studies incorporating these modalities would further refine the integration of LAP into the broader diagnostic landscape of MASLD.

This study also has some limitations. First, as a retrospective single-center study involving Chinese T2DM inpatients, the direct extrapolation of these findings to the general population requires caution. Notably, the prevalence of MASLD in our hospitalized cohort was 65.7% (325/495), which is substantially higher than in the general public. This selection bias, inherent to tertiary hospital settings, may influence the observed discriminative performance of log₁₀LAP. However, characterizing such a high-risk clinical cohort enhances the relevance of our results to settings where precision screening is most urgently needed. Second, despite multivariable adjustments and E-value analysis, residual confounding from unmeasured variables—such as physical activity, dietary patterns, or genetic factors—cannot be entirely excluded. Third, the cross-sectional nature of the data precludes the establishment of definitive causality. Furthermore, while abdominal ultrasonography is a validated tool for large-scale screening, its lower sensitivity for mild steatosis compared to MRI-PDFF may lead to an underestimation of early-stage MASLD. Future prospective, multi-ethnic longitudinal studies incorporating advanced imaging are warranted to further validate these findings. Fourthly, the optimal sex-specific LAP cut-off values identified in this study (20.4 for males and 27.0 for females) were derived from a single-center retrospective T2DM cohort. While internal validation using bootstrapping and external comparison with the NHANES database demonstrated relative stability, these thresholds should be regarded as preliminary reference values rather than universal diagnostic standards. Due to variations in dietary habits, metabolic backgrounds, and ethnicity, these cut-offs may not be directly applicable to all populations. Therefore, further validation in large-scale, prospective, multi-center studies is essential to refine these thresholds and confirm their clinical utility in broader, diverse clinical practices.

Despite these limitations, this study provides methodologically rigorous evidence that log₁₀LAP is a potent and reliable predictor of MASLD risk, supporting its integration into routine metabolic screening protocols for high-risk diabetic populations.

Conclusions

This retrospective clinical study suggests that log₁₀LAP is a clinically accessible indicator associated with prevalent MASLD among hospitalized patients with type 2 diabetes. Our findings reveal a significant linear association between log₁₀LAP and MASLD, demonstrating that the likelihood of MASLD increases across the metabolic spectrum within this selected cohort. Despite this consistent trend, the implementation of sex-specific reference thresholds (20.4 for men and 27.0 for women) may assist in risk reclassification and provide incremental clinical net benefit, as validated by decision curve analysis. By capturing the synergistic burden of visceral adiposity and lipid dysregulation, log₁₀LAP represents a potential non-invasive adjunct for identifying individuals at higher risk of MASLD, providing a preliminary foundation for future prospective studies to refine screening and intervention strategies in diabetic populations.

Electronic Supplementary Material

Below is the link to the electronic supplementary material.

Abbreviations

ALT

Alanine aminotransferase

ARB

Angiotensin receptor blocker

AST

Aspartate aminotransferase

AUC

Area under the receiver operating characteristic curve

BMI

Body mass index

CI

Confidence interval

Cr

Creatinine

FBG

Fasting blood glucose

FIB-4 index

Fibrosis-4 index

IDI

Integrated discrimination improvement

IQR

Interquartile range

LAP

Lipid accumulation product

log₁₀LAP

Log-transformed lipid accumulation product

MASLD

Metabolic dysfunction-associated steatotic liver disease

NRI

Net reclassification improvement

OR

Odds ratio

PLT

Platelet count

RCS

Restricted cubic spline

SD

Standard deviation

T2DM

Type 2 diabetes mellitus

TyG index

Triglyceride-glucose index

UA

Uric acid

VD

Vitamin D

WBC

White blood cell count

Author contributions

XXQ: data collection, data analysis, manuscript writing. GHL, XMY, JLH, QZC: data analysis, manuscript editing.HQT, JX, MHJ: data collection. WJ, QPX: project development, manuscript editing. All authors have read and approved this manuscript.

Funding

No funding was received for conducting this study.

Data availability

The data that support the findings of this study were abstracted from the electronic medical record system of the Second Affiliated Hospital and Yuying Children’s Hospital of Wenzhou Medical University. Given the sensitive nature of clinical patient information and institutional privacy policies, these data are not publicly available. However, the datasets used and analyzed during the current study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

The study was conducted in accordance with the Declaration of Helsinki and was approved by the Ethics Committee of the Second Affiliated Hospital and Yuying Children’s Hospital of Wenzhou Medical University (No. 2025-K-270-01). Given the retrospective nature of the study and the use of de-identified clinical data, the requirement for written informed consent was waived by the Ethics Committee.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.