Development and external validation of an interpretable machine learning model for identifying prevalent metabolic dysfunction-associated steatotic liver disease in patients with early-onset type 2 diabetes

Objective Patients with early-onset type 2 diabetes mellitus (EOT2D) are at high risk of metabolic dysfunction-associated steatotic liver disease (MASLD), yet interpretable tools for estimating prevalent MASLD in this population are lacking. We developed and externally validated an interpretable machine learning model using routinely available clinical variables to prompt consideration of liver ultrasonography at the index visit. Methods We included 1,322 outpatients and inpatients with EOT2D (907 [68.6%] with prevalent MASLD), divided 7:3 into training (n = 925) and internal test (n = 397) sets. After multiple imputation by chained equations (m = 5), 68 candidate variables underwent least absolute shrinkage and selection operator regression, Boruta selection, and collinearity screening (variance inflation factor < 5). Eight machine learning models were tuned by 10-fold cross-validation. The final model was selected on the basis of internal discrimination and calibration, interpreted with Shapley additive explanations (SHAP), and externally validated in 515 patients (68.0% with MASLD) from four affiliated community hospitals in the same region. Results Twelve variables were retained. The extreme gradient boosting (XGBoost) model matched the leading algorithms in discrimination and maintained stable calibration. It achieved an AUROC of 0.773 (95% CI 0.721–0.824; AUPRC 0.864) in the internal test set and 0.753 (95% CI 0.708–0.797; AUPRC 0.852) in the external cohort, with calibration slopes of 1.03 and 0.93 and Brier scores of 0.169 and 0.180. SHAP identified total protein, globulin, albumin, and indirect bilirubin as the leading contributors. Complete-case analysis (AUROC 0.751), five random splits (0.764 ± 0.023), and nested cross-validation (0.755) were consistent with the main analysis. A free web calculator provides individualised probabilities. Conclusions An XGBoost model based on 12 routinely available variables estimated prevalent MASLD in EOT2D with moderate discrimination and good calibration. External validation was institutional, geographic, and temporal but confined to one region. The model is intended to prompt consideration of liver ultrasonography during routine diabetes care and to complement ultrasound and clinical judgment. Prospective implementation studies are required before clinical adoption.

Authors

Institutions

Publication Details

Journal
Frontiers in Endocrinology
Published
2026-11-13
DOI
https://doi.org/10.3389/fendo.2026.1998347
Primary Topic
Liver Disease Diagnosis and Treatment
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Development and external validation of an interpretable machine learning model for identifying prevalent metabolic dysfunction-associated steatotic liver disease in patients with early-onset type 2 diabetes

张雪娟, Shutong Du, Hongjian Jia, Yan Wang et al.
6 links
Frontiers in Endocrinology
Liver Disease Diagnosis and Treatment
article

Development and external validation of an interpretable machine learning model for identifying prevalent metabolic dysfunction-associated steatotic liver disease in patients with early-onset type 2 diabetes

张雪娟, Shutong Du, Hongjian Jia, Yan Wang, Tong Chen, Yunzhao Sun
article en

Abstract

Objective Patients with early-onset type 2 diabetes mellitus (EOT2D) are at high risk of metabolic dysfunction-associated steatotic liver disease (MASLD), yet interpretable tools for estimating prevalent MASLD in this population are lacking. We developed and externally validated an interpretable machine learning model using routinely available clinical variables to prompt consideration of liver ultrasonography at the index visit. Methods We included 1,322 outpatients and inpatients with EOT2D (907 [68.6%] with prevalent MASLD), divided 7:3 into training (n = 925) and internal test (n = 397) sets. After multiple imputation by chained equations (m = 5), 68 candidate variables underwent least absolute shrinkage and selection operator regression, Boruta selection, and collinearity screening (variance inflation factor < 5). Eight machine learning models were tuned by 10-fold cross-validation. The final model was selected on the basis of internal discrimination and calibration, interpreted with Shapley additive explanations (SHAP), and externally validated in 515 patients (68.0% with MASLD) from four affiliated community hospitals in the same region. Results Twelve variables were retained. The extreme gradient boosting (XGBoost) model matched the leading algorithms in discrimination and maintained stable calibration. It achieved an AUROC of 0.773 (95% CI 0.721–0.824; AUPRC 0.864) in the internal test set and 0.753 (95% CI 0.708–0.797; AUPRC 0.852) in the external cohort, with calibration slopes of 1.03 and 0.93 and Brier scores of 0.169 and 0.180. SHAP identified total protein, globulin, albumin, and indirect bilirubin as the leading contributors. Complete-case analysis (AUROC 0.751), five random splits (0.764 ± 0.023), and nested cross-validation (0.755) were consistent with the main analysis. A free web calculator provides individualised probabilities. Conclusions An XGBoost model based on 12 routinely available variables estimated prevalent MASLD in EOT2D with moderate discrimination and good calibration. External validation was institutional, geographic, and temporal but confined to one region. The model is intended to prompt consideration of liver ultrasonography during routine diabetes care and to complement ultrasound and clinical judgment. Prospective implementation studies are required before clinical adoption.

Frontiers in EndocrinologyVol. 17
Affiliated Hospital of Qingdao University (CN)
Good health and well-being
Liver Disease Diagnosis and Treatment
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon