Answers are grounded in this paper and cite their sources.
Background and objective Malnutrition occurs in 30%–80% of patients with cirrhosis, increasing mortality, hospital stay, and disease burden. Early identification is crucial, but standardized and practical tools remain limited. This study aimed to develop and validate an interpretable machine learning (ML) model for assessing malnutrition in cirrhosis. Methods Patients with cirrhosis hospitalized at the Affiliated Hospital of Zunyi Medical University from October 2020 to November 2023 formed the derivation cohort ( n = 405), while those from August to December 2025 were prospectively enrolled as the temporal validation cohort ( n = 163). Malnutrition was defined according to the GLIM criteria. Features were selected via least absolute shrinkage and selection operator (LASSO) regression. Six ML models were developed: logistic regression (LR), random forest (RF), support vector machine with RBF kernel (RBF-SVM), elastic net (EN), gradient boosting machine (GBM), and soft voting ensemble (SVE). The final model was selected by a prespecified composite ranking of the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and F1 score in internal validation, and evaluated in the temporal validation cohort. SHAP (SHapley Additive exPlanations) was used for interpretability. Results Of 568 patients, malnutrition prevalence was 60.49% (245/405) in the derivation cohort and 61.35% (100/163) in the temporal cohort. LASSO identified eight predictors: albumin (ALB), dry-weight-adjusted BMI, arm muscle circumference (AMC), spontaneous bacterial peritonitis (SBP), sex, arm circumference (AC), Child-Pugh grade, and alkaline phosphatase (ALP). The RBF-SVM model ranked first in the prespecified composite ranking. In internal validation, its AUC was 0.874 (95% CI: 0.810–0.939), accuracy 0.843, sensitivity 0.836, specificity 0.854, PPV 0.897, NPV 0.774, and F1 score 0.865. In temporal validation, the AUC was 0.832 (95% CI: 0.765–0.899), accuracy 0.785, sensitivity 0.820, specificity 0.730, PPV 0.828, NPV 0.719, and F1 score 0.824. The Hosmer-Lemeshow test indicated acceptable calibration ( p = 0.230), and decision curve analysis suggested net clinical benefit. Conclusion The RBF-SVM model showed good discrimination and acceptable calibration for malnutrition in patients with cirrhosis, with performance maintained in temporal validation. Built on eight routine indicators, it may serve as a candidate decision-support tool for clinical malnutrition assessment.
No takes yet. Share an insight, caveat, or question.
李秋香 et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: