Malnutrition significantly impacts the prognosis of elderly patients with stable chronic obstructive pulmonary disease (COPD). The objective of this study was to develop and validate an automated machine learning (AutoML) framework for predicting malnutrition risk in this population. Data from the National Health and Nutrition Examination Survey (NHANES) 2007–2012 were utilized for model development (n = 710). An independent clinical cohort (n = 330) from the First Hospital of Shanxi Medical University (China) served as the external validation set. Malnutrition status was defined according to the Controlling Nutritional Status (CONUT) score. After feature selection via Least Absolute Shrinkage and Selection Operator (LASSO) regression, the H2O AutoML platform was employed to train and compare models across 6 algorithm families. Model performance was evaluated using the Area Under the Receiver Operating Characteristic Curve (AUC), sensitivity, and specificity. Model interpretability was assessed via SHapley Additive exPlanations (SHAP) analysis. Malnutrition prevalence in the development cohort was 30.99%. Among 52 trained models, a Gradient Boosting Machine (GBM) demonstrated the highest predictive performance. In the internal validation set, the GBM achieved an AUC of 0.813 (95% CI: 0.747–0.872). In the external validation cohort, the model yielded an AUC of 0.830 (95% CI: 0.786–0.875) with a sensitivity of 0.911. SHAP analysis identified fasting glucose, serum creatinine, comorbidity count, hemoglobin, body mass index (BMI), triglycerides, and age as the most influential predictors. The GBM model, developed through an AutoML framework, demonstrates robust predictive performance and generalizability across geographically and ethnically diverse populations. The deployment of a web-based risk calculator facilitates early screening and supports personalized nutritional management in clinical settings.
Guo et al. (Fri,) studied this question.