A LASSO-based nomogram for predicting sarcopenia-related nutritional risk in esophageal cancer patients: model development, validation, and an interactive clinical decision tool
Abstract
Objective This study aims to construct an interpretable predictive model for esophageal cancer sarcopenia-related nutritional risk based on multidimensional clinical indicators and develop an interactive clinical decision support system. Methods The study cohort comprised 202 patients with a confirmed diagnosis of esophageal cancer who were retrospectively recruited from Taikang Xianlin Drum Tower Hospital. The study period spanned from January 2023 to December 2025. The sample was split into training and validation datasets (7:3), respectively. Sarcopenia nutritional risk was defined as NRS-2002 score ≥3. Predictors were screened using LLogistic and LASSO regression to construct a model. Model performance was evaluated using ROC, calibration, decision curve analysis (DCA), and CIC curves. Model interpretability was analyzed using SHAP values. An interactive online prediction system was developed based on R Shiny. Results Age, sex, BMI, diabetes, intervention, prothrombin time (PT), and clinical stage (cStage) were predictors of sarcopenia-related nutritional risk. The AUC was 0.864 (0.778–0.940) in the training set and 0.802 (0.696–0.898) in the validation set. The calibration curves showed satisfactory goodness of fit (P > 0.05), and DCA indicated clinical net benefit within the threshold range of 0.05–0.70. SHAP analysis revealed that BMI, PT, and gender were key contributing factors. The interactive prediction system “Sarcopenia Intelligence” was successfully developed (https://ytky622.shinyapps.io/Sarcopenia_Predictor_2026/). Conclusion The model demonstrates favorable discrimination, calibration, and clinical utility. Combined with SHAP interpretability and the interactive system, it provides a convenient and precise tool for early screening and individualized intervention of sarcopenia in patients with esophageal cancer.