Frontiers in Nutrition · Nov 19, 2026

Development and Internal Validation of a Bedside Case-Finding Model for Low Muscle Mass in Middle-Aged and Older Hospitalized Patients with Cancer: A Multicenter Cross-Sectional Study

By Shu An Wang, Qingqing Deng, Zimeng Zhu et al.

Background: Low muscle mass is common in patients with cancer, but access to bioelectrical impedance analysis (BIA) and other objective muscle-mass measurements varies across hospitals. We aimed to develop and internally validate a bedside case-finding model based exclusively on non-BIA predictors to identify hospitalized patients who should be prioritized for objective muscle-mass assessment according to the 2025 Asian Working Group for Sarcopenia (AWGS) criteria. Methods: This multicenter cross-sectional prediction-model study included hospitalized patients with cancer aged ≥50 years with an ascertainable BIA-defined outcome. Candidate predictors were prespecified in three hierarchical sets. Missing predictor values were addressed by multiple imputation nested within hospital-grouped five-fold cross-validation. Gradient boosting, L2-penalized logistic regression, and random forest were compared. Model selection considered discrimination, calibration, overall prediction error, clinical utility, predictor availability, and transparency. Results: Among 1,058 patients from 36 hospitals, 432 (40.8%) had BIA-defined low muscle mass. Under gradient boosting, the ROC-AUCs of Models A, B, and C were 0.803, 0.819, and 0.817, respectively. Adding calf circumference and handgrip strength modestly improved performance, whereas adding albumin, hemoglobin, and ln(SII) did not. The selected cancer-category-free L2-penalized logistic model achieved a ROC-AUC of 0.821 (95% CI, 0.780–0.856), a PR-AUC of 0.734, a Brier score of 0.170, and a calibration slope of 0.939. At the exploratory threshold of 0.382, sensitivity was 80.6%, specificity 72.0%, positive predictive value 66.5%, negative predictive value 84.3%, and the F1 score 0.729. These classification metrics are exploratory, not externally validated clinical-rule performance. Conclusions: A bedside model comprising age, sex, body mass index, mid-upper arm circumference, NRS 2002, reduced intake or absorption, calf circumference, and handgrip strength showed good internal discrimination and acceptable calibration. The model does not diagnose sarcopenia or malnutrition. Its full specification, independent external validation, locally appropriate decision thresholds, and prospective clinical-impact evaluation are needed before clinical implementation.

The paperJournal siteBiomarkers

AgingResearch.News