Abstract
<sec> <title>BACKGROUND</title> <p>Low muscle mass may develop during midlife before overt weakness or functional impairment, but routine identification is limited by the cost and availability of dual-energy x-ray absorptiometry (DXA). Simple anthropometric screening could help identify adults who warrant confirmatory assessment. However, low muscle mass is defined differently across commonly used frameworks, which may affect both who is classified and how prediction-model performance is interpreted.</p> </sec> <sec> <title>OBJECTIVE</title> <p>This study aimed to develop and temporally validate a screening model using routinely available anthropometric measurements to identify DXA-defined low muscle mass in adults aged 40–59 years. We additionally evaluated performance under both appendicular skeletal muscle mass index (ASMI) and Foundation for the National Institutes of Health (FNIH) definitions and quantified the contribution of predictors mathematically embedded within each definition.</p> </sec> <sec> <title>METHODS</title> <p>Data was obtained from the 2013–2014 National Health and Nutrition Examination Survey (NHANES) for model development (N=1,556) and the independently collected 2017–2018 cycle for temporal validation (N=1,130). Separate logistic regression models were developed for ASMI- and FNIH-defined low muscle mass using age, sex, body mass index, waist circumference, and standing height. Performance was assessed using stratified cross-validation, calibration, and decision-curve analysis. The operating threshold was prespecified to achieve 90% sensitivity in the development cohort and applied unchanged to the validation cohort. Definition-linked predictors were subsequently removed in ablation analyses to assess their contribution to discrimination.</p> </sec> <sec> <title>RESULTS</title> <p>Cross-validated AUROC was 0.953 for ASMI and 0.942 for FNIH, with similar performance during temporal validation (0.954 and 0.940, respectively). At the frozen screening threshold, external sensitivity was 91.4% for ASMI and 91.9% for FNIH, with negative predictive values of 98.9% and 98.4%, respectively. Despite similar prevalence, agreement between ASMI and FNIH was minimal (Cohen κ=0.008), indicating that the definitions identified largely different individuals. When definition-linked predictors were removed, ASMI AUROC decreased from 0.953 to 0.869, whereas FNIH declined from 0.942 to 0.698, indicating substantially greater dependence of FNIH performance on variables incorporated into its own definition. External calibration remained stable for ASMI but showed greater drift for FNIH.</p> </sec> <sec> <title>CONCLUSIONS</title> <p>Routine anthropometric measurements can provide strong discrimination and high negative predictive value for DXA-defined low muscle mass in middle-aged adults, supporting a potential role as a first-stage rule-out and triage tool for targeted DXA assessment. However, prediction performance and the individuals identified depend strongly on the low-muscle-mass definition used. Prospective validation in clinical populations is required before implementation.</p> </sec>