Prediction models for sarcopenia risk in type 2 diabetes mellitus: A systematic review and meta-analysis.
Weiju Tang, Rui Luo, Yue Luo et al.
✦ AI-curated · Sources linked
In 30 seconds
This systematic review and meta-analysis assessed multivariable models predicting sarcopenia or low muscle mass in adults with type 2 diabetes. It included 26 observational studies published between 2020 and 2025, mostly cross-sectional or retrospective studies from China. Pooled discrimination was promising: training-set AUC was 0.885 (95% CI 0.858-0.913), and internal-validation AUC was 0.859 (95% CI 0.812-0.905). External-validation evidence was sparse; only two studies contributed, with a pooled AUC of 0.945 (95% CI 0.923-0.966), which the authors advise interpreting cautiously.
Key findings
- Seventeen studies contributed training-set results; pooled AUC was 0.885 (95% CI 0.858-0.913; I 2 = 93.10%).
- Eleven studies contributed internal-validation results; pooled AUC was 0.859 (95% CI 0.812-0.905; I 2 = 94.90%).
- Only two studies contributed external-validation AUCs; pooled AUC was 0.945 (95% CI 0.923-0.966; I 2 = 0.00%).
- Most models used logistic regression, and external validation was limited.
Why it matters
Sarcopenia prediction models may help identify adults with type 2 diabetes who warrant further assessment, and pooled results suggest potentially useful discrimination. However, high heterogeneity in training and internal-validation estimates and the small external-validation evidence base mean that reported performance may not transfer reliably across settings. Methodological quality and reporting therefore matter when interpreting these models.
What to keep in mind
The evidence was predominantly from cross-sectional or retrospective studies, largely in China. External validation was limited to two studies, and heterogeneity was high for training and internal-validation AUCs.