Prediction models for sarcopenia risk in type 2 diabetes mellitus: A systematic review and meta-analysis.

Objective

To systematically summarise prediction models for sarcopenia in patients with type 2 diabetes mellitus (T2DM), appraise their methodological quality and reporting, and synthesise their discriminative performance.

Methods

PubMed, Web of Science, Embase, Cochrane Library, CNKI, Wanfang and VIP were searched from inception to 22 December 2025 for observational studies developing and/or validating multivariable models (≥2 predictors) for sarcopenia or low muscle mass in adults with T2DM. For studies reporting the area under the receiver operating characteristic curve (AUC), random- or fixed-effects meta-analyses were conducted to pool AUCs for training, internal validation and external validation datasets.

Results

Twenty-six studies published between 2020 and 2025 were included, mostly cross-sectional or retrospective studies from China, with one based on US NHANES data. Most models used logistic regression; a few applied machine learning methods. Predictors commonly included age, sex, anthropometric indices, diabetes duration and control, biochemical and nutritional indicators, and muscle-related measures. Internal validation was frequently performed, whereas external validation was limited. Seventeen studies contributed training-set AUCs (pooled AUC 0.885, 95% CI 0.858-0.913; I2 = 93.10%), and eleven contributed internal validation AUCs (pooled AUC 0.859, 95% CI 0.812-0.905; I2 = 94.90%). Only two studies contributed external validation AUCs for pooling, yielding a pooled AUC of 0.945 (95% CI 0.923-0.966; I2 = 0.00%); this result should be interpreted cautiously because of the very limited external validation evidence.

Conclusions

Existing prediction models for sarcopenia or low muscle mass in T2DM patients show promising discrimination, but limitations in analysis methods, reporting and external validation may affect their robustness and generalisability. Future work should strengthen methodological rigor and reporting in line with PROBAST and TRIPOD to facilitate reliable clinical use.