Eur J Radiol. 2026 Jun 10;203:112991. doi: 10.1016/j.ejrad.2026.112991. Online ahead of print.
ABSTRACT
RATIONALE AND OBJECTIVES: Sarcopenia is a significant predictor of adverse outcomes in cardiovascular diseases. However, most prior studies have focused on lumbar muscle indices and have overlooked the prognostic relevance of site-specific muscle loss in peripheral artery disease (PAD). This study aimed to evaluate whether machine learning (ML) models incorporating the lower-leg muscle index (LMI) provide superior prediction of long-term mortality in PAD patients undergoing revascularization.
MATERIALS AND METHODS: A retrospective cohort of 208 consecutive PAD patients hospitalized for planned lower-extremity revascularization was analyzed. Preoperative clinical data and CT angiography-derived muscle indices were collected. Feature selection employed LASSO, mRMR, random forest, and gradient boosting; the number of variables was limited by the 10 events per variable rule. Multivariate Cox regression assessed associations between muscle indices and all-cause mortality. Eight ML models were validated, and performance was evaluated with C-index and AUC. Calibration and decision curve analysis were also performed.
RESULTS: During a mean follow-up of 29.5 ± 15.5 months, 44 deaths were recorded. Non-survivors had more comorbidities, inflammation, and lower-limb muscle loss. Among muscle indices, LMI showed the highest discrimination for mortality (AUC 0.71), outperforming abdominal indices. Higher LMI was associated with lower mortality risk (HR = 0.94, 95 % CI 0.92-0.97, p = 0.004) across all subgroups. Gradient boosting yielded the best predictive performance, with C-indexes of 0.847, 0.809, and 0.823 at 1, 3, and 5 years, respectively. Calibration and decision curve analysis demonstrated model robustness and clinical utility.
CONCLUSIONS: Lower-leg muscle index is a superior and accessible predictor of long-term mortality in PAD, exceeding conventional abdominal muscle indices. ML models using LMI enable individualized risk stratification and may improve post-revascularization management.
PMID:42320142 | DOI:10.1016/j.ejrad.2026.112991