A machine learning-derived sarcopenia index is associated with survival and nonrelapse mortality in DLBCL.

abstractBody composition analysis (BCA) provides an objective assessment of metabolic states, but its prognostic value in diffuse large B-cell lymphoma (DLBCL) remains unclear. We applied machine learning-supported BCA to computed tomography imaging from patients with newly diagnosed DLBCL enrolled in the prospective phase 3 PETAL trial to quantify radiologic sarcopenia. We assessed BCA results in relation to survival after first-line immunochemotherapy, treatment-related hematologic toxicities, and molecular disease features. Patients in the lowest tertile of normalized skeletal muscle mass exhibited inferior survival after adjustment for established risk factors. Cause-specific time-to-event analyses revealed that sarcopenia was not associated with lymphoma-specific death but strongly predicted nonrelapse mortality, indicating a potential role as a biomarker of host vulnerability. Consistent with these findings, sarcopenic patients had a higher probability of experiencing hematologic toxicity during immunochemotherapy, and sarcopenia was the only independent risk factor for higher-grade hematotoxicity in multivariable analyses. Longitudinal BCA revealed inferior survival in patients with early muscle loss during therapy. Baseline sarcopenia and treatment-emergent muscle loss were not correlated, suggesting that these represent distinct biological phenomena, and only a small fraction of the interindividual variability in muscle mass could be attributed to age and lymphoma burden. Both phenotypes were independent of DLBCL molecular clusters, and no recurrently mutated gene was associated with lower skeletal muscle mass. Taken together, our results establish baseline sarcopenia and treatment-emergent muscle loss as orthogonal risk factors for adverse outcomes in DLBCL, supporting the evaluation of BCA for risk stratification in personalized lymphoma therapy.