Early risk stratification for sarcopenia in a pan-cancer cohort using quantile regression forests on handgrip strength.

e13585 Background: Sarcopenia, characterized by loss of muscle mass, strength, and physical function, is associated with poorer oncologic outcomes, treatment intolerance, and reduced quality of life. Handgrip strength (HGS) is a low-cost, scalable screening tool, yet oncology-specific reference standards are lacking. We evaluated a quantile regression forest (QRF) model to enable individualized, percentile-based risk stratification for low HGS in patients with solid tumors, aiming at early identification of patients at risk of sarcopenia rather than diagnosis of established disease. Methods: Eligible participants were cancer patients referring to Radio-Chemotherapy service at Fondazione Policlinico Gemelli in Rome. Data collection included demographic characteristics, anthropometric measures, cancer type, and handgrip strength. All variables are collected at baseline and at predefined follow-up visits. Median and interquartile range (IQR) were reported as summary statistics. Quantile regression forest (QRF) algorithm was applied on this sample and descriptive analysis was performed based on QRF findings. Results: Data from 174 patients were included in this analysis. Median age was 61 (IQR 51-68), a total of 127 patients were females (73%) and 47 (27%) males. 70 patients (40%) were aged 65 years or older. The three most frequent malignancies were breast (63, 36%), gynecological (35, 20%) and colorectal cancers (30, 17%). Median handgrip strength in females was 20 kg (16.0-25.0) and 31 kg for males (24.0-40.0). Applying QRF, data showed that a total of 94 subjects (54%) performing handgrip strength fell below their 25 th quantile prediction, and a total of 62 subjects fell below their 10 th predicted quantile. As for the tumor type, 51% of breast patients were under their 25 th quantile and 39% below their 10 th quantile, while gynecologic cancers showed 60% of patients below their 25 th quantile and 23% under their 10 th quantile. In colorectal cancer, 53.3% of patients fell below their 25 th predicted quantile and 30% showed handgrip strength below their 10 th quantile. Notably the median difference between the 50 th predicted percentile and the actual value was -3.8 Kg (-8.47 to -0.40 kg). A total of 134 subjects were identified with a negative difference between the actual handgrip and their predicted 50 th percentile. Conclusions: QRF-based, percentile-referenced modeling of HGS provides individualized, scalable risk stratification to flag cancer patients at risk of functional decline and sarcopenia, potentially enabling earlier supportive interventions (targeted exercise, nutritional optimization, and geriatric assessment in older adults) before overt sarcopenia develops. Ongoing cohort expansion and longitudinal follow-up will assess associations with toxicity, treatment modification, and clinical outcomes.