Application of quantile regression forest for early detection of sarcopenia in oncogeriatric patients.

Abstract

e13854 Background: Sarcopenia, characterized by the progressive loss of muscle strength and mass, adversely affects oncological outcomes, treatment tolerance, and quality of life in elderly patients. In our previous work, we developed a quantile regression forest (QRF) model trained on a cohort of 11,019 healthy individuals from the Lookup 8+ project to predict handgrip strength. In this study, we applied this model to a cohort of 54 elderly oncological patients to assess its potential for identifying early signs of muscle strength reduction. Methods: Eligible participants were oncological patients aged 65 years or older referred for a Comprehensive Geriatric Assessment (CGA) at Fondazione Policlinico Gemelli in Rome (CGA cohort). Handgrip strength was assessed using a hydraulic dynamometer by trained investigators. The QRF model was retrained to predict handgrip strength based on gender and age. Patients’ actual handgrip values were compared to predicted quantiles. Probable sarcopenia was defined according to the European Working Group on Sarcopenia in Older People 2 (EWGSOP2) criteria. Median and interquartile range (IQR) were reported as summary statistics, correlations were assessed using Spearman’s rho (ρ). The root mean squared error (RMSE) was calculated for the CGA cohort, with 95% confidence intervals (95% CI) derived from 5,000 bootstrap replicates. Results: The QRF model incorporating only gender and age achieved a mean cross-validated R² of 0.60 across the 5 folds of the original cohort. Of the 54 patients included in the CGA cohort, 53 (98%) were male, with prostate cancer as the most common tumor type. Median age was 77 years (IQR 70–80), and median handgrip strength was 32 kg (IQR 28–39). Handgrip strength showed a negative correlation with age (ρ = -0.22), 5-repetition chair stand test (ρ = -0.49), and number of non-oncological medications (ρ = -0.40). Handgrip strength was positively correlated with social support (ρ = 0.48). According to EWGSOP2 criteria, 13 patients (24%) were classified as having probable sarcopenia.The RMSE of the QRF model in the CGA cohort was 8.2 kg (95% CI 6.0–10.5). Among the patients, 29 (54%) had a negative difference between the predicted (50th percentile) and actual handgrip strength, with a median difference of -5.4 kg (IQR -7.3 to -2.2 kg). Additionally, 17 patients (31%) had actual handgrip values below the 25th predicted quantile, and 6 patients (11%) fell below the 10th predicted quantile. Conclusions: This study highlights the potential utility of a quantile regression approach in predicting handgrip strength to identify oncogeriatric patients at risk for sarcopenia. Handgrip strength evaluation offers a simple yet effective tool for screening, providing a non-invasive approach for sarcopenia detection. Cohort expansion and radiological confirmation (using Dual X-ray absorptiometry) of sarcopenia are ongoing.