C110-18 CT-derived Body Composition Biomarkers Predict Delayed Bone Metastasis in Advanced Non-small Cell Lung Cancer

Abstract Rationale Bone metastasis is a common and debilitating complication of advanced non-small cell lung cancer (NSCLC), worsening both quality of life and survival. Yet, tools for pre-treatment risk stratification remain limited. Current assessments rely primarily on clinicopathologic factors, laboratory tests, and imaging modalities that typically detect metastasis only after they develop, emphasizing management rather than prevention. Early noninvasive identification of patients at high risk could enable timely bone-targeted therapies, closer surveillance, and personalized treatment planning. We investigate whether quantitative, pre-treatment CT biomarkers of body composition and thoracic anatomy can predict the risk and timing of bone metastasis in advanced NSCLC. Methods We retrospectively analyzed 126 patients with advanced NSCLC, including 19 who developed delayed bone metastases (90 days after initial diagnosis). Fully automated deep learning algorithms were used to quantify the volume and density of five body composition tissues and thoracic anatomical features on pre-treatment CT scans (Figure 1.a). Associations between individual features and time to bone metastasis were assessed using univariate Cox regression. A multivariate LASSO-Cox model with five-fold cross-validation was then constructed for prognostic prediction. Model performance was evaluated using the area under the time-dependent receiver operating characteristic curve (AUC) at 1-, 3-, and 5-year intervals and calibration by the Brier score. Results Higher subcutaneous and visceral fat density were significant predictors of increased bone metastasis risk (HR = 1.04 95% CI 1.01-1.07, p = 0.011; HR = 1.02 95% CI 1.00-1.05, p = 0.016). No clinical variables showed significant univariate associations. In multivariate analysis, a four-feature LASSO-Cox model (vein volume, visceral and subcutaneous fat density, muscle-to-fat ratio) was significant (likelihood ratio = 11.4, p = 0.02). The model achieved time-dependent AUCs of 0.64 (95% CI 0.45-0.89), 0.76 (95% CI 0.61-0.91), and 0.73 (95% CI 0.56-0.90) at 1-, 3-, and 5-years, respectively, with corresponding Brier scores of 0.08 (95% CI: 0.03-0.12, IPA: 0.04), 0.13 (95% CI: 0.08-0.18, IPA: 0.23) and 0.15 (95% CI: 0.09-0.20, IPA: 0.22). Higher vein volume trended toward increased metastasis risk (p = 0.10), whereas a higher muscle-to-fat ratio suggested a potential protective effect (p = 0.12). Conclusion Automated analysis of pre-treatment CT-derived body composition and thoracic features showed significant prognostic value for predicting the timing of bone metastasis in advanced NSCLC. This noninvasive approach enables early identification of patients at increased risk for skeletal spread, supporting personalized treatment planning and targeted surveillance. Funding Sources NIH R01HL174570 and R01CA237277 This abstract is funded by: NIH R01HL174570 and R01CA237277