A model integrating six CT-derived systemic biomarkers and four clinical factors predicted lung cancer risk (C-index 0.720), outperforming PLCOm2012, LCRAT, and Pittsburgh Predictor.
Does a risk prediction model integrating systemic CT biomarkers and clinical factors improve the prediction of longitudinal lung cancer risk compared to established risk models in lung cancer screening participants?
Systemic CT-derived biomarkers, including cardiovascular and body composition metrics, significantly improve long-term lung cancer risk prediction beyond traditional nodule-based and clinical models.
Absolute Event Rate: 0% vs 0%
Abstract Rationale Current lung cancer screening and AI tools rely primarily on pulmonary nodule detection and clinical factors (e.g., age, smoking) for risk stratification. Consequently, they offer limited value to the majority of screened individuals who lack suspicious nodules yet remain subject to annual screening. We hypothesized that systemic host characteristics, such as lung, vascular, and body composition health, are predictive of long-term lung cancer risk independent of nodule presence. Methods We analyzed baseline low-dose computed tomography (LDCT) scans from 15,966 participants in the National Lung Screening Trial (NLST). Systemic biomarkers across body tissue composition, lung/airway morphology, and cardiovascular domains were quantified using validated AI algorithms and analyzed using the Cox proportional hazards model to predict continuous, longitudinal lung cancer risk, with hazard ratio (HR) reported. Performance was evaluated using 10-fold cross-validation across several metrics, including C-index, AUC, calibration plots, and decision curve analysis, and benchmarked against established risk models (PLCOm2012, LCRAT, and Pittsburgh Predictor). Results Multivariable analysis identified intrapulmonary vein volume (HR = 1.35, 95% CI: 1.26-1.45), aortic calcification (HR = 1.13, 95% CI: 1.09-1.18), and bone ratio (HR = 1.11, 95% CI: 1.02-1.21) as significant risk factors for lung cancer. Conversely, extrapulmonary artery volume (HR = 0.78, 95% CI: 0.72-0.85), skeletal muscle density (HR = 0.86, 95% CI: 0.80-0.93), and airway ratio (HR = 0.86, 95% CI: 0.81-0.92) were associated with reduced risk (all p 0.05). These biomarkers capture distinct complementary host susceptibility pathways. Lower skeletal muscle density may reflect myosteatosis and metabolic vulnerability. Vascular remodeling may be indicative of subclinical cardiopulmonary stress and impaired reserve. A lower airway ratio likely serves as a structural surrogate for airflow limitation and reduced clearance of inhaled carcinogens. Together, these markers characterize a vulnerable, pro-tumorigenic host environment independent of smoking history. A model integrating the six imaging biomarkers and four clinical factors achieved strong discrimination (C-index = 0.720, 95% CI: 0.688-0.751), outperforming PLCOm2012, LCRAT, and the Pittsburgh Predictor (Fig. 1). At 6 years, it achieved an AUC of 0.719, exceeding PLCOm2012 and the Pittsburgh Predictor. The model also demonstrated superior calibration across the risk spectrum, supporting reliable risk estimation for both high- and low-risk individuals, and delivered the greatest net clinical benefit across screening-relevant thresholds in decision curve analysis. Conclusion This study demonstrates that systemic CT biomarkers provide a stable, long-term signature for predicting future lung cancer risk independent of nodule presence. By enabling more precise risk stratification, our model supports personalized screening strategies, potentially safely extending screening intervals for individuals at low risk. This abstract is funded by: NIN R01CA237277
Pu et al. (Mon,) reported a other. A model integrating six CT-derived systemic biomarkers and four clinical factors predicted lung cancer risk (C-index 0.720), outperforming PLCOm2012, LCRAT, and Pittsburgh Predictor.