LAMPAD: An Integrated Circulating Tumor DNA-Based Model for Predicting Potential Cure in Patients With Resected NSCLC

Evaluation of molecular residual disease (MRD) status in patients with NSCLC after surgery primarily relies on circulating tumor DNA (ctDNA) analysis. However, given the narrow postoperative window (4-6 wk) for adjuvant therapy, the approximately 70% false-negative rate of single-time point ctDNA landmark detection severely limits its utility.

Here, we introduce LAMPAD, an XGBoost-Cox model that incorporates an additional preoperative time point alongside the standard postoperative landmark. By leveraging ctDNA quantification from both time points, it refines prognostic stratification among patients with landmark undetectable MRD, specifically identifying those who are truly disease free.

The LAMPAD model, incorporating the top five features ranked by Shapley additive explanations analysis-baseline ctDNA level, TNM stage, landmark cell-free DNA (cfDNA) concentration, baseline cfDNA concentration, and baseline ctDNA status-was trained on 163 patients with stages I to III NSCLC with landmark undetectable MRD. The model effectively stratified patients into low-risk (2-y disease-free survival [DFS]: 97.8%) and high-risk (2-y DFS: 71.6%) groups (hazard ratio = 0.11, 95% confidence interval: 0.06-0.21, p < 0.001). LAMPAD demonstrated consistent performance across both fixed-panel and personalized ctDNA-MRD approaches in multiple validation NSCLC cohorts (pooled 2-y DFS: 94.3% versus 72.4% for low- versus high-risk groups; hazard ratio = 0.18, 95% confidence interval: 0.13-0.25, p < 0.001). Preoperative blood test markedly contributed to the LAMPAD model, with methylation analysis revealing elevated immune-derived and lung-derived cfDNA in high-risk patients, suggesting systemic immune involvement in risk stratification.

Overall, the LAMPAD model outperforms single-time point postoperative ctDNA detection by effectively discriminating true negative patients, thereby offering a more reliable prognostic tool for identifying low-risk patients with potential for cure.