Decoding circulating tumor DNA to identify durable benefit from immunotherapy in lung cancer

Objectives: Predicting the outcome of immunotherapy-treated non-small cell lung cancer (NSCLC) patients is challenging. Measuring circulating tumor DNA (ctDNA) in plasma is promising, but its application for outcome delineation needs further refinement. Since most information from the next-generation sequencing (NGS) panel is typically left unused, we aim to integrate more information. Materials and Methods: Patient and ctDNA data were compiled from five published studies involving advanced NSCLC. Plasma samples collected prior (t0) and early during (t1) immunotherapy were selected, tracking the changes of the highest t0 variant per gene. Durable benefit (DB, defined as progression free survival ≥ ½ year) was predicted. Performance was quantified using the integrated receiver operating characteristic curve (ROC AUC) and compared with the traditional molecular response (MR). Results: A total of 365 patients were pooled. Seven recurrently mutated genes were selected which optimally predicted DB (ROC AUC: 0.77-0.11+0.10), outperforming the MR predictor (with a ROC AUC: 0.64-0.11+0.11). Inclusion of patient characteristics led to a slight further improvement (ROC AUC: 0.80-0.10+0.09). The model performed satisfactory across all ctDNA platforms despite differences in panel size and content. Conclusion: Relative to a non-informative classifier (ROC AUC: 0.5), a twofold improvement in predictive value was achieved compared to MR by an integration of changes across seven selected genes in immunotherapy-treated NSCLC patients, whilst being broadly applicable across ctDNA NGS panels