9525 Background: First-line immunotherapy (IO) and chemo-immunotherapy (chemo-IO) are approved for PD-L1-expressing advanced non-small cell lung cancer (NSCLC), but PD-L1 expression does not reliably predict therapeutic response. Therefore, accurate on-therapy response assessment is needed to guide clinical decision-making. Conventional radiographic imaging is the gold-standard, but may not capture the nature and timing of an immune-mediated response. We investigated whether comprehensive longitudinal circulating tumor DNA (ctDNA) dynamics could enhance prediction of clinical response. Methods: We conducted targeted error correction sequencing (TEC-Seq) of longitudinal plasma specimens and matched white blood cells (WBCs) in patients with NSCLC treated with first-line IO or chemo-IO. Plasma ctDNA variants were identified by filtering out clonal hematopoiesis (CH) and germline (GL) variants found in matched WBCs. Clinical and pathological data, including PD-L1 tumor proportion score (TPS), and imaging response by RECIST1.1 were assessed. ctDNA dynamics were modeled to predict durable clinical benefit (at 6 months), progression-free survival (PFS), and overall survival (OS). Results: A total of 143 longitudinal plasma and 24 white blood cell samples underwent TEC-Seq for 31 patients with NSCLC treated with IO (n = 17) or chemo-IO (n = 14). ctDNA variants were found in 77% (n = 24) of patients after filtering out CH and GL variants, which comprised 53% (n = 196 of 373) of all variants. Molecular response, signified by elimination of ctDNA variants, was detected in 32% (n = 10) of patients and associated with improved PFS (p = 0.0004, log rank) and OS (p = 0.017, log rank). Time to molecular response was shorter than time to best RECIST response (median 3 vs. 7.71 weeks, p = 0.048, Mann-Whitney U test). A logistic regression model incorporating molecular response, recrudescence, or emergence of new variants predicted durable clinical benefit (sensitivity 84%, specificity 76%, AUC 0.88) better than PD-L1 TPS (AUC 0.67, p = 0.046, bootstrap method). Conclusions: Comprehensive modeling of ctDNA variant dynamics predicts clinical outcome independent of PD-L1 status in patients with advanced NSCLC treated with first-line IO or chemo-IO. Verification of bona fide ctDNA variants by matched WBC sequencing is essential. Molecular response can be identified earlier than imaging response and could enable on-therapy decision-making to alter clinical outcomes.