Background: Stage II-III non-small cell lung cancer (NSCLC) patients receive adjuvant chemotherapy after surgery as standard-of-care treatment, despite an absolute 5-year disease free survival benefit of only 5.8%. Stage I patients do not receive adjuvant treatment but may benefit. Identification of residual disease by circulating tumor DNA (ctDNA) after curative intent therapy may allow more accurate decisions. Methods: Tumor tissue and serial plasma samples were collected from stage I-III NSCLC patients enrolled in the LEMA trial (NCT02894853). Somatic mutations identified by tumor exome sequencing were used to design patient-specific RaDaR assays, targeting up to 48 variants per patient, to analyze ctDNA in serial plasma samples. Results were compared and combined with an independent dataset (LUCID; Gale et al, Annals Oncol 2022). Results: In LEMA, 129 patients (53% stage I, 18% stage II and 29% stage III) were treated with curative intent by surgery (n=117), chemoradiation (n=8) or stereotactic radiotherapy (n=4) and followed for a median of 3.4 years. The LUCID dataset included 88 patients (49%/28%/23% in stage I/II/III). Results in the two cohorts were highly concordant (Table) and were assessed in the combined dataset of 808 serial plasma samples. Before treatment, ctDNA was detected in 22%, 72% and 90% of patients with stage I, II and III disease. ctDNA was detected post-treatment (at least one positive samples ≥14 days after the end of therapy) in 41/65 (63%) of patients who developed radiographic recurrence of their primary cancer, and preceded recurrence by a median of 204 days. For recurrence prediction by ctDNA detection post-treatment, the sensitivity, specificity, positive and negative predictive value (PPV and NPV) in stage I patients were 53%, 99%, 91% and 90%. In stage II-III patients, sensitivity was higher but specificity and NPV were lower (67%, 93%, 91% and 74%, respectively). ctDNA detection after treatment was associated with shorter recurrence-free survival (hazard ratio (HR) 11.5, P=1.8x10^-32) and overall survival (HR 8.1, P=1.2x10^-17). Conclusions: ctDNA detection by RaDaR assays predicted recurrence in two independent cohorts, confirming the potential to identify patients for adjuvant treatment. In stage I patients with positive ctDNA, adjuvant treatment could be offered with confidence due to high specificity of 99% and PPV of 91%. In stage II-III patients, the primary cancer recurred in a quarter of patients with no ctDNA detected post-treatment (NPV 74%). Our results help identify the potential role for ctDNA analysis as a decision support tool for adjuvant therapy in NSCLC.