Preclinical circulating tumor DNA (ctDNA) shedding duration and prognostic implications of modeling 3669 participants from American Cancer Society Cancer Prevention Study-3 (CPS-3) and Circulating Cell-free Genome Atlas substudy 3 (CCGA3).

3060 Background: Early detection of cancer occurs in the preclinical phase, when development of cancers is unobservable. Shedding of ctDNA has previously been associated with aggressive cancers, suggesting these tumors may develop rapidly before diagnosis. Assays of ctDNA in blood samples from prospective cohort studies at various lead times (interval between blood draw and diagnosis) may provide insight into the timescales of preclinical cancer development, as well as prognosis. Methods: We used a targeted methylation assay and the associated locked classifier for cancer signal detection (CSD) on samples from two large prospective biobank studies: 1) CPS-3, where blood draw at enrollment (2006-13) occurred ≤3 years (Ys) before clinical diagnosis (n=1065); 2) CCGA3 (NCT02889978), where blood draw (2016-19) was concurrent with clinical diagnosis (n=2604). We constructed a model for detectability including study, inherent shedding of ctDNA at clinical diagnosis, and effect of lead time by cancer type and stage. We additionally evaluated prognosis stratified by ctDNA status (CSD positivity), compared to expected cancer-specific survival (CSS) matched by age, cancer type, and stage for ≤9 Ys of follow-up, noting that ctDNA status may change during lead time. Results: Estimated inherent ctDNA detectability did not differ between CCGA3 and CPS-3 studies (95% confidence interval [CI] for difference is [-8.2%, 4.6%] and spans zero). Median [CI] detection timescales for exponential decay of detection by time were 0.75 [0.47, 1.3] Ys for localized stage, 0.89 [0.61, 1.33] Ys for regional, and 1.2 [0.84, 1.67] Ys for distant, consistent with a published model, and ranged by type from 0.49 [0.26, 0.88] Ys for pancreas to 2.45 [1.14, 4.87] for lymphoma. For CPS-3, estimated CSS at 9 Ys from blood draw was 54% for ctDNA+ and 79.5% for ctDNA- participants. Hazard ratios (HR) [CI] for ctDNA+ vs ctDNA- alone predicting CSS were 7 [4.2, 12.2], 4.2 [2.9, 6], and 3.5 [2.5, 4.7] for those diagnosed up to 1, 2, and 3 Ys after blood draw, respectively. Compared to synthetic controls for age, cancer type, stage, and lead time in Surveillance, Epidemiology, and End Results (SEER) data, CSS was better in each group: ctDNA+ HR 0.63 (CI: 0.54, 0.74); ctDNA- HR 0.45 (CI: 0.22, 0.74). Conclusions: These results provide estimates for the average detection window in tumors shedding ctDNA across multiple cancer sites and stages, including cancer types with no current screening paradigm. Taking lead time into account, preclinical ctDNA status selected clinically significant cancers, limiting potential for overdiagnosis. Additionally, long-term survival remained better than in SEER data, suggesting that ctDNA+ cancers can be survivable. Results require confirmation in future prospective analyses.