Methylated DNA biomarkers and incident cancer in the American Cancer Society (ACS) Cancer Prevention Study-3 (CPS-3) cohort.

3004 Background: Multi-cancer early detection (MCED) tests are being developed to complement current single-cancer screening, and are based on next-generation sequencing of cell-free DNA in blood plasma with subsequent analysis of genomic features by machine-learning algorithms. To evaluate the detectability of cancer signals in individuals without known cancer at the time of blood draw, we assessed pre-diagnostic signals using a previously validated locked, methylation-based assay in plasma samples from the ACS CPS-3 cohort. Methods: CPS-3 is a prospective cohort study of cancer incidence and mortality. Participants (N=294,000) donated a blood sample (36 mL) at enrollment (2006-2013) and are followed for cancer outcomes via routine linkages with population-based cancer and death registries. Selected participants diagnosed with cancer within 3 years (Y) of blood donation and an equivalent number of matched non-cancer participants were chosen for processing. Despite differences in sample collection (tube, protocol) and long storage time in freezer, standard requirements for MCED sample quality were applied. Primary outcome: demonstrate that cancer signal was detectable via the MCED test up to 3Y prior to conventional diagnosis. Secondary outcomes: assess detectability based on time from blood draw and other clinical factors. Sampling weights were considered in the statistical analyses. Results: Of 268,726 CPS-3 participants, 2,850 were selected and assayed by cohort sampling, and 2,754 were included in the primary analysis set. CPS-3 cohort demographics and cancer incidence patterns compared to Surveillance, Epidemiology, and End Results suggested that this cohort is younger and healthier than the U.S. population. The test negative rate was 100% among non-cancers. The test positive rate (TPR) stratified by time between blood draw and clinical diagnosis was 2.3% (3Y), 6.0% (2Y), 15.2% (1Y). For eventually fatal cancers, TPR was 5.1% (3Y), 21.6% (2Y), 55.1% (1Y) vs 1.8%, 4.1%, 10.2% for non-fatal cancers. By stage, within 1Y, TPR was 73.7% for Distant at 4-6 months (M) and 94.1% at 1-3M vs 26.7% for Regional at 4-6M and 71.0% at 1-3M. The overall accuracy of predicting cancer signal origin was 95% (1Y). Many cancers were detected 3Y prior to clinical diagnosis: for example, a case of colon/rectum cancer at Regional stage was detected at 991 days. Conclusions: The ACS CPS-3 cohort includes persons at lower-than-average risk of cancer but offered a biospecimen repository with sufficient plasma samples to assess the preclinical detectability of cancer signals by retrospective evaluation. Rapid changes in detectability over time underscore the short preclinical time scales for aggressive cancers to develop, and therefore to be detected. These insights into cancer biology reinforce the importance of pre-diagnostic samples in the study of cancer signals for early detection.