Evaluation of a plasma cell-free DNA methylation-based multi-cancer detection test.

Abstract

10550 Background: Blood-based multi-cancer detection (MCD) tests hold promise for early cancer detection. MCD tests should demonstrate high specificity, clinically meaningful sensitivity, and provide information to guide clinical diagnostic evaluation. Methods: We developed an MCD test that leverages a next-generation sequencing epigenomics hybrid capture assay (Shield) to measure DNA methylation in regions differentially methylated across solid tumor cancers. We then implemented a two-step classification algorithm. First, a regression model (multi-cancer classifier) was trained to distinguish cancer from non-cancer samples at 98% target specificity. Next, a cancer signal of origin (CSO) model was trained to categorize 10 solid tumors (Table 1) in samples predicted to be cancer. The test was evaluated in a blinded case-control cohort of plasma samples (3mL; Streck) from adults with treatment naïve cancer, or who were self-reported cancer free with 1 year of follow-up (NCT05334069). Results: Final evaluable dataset was 962 participants (excludes 31 (3.1%) that failed quality control). Median age was 62 years (range: 40-78). 55% were female. Self-reported race was 81% White. Observed specificity was 98.6% (436/442). Sensitivity for the 10 cancer types included in the CSO model was 59.7% (224/375) and was 55.7% (263/472) overall when incorporating 4 solid tumor types not included in the CSO model. Primary or secondary CSO prediction was 92% accurate for the cancer types included (Table 1). Conclusions: In this cohort, this MCD test shows 59.7% sensitivity, 98.6% specificity, and 92% primary or secondary CSO accuracy for the cancer types included. Integrating an MCD test with an FDA approved blood-based CRC screening test may increase the MCD clinical value in the intended use population, which should be studied further. A version of this MCD test is being studied in a prospective, interventional study evaluating MCD testing feasibility. Overall and per cancer sensitivity and CSO accuracy results. Overall Sensitivity % Sensitivity Stage I/II, % Sensitivity Stage III/IV, % Primary or Secondary CSO Accuracy, % All Samples, 472 56% 31% 81% - Cancers included in CSO caller, 375 60% 35% 84% 92% Bladder, 13 62% 44% 100% 75% Breast, 86 45% 17% 88% 97% Colorectal, 41 83% 53% 100% 91% Esophageal-Stomach, 25 96% 100% 95% 96% Hepatocellular, 16 94% 100% 92% 67% Lung, 57 67% 41% 93% 97% Ovarian, 20 70% 80% 67% 100% Pancreas, 59 68% 50% 96% 93% Prostate, 59 21% 3% 41% 92% Cancers not included in CSO caller, 97 40% 18% 65% Endometrial, 29 38% 12% 75% Head & Neck, 15 80% 100% 63% Kidney, 44 34% 0% 71% Melanoma, 9 11% 0% 20%