Gastric cancer remains a global health challenge due to the difficulty of detecting it early in asymptomatic, high-risk populations. Current invasive diagnostic methods are impractical for widespread screening. Liquid biopsy using circulating tumor DNA (ctDNA) shows promise, but early detection is hindered by the low abundance and heterogeneity of ctDNA. We developed a multimodal cfDNA assay integrating methylation, fragmentomic, and hotspot mutation profiling from a single blood draw to detect gastric cancer-specific molecular signatures. Using these signatures, a machine-learning model was trained on a discovery cohort of 110 nonmetastatic GC patients and 119 healthy controls, then validated on an independent cohort of 58 patients and 65 controls. The ensemble model achieved an AUC of 0.87 (95% CI: 0.80–0.93), with 70.7% sensitivity and 92.3% specificity for detecting nonmetastatic GC. Incorporating hotspot mutation profiling increased overall sensitivity to 75.9% without affecting specificity. Compared to a previous multi-cancer model, our ensemble model showed improved sensitivity across all stages, particularly for early-stage GC (72.7% vs. 36.4%). This multimodal cfDNA assay provides a minimally invasive and effective strategy for early GC detection, making it a potential screening tool for high-risk populations. This study presents a novel multimodal cfDNA assay that combines methylation, fragmentomic, and hotspot mutation profiling, achieving 75.9% sensitivity and 92.3% specificity for early gastric cancer detection.
Võ et al. (Mon,) studied this question.