Development and validation of a novel blood-based biomarker for gastric cancer triage in chronic dyspepsia.

Global implementation of gastric cancer (GC) screening in chronic dyspepsia populations faces challenges due to the high number-needed-to-scope (NNS) for oesophagogastroduodenoscopy. Routine blood tests (RBT) have limited utility for GC screening but offer potential for risk stratification when repurposed through machine learning. This study develops and validates a machine-learning-integrated biomarker (RBT-GC) that uses opportunistic triage to optimise endoscopy resource allocation. The team analysed 20 years of territory-wide retrospective data (2000-2020) from the Hong Kong Hospital Authority. 24 RBT and demographic features from 210,463 subjects (3071 cases) between 2000 and 2015 were used in training. An independent cohort of 90,479 subjects (2066 cases) from 2016 to 2020 was used in validation. The RBT-GC model successfully stratified validation cohort (2.3% baseline GC prevalence) into low-risk (0.3% prevalence), intermediate-risk (1.9%) and high-risk (14.0%) categories. The model detected (1276 cases) 12x more than CEA (102 cases) and 30x more than CA19.9 (42 cases). The application of opportunistic RBT-GC risk stratification reduced the NNS from 44 to 7 in the high-risk category of validation cohort. This machine learning approach repurposed standard blood tests into an opportunistic, affordable, scalable triage tool to alleviate endoscopic burdens across healthcare systems.