Abstract
<title>Abstract</title> <p>Gastrointestinal malignancies pose a major public health challenge in China, where the majority of patients are diagnosed late and face poor survival outcomes. A clinlabomics data-based auxiliary screening tool could address this gap, yet previous studies lack the large-scale, multicenter validation necessary to support clinical translation. This study is a multicenter retrospective analysis involving 115,032 participants from four hospitals between January 1, 2019, and April 30, 2026. The cohort from Sichuan Cancer Hospital (SCH) was designated as the discovery set. After systematic comparison of various machine learning and deep learning models, glmnet was ultimately selected as the optimal predictive model. The final tool, Gastrointestinal Cancers Seek (GaSeek), was subsequently validated in three independent external cohorts, as well as in three real-world clinical cohorts. This GaSeek achieved an area under the receiver-operating characteristic curve (AUC) of 0.960 (95% CI: 0.952-0.968) in the internal validation cohort and an AUC of 0.852 (95% CI: 0.843-0.860) in the overall external validation cohort. Analysis of feature importance identified albumin (ALB) and hemoglobin (HGB) as the most influential predictors. Furthermore, the model demonstrated robust performance in detecting early-stage gastrointestinal cancers and was proficient in screening cases with high confidence for positive results. Using an appropriate threshold, GaSeek not only screened out gastrointestinal cancers but also identified other endoscopically detectable digestive tract tumors in real-world cohorts. We used “Screening Efficiency Gain (SEG)," defined as the ratio of the detection rate in the high-risk group to that in the overall cohort, to evaluate the improvement. In the SCH real-world cohort, whose non-cancer participants were healthy individuals, GaSeek delivered a high SEG of 15.70, showing excellent screening efficiency. The GaSeek was deployed as a publicly accessible web application (https://weiyanghe520.shinyapps.io/gastrointestinal-cancer-predictor/). It effectively identifies individuals at high risk for digestive tract tumors and could improve the efficiency of endoscopic screening for digestive tract tumors through risk stratification. This study has been registered with the Chinese Clinical Trial Registry, registration number: ChiCTR2500107671, registration date: August 15th, 2025, (https://www.chictr.org.cn/showproj.html?proj=276534).</p>