Abstract 6324: A proteomics-based model improves the accuracy of upper gastrointestinal cancer prediction

Abstract Background: Upper gastrointestinal cancers (UGI), including esophageal cancer (EC), gastroesophageal junction cancer (GEJ), and gastric cancer (GC), pose a major global health burden, with prognosis heavily dependent on early diagnosis. Current screening strategies inadequately identify high-risk individuals. While advances in proteomics offer promise for risk assessment and early detection, their application in UGI cancer screening remains underexplored. Methods: We conducted a prospective analysis within the UK Biobank. Participants were randomly divided into training and testing sets at a 7:3 ratio. Among proteins measured using the Olink Explore 3072 panel, those with >30% missing values were excluded, resulting in a final analysis set of 2,920 proteins. The associations of proteins with UGI cancer were identified by Cox proportional hazards models (FDR < 0.05). L1-penalized LASSO-Cox regression with ten-fold cross-validation was then applied to select candidate proteomic biomarkers. Then, two risk prediction models (a simple model based on epidemiological factors and an integrated model that further incorporated proteins) were developed and internally validated. The discrimination of the models was assessed using the area under the curve (AUC) and compared via DeLong's test. Results: After excluding participants with baseline cancer or missing data, 48,366 individuals were included (median follow-up: 14.7 years; 261 UGI cancer cases). We identified 912 proteins significantly associated with UGI cancer (830 risk-related, 82 protective). LASSO-Cox regression refined this set to 46 proteins (31 risk, 15 protective), including five novel risk-related proteins—TEX101, MYBPC1, KIR2DL3, CLSTN2, and ADAMTS4—while the remaining 41 have been previously reported in the literature. Compared to the traditional model, which incorporated a novel high-risk status variable (defined as age ≥45 years plus at least one of: smoking history, heavy alcohol consumption, precancerous lesions, Helicobacter pylori infection, or an unhealthy diet), the integrated model demonstrated superior diagnostic performance, with significantly higher AUCs in both the training set (0.83 [95% CI: 0.80-0.86] vs. 0.70 [0.67-0.73]) and the testing set (0.81 [95% CI: 0.77-0.85] vs. 0.69 [0.65-0.74]; DeLong's test, p < 0.05). Conclusion: We developed an integrated model that combines 46 plasma protein biomarkers with traditional epidemiological factors. This approach demonstrated superior performance over the traditional model in identifying UGI cancer. Further prospective studies in diverse regions and populations are needed to validate the generalizability of this promising early screening strategy. Citation Format: Xinyu Liu, Zhangyan Lyu, Kexin Chen. A proteomics-based model improves the accuracy of upper gastrointestinal cancer prediction [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 6324.