Nuclear magnetic resonance-based metabolomics and risk of pancreatic cancer: a prospective analysis in the UK Biobank.

Abstract

Background: Plasma metabolite levels in patients with pancreatic cancer (PC) have changed, but the relationship between the altered plasma metabolites and the risk for PC occurrence is not fully clear, as well as the predictive value of the specific metabolites.

Methods: In this study, we obtained the metabolomics data of 243,145 people from the UK Biobank. An extreme gradient boosting (XGBoost) model, least absolute shrinkage and selection operator (Lasso) regression, and covariate-adjusted Cox proportional hazard regression models were used to evaluate the relationship between metabolites and PC risk. We also evaluated conventional risks, metabolites, and combination models for PC risk by comparing the area under the receiver operating characteristic curve (AUC).

Results: The average follow-up time was 13.8 (± 2.1) years; 1,026 of 243,145 participants developed PC. Fourteen metabolites were significantly associated with PC, including glucose-related metabolites, lipids, lipoproteins, and amino acids. Increased PC risk was associated with citrate, glucose, and the percentage of triglycerides to total lipids in intermediate-density lipoprotein or small low-density lipoprotein. Glycine, histidine, cholesterol, and cholesterol ester subclasses were associated with lower PC risk. Predicting PC risk improved when the newly identified metabolites were added to conventional PC risk factors (AUC: 0.705 vs 0.711, p = 0.037). The Kaplan-Meier cumulative incidence curves showed that these metabolites were good predictors of PC risk (all log-rank p < 0.05).

Conclusion: We identified novel metabolites that were significantly associated with the occurrence of PC, which may aid in the early diagnosis of PC.