Abstract 3024: Development of a plasma metabolomics prediction model for early detection of ovarian cancer using prospectively collected blood samples in two independent cohorts

Abstract

Abstract Background: Ovarian cancer has poor survival due to more than 60% of patients being diagnosed at advanced stage. Therefore, discovery of novel, non-invasive early detection biomarkers has high potential to improve outcomes. Here, using prospectively collected blood samples, we developed a model using plasma metabolite levels to discriminate ovarian cancer cases from controls and applied it to an independent dataset. Method: We examined 251 lipid and lipid-related metabolites measured using a liquid chromatography tandem mass spectrometry method (Broad Institute, Cambridge, MA) in plasma collected up to three years prior to ovarian cancer diagnosis and controls matched on age, blood collection characteristics, and menopausal status at blood draw and diagnosis in the Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial (PLCO; n=100, testing dataset) and the Nurses’ Health Studies (NHS; n=98, replication dataset). We used a paired t-test to identify metabolites with a significant difference (p<0.05) between ovarian cancer cases and controls. Then we used LASSO to identify metabolites that best discriminated ovarian cancer cases from controls in PLCO. We estimated the area under the receiver operating characteristic curve (AUC) and 95% confidence interval (CI) to assess model performance in discriminating ovarian cancer cases from controls when adding the selected metabolite score to a model based on CA125 using the clinical cutoff of 35U/mL, in PLCO and NHS. Results: Ten metabolites were significantly different between ovarian cancer cases and controls (p<0.05) in PLCO. LASSO selected 9 of the 10 metabolites into a prediction model including CA125 [i.e., C16:0 cholesteryl ester (CE), C22:6 CE, C30:0 phosphatidyl choline (PC), C32:1 PC, C40:10 PC, C38:2 phosphatidylethanolamine, C45:0 triglyceride, acetaminophen, 4-hydroxyhippurate]. In PLCO, adding a metabolite score derived from the 9 LASSO selected metabolites to a model with CA125 alone resulted in a significant (p<0.01) increase in AUC (95%CI) from 0.66(0.59-0.72) to 0.87(0.80-0.94). In NHS, adding the metabolite score to a model with CA125 alone also resulted in a significant (p=0.05) increase in AUC from 0.63(0.56-0.71) to 0.72(0.61-0.83). Conclusion: Using prospectively collected blood samples, we created a metabolite score which significantly increased the AUC compared to a model with CA125 alone and discriminated ovarian cancer cases from controls in the training dataset (PLCO). We replicated our findings in an independent test dataset (NHS). Further investigations assessing the performance of other statistical methods for replication (e.g., bootstrapping) and model building (e.g., random forest) are ongoing. These results suggest plasma metabolites represent promising novel early detection biomarkers for ovarian cancer. Citation Format: Oana A. Zeleznik, Julian Avila-Pacheco, Daniel W. Cramer, Britton Trabert, Clary B. Clish, Shelley Tworoger, Kathryn L. Terry, Naoko Sasamoto. Development of a plasma metabolomics prediction model for early detection of ovarian cancer using prospectively collected blood samples in two independent cohorts [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 3024.