Abstract Purpose: There is no consensus on screening for ovarian cancer in high-risk women because no effective biomarker or screening method exists, especially for early-stage cancer patients. The available methods, such as monitoring of CA125 levels, transvaginal ultrasound, and combination of pelvic examination are associated with high false-positive results, leading to additional unnecessary imaging tests, emotional burdens, and risk of invasive surgery for women without cancer. An effective, reliable, and non-invasive method for early detection of ovarian cancer is thus critically needed. In this research, weThus, we aim to identify a distinctive signature of exosomal small RNAs to serve as a novel biomarker for detecting early-stage ovarian cancer in high-risk population. Methods: Women of all ages and ethnical backgrounds were included in this study. Blood samples were collected in the clinic from women with germline BRCA mutations. A total of 30 patients with ovarian cancer before surgery, and 30 age-matched (mean age = 55) healthy women as controls were used. Among the 30 thirty cases, 11 patients had early (stage I) ovarian cancer. Identification of differential exosome small non-coding RNA copy numbers in plasma samples from BRCA1 mutation carriers with and without ovarian cancer was performed by next-generation RNA-Sseq. We randomly split 60 samples into two groups: training set (20 cancer, 20 normal) and , testing set (10 cancer, 10 normal). We apply 10-fold cross-validation to determine hyperparameter of five machine learning models and use the ensemble learning method to evaluate the predictive value of the small RNAs in the exosome by integrating the five machine learning models. Results: The ensemble learning model achieved the testing accuracy of 85% with the sensitivity of 70% and the specificity of 100% by using small RNA and CA125 abundance level as input features. While uUsing CA125 alone, the testing accuracy equals to 90% with the sensitivity of 80% and the specificity of 100%. For the 11 eleven early-stage patient samples, our ensemble learning model made 2 false negative predictions with a total accuracy of 95.1% (sensitivity=81.8%, specificity=100%) while using CA125 alone had five 5 false negative predictions with a total accuracy of 87.8% (sensitivity=54.5%, specificity=100%). All 30 control samples had been correctly classified. Conclusions: We found that ensemble learning model using combining both exosome small RNAs and CA125 will not improve the prediction accuracy for late-stage cancer patients compared to CA125 alone but will significantly improve the prediction accuracy for early-stage patients. Our research findings may indicate a two-pronged strategy in assessing ovarian cancer risks using existing CA125 test followed by small RNA features to detect early-stage ovarian cancer. We are now collecting a larger number of early-stage cancer patient samples for validation. Citation Format: Yuliang Cao, Cun Han, Jianting Sheng, Samuel Mok, Stephen T.C. Wong. Exosomal non-coding RNAs as a biomarker to improve early ovarian cancer detection [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 3333.
Read full abstract