Jonathan Douissard, MD1, Frederic Ris, MD1, Minia Hellan, MD2, James Ouellette, MD2, Thibaud Koessler, MD1, Nicolas C Buchs, MD1, Leo Bueler, MD, PhD1, Frederic Triponez, MD, PhD1, Christian Toso, MD, PhD1. 1University Hospital of Geneva, Switzerland, 2Wright State University, Ohio, USA
OBJECTIVE OF THE TECHNOLOGY: Standard screening recommendations are available for only a few cancers and most malignancies are discovered through incidental findings or after the presentation of symptoms. Early detection could lead to higher rates of surgical resection with curative intent and should increase overall treatment success. Previous research has shown that circulating tumor DNA analyses in the blood (liquid biopsy) could lower the threshold for cancer discovery before initial treatment, as well as assess the efficacy of treatment, and detect eventual recurrence.
METHODS: 10 ml of peripheral blood was drawn from newly diagnosed patients with pancreatic or colorectal cancer, and a control cohort of patients without a known malignancy. DNA was extracted through a custom DNA extraction protocol using Qiagen circulating free nucleic acid extraction kits with 4 ml of plasma. Optimized next generation sequencing (ONGS) was applied to detect elevations of pre-selected mutations that have been described to be associated with cancer. Raw data was analyzed using 2 different machine learning methods: Method 1 used 16 unblinded initial samples (including cancer and non-cancer samples) to train algorithm. The algorithm was then applied in a blinded fashion to 24 additional samples. Sensitivity, specificity and accuracy of the algorithm in detecting cancer mutation patterns of these 24 samples were assessed. In method 2, after unmasking all of the clinical data, machine learning was tested for its ability to discriminate between cancer and non-cancer samples.
PRELIMINARY RESULTS: The pancreatic cancer cohort included 5 patients: 1 early intra-ductal carcinoma, 1 stage II adenocarcinoma, 2 stage III adenocarcinoma and 1 stage IV adenocarcinoma. The colorectal cancer cohort included 6 patients with 1 stage I, 3 stage II and 2 stage III adenocarcinomas. 29 participants without cancer diagnosis were included in the control group. The predictive machine learning algorithm differentiated between cancer and non-cancer samples at an overall accuracy of 0.9583, a specificity of 1 and a sensitivity of 0,74. After unmasking the clinical diagnoses, the machine learning algorithm showed a 100% accuracy in distinguishing cancer patterns from non-cancer patterns.
CONCLUSIONS AND FUTURE DIRECTIONS: Despite a very limited data set, this new approach which combines custom DNA extraction, ONGS and machine learning appears to deliver a clinically relevant accuracy for detecting colorectal and pancreatic cancer through a single peripheral blood sample. As such, it might be a useful tool for early and accurate cancer detection. Furthermore, the ease of use of this minimally invasive procedure could make it a powerful method to monitor treatment response including completeness of surgical resection and early detection of disease recurrence. More clinical data is needed to determine the precise role of this approach in cancer care.
Presented at the SAGES 2017 Annual Meeting in Houston, TX.
Abstract ID: 98921
Program Number: ETP729
Presentation Session: Emerging Technology Poster Session (Non CME)
Presentation Type: Poster
sages_adbutler_leaderboard