Aaron M Praiss1, Yongmei Huang2, Caryn M St Clair3, Ana I Tergas4, Alexander Melamed3, Fady Khoury-Collado3, June Y Hou3, Jianhua Hu5, Chin Hur4, Dawn L Hershman4, Jason D Wright6. 1. Columbia University, Vagelos College of Physicians and Surgeons, United States of America; NewYork-Presbyterian Hospital, United States of America. 2. Columbia University, Vagelos College of Physicians and Surgeons, United States of America. 3. Columbia University, Vagelos College of Physicians and Surgeons, United States of America; Herbert Irving Comprehensive Cancer Center, United States of America; NewYork-Presbyterian Hospital, United States of America. 4. Columbia University, Vagelos College of Physicians and Surgeons, United States of America; Joseph L. Mailman School of Public Health, Columbia University, United States of America; Herbert Irving Comprehensive Cancer Center, United States of America; NewYork-Presbyterian Hospital, United States of America. 5. Columbia University, Vagelos College of Physicians and Surgeons, United States of America; Joseph L. Mailman School of Public Health, Columbia University, United States of America; Herbert Irving Comprehensive Cancer Center, United States of America. 6. Columbia University, Vagelos College of Physicians and Surgeons, United States of America; Herbert Irving Comprehensive Cancer Center, United States of America; NewYork-Presbyterian Hospital, United States of America. Electronic address: jw2459@columbia.edu.
Abstract
OBJECTIVE: We used a novel machine learning algorithm to develop a precision prognostication system for endometrial cancer. METHODS: The Ensemble Algorithm for Clustering Cancer Data (EACCD) unsupervised machine learning algorithm was applied to women with endometrioid endometrial cancer in the Surveillance, Epidemiology, and End Results database from 2004 to 2015. The prognostic system was created based on TNM stage, grade, and age. The concordance (C-index) was used to cut dendrograms and create prognostic groups. Kaplan-Meier cancer-specific survival was employed to visualize the survival function of EACCD-based prognostic groups and AJCC groups. RESULTS: A total of 46,773 women were identified. Using the machine learning algorithm with TNM stage, grade, and three age groups, eleven prognostic groups were generated with a C-index of 0.8380. The five-year survival rates for the eleven groups ranged from 37.9-99.8%. To simplify the classification system further, using visual inspection of the data we created a modified EACCD grouping, and combined the top six survival groups into three new prognostic groups. The new five-year survival rates for these eight modified prognostic groups included: 99.1% for group 1, 96.5% for group 2, 92.2% for group 3, 84.8% for group 4, 72.7% for group 5, 61.1% for group 6, 52.6% for group 7, and 37.9% for group 8. The C-index for the modified eight prognostic groups was 0.8313. CONCLUSION: This novel machine learning algorithm demonstrates improved prognostic prediction for patients with endometrial cancer. Using machine learning for endometrial cancer allows for the integration of multiple factors to develop a precision prognostication system.
OBJECTIVE: We used a novel machine learning algorithm to develop a precision prognostication system for endometrial cancer. METHODS: The Ensemble Algorithm for Clustering Cancer Data (EACCD) unsupervised machine learning algorithm was applied to women with endometrioid endometrial cancer in the Surveillance, Epidemiology, and End Results database from 2004 to 2015. The prognostic system was created based on TNM stage, grade, and age. The concordance (C-index) was used to cut dendrograms and create prognostic groups. Kaplan-Meier cancer-specific survival was employed to visualize the survival function of EACCD-based prognostic groups and AJCC groups. RESULTS: A total of 46,773 women were identified. Using the machine learning algorithm with TNM stage, grade, and three age groups, eleven prognostic groups were generated with a C-index of 0.8380. The five-year survival rates for the eleven groups ranged from 37.9-99.8%. To simplify the classification system further, using visual inspection of the data we created a modified EACCD grouping, and combined the top six survival groups into three new prognostic groups. The new five-year survival rates for these eight modified prognostic groups included: 99.1% for group 1, 96.5% for group 2, 92.2% for group 3, 84.8% for group 4, 72.7% for group 5, 61.1% for group 6, 52.6% for group 7, and 37.9% for group 8. The C-index for the modified eight prognostic groups was 0.8313. CONCLUSION: This novel machine learning algorithm demonstrates improved prognostic prediction for patients with endometrial cancer. Using machine learning for endometrial cancer allows for the integration of multiple factors to develop a precision prognostication system.
Authors: Matthew Hueman; Huan Wang; Zhenqiu Liu; Donald Henson; Cuong Nguyen; Dean Park; Li Sheng; Dechang Chen Journal: Thorac Cancer Date: 2021-03-13 Impact factor: 3.500
Authors: Philip M Grimley; Zhenqiu Liu; Kathleen M Darcy; Matthew T Hueman; Huan Wang; Li Sheng; Donald E Henson; Dechang Chen Journal: Acta Obstet Gynecol Scand Date: 2021-03-18 Impact factor: 4.544