Prognosticating outcome in pancreatic adenocarcinoma with the use of a Machine Learning Algorithm

By: Zarrukh Baig, Dr. Nawaf Abu-Omar, Rayyan Khan, Carlos Verdiales Castro, Ryan Frehlick, Dr. Fangxiang Wu, & Dr. Yigang Luo

Objectives: 1. To identify prognostic factors that predict survival < 2 years in patients undergoing Whipples Surgery for pancreatic cancer. 2. To create a Support Vector Model (SVM) machine learning algorithm that utilizes prognostic factors to predict survival for patients undergoing Whipples Surgery for pancreatic cancer.

Pancreatic cancer is a lethal condition with a poor prognosis. Pancreaticoduodenectomy, also known as the Whipple procedure, is the treatment of choice in patients who have a resectable tumour, which is associated with severe morbidity. The purpose of this project was to identify prognostic features in resectable pancreatic head adenocarcinoma and use these features to develop a machine learning algorithm that prognosticates survival post-Whipples. A retrospective chart review of 93 patients who underwent a Whipples procedure was performed. The patients were analyzed in 2 groups: Group 1 (n=38) comprised of patients who survived < 2 years and Group 2 (n=55) comprised of patients who survived > 2 years. Using a student’s t-test, chi-squared analysis, and a recursive feature elimination test 10 categorical features and 2 continuous features (12 total) were selected from 50 to be statistically significantly (p<0.05) in prognosticating survival < 2 years. These include: 1) presence of type 2 diabetes, 2) family history of any type of cancer, 3) number of family members with cancer, 4) bile duct stricturing, 5) perineural involvement of tumor, 6) positive margins on resection, 7) resection of portal tissue, 8) tumor size on MRI, 9) tumor size on pathology, 10) neoadjuvant chemotherapy, 11) adjuvant chemotherapy and 12) recurrence of cancer. These 12 features were used to train a Support Vector Model (SVM) algorithm. The algorithm was created using 90% of the data and validated using 10% testing data. The algorithm obtained 78% accuracy, 56.83% sensitivity, and 91.94% specificity in predicting survival < 2 years using the 12 prognosticating factors. A machine learning algorithm that prognosticates survival can be a useful tool to individualize treatment plans for patients with pancreatic cancer and prevent futile surgery.

Share this story