Cholangiocarcinoma survival prediction using machine learning algorithms


Year:

Session type:

Theme:

Mohamed Osman1
1Faculty of Medicine, Zagazig University, Egypt

Abstract

Background

Cholangiocarcinoma is the most common malignancy of biliary tract and the second most common primary hepatic malignancy. The incidence and mortality of cholangiocarcinoma are increasing worldwide. Predicting cholangiocarcinoma survival is difficult due to different primary sites, treatments and undefined risk factors. Reliable predictions can help in personalizing care and good treatment and management. Here, we test the ability of machine learning to predict survival of cholangiocarcinoma.

Method

Patients with cholangiocarcinoma were identified through the Surveillance, Epidemiology and End Results database from 2010-2013. Patients’ data were extracted including: age, sex, race, primary site, TNM stage, grade, size, extension, lymph nodes, metastasis, cancer sequence number, surgery, radiotherapy, chemotherapy, radiation sequence with surgery, state, county and survival months. Data were randomly divided into; a training set (80%) and a validation set (20%). Machine learning algorithms were used to predict survival.

Results

A total number of 1,095 patients were identified with a median survival of 15 months. The most common primary tumor sites were intrahepatic bile duct (52.6%), extrahepatic bile duct (27.7%), and liver (16.3%). Random Forests algorithm achieved better results compared to other tested models. For evaluating model performance, the Area Under the Receiver Operating Characteristic Curve (AUC) was calculated. Random Forests yielded AUCs of 81.5% at 6 months, 87.1% at 12 months and 80.6% at 24 months. Sensitivity of the trained model were 83%, 80% and 76% at 6, 12 and 24 months, respectively. The model achieved an average accuracy of 82.6%, 80.4% and 75.8% at 6, 12 and 24 months, respectively. The most important characteristics which influenced the model performance were: surgery, age, and tumor size.

Conclusion

Supervised machine learning algorithms achieved a good performance of predicting survival of patients with cholangiocarcinoma. Improved performance in survival prediction can help in making better treatment decisions and planning social and care needs.