310P - A study on the prediction of recurrence site of endometrial cancer using various machine learning techniques

Background

This study investigated site-specific differences in clinical factors for recurrence in patients who were newly diagnosed and treated for endometrial cancer. Several machine learning algorithms were adapted to predict the recurrence of patients.

Methods

Electronic medical records’ data were retrieved from January 2006 to December 2018 for patients who were diagnosed with endometrial cancer at the XXX in Korea. Recurrence sites were classified as local, regional, or distant. We employed various machine learning algorithms, including logistic regression models (LR), random forest (RF), support vector machine (SVM) and artificial neural network (ANN), and assessed their prediction performances by cross-validation. Since our problem is an imbalanced multi-classification problem, the average score of AUC (area under curve) for each class obtained from one-vs-rest strategy was used for evaluating each machine learning algorithm.

Results

The data of 611 patients were selected for analysis; there were 20, 12, and 25 local, regional, and distant recurrence, respectively, and 554 patients had no recurrence. Random forest showed the best performance (0.8587) in prediction accuracy. Other algorithms followed with 0.7790 (LR), 0.7398 (ANN) and SVM (0.7119). The most important variables in Random Forest were invasion depth, age and size, in order.

Conclusions

We identified different risk factors specific for each type of recurrence site. Using these risk factors, we suggest that individually tailored adjuvant treatments be introduced for patients.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

The authors.

Funding

Has not received any funding.

Disclosure

All authors have declared no conflicts of interest.