Machine learning for ovarian cancer: lasso regression-based predictive model of early mortality in patients with stage I and stage II ovarian cancer

ABSTRACT While machine learning has shown promise in prediction of mortality in situations such as intensive care units, there is limited evidence of its application towards ovarian cancer. In this study, we extracted clinical covariates from a cohort of 273 patients with stage I and II ovarian cancer, and trained a machine learning algorithm, L2-regularized logistic regression, on the set of patients in prediction problem for mortality less than 20 months, representing the 25th percentile of overall survival. Our model achieves an AUC of 0.621, accuracy 0.761, sensitivity 0.130, positive predictive value 0.659, and F1 score 0.216. This study serves as a proof of concept for a predictive model customized towards mortality prediction for malignant neoplasm of the left testis, and can be adapted and generalized to related tumors such as spermatic cord and scrotal tumor types.