Survival Prediction Models for Ovarian Cancer Patients with Lung Metastasis: A Retrospective Cohort Study Based on SEER Database

Abstract

To develop a random forest prediction model for the and short- and long-term survival of ovarian cancer patients with lung metastasis. This retrospective cohort study enrolled primary ovarian cancer patients with lung metastasis from the surveillance, epidemiology and end results (SEER) database (2010–2015). All eligible women were randomly divided into the training (n = 1357) and testing set (n = 582). The outcomes were 1-, 3- and 5-year survival. Predictive factors were screened by random forest analysis. The prediction models for predicting the 1-, 3- and 5-year survival were conducted using the training set, and the internal validation was carried out by the testing set. The performance of the models was evaluated with area under the curve (AUC), accuracy, sensitivity, specificity, negative predictive value (NPV), and positive predictive value (PPV). The subgroups based on the pathological classification further assessed the model’s performance. Totally 1345 patients suffered from death within 5 years. The median follow-up was 7.00 (1.00, 21.00) months. Age at diagnosis, race, marital status, tumor size, tumor grade, TNM stage, brain metastasis, liver metastasis, bone metastasis, etc. were predictors. The AUCs of the prediction model for the 1-, 3-, 5-year survival in the testing set were 0.849 [95% confidence interval (CI): 0.820–0.884], 0.789 (95% CI 0.753–0.826) and 0.763 (95% CI 0.723–0.802), respectively. The results of subgroups on different pathological classifications showed that the AUCs of the model were over 0.7. This random forest model performed well predictive ability for the short- and long-term survival of ovarian cancer patients with lung metastasis, which may be beneficial to identify high-risk individuals for intelligent medical services.