Development and validation of a deep learning model for individualized survival prediction in advanced cervical...

Clinical staging remains the most commonly used systems for advanced cervical cancer in clinical practice and cannot provide satisfactory prognostication. Deep learning model could capture complex and nonlinear relationships of data, which has not been fully explored in cancer survival analyses. This is the first study to take advantage of the advanced cervical cancer database as well as a deep learning method to develop a novel prognostic model for such patients.

1,143 advanced cervical cancer patients from inhouse (Qilu Hospital of Shandong University and Shandong First Medical University Affiliated Cancer Hospital) were enrolled in this study. The patients from inhouse were randomly split into a training set (n = 914, 80%) and testing set (n = 229, 20%). 49 selected epidemiologic, clinical and hematologic variables were used for model establishment. We developed a deep survival learning model (DSLM) and tested it for individualized survival prediction. Finally, a novel personalized prognostic model was explored. We also compared DLSM model with four other baseline models (CPH, RSF, LMTM and SVM) to prove DLSM is better. In addition, we selected 3495 patients in the SEER database for external verification.

The novel prognostic model based on DSLM showed better performances than the conventional clinical staging system (AUROC of 0.8603 and 0.5403, respectively). Personalized survival curves for advanced cervical cancer patients also showed notably different survival slopes.

Our study developed a novel, practical, personalized prognostic model for advanced cervical cancer, this individualized survival prediction system may also help clinicians to design more balanced and reasonable clinical trials.