A web-based prediction model for overall survival of elderly patients with epithelial ovarian cancer: A population-based study.

Abstract

e17570 Background: Epithelial ovarian cancer (EOC) has been extensively studied. However, no prediction model has been carried out on the prognosis of elderly patients. Our study aims to explore survival patterns in elderly EOC patients. Methods: Data were extracted from the Surveillance, Epidemiology, and End Results (SEER) database. Patients over 65 with EOC between 2004 and 2014 were included for model training and validation, while 2015 for external validation. COX proportional-hazards regression model was used to identify risk factors and applied them to the prediction model. We then evaluated performance of our model through receiver operating characteristic (ROC) curve, calibration curve and decision curve analysis (DCA). Risk stratification system was used to reflect the survival. To facilitate daily use, we developed an online application. Results: A total of 6141 patients were included. COX regression models indicated 10 independent risk factors. Our model was proved to have good performance with AUC of ROC curve: 0.768(95CI:0.749-0.788), 0.748 (95CI:0.733-0.763), 0.766 (95CI:0.750-0.781) for 1-, 3-, 5-year OS in the training cohort and 0.776 (95CI:0.757-0.796), 0.760 (95CI:0.745-0.775), 0.784 (95CI:0.769-0.799) for CSS. In the validation cohort, 0.800 (95CI:0.773-0.827), 0.770 (95CI:0.748-0.793), 0.787 (95CI:0.765-0.810) for OS and 0.808 (95CI:0.780-0.836), 0.780 (95CI:0.758-0.803), 0.801 (95CI:0.779-0.824) for CSS. In the risk stratification system, the low-risk group had the highest survival rate. Almost all patients in the low- and intermediate-risk groups underwent surgery, about 1/3 patients in the high-risk group did not receive surgery. Their survival rates were not optimistic, and patients over 85 were more likely to be classified as high-risk. Conclusions: A predictive model was constructed to evaluate survival for elderly EOC patients and demonstrated good predictive value that we can apply to help physicians make clinical decisions and plan treatment. [Table: see text]