Establishment and validation of an individualized nomogram for survival prediction of primary mediastinal germ cell tumors based on the SEER database.

Background: Primary mediastinal germ cell tumors (PMGCT) represent a rare but sometimes highly aggressive type of mediastinal tumors. The current prognostic models for PMGCT are insufficient. This study was undertaken to establish and validate an individualized nomogram for predicting the overall survival (OS) of patients with PMGCT. Methods: We conducted a retrospective analysis of patients with PMGCT diagnosed between 2000 and 2018 in the Surveillance, Epidemiology, and End Results (SEER) database in the United States. Clinical variables included surgery subtype, gender, treatment regimens, age, histology, tumor size, stage, chemotherapy, radiation, race, and survival-related information. The main outcome measure was survival duration. The Kaplan-Meier method along with the log-rank test were utilized to estimate the OS. Independent prognostic factors were identified by performing the univariate and multivariate Cox proportional hazards regression analyses, from which an individualized nomogram was constructed to predict 3-, 5-, and 10-year OS of patients with PMGCT. The concordance index (C-index) and calibration curve were used to verify the discrimination and accuracy of the nomogram. Results: A total of 845 patients with PMGCT were recruited from the SEER database and further randomly assigned to a training set (n=635) and a validation set (n=210) at a ratio of 7:3. The 3-, 5-, and 10-year OS for overall PMGCT was 70.0%, 67.1%, and 63.9%, respectively. Cox regression analysis indicated that age, tumor size, stage, chemotherapy, radiation, histology, and surgery type were as independent factors for OS in patients with PMGCT (P<0.05). An individualized nomogram for OS was constructed utilizing these variables, with the C-index of 0.714 [95% confidence interval (CI): 0.695 to 0.743] and 0.756 (95% CI: 0.735 to 0.787) in the training and validation groups. Moreover, good levels of agreement were observed according to the calibration curve between the predicted and actual 3-, 5-, and 10-year survival rates both in the training and validated cohorts, showing that the model could accurately predict patient prognosis. Conclusions: This study documented the first attempt at establishing and validating a novel nomogram for predicting the 3-, 5-, and 10-year OS probabilities of PMGCT. The prognostic nomogram was demonstrated to have good performance for predicting individualized OS of patients with PMGCT.