Elderly patients with newly diagnosed glioblastoma: can preoperative imaging descriptors improve the predictive power of a survival model?

The purpose of this study was to identify independent prognostic factors among preoperative imaging features in elderly glioblastoma patients and to evaluate whether these imaging features, in addition to clinical features, could enhance the predictive power of survival models. This retrospective study included 108 patients ≥65 years of age with newly diagnosed glioblastoma. Preoperative clinical features (age and KPS), postoperative clinical features (extent of surgery and postoperative treatment), and preoperative MRI features were assessed. Univariate and multivariate cox proportional hazards regression analyses for overall survival were performed. The integrated area under the receiver operating characteristic curve (iAUC) was calculated to evaluate the added value of imaging features in the survival model. External validation was independently performed with 40 additional patients ≥65 years of age with newly diagnosed glioblastoma. Eloquent area involvement, multifocality, and ependymal involvement on preoperative MRI as well as clinical features including age, preoperative KPS, extent of resection, and postoperative treatment were significantly associated with overall survival on univariate Cox regression. On multivariate analysis, extent of resection and ependymal involvement were independently associated with overall survival and preoperative KPS showed borderline significance. The model with both preoperative clinical and imaging features showed improved prediction of overall survival compared to the model with preoperative clinical features (iAUC, 0.670 vs. 0.600, difference 0.066, 95% CI 0.021–0.121). Analysis of the validation set yielded similar results (iAUC, 0.790 vs. 0.670, difference 0.123, 95% CI 0.021–0.260), externally validating this observation. Preoperative imaging features, including eloquent area involvement, multifocality, and ependymal involvement, in addition to clinical features, can improve the predictive power for overall survival in elderly glioblastoma patients.