Validation and optimization of a web-based nomogram for predicting survival of patients with newly diagnosed glioblastoma

Purpose: To optimize and validate a current (NRG [a newly constituted National Clinical Trials Network group through National Surgical Adjuvant Breast and Bowel Project [NSABP], the Radiation Therapy Oncology Group [RTOG] and the Gynecologic Oncology Group (GOG)]) nomogram for glioblastoma patients as part of continuous validation.

Methods: We identified patients newly diagnosed with glioblastoma who were treated with temozolomide-based chemoradiotherapy between 2006 and 2016 at three large-volume hospitals. The extent of resection was determined via postoperative MRI. The discrimination and calibration abilities of the prediction algorithm were assessed; if additional factors were identified as independent prognostic factors, updated models were developed using the data from two hospitals and were externally validated using the third hospital. Models were internally validated using cross-validation and bootstrapping.

Results: A total of 837 patients met the eligibility criteria. The median overall survival (OS) was 20.0 (95% CI 18.5-21.5) months. The original nomogram was able to estimate the 6‑, 12-, and 24-month OS probabilities, but it slightly underestimated the OS values. In multivariable Cox regression analysis, MRI-defined total resection had a greater impact on OS than that shown by the original nomogram, and two additional factors-IDH1 mutation and tumor contacting subventricular zone-were newly identified as independent prognostic values. An updated nomogram incorporating these new variables outperformed the original nomogram (C-index at 6, 12, 24, and 36 months: 0.728, 0.688, 0.688, and 0.685, respectively) and was well calibrated. External validation using an independent cohort showed C‑indices of 0.787, 0.751, 0.719, and 0.702 at 6, 12, 24, and 36 months, respectively, and was well calibrated.

Conclusion: An updated and validated nomogram incorporating the contemporary parameters can estimate individual survival outcomes in patients with glioblastoma with better accuracy.