Validation of a New Clinical-Genetic Recursive Partitioning Analysis of High-Grade Glioma Using RTOG 0525, 0513 and 0131

• Published in: International journal of radiation oncology, biology, physics Vol. 111; no. 3; p. e601
• Main Authors: Sutera, P., Flickinger Sr, J.C., Wang, H., Heron, D.E.
• Format: Journal Article
• Language: English
• Published: Elsevier Inc 01.11.2021
• ISSN: 0360-3016; 1879-355X
• DOI: 10.1016/j.ijrobp.2021.07.1605

Despite advances in our understanding/treatment of patients with high grade glioma, the current model for prognostication is based on 40-year-old data. We previously generated a new model incorporating clinical and genetic factors utilizing data from patients treated from 2004-2017. Here we sought to validate our prognostic model within data from prospective trials and compare its accuracy against the original RTOG-RPA. Data from RTOG 0525, RTOG 0513 and RTOG BR-0131 was requested through an NRG Ancillary Project application. Subjects were categorized by both RTOG RPA survival class and our new-RPA survival class. We generated Kaplan-Meier survival curves for classes from each model. To compare the accuracy and heterogeneity of each model, we calculated the mean prediction errors and interquartile ranges (IQR) of overall survival (OS). We then compared the two models for their predictive ability and homogeneity within each survival class. 960 patients were included in our validation dataset. Using the validation dataset, Kaplan-Meier survival curves for each terminal class were plotted. Log rank tests demonstrated no significant differences between the predicted and observed survival curves of the new RPA for Class 1 (P = 0.39), Class 2 (P = 0.59), Class 4 (P = 0.75), Class 5 (P = 0.18), Class 6 (P = 0.50), and Class 3/5 (P = 0.12). Mean prediction error of median survival was 4.5 months vs 1.9 months for the RTOG- and new-RPA, respectively. Mean Kaplan-Meier OS IQR was 38.1 months and 17.8 months for the RTOG- and new-RPA, respectively. We demonstrated that both the RTOG-RPA and our new-RPA classes maintain their relative prognostic significance. Although both models generate 6 distinct survival classes, in this partial validation our updated model more accurately predicts median survival for each class with significantly greater class homogeneity. Additionally, our proposed model utilizes exclusively objective measures decreasing the risk of inter-observer classification variability. P. Sutera: None. J.C. Flickinger: None. H. Wang: None. D.E. Heron: None.