Joeky Senders

103 Glioblastoma survival calculator academic realm, 29 lack of interpretability and computational inefficiency hinders their deployment in the clinical realm. When evaluating models for clinical deployment, we recommend evaluating fitted models on several criteria rather than a singular focus on prediction performance since factors unrelated to prediction performance (such as interpretability or applicability) can exclude high-performing models from clinical deployment. Although the AFT model was selected due its high overall performance, the difference in prediction performance was not always clinically meaningful, thereby emphasizing the importance of taking into account these secondary metrics as well. Furthermore, the prediction performance can change as the number and nature of the input features change. For example, the assembly of multimodal data including radiogenomics data might call for alternative analytical approaches in the near future. Prognostication is and always has been aimed at a moving target and future factors impacting clinical course cannot bemodeled, most importantly advances in clinical care. Prediction performance therefore remains an asymptotic ideal for which perfection will never be reached. Future research should focus on developing clinically meaningful and interpretable prediction tools. Improving the end-user transparency regarding the underlying predictive mechanisms and the inherent limitations allows for a safe and reliable implementation of survival prediction tools in clinical care. Conclusion This study provides a framework for the development of survival prediction tools in cancer patients, as well as an online calculator for predicting survival in glioblastoma patients. Future efforts should focus on developing additional algorithms that can train on right-censored survival data, improve the granularity of population-based registries, and externally validate the proposed prediction tool. Supplementary material Supplementary tables and figures available online at: https://academic.oup.com/neurosurgery/article/86/2/E184/5581744#supplementary - data