Joeky Senders

97 Glioblastoma survival calculator TABLE 2. Discriminatory performance for all time-to-event, continuous, and binary survival models according to the (integrated) concordance index. C-index (95%-CI) Integrated C-index Overall survival 1Y-survival status Time-to-event Models Accelerated Failure Time 0.70 (0.70-0.70) 0.70 (0.70-0.70) 0.70 (0.70-0.70) CPHR 0.69 (0.69-0.70) 0.69 (0.69-0.70) 0.69 (0.69-0.70) Boosted Decision Tree Survival 0.69 (0.69-0.70) 0.69 (0.69-0.70) 0.69 (0.69-0.70) Random Forest Survival 0.68 (0.68-0.68) 0.69 (0.69-0.69) 0.68 (0.68-0.68) Recursive Partitioning 0.68 (0.68-0.68) 0.68 (0.68-0.68) 0.68 (0.68-0.68) Continuous and binary Models Gradient Boosting 0.70 (0.70-0.70) 0.70 (0.70-0.70) NA Regularized GLM 0.70 (0.70-0.70) 0.70 (0.70-0.70) NA GLM 0.70 (0.70-0.70) 0.70 (0.70-0.70) NA Support Vector Machines 0.70 (0.70-0.70) 0.69 (0.69-0.69) NA Multilayer Perceptron 0.61 (0.61-0.61) 0.69 (0.69-0.69) NA Naïve Bayes a NA 0.69 (0.69-0.69) NA Random Forest 0.69 (0.69-0.69) 0.69 (0.69-0.69) NA Extreme Gradient Boosting 0.68 (0.68-0.68) 0.68 (0.68-0.68) NA K-Nearest Neighbors 0.67 (0.67-0.67) 0.68 (0.67-0.68) NA Bagging 0.67 (0.66-0.67) 0.66 (0.66-0.66) NA Abbreviations: 1Y=one year; C-index=concordance index; CI=confidence interval; CPHR=cox proportional hazards regression; GLM=generalized linear models; NA=not available a Naïve Bayes fits to categorical data only. Secondary metrics Secondary metrics related to model deployment and clinical implementation varied across all models (Table 3). AFT, CPHR, and (regularized) generalized linear models were the only models with inferential utility. AFT, CPHR, boosted decision trees survival, recursive partitioning, and random forest survival were the only models that can analyze time-to-event data and thus compute subject-level survival curves. The application loading time varied between 0.2 seconds and 45 minutes. The 100-fold bootstrapped prediction time varied between 1.9 seconds and four minutes on a single central processing unit.