Fig. 6.

Original random survival forest approach with feature selection/engineering outperforms other common modeling approaches. (A) Original random survival forest EFS predictions for the test cohort versus random survival forest with recursive feature elimination and multivariable Cox proportional hazards models of EFS for the same cohort. (B) Test cohort EFS predictions for a random survival forest generated using current gold-standard prognostic variables and variable importance (assessed by variable permutation) in that model.