A risk classification tree model identified a subgroup of ultra-high-risk patients with the highest risk of experiencing therapy failure and disease progression. (A) A classification tree based on Sokal risk score and the BIM and ASXL1 variant genotypes to predict 48-month FFS was generated by recursive partitioning using the Rpart package. The final classification tree produced 7 terminal nodes, which were readily distinguished into 4 risk categories; favorable, low, high and ultra-high. Each node box displays the relative risk of the node compared with the whole population, the number of events/sample size at that node, and the percentage of observations used at that node. (B) Kaplan-Meier FFS survival plot based on the classification tree risk categories. (C-F) Cumulative incidence of (C) EMR, (D) MMR, (E) MR⁴, and (F) MR^4.5 as stratified by the classification tree risk groups, as defined in panel A. CI, 95% confidence interval.