Survival lasso models and curves according to clinical features, mutational status, and chromosomal aberrations in the entire CBF-AML cohort. Optimized models for OS (A) and RFS (D) with a hazard ratio (HR) <1.0 indicating a lower risk and a HR >1.0 a higher risk of death and relapse, respectively. (A) The OS model has a cross-validation C-index of 0.63; 295 patients were included in the analysis [eg, patients with t(8;21) AML or trisomy 8 had the highest risk of death compared with patients with evidence of clonal heterogeneity or WT1 mutations who had a more favorable prognosis]. (D) The RFS model has a cross-validation C-index of 0.61; 261 patients were included in the analysis. KM estimates for OS (B,C) and RFS (E,F) given levels of competing predictors were generated using the original data set (B,E: apparent estimates) and a fivefold cross-validation (C,F). Because predictors represent continuous variables (eg, linear predictor of Cox models), survival curves were derived computing nearest-neighbor estimate of bivariate distribution of survival times and predictor levels. For illustration purposes, survival curves for low, medium, and high levels of the predictor were derived from the modified KM estimator, determined as the survival curve estimate for the neighborhood of the smallest, median, and largest values of the predictor, respectively. The modified KM estimator thereby reflects the discriminatory properties of the model.