Outcome predictions for AML. For development of a methylation-based outcome predictor using SuperPC, the 182 AML samples were separated into a training set (n = 89) and an independent test set (n = 93). (A) Kaplan-Meier survival analysis of the training set with the good (black line) and poor (red line) outcome group showing a significant difference in overall survival (P < .01; log-rank test). (B) The same analysis for the test sample set, where the good prognosis group was still associated with significantly longer survival times compared with the poor prognosis group (P = .028; log-rank test). (C) Kaplan-Meier analysis of the predictor using an additional independent set of 74 AML samples. For this validation set, predicted good and poor outcome classes were again significantly correlated with patient survival (P < .001; log-rank test). (D-F) A comparison of the outcome predictors based on methylation data (MP; D) and on gene expression data (EP; E). (F) Insight on the concordance of the methylation-based and expression-based predictor. The samples representing the black line belong to an outcome group (group GG) where both methods predict good survival. For the red line, both predictors label samples with poor survival outcome (group PP). For samples on the blue and green lines at least one method (expression or methylation) predicted poor outcome. In detail, the green line represents samples where MP predicted good survival, whereas EP predicted poor survival (group GP), and the blue line represents samples where MP predicted poor survival and EP predicted good survival (group PG). For both of these sample groups with discordant survival prediction, the probability of survival was in fact markedly inferior to the predictions providing concordant results in both models.