Comparison between our gene expression-based classification and other prognostic models. (A) A forest plot of odds ratios for detecting genetic lesions in the IMP subgroup. P values are based on Fisher’s exact test. The square sizes were inversely proportional to the confidence intervals of the estimated odds ratio. (B) A boxplot shows distribution of log odds ratio for the predicted IMP subgroup that was calculated based on our regression model using expression levels of 9 genes. Boxes correspond to healthy adults (n = 3), patients with MDS or myelodysplastic/myeloproliferative neoplasm (MDS/MPN; n = 107), those with AML with myelodysplasia-related changes (AML-MDS; n = 7), and those in a published external AML cohort from The Cancer Genome Atlas (n = 179). *P < .05 (Bonferroni adjusted); **P < .01 (Bonferroni adjusted); ***P < .001 (Bonferroni adjusted). (C) HR of death or leukemic transformation comparing the EMK and IMP subgroups. Number of patients in the EMK and IMP subgroup, HR, its 95% CI, and P values are shown for each IPSS-R risk category. (D) Cumulative incidence of leukemic transformation in MDS patients stratified by categories of IPSS-R.