Fig. 2: Cumulative incidence of therapy failure for the three risk models and ELTS score.

The cumulative incidence of treatment failure was calculated using the predictive score developed by Zhang et al. (Model 1). Model 1 was applied to the group of 601 patients selected from the CML Observatory using the selection criteria described by Zhang et al. (A) and also to the same patients after their classification into two groups: first-line (1 L) imatinib (B) and first-line 2G-TKI (C). Then, the model was reconstructed using the procedure detailed by Zhang et al. (Supplementary Information) using the six key variables identified in their study (hemoglobin, blasts, age, sex, high-risk ACAs, spleen size) (Model 2). This new model was used to calculate the risk of failure in the whole cohort (D), and in the first-line imatinib (E) and first-line 2G-TKI groups (F). In Model 3, from the initial twelve candidate variables used by Zhang et al., four variables (sex, high-risk, leukocytosis, and comorbidities) were selected to optimize the model fit. This revised model was applied to the whole cohort (G), and to the first-line imatinib (H) and first-line 2G-TKI groups (I). In parallel, the ELTS score was calculated in the whole cohort (J), first-line imatinib group (K), and first-line 2G-TKI group (L). P-values were computed with the omnibus Gray test.