New predictive model for treatment failure in chronic myeloid leukemia

A new model accurately predicts for which patients with chronic myeloid leukemia (CML) initial tyrosine kinase inhibitor (TKI) therapy will fail. This includes imatinib and second-generation TKIs. The model also offers physicians a new tool that may enable more personalized treatment approaches to CML according to a study published in Blood.¹

Developed by researchers in China, the model incorporates clinical covariates associated with TKI therapy failure to stratify patients into low-, intermediate-, and high-risk subgroups with significantly different cumulative incidences of treatment failure. The clinical covariates include sex, age, hemoglobin concentration, blood blast percentage, spleen size, and additional high-risk chromosomal abnormalities in Philadelphia chromosome–positive cells. The model was first developed using data from a single-center test cohort of 1955 patients with chronic-phase CML who were receiving as their initial treatment either a first-generation TKI (imatinib) or a second-generation TKI (nilotinib, dasatinib, or flumatinib), and it was then validated in 3454 patients from 76 other centers.

The model showed good predictive accuracy, as demonstrated by a high time-dependent area under the receiver operating characteristic curve (AUROC). Specifically, the model showed good prediction sensitivity and specificity with 1-, 3-, and 5-year AUROC scores of 0.83, 0.84, and 0.84, respectively (training set), and 0.77, 0.79, and 0.80, respectively (validation set). The AUROC scores range from 0 to 1, with 1 indicating a perfect performance and 0.5 indicating random guessing.

These AUROC values indicate better prediction discrimination than those offered by the Sokal and EUTOS long-term survival (ELTS) scores, the most widely used scores for guiding initial TKI therapy in patients with chronic-phase CML and predicting CML-related survival.

The authors say that they see their model as being used in conjunction with the Sokal and ELTS scores, not as a replacement, to further stratify patients to make risk assessment more precise.

“Using this model, physicians can better predict which patients are at high risk of therapy failure and make a more informed decision regarding the choice of the appropriate initial TKI,” says the senior author of the study, Qian Jiang, MD, professor and deputy chair of the Department of Hematology at Peking University People’s Hospital in Beijing, China.

Dr Jiang says that the model is ready for clinical use and that it can be readily applied in the clinical setting because the covariates on which the model is based are easily collected at the time of CML diagnosis.

Although the model is robust, he underscores that further validation is needed in Western populations to ensure its broad applicability and effectiveness across different demographic groups.