Do We Need A CML-Specific Bleeding Tool? Concerns About The Proposed BAT-CML

To the Editor,

The recent systematic review by Gameil et al., “Evaluating the Safety of Combining Tyrosine Kinase Inhibitors with Anticoagulants in Chronic Myeloid Leukaemia: A Systematic Review”, was quite interesting. As chronic-phase CML has become a long-term condition for many patients, the practical question of how-to anticoagulated individuals receiving TKIs is increasingly relevant. The authors’ addressing a clinically prevalent dilemma in an area with limited prospective data, and their proposal of the “Bleeding Assessment Tool for CML (BAT-CML)” as a clinically useful risk classification tool, will inevitably be a significant topic of discussion. ¹

In the present review, the underlying evidence consists of two short pharmacokinetic studies conducted in in healthy volunteers and limited number of case reports, conference abstracts, and small retrospective series. These studies differ substantially in design, clinical indication (atrial fibrillation, venous thromboembolism or prophylaxis), type and dose of anticoagulant, TKI exposure, and baseline patient characteristics. Such heterogeneity, combined with the very small number of clinical events, does not provide a sufficient basis for developing a formal risk “tool” or assigning weighted risk factors. The authors acknowledge that the BAT-CML “requires testing and validation,” however, they present this scoring as a unique tool and by placing it alongside validated scores, they imply a level of clinical applicability that is not supported by current evidence.

The use of “Bleeding Assessment Tool (BAT)” in the scoring name causes significant concerns in terminology. “(BAT)” is strongly associated with the International Society on Thrombosis and Haemostasis (ISTH) standardized bleeding assessment framework. The ISTH-BAT was developed to provide a structured and reproducible method for documenting bleeding history and phenotypic severity in patients with suspected inherited or acquired bleeding disorders, rather than to predict future bleeding risk during anticoagulation therapy. ² Using the term “BAT” for an unvalidated and largely qualitative list of risk modifiers may therefore create conceptual confusion and blur the distinction between diagnostic bleeding assessment and prognostic risk modelling. As currently presented, BAT-CML is more accurately described as a preliminary checklist of potential risk factors. It lacks predefined outcomes, scoring thresholds, and performance metrics, and therefore does not meet the criteria of a formal assessment tool or prediction score.

Also, the proposed BAT-CML does not clearly demonstrate what it adds beyond established bleeding risk frameworks. Its included variables—such as advanced age, renal impairment, polypharmacy, critical illness, and advanced disease stage—are well-recognized predictors of bleeding in anticoagulated patients and are already incorporated into validated tools. For instance, the VTE-BLEED score includes age and renal dysfunction and has been validated for bleeding risk stratification in real-world VTE cohorts. Similarly, the RIETE score predicts early major bleeding using overlapping variables such as age, anaemia, cancer, and renal function, while HAS-BLED has also shown utility in identifying high-risk VTE patients. ³ If the goal is to justify a CML-specific prediction model, the review should specify which clinical decision the model is intended to change, and it should define candidate predictors that are truly specific to CML or TKI therapy rather than general comorbidity markers. As presented, BAT-CML lacks explicit thresholds, a reproducible scoring rule, and linkage to a prespecified outcome, which means it cannot be implemented, compared, or tested in a reliable way. ⁴

Additionally, evaluations made in the article about anticoagulant choices such as “apixaban emerged as the safest choice” and the suggestion that LMWH is a “safer option” are not supported by comparative evidence in CML studies. Although some observational and meta-analytic data in broader anticoagulated populations suggest lower major bleeding rates with apixaban compared with rivaroxaban, these findings are derived from indirect or non-randomized comparisons and are not specific to CML. Given the very small sample sizes, short follow-up, and likely publication bias in the reviewed CML literature, the absence of reported major bleeding cannot be interpreted as safety equivalence, nor can it reliably differentiate between agents. Similarly, the observation of “no thrombotic events” should be framed as insufficient evidence rather than proof of low thrombotic risk. ⁵ Especially in CML patients receiving nilotinib or ponatinib, linked to vascular events.

A major concern in this modelling study is that development of a clinical risk score requires a rigorous and transparent framework grounded in established prediction model methodology. This includes clear definition of the target population, outcome, and time horizon, along with systematic identification of candidate predictors within an adequately powered derivation cohort. Both internal (e.g., bootstrapping or cross-validation) and external validation are mandatory before clinical use. Without these safeguards, proposed scores should be considered exploratory rather than ready for implementation.

In summary, individualized clinical judgment remains the cornerstone of anticoagulation management in CML. The available data are insufficient to support the adoption of a CML-specific bleeding score. BAT-CML should be regarded as a conceptual framework rather than a clinically applicable tool. Definitive guidance will require prospective, multicentre studies with rigorous methodology and validated outcomes.