Introduction

Direct oral anticoagulants (DOACs) are the standard treatment in individuals with atrial fibrillation (AF), and are increasingly used also in cancer patients with this common arrhythmia.^1,2

The risk of AF is relatively low among breast cancer patients; however, the adjuvant hormonal therapy can interact with DOACs, increasing the risk of bleeding.³ Evaluation of bleeding risk in cancer patients with AF treated with DOACs is challenging.⁴ The HAS‑BLED score was suggested for this purpose in the 2022 European Society of Cardiology (ESC) guidelines on cardio‑oncology,⁵ in contrast to the 2024 ESC AF guidelines, which recommended focusing on the assessment of individual risk factors for bleeding.⁶ In 2023, the DOAC score, based on factors such as age, creatinine clearance, underweight, comorbidities, bleeding history, and drug interactions, was validated to predict bleeding in individuals receiving DOACs.⁷ Mei et al⁸ analyzed a cohort of patients with AF treated with DOACs, 8.8% of whom were cancer patients, and showed that the HAS‑BLED score outperformed the DOAC score in terms of identifying individuals at a very low risk of bleeding; however, reclassification analysis failed to demonstrate any difference between the 2 scores.

Given the growing interest in biomarkers as predictors of bleeding in AF, the Age / Biomarkers / Clinical history (ABC)-bleeding score has been developed, which incorporates age, biomarker levels (growth differentiation factor‑15 [GDF‑15], high‑sensitivity cardiac troponin T [hs‑cTnT], hemoglobin), and bleeding history. It has been reported that this scoring system may outperform the HAS‑BLED score,⁹ although data on specific cancer populations are limited.¹⁰ In their post hoc analysis of the AVERT (Apixaban to Prevent Venous Thromboembolism in Patients with Cancer) trial, predominantly including patients with gynecologic cancer (25.8%) and lymphoma (25.3%), Mulder et al¹¹ showed that GDF‑15 alone could predict bleeding among oncologic patients treated with apixaban.

To our knowledge, no previous study evaluated the predictive value of biomarkers and bleeding scales (including the DOAC score) specifically in breast cancer patients. Therefore, we investigated this issue in a cohort of women with this malignancy and AF.

Patients and methods

We studied 60 consecutive adult outpatients with breast cancer and AF diagnosed for the first time, who were referred to our center between 2015 and 2020 and were treated with DOACs. The study population was presented previously.³ Briefly, the patients were eligible if they were on hormonal therapy including tamoxifen, aromatase inhibitors, or sequential tamoxifen for 2–3 years, followed by 2–3 years of aromatase inhibitors. The key exclusion criteria were heart failure (New York Heart Association class III–IV), prior stroke, recent myocardial infarction, venous thromboembolism, presence of any valve prosthesis, prior major bleeding, platelet count below 100 × 10³/μl, estimated glomerular filtration rate (eGFR) below 50 ml/min/1.73 m², and significant drug–drug interactions, particularly with carbamazepine and azole antifungal drugs.

AF diagnosis and classification were performed according to the ESC guidelines.¹

The study protocol was approved by the ethics committee at the Regional Medical Chamber (135/KBL/OIL/2013), and all participants provided written informed consent.

Laboratory investigations were conducted using standard techniques. Serum levels of GDF‑15, N‑terminal pro–B‑type natriuretic peptide (NT‑proBNP), and hs‑cTnT were measured using electrochemiluminescence immunoassays (Roche Diagnostics, Mannheim, Germany). The cutoff values of NT‑proBNP below 125 ng/l and GDF‑15 below 1000 ng/l were applied. The bleeding risk for each patient was assessed at baseline using the DOAC, HAS‑BLED, and ABC‑bleeding scores.^7-9

The patients were followed since the initiation of DOAC treatment by phone calls and ambulatory visits at least once every 6 months, until the occurrence of bleeding, death, or the end of the follow‑up period (September 30, 2021). We recorded symptomatic ischemic cerebrovascular events, systemic thromboembolism, and bleeding (major and clinically relevant nonmajor bleeding [CRNMB]) as defined by the International Society on Thrombosis and Haemostasis.^12,13

Results

Patient characteristics are shown in Table 1. The median (IQR) CHA₂DS₂-VASc score was 2 (2–4) points. All women were treated with DOACs (apixaban, rivaroxaban, or dabigatran), including 35% with reduced‑dose regimens. The most commonly used DOAC was apixaban.

AF was diagnosed after a median of 14.5 (range, 8–25) months since the cancer diagnosis. Using the DOAC score, 3 patients (5%) were categorized as having a high or very high risk of bleeding (score ≥8), 7 (12%) as having a moderate risk (score of 6 or 7), and 50 (83%) as having a very low or low risk (score <⁠6). Based on the HAS‑BLED score, 12 patients (20%) were assigned to a high‑risk category (score ≥3), while 16 and 19 patients (80% in total) scored 0 and 1 point, respectively, which placed them in a low‑risk category (score <⁠3). Using the ABC‑bleeding score, the patients were assigned to either a high‑risk (score of ≥2%; n = 23 [38.3%]), or a low‑to‑moderate risk category (score <⁠2%; n = 37 [61.7%]).

All the evaluated laboratory parameters (GDF‑15, hs‑cTnT, and NT‑proBNP) positively correlated with age (all P <⁠0.05). There were no significant associations between GDF‑15, hs‑cTnT, and NT‑proBNP values and body mass index or comorbidities, with 2 exceptions: higher hs‑cTnT levels were observed in the patients with liver injury, while those with coronary artery disease (CAD) had elevated GDF‑15 values. The latter biomarker also correlated with the DOAC score (R = 0.45; P <⁠0.001), HAS‑BLED score (R = 0.42; P <⁠0.001), and CHA₂DS₂-VASc score (R = 0.43; P <⁠0.001). Similar though weaker associations with the 3 scores were observed for hs‑cTnT (R = 0.4; P = 0.001, R = 0.36; P = 0.005, and R = 0.29; P = 0.02, respectively). eGFR correlated inversely with the ABC‑bleeding score (R = –0.45; P <⁠0.001), hs‑cTnT (r = –0.37; P = 0.004), and GDF‑15 (r = –0.5; P <⁠0.001). Regarding associations among the 3 biomarkers assessed, the only one observed was a positive strong association between hs‑cTnT and GDF‑15 (r = 0.66; P <⁠0.001). There were no significant associations between biomarker levels and prior chemotherapy or the presence of metastatic disease.

During a median 45.5 (40.5–55) months of follow‑up, we recorded 11 episodes of bleeding in 11 patients (18.3%), including major bleeding in 7 individuals (11.7%; all gastrointestinal bleeds). Three major bleedings occurred during infection, and 4 while on treatment with nonsteroidal anti‑inflammatory drugs. CRNMBs were documented in 4 patients (6.7%). Bleeding was associated with liver injury, aspirin use, lower eGFR, and higher GDF‑15 and hs‑cTnT levels, as well as with higher DOAC, HAS‑BLED, and ABC‑bleeding scores (Table 1).

An increase in hs‑cTnT level by 1 ng/l was associated with a 30% higher risk of bleeding, whereas an increase in GDF‑15 level by 100 ng/l increased the bleeding risk by 32% (Supplementary material, Table S1). An increase in the DOAC score by 1 point was associated with a 27% higher risk of all bleeding events (HR, 1.27; 95% CI, 1.06–1.51). For the HAS‑BLED score, a 1‑point increase was associated with a 53% higher risk of bleeding (HR, 1.53; 95% CI, 1.1–2.14; Supplementary material, Table S1). The patients with a high bleeding risk based on the ABC‑bleeding score (a score of ≥2%) had a 9‑fold higher risk of bleeding, as compared with the patients with low‑to‑moderate risk (score <⁠2%), and the risk of bleeding events increased nearly 5‑fold for each additional percent scored (HR, 4.85; 95% CI, 2.29–10.29). The analysis of ROC curves showed strong discriminatory power of bleeding prediction for GDF‑15 (cutoff value, 2667 ng/l), hs‑cTnT (cutoff value, 14 ng/l), and the ABC‑bleeding score (cutoff value, 2%; Supplementary material, Table S1 and Figure S1). However, the discriminatory power of hs‑cTnT was slightly attenuated in sensitivity analysis after restricting the sample to the patients free of liver disease. Sample restriction to the patients without CAD and those with eGFR equal to or greater than 60 ml/min/1.73 m² marginally affected the predictive power of hs‑cTnT and GDF‑15.

Fair discriminatory power was observed for the DOAC score, HAS‑BLED score, and eGFR (Supplementary material, Table S1), but not for NT‑proBNP. Comparison of the ROC curves using the Hanley test showed that only GDF‑15 had better discriminatory power for bleeding prediction than the clinical bleeding scores.

During the follow‑up, a total of 6 thromboembolic incidents were recorded (6 patients [10%]), including transient ischemic attack in a patient on reduced‑dose apixaban, stroke in a patient on reduced‑dose rivaroxaban, and episodes of venous thromboembolism in 4 patients (2 on reduced‑dose dabigatran and 1 during interruption in anticoagulation). The patients with thromboembolic incidents had elevated levels of NT‑proBNP, but not of other biomarkers.

Overall, 7 patients died during follow‑up, mainly due to cancer progression. None of the deaths was related to bleeding or stroke.

Discussion

To our best knowledge, this study is the first to evaluate biomarkers and risk scores as potential predictors of bleeding in women with breast cancer who developed AF during adjuvant hormonal therapy. We observed that GDF‑15 was the strongest predictor of bleeding in women with breast cancer and AF on DOACs, which might help guide antithrombotic prevention.

GDF‑15 is a cell regulatory protein involved in chronic inflammation in cancer patients.¹¹ It was shown to predict major bleeding in the ARISTOTLE (Apixaban for Reduction in Stroke and Other Thromboembolic Events in Atrial Fibrillation)¹⁴ and AVERT trials.^11,15 The association between GDF‑15 and bleeding risk may stem from the fact that its levels increase in response to cellular stress and cell vulnerability, which may predispose to bleeding following tissue damage.

Furthermore, GDF‑15 inhibits platelet activation through a mechanism similar to the blockade of glycoprotein IIb/IIIa, impairing clot formation.¹⁶ Its effect on bleeding may be influenced by platelet function and count, as well as the presence of comorbidities, such as liver disease.

However, in our study, as few as 4 patients with bleeding had liver injury, while a total of 11 women experienced bleeding. Therefore, in our opinion, biomarkers appear to perform better in terms of identifying breast cancer patients on DOACs who are prone to bleeding, as compared with clinical bleeding scores.

Recent results from the CAT‑BLED (Vienna Cancer, Thrombosis, and Bleeding) study,¹⁷ including patients initiating systemic anticancer therapies (9.6% with breast cancer), showed that elevated GDF‑15 levels were independently associated with an increased risk of major bleeding. However, unlike in our population, individuals with AF comprised only 5.9% of the cohort, and those receiving therapeutic anticoagulation for various reasons accounted for 15%. Interestingly, Arfsten et al¹⁸ reported a relationship between GDF‑15 and cancer progression, and showed that increased GDF‑15 levels were associated with systemic inflammation and subclinical cardiac dysfunction. Therefore, it might be speculated that apart from an elevated bleeding risk, the patients with high GDF‑15 levels could be at a risk of cancer progression. However, this issue was beyond the scope of our study.

Chemotherapy and other anticancer therapies may also contribute to an increase in GDF‑15 levels, but additional data are required to determine whether GDF‑15 can serve as a reliable biomarker of cardiotoxicity.¹⁹

To our knowledge, the present study is the first to report the relationship between hs‑cTnT level elevation and a higher bleeding risk in breast cancer patients. In contrast, numerous studies have extensively examined the role of hs‑cTnT in the context of cardiotoxicity, particularly among patients receiving chemotherapy or radiotherapy.²⁰ The current findings suggest that this biomarker might also be useful in predicting bleeding in specific cancer subgroups; however, further mechanistic and clinical investigations are warranted to confirm this.

In our study population, the women with CAD were found to have elevated GDF‑15 levels. It might be speculated that AF and hormonal therapies contribute to this phenomenon, given the fact that both of these factors enhance oxidative stress, especially in low‑grade inflammation, which stimulates the production and secretion of GDF‑15 by cardiomyocytes.^5,21

Given a strong positive correlation between GDF‑15 and hs‑cTnT, it is likely that in breast cancer women on hormonal therapy, both of these biomarkers act in tandem, reflecting complex vascular injury and enhanced oxidative stress.²² Further studies are needed to explore this concept.

Our results indicate that in women with breast cancer on hormonal therapy who are also receiving a DOAC, the ABC‑bleeding score seems to outperform other scoring systems. However, the relation was significant in the Cox proportional hazards model but not in the comparison of the predictive power. Importantly, the strong predictive capability of the ABC‑bleeding score most likely resulted from the inclusion of GDF‑15 and hs‑cTnT levels in its calculation.