In our study, a total of 203 patients treated for cancer-associated thrombosis were identified during the 10-year study period. The incidence of bleeding events within six months of anticoagulation was 25.1% (51/203). This is notably higher than rates reported in pivotal randomised trials, including SELECT-D (12.1%)9, Hokusai-VTE Cancer [10] (16.3%), and Caravaggio [13] (10.9%). Differences in study design, patient selection, and anticoagulant type could also account for part of the discrepancy. Clinical trials often exclude patients with high bleeding risk, such as those with severe thrombocytopenia or recent major bleeding. In our study, we also determined whether severe thrombocytopenia was a risk factor for bleeding [9, 10, 13]. Furthermore, unlike those trials that focus on one or two anticoagulants, our real-world study encompassed a broader spectrum, including warfarin, parenteral anticoagulants (enoxaparin, fondaparinux) and multiple DOACs [1, 2, 8].

On the other hand, when compared with retrospective real-world cohorts, our findings are most comparable to Poenou et al., who reported a 6-month cumulative bleeding incidence of 26.7% among 110 cancer patients treated for VTE, whereby this diversity better reflects real-world prescribing among patients with cancer-associated thrombosis [14]. In contrast, another retrospective study by the TROLL Registry documented substantially lower rates, with a 12.7% bleeding incidence, among patients receiving LMWH (85.8%)15. These differences may be attributed to variations in anticoagulation use and ethnic background. In addition, similar to the landmark trials, the TROLL registry was conducted among a Western population, whereas our study population was predominantly Asian [15]. Ethnic variation may explain the discrepancy with our study as ethnic and genetic factors are known to influence anticoagulant metabolism and bleeding risk, particularly for DOACs and VKAs, as in differences in pharmacogenetics, cancer biology, comorbidities and healthcare systems have been shown to influence bleeding risk across populations [16].

Regarding anticoagulant distribution, our study showed a distinct pattern compared with the TROLL registry [15]. DOACs were the predominant agents used (66%) followed by parenteral anticoagulants (33.3%) and lastly warfarin (5.9%). Since this was a retrospective study, the exact reasons for choosing DOACs over LMWH were not documented. This contrasts with the TROLL Registry, where LMWH accounted for 85.8% of the initial anticoagulants used and DOACs for only 7.7%, reflecting an earlier treatment era (2005–2022) when LMWH was the recommended standard for CAT [15]. Nonetheless, the TROLL data showed a progressive increase in DOAC use over time, particularly in the later years of registry enrolment, which parallels the trend observed in our cohort [15]. This shared trajectory highlights the global shift towards DOAC-based therapy, aligning with contemporary international guidelines that now endorse DOACs as a preferred option for CAT treatment [4, 17, 18].

In relation to bleeding patterns, bleeding in our cohort occurred most frequently among patients treated with warfarin, while DOAC users experienced the lowest proportion of bleeding events. Regarding the site of bleeding, the gastrointestinal tract was the most common location (45%), with similar distribution between major and CRNMB. This finding is consistent with previous randomised studies, in which GI bleeding predominated, particularly among patients receiving DOACs [9, 10]. Similarly, the retrospective study by Poenou et al. reported that 31% of all bleeding events were gastrointestinal in origin; however, the study focused predominantly on patients receiving LMWH and unfractionated heparin, which may limit direct comparison with contemporary DOAC-based cohorts [14]. Nevertheless, data from the TROLL Registry demonstrated a comparable pattern, with 33.3% of major bleeding events occurring at GI sites [15]. This can be due to several pathophysiological mechanisms, including tumour infiltration of the GI wall or mucosal ulceration, which can subsequently predispose to bleeding [19]. In our study, five bleeding events occurred at the tumour site in patients with GI malignancies, all of whom were on DOACS. Three of the events occurred before 2020–2021, when international guidelines began recommending against the use of DOACs in patients with luminal GI and GU cancers. These findings emphasise the importance of individualised anticoagulant selection and close monitoring in patients with GI malignancies or pre-existing mucosal pathology.

We further analysed the timing of bleeding events from the initiation of treatment. In the study, we found that most bleeding events occurred within the first 3 months of treatment, with fewer events thereafter. This pattern is consistent with previous studies, which have shown that the risk of bleeding is highest during the early treatment phase [14]. These findings suggest that early anticoagulation is a period of heightened risk, likely reflecting advanced tumour burden, treatment-related mucosal injury, and the effects of high-intensity initial anticoagulation, thereby highlighting the need for close monitoring during the initial phase of therapy [20, 21].

With respect to anticoagulation eligibility, a total of 124 patients were not started on anticoagulation and were therefore excluded from this analysis, The most frequent reasons for this were high bleeding risk, advanced frailty or patient refusal This is consistent with findings from retrospective studies, such as the TROLL Registry, where approximately 10% of patients with cancer associated thrombosis did not receive anticoagulation, mainly due to recent major bleeding, thrombocytopenia or limited life expectancy [15]. As in another study, several patients were excluded due to contraindications to anticoagulants or palliative status, highlighting a shared real-world challenge in managing fragile populations [14]. This approach aligns with international guidelines that emphasise that treatment decisions should be carefully individualised, balancing thrombotic and bleeding risks with the overall clinical prognosis and the patient’s preference [4, 18].

Another important outcome of this study was the evaluation of risk factors for bleeding events. In this study, prolonged INR at the time of bleeding (defined as an INR > 1.5) emerged as a significant associated risk factor for bleeding (OR 2.42, p = 0.016), accentuating the role of impaired haemostasis in amplifying anticoagulant-associated bleeding. However, there are limited retrospective studies that specifically assessed INR derangement as a risk factor for bleeding in cancer-associated thrombosis, making direct comparison challenging. One of the few studies addressing this relationship was conducted among 75 cancer patients on oral anticoagulants (warfarin), whereby 20% experienced bleeding and 30% were over-anticoagulated (INR > 4), alongside a trend towards higher INR values and older age among those who had bleeding complications [22]. Similarly, Bhardwaj et al. found that even mildly elevated INR levels predicted bleeding in patients treated with DOACs for venous thromboembolism, although the study did not distinguish between cancer and non-cancer populations [23]. In contrast, most available evidence has focused on clinical or treatment-related predictors of bleeding. In the COMMAND-VTE, chronic kidney disease was recognised as an independent predictor of bleeding [24]. Another study by the RIETY Registry otherwise identified recent major bleed, anaemia and metastatic disease as significant predictors of bleed among anticoagulated cancer patients [25]. The lack of significance in our study may reflect the limited number of cases recruited; the association between prolonged INR and bleeding provides valuable insight into the potential role of laboratory markers in identifying patients at higher risk.

This study has several limitations. Its retrospective, single-centre design may introduce selection bias and limit the generalisability of findings. Due to the retrospective nature of data collection, certain variables, such as smoking history, body mass index (BMI), and Eastern Cooperative Oncology Group (ECOG) performance status, were not consistently available or documented. These parameters may represent important contributors to bleeding risk and limit the ability to fully adjust for potential confounders or to identify additional predictors of bleeding in this cohort. Furthermore, bleeding events may have been under-reported due to incomplete documentation or management outside of the study centre, leading to an underestimation of bleeding incidence. In this study, the modest sample size restricted the statistical power to detect associations between associated risk factors and bleeding outcomes. The use of simple logistic regression, driven by sample size constraints, may have led to residual confounding due to the inability to adjust for multiple variables. In addition, transitions between anticoagulant therapies, such as switching from LMWH to DOACs, were not recorded, limiting assessment of their potential impact on bleeding outcomes. Lastly, treatment allocation was not randomised and reflected clinician preference, leading to heterogeneity in anticoagulant choice and potential confounding by indication.

The study’s strength lies in its inclusion of multiple anticoagulant classes, which provides a comprehensive picture of treatment patterns and outcomes beyond the narrower scope of randomised trials, which typically include a limited number of agents. This study also represents one of the few available cohorts of cancer-associated VTE managed with anticoagulation in an Asian population, addressing a gap in the literature where most trials were conducted among Western populations.