Elimination of Modifiable Risk Factors for Anticoagulant-Related Bleeding in Patients with Cancer Reduces the Probability of Bleeding

Background: Patients with cancer are at increased risk of bleeding because of their underlying malignancy, cancer-directed therapies, and associated comorbid conditions. Anticoagulant therapy for treatment of cancer-associated venous thromboembolism (VTE) further increases the risk of bleeding. However, patients with cancer also have high risk of recurrent VTE and thus are recommended long term anticoagulant therapy. Identifying and potentially eliminating modifiable risk factors could reduce bleeding events without eliminating anticoagulant therapy.

Aims: We aimed to quantify the proportion of anticoagulant-related bleeding cases attributable to candidate modifiable risk factors by determining the population attributable fraction (PAF) of each risk factor.

Methods: We identified patients with newly diagnosed cancer-associated VTE and a new prescription for anticoagulant therapy between 2012 and 2020 in the US Veterans Administration healthcare system. To optimize identification of incident VTE events, we excluded patients with outpatient prescription for anticoagulant therapy within the six months preceding cancer-associated VTE. Using the methods of Fine and Gray, we measured the association between candidate risk factors and bleeding requiring hospitalization. Anticoagulant related bleeding was identified using previously validated international classification of disease codes. Candidate risk factors with a two-sided p value < 0.05 or with a p value < 0.10 and consistent with prior literature were offered into a multivariable model. Using backward elimination, we retained risk factors in the multivariable model until all remaining factors had a two-sided p value < 0.05 or a p value < 0.10 if consistent with prior literature. We identified modifiable factors (factors that potentially could be reversed). We adapted the Samulesen (2008) method and used cumulative incidence function (CIF) to account for the competing event of non-bleeding associated mortality. We estimated CIF functions for each subject in the initial population and obtained average CIF estimates over the entire population, and estimated CIF functions for each subject in the same population with hypothetical elimination of the modifiable risk factor and averaged the estimates. All analyses were done using SAS 9.4 and R statistical software.

Results: We identified a cohort of 11,731 patients with cancer-associated VTE. The most common malignancies included lung (23%), gastrointestinal [GI] (14%), and prostate (11%). Most patients received anticoagulant therapy with low molecular weight heparin (47.7%), followed by direct oral anticoagulants (26.8%), vitamin K antagonists (24.5%), and fondaparinux (1%). The probability of bleeding before death at 1-year was 8.5%. The most frequent sites of clinically significant bleeding included GI tract (57.2%), genitourinary tract (excluding microscopic) (20.2%), and intracranial hemorrhage (10.4%). Potentially modifiable risk factors associated with an increased risk of anticoagulant-related bleeding at 1-year included: alcohol abuse (subdistribution hazard ratio [sHR] 1.30; 95% CI: 1.02-1.65), anemia (hemoglobin < 10g/dL) (sHR 1.27; 95% CI: 1.11-1.46), antiplatelet prescription (sHR 1.16; 95% CI: 0.98-1.37), and uncontrolled hypertension [HTN] (sHR 1.19; 95% CI: 1.04-1.37).

Anemia had the highest PAF of anticoagulant-related bleeding at 12.2% at 1-year. This was followed by uncontrolled HTN with a calculated PAF of 7.9% at 1-year. Elimination of anemia and uncontrolled HTN together resulted in a 19% reduction in the probability of bleeding at 1-year after the start of anticoagulant therapy (6.9% versus the 8.5% observed). Elimination of all 4 modifiable risk factors (alcohol abuse, anemia, antiplatelet prescription, and uncontrolled HTN) resulted in a 23% reduction in probability of bleeding at 1-year after the start of anticoagulant therapy (6.6% versus 8.5%).

Conclusion(s): Our study found that addressing modifiable risk factors for anticoagulant-related bleeding in patients with cancer has potential to reduce bleeding and thus facilitate continuation of anticoagulant therapy in this high-risk patient population.