Comparative Risk Assessment of Severe Uterine Bleeding Following Exposure to Direct Oral Anticoagulants: A Network Study Across Four Observational Databases in the USA


These research results are from a retrospective, real-world, observational study assessing the risk of severe uterine bleeding among women newly exposed to direct oral anticoagulants or warfarin. This web-based application provides an interactive platform to explore all analysis results generated as part of this study, as a supplement to the full manuscript published in the journal Drug Safety.


Introduction: Antithrombotic therapies are associated with increased bleeding risk. Abnormal uterine bleeding data have been reported in clinical trials of patients with venous thromboembolism (VTE), but data are limited for patients with atrial fibrillation (AF).

Objective: Using real-world data from 4 US healthcare databases (October 2010 to December 2018), we compared the occurrence of severe uterine bleeding among women newly exposed to rivaroxaban, apixaban, dabigatran, and warfarin stratified by indication.

Methods: To reduce potential confounding, patients in comparative cohorts were matched on propensity scores (PS). Treatment effect estimates were generated using Cox proportional hazard models for each indication, in each database, and only for pairwise comparisons that met a priori study diagnostics. If estimates were homogeneous (I2 <40%), meta-analysis across databases was performed and pooled hazard ratios (HR) reported.

Results: Data from 363,919 women newly exposed to a DOAC or warfarin with a prior diagnosis of AF (60.8%) or VTE (39.2%) were analyzed. Overall incidence of severe uterine bleeding was low in the populations exposed to direct oral anticoagulants (DOACs), although relatively higher in the younger VTE population versus the AF population (unadjusted incidence rates: 2.8-33.7 vs 1.9-10.0 events/1,000 person-years). In the PS-matched AF population, a suggestive, moderately increased risk of severe uterine bleeding was observed for rivaroxaban relative to warfarin (HRs and 95% confidence intervals ranging from 0.83 [0.27-2.48] to 2.84 [1.32-6.23] across databases with significant heterogeneity), apixaban (pooled HR: 1.45 [0.91-2.28]), and dabigatran (2.12 [1.01-4.43]), which were sensitive to the time-at-risk period. In the PS-matched VTE population, a consistent, increased risk of severe uterine bleeding was observed for rivaroxaban relative to warfarin (2.03 [1.19-3.27]) and apixaban (2.25 [1.45-3.41]), which were insensitive to the time-at-risk period.

Conclusion: For women who need antithrombotic therapy, personalized management strategies with careful evaluation of benefits and risks are required.

Key points:

  • Using real-world data, the incidence of severe uterine bleeding was low but varied by antithrombotic therapy and occurred more often in patients with VTE versus those with AF.
  • Compared with warfarin, apixaban, and dabigatran, rivaroxaban was associated with a moderately increased risk of severe uterine bleeding in AF patients that was sensitive to the time-at-risk period analyzed and a more consistently increased risk in VTE patients regardless of time-at-risk period analyzed.
  • Careful evaluation of the benefits and risks of antithrombotic therapy in women requires consideration of factors including age, comorbidities, treatment duration, and indication for use.

  • Below are links for study-related artifacts that have been made available as part of this study:

    Table 3. Fitted propensity model, listing all coviates with non-zero coefficients. Positive coefficients indicate predictive of the target exposure.
    Figure 2. Preference score distribution. The preference score is a transformation of the propensity score that adjusts for differences in the sizes of the two treatment groups. A higher overlap indicates subjects in the two groups were more similar in terms of their predicted probability of receiving one treatment over the other.
    Figure 4. Systematic error. Effect size estimates for the negative controls (true hazard ratio = 1) and positive controls (true hazard ratio > 1), before and after calibration. Estimates below the diagonal dashed lines are statistically significant (alpha = 0.05) different from the true effect size. A well-calibrated estimator should have the true effect size within the 95 percent confidence interval 95 percent of times.