Haemorrhagic event reporting with concomitant gemcitabine and warfarin: a VigiBase disproportionality analysis

Case reports suggest that concomitant gemcitabine and warfarin may lead to clinically important over-anticoagulation and bleeding, but population-level characterization remains limited. We evaluated whether concomitant gemcitabine and oral anticoagulants, particularly warfarin, were associated with disproportionate reporting of haemorrhagic events in VigiBase, the WHO global database of individual case safety reports.

We performed a retrospective pharmacovigilance study of VigiBase reports from 1 January 1994 to 28 April 2025. Haemorrhagic events were defined using the Standardised MedDRA Query "Haemorrhages". For each anticoagulant, we fitted multivariable logistic regression models including a gemcitabine × anticoagulant interaction term, adjusting for age group, sex, WHO region, and bleeding-related co-medications. A sensitivity analysis was restricted to serious reports. Additional analyses adjusted for calendar period. Time-to-onset and dechallenge/rechallenge were summarised where available.

Among 6,191,806 eligible reports, 707 described concomitant gemcitabine and warfarin. The gemcitabine-warfarin interaction term was significantly positive (interaction reporting odds ratio 1.31, 95% confidence interval 1.07-1.60; p = 0.010) and remained positive in serious reports (1.36, 1.09-1.69; p = 0.006). Interaction-term estimates for direct oral anticoagulants were inconsistent or not robust in sensitivity analyses. Time-to-onset was available for 29 haemorrhagic reports, with a median of 15 days (interquartile range 10-52; range 4-529 days). Positive dechallenge was recorded in 17 reports; no positive rechallenge was recorded.

This VigiBase analysis identified disproportionate haemorrhagic-event reporting with concomitant gemcitabine and warfarin. These findings should be interpreted in light of the limitations inherent to spontaneous reporting, including under-reporting, reporting biases, missingness, and residual confounding.