OBJECTIVES: To investigate the impact of accounting for data on competing risks reported in different formats in meta-analysis. BACKGROUND: When there are competing risks, studies may report results in different formats. For example, in randomized controlled trials (RCTs) of therapies for the prevention of stroke in atrial fibrillation, some RCTs report numbers of patients whose first event was stroke, others total number of strokes, and others number of patients with at least one stroke. In meta-analysis and network meta-analysis (NMA), it is common not to adjust for competing risks or differently reported data. METHODS: We compared adjusted and unadjusted analyses in an illustrative example of NMA of anticoagulant therapies, including Apixaban, for prevention of stroke in atrial fibrillation. We artificially increased event rates to explore cases when unadjusted analyses are likely to be biased. We confirmed our findings using asymptotic calculations. RESULTS: Across RCTs, Apixaban 5mg twice daily had an absolute observed stroke rate per patient years at risk of 0.011. Unadjusted NMA gave odds ratio (OR) relative to Warfarin of 0.91 (0.73, 1.13) while the adjusted model gave hazard ratio (HR) of 0.90 (0.73, 1.11). Differences remained small when rate was increased to 0.11 but a 0.22 rate gave unadjusted OR 0.87 (0.82, 0.93) and adjusted HR 0.92 (0.87, 0.96), showing bias in the unadjusted analysis. The observed clinically relevant bleeds rate was 0.021; unadjusted OR was 0.81 (0.70, 0.95) while adjusted HR was 0.81 (0.70, 0.95). Increasing rate to 0.21 gave unadjusted OR 0.73 (0.69, 0.78) and adjusted HR 0.80 (0.76, 0.84), again showing bias. Other treatments showed similar bias if rates exceeded 0.20 but this depends on competing event rates and numbers of RCTs reporting each data type. CONCLUSIONS: We identified cases in which competing risks and differently reported data can bias estimates from unadjusted evidence synthesis.