Cause of death and competing risk in patients with device-detected atrial fibrillation treated with apixaban or aspirin: Insights from the ARTESiA trial

Introduction

The ARTESiA trial found that, in patients with device-detected atrial fibrillation (AF), apixaban, as compared to aspirin, reduced stroke or systemic embolism. In the primary analysis, follow-up was censored at the time of AF progression or at the time of death, whichever came first.

Purpose

We aimed to describe causes of death in ARTESiA participants who died during follow-up and to test different assumptions regarding the competing risk of death on the primary results.

Methods

Two adjudicators blinded to treatment allocation classified deaths occurring during follow-up using pre-specified criteria. In the intention-to-treat population, we used different models to compare the efficacy of apixaban to aspirin. We tested a Fine and Gray model. We performed multiple imputations using Cox models that imputed a different hazard ratio (HR) of baseline risk of stroke/systemic embolism among trial participants with a competing death event as compared to those without. We censored participants at the time of AF progression (device-detected AF ≥24 hours or clinical AF).

Results

Over a mean follow up of 3.5+1.8 years, 644 of 4,012 (16.1%) trial participants died prior to experiencing a stroke or systemic embolism. There were numerically more competing deaths in the apixaban arm (341/2,015 [16.9%] versus 303/1,997 [15.2%]). The largest proportion of the 896 total deaths were non-cardiovascular (49%), with cancer (15.6% of all deaths) and infection/sepsis (15.4%) predominating. Among cardiovascular deaths (32.3% of all deaths), heart failure (14.8% of all deaths) and sudden death (8.6%) were the most common causes. The cause of death was unknown in 18.1% of cases. Compared to patients without a competing death event, those with a death event had a higher mean CHA2DS2-VASc score: 4.21±1.21 vs 3.87±1.10, p <0.001.

Apixaban, as compared to aspirin, significantly reduced stroke or systemic embolism under each assumption tested. The base Cox model (censoring at death) rendered a treatment effect of: HR 0.63 (95% confidence interval [CI] 0.45-0.88). The Fine and Gray model (imputing lost follow-up time but no events) rendered a treatment effect of HR 0.63 (95% CI 0.45-0.88). A Cox model imputing lost follow-up time with an HR of 2 for the baseline risk of stroke/systemic embolism rendered a treatment effect of HR 0.73 (95% CI 0.54-0.99). A Cox model with imputing an HR of 3 for the baseline risk of stroke/systemic embolism rendered a treatment effect of HR 0.71 (95% CI 0.54-0.94).

Conclusion

Among patients with device-detected AF who died during follow-up in ARTESiA, the largest proportion of deaths were non-cardiovascular. The most common causes of cardiovascular death were not anticoagulation-sensitive conditions. The observed reduction in stroke or systemic embolism with apixaban in ARTESiA is robust against a number of assumptions regarding the competing risk of death.