Apixaban Benefit in Subclinical Atrial Fibrillation Accounting for Competing Risks: A Reanalysis of the ARTESIA Trial
Discuss this preprint
Start a discussion What are Sciety discussions?Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Background
It is uncertain if anticoagulants provide a net benefit in subclinical AF (SCAF). The ARTESIA trial showed apixaban reduced the relative hazard of stroke/systemic embolism (SSE) in SCAF, but did not report absolute risk reduction (ARR). The reported Kaplan-Meier analysis did not account for the competing risk of death or 24-hour AF events. We reanalyzed ARTESIA accounting for competing risks to determine the ARR of apixaban vs. aspirin.
Methods
Individual time-to-event and time-to-censoring data were extracted from the published Kaplan-Meier curve. ARTESIA classified deaths and 24-hour AF events as censoring events. We probabilistically reclassified them to competing events to estimate the ARR. We used the Aalen-Johansen estimator to estimate the cumulative risk of stroke/systemic embolism at 3.5 years, accounting for competing events. We compared these results to the Kaplan-Meier estimator.
Result
After reclassification of deaths and 24-hour AF to competing events, there were 1111 censoring and 852 competing events in the apixaban arm, and 1100 censoring and 816 competing events in the aspirin arm. Compared to the Aalen-Johansen estimator, the Kaplan-Meier estimator overestimated the cumulative SSE risk at 3.5 years with apixaban (2.64% vs. 2.20%, difference in estimates 0.43 percentage points; 95% CI 0.27 to 0.62) and with aspirin (4.58% vs. 3.82%, difference in estimates 0.76 percentage points; 95% CI 0.56 to 1.01). Compared to the Aalen-Johansen estimator, the Kaplan-Meier estimator overestimated the ARR of SSE risk at 3.5 years with apixaban compared to aspirin (1.94% vs. 1.61%, difference in estimates 0.33 percentage points; 95% CI, 0.06 to 0.64).
Conclusion
In SCAF, apixaban reduced the 3.5-year risk of SSE by 1.61%, which would be overestimated by 20% if death and >24-hour AF events are treated as censoring rather than competing events.
