|
Abstract:
|
Meta-analysis for time-to-event outcomes commonly uses hazard ratio (HR) or relative risk (RR) as a between-group summary measure. However, as the issues of HR have been discussed elsewhere, the meta-analysis combining HR will inherit those issues. Also, since the approach using RR handles the censored time-to-event outcome as a binary outcome, it may not be appropriate. An example of the latter approach is seen in a recent publication by Ruff and colleagues (Lancet, 2014) to compare stroke or systemic embolic events (SEE) between novel oral anticoagulants (NOAC) and warfarin. They reported NOAC reduced the risk by 19%. However, clinical interpretation of the 19% might be difficult due to the issues of their analytic approach. We propose an alternative procedure that uses a robust and more clinically interpretable metric. We reanalyzed the same data as were used by Ruff et al., using difference in restricted mean survival time (RMSTd) with a 2-year of the truncation time. We found that the RMSTd (NOAC minus warfarin) for SEE was merely 3 days. Although Ruff et al. reported the 19% reduction in RR, the treatment effect might be almost negligible clinically.
|
