The importance of time-to-event analysis in measuring the prognostic impact of coronary flow reserve

This commentary refers to ‘Coronary flow reserve and cardiovascular outcomes: a systematic review and meta-analysis’, by M.A. Kelshiker et al., https://doi.org/10.1093/eurheartj/ehab775 and the discussion piece ‘An attempt to better show some results such as the comparison of mortality and major adverse cardiovascular events between the abnormal and normal coronary flow reserve cohorts’, by H. Zhu et al., https://doi.org/10.1093/eurheartj/ehac166.

We thank Zhu et al.1 for their attention to our recent paper and particularly to the reporting of event rates when determining the prognostic association between coronary flow reserve (CFR) and long-term clinical outcomes — mortality and major adverse cardiovascular events (MACE).2 We would like to take this opportunity to touch upon event rate analysis again and address the errors they expertly identified.

We agree completely that significance testing of crude event rates adds little, but that their comparison can be intuitive and illustrative. We made the hazard ratio analysis the primary one to avoid bias that might arise from unequal follow-up duration in the two observational groups. We thank Zhu et al. for identifying errors in the translation of data points from the Supplementary material online, Table S4 to the main text. Here are the correct figures: our event-rate analysis identified 28 studies with published event rates for MACE and 13 studies with published event rates for mortality. The weighted mean event rate for mortality was 27.4% amongst 8377 patients with abnormal CFR vs. 6.8% amongst 12 022 patients with normal CFR. The weighted mean event rate for MACE was 30.9% amongst 5465 patients with abnormal CFR vs. 11.5% amongst 8447 patients with normal CFR.

Zhu et al. rightly note that HRs represent a continuous index of risk, with higher values representing greater clinical hazard. Therefore, for simplicity, we elected not to present the inverse of the HRs, and so axes were consistently and systematically labelled according to the prognostic clinical favorability of normal CFR or IMR.

Our goal in this study was to provide the broadest possible analysis of this topic, considering as much of the published data as can be effectively meta-analysed. The finding that abnormal CFR predicts poor outcomes is overwhelmingly seen and largely robust across subgroups. We are grateful to Zhu et al. for identifying these errors in our text but also glad that these oversights did not undermine the core findings. We look forward to further debate on this topic.