,
34 tweets,
12 min read
Read on Twitter
After a few days to go over the #EXCEL 5 year results published in @NEJM and thinking hard about @DrMoritzWvB tweet on IC and cardiac surgery seeing things quite differently, here are a few thoughts. Glad to hear your thoughts
@tssmn @ctsnetorg @EACTS
1/30
@tssmn @ctsnetorg @EACTS
1/30
2/ Congratulations to the investigators for an outstanding RCT! Important question and results, which will have an impact on guidelines. The protocol was powered for non-inferiority of the composite outcome at 3 years, and congratulations for following these patients to 5 years
3/ why were p-values for secondary outcomes reported selectively? e.g. for revascularization, but not others?
4/ The authors change their way of reporting depending on whether the outcome favors non-inferiority or not (i.e. superiority of CABG). For the primary endpoint, notice the p-value 
5/ for the secondary outcome, no p-value
6/ But wait, that 95% CI remains above 1. This means it must be significant. Why not state it? Oh, right, it’s not non-inferior (oops, getting into double negatives).
7/ As @dompagano showed, many of us don’t understand 95% CI alone.
Adding a p-value (even if the trial wasn’t powered for the secondary outcome) would make getting the point easier.
Adding a p-value (even if the trial wasn’t powered for the secondary outcome) would make getting the point easier.
8/ Language matters. Spin in CV research is widespread (medscape.com/viewarticle/91…) and it’s interesting to look at how these results are worded.
9/ For the primary and secondary outcomes, the result is stated blankly: “the outcome occurred in X% of PCI and Y% of CABG group”, with only the p-value to guide the novice reader. A quick read would think: “wow, no difference in both primary and secondary outcomes at 5 years”.
10/ Let’s correct that for the secondary outcome: “the secondary composite outcome […] at 5 years occurred more frequently in the PCI group (31.3%) than the CABG group (24.9%, OR 1.39, 95% CI 1.13-1.71, p = 0.002)”.
11/ The same applies to the individual events. For example, all-cause mortality
12/ Wait, but, the KM curve of overall mortality tells a different story: OR 1.38, 95% CI 1.03-1.85, p-value isn’t given, but < 0.05!
13/ So let’s correct the text: “all-cause mortality at 5 years occurred more frequently in the PCI group (13%) than the CABG group (9.9%, OR 1.38, 95% CI 1.03-1.85, p < 0.05).”
14/So we have corrected the secondary outcome and overall survival, that both favor CABG
15/ The remainder of the results of individual events (stroke, MI) are reported as would be expected, saying which is more or less frequent in each group. Why the change, and not for the other measures above?
16/ The choice of a composite endpoint is also problematic. As studied, this is used to muddle or game the results. bmj.com/content/341/bm…
All-cause mortality is the only outcome that can’t be gamed
All-cause mortality is the only outcome that can’t be gamed
17/ @VPrasadMDMPH et al. have shown that increasingly surrogate endpoints are used for the approval of new drugs in oncology (jamanetwork.com/journals/jamai…), without any improvement in overall survival. The same applies to cardiovascular research
18/ Not only did EXCEL use a composite outcome, it strayed from previous studies, and didn’t include repeat revascularization. Here is the justification
19/ Yet in the discussion, repeat revascularization is associated with MI and mortality.
19/ Why not follow the design of previous trials (if you are going to avoid using a simple hard outcome: all-cause mortality) and include it? Oh, because the result would be not non-inferiority
20/ The authors acknowledge the differences between treatment, focusing on the primary outcome over time. Amen!
21/ A major driver of non-inferiority of the primary outcome was MI, namely an excess of perioperative MI in the CABG group. This was balanced out by a flat rate of late MI, compared to an increasing number of MI late after PCI
22/ The definition of MI was defined differently than most trials comparing PCI and CABG
23/ As a speciality, we need to do better, as there are no universally validated criteria for diagnosis periopertive MI. CABG will inevitably do some myocardial ischemia (the aorta is cross-clamped!), despite our best cardioplegia
25/ In SYNTAX, MI was defined as
26/ How did PCI and CABG compare in terms of MI in SYNTAX? CABG was better!
27/ Or in a KM curve (yellow is PCI, blue CABG):
28/ So by using a definition of perioperative MI that places the bar unusually low (CK-MB or troponin >1x upper limit) after 72 hours, Bingo! CABG suddenly had many more MI in #EXCEL. This was the main driver of the primary endpoint being not significantly different btw groups
29/ The choice of MI definition is justified by the fact that it has been shown to be predictive of late death. Yet, the primary outcome wasn’t mortality, but this muddled composite.
30/ Final point: why won’t the authors share their data?
31/ It would be interesting to have the dataset re-analyzed by an independent group of statisticians and see if the results and interpretation are similar
medscape.com/viewarticle/90…
medscape.com/viewarticle/90…
Wow, that was a looooong thread. But quite a bit to say about #EXCEL 5 years. It seems the #medicalconservative would have pause to take the results as presented.
Glad to hear from @GreggWStone @drjohnm @adamcifu @ADAlthousePhD for your input
Glad to hear from @GreggWStone @drjohnm @adamcifu @ADAlthousePhD for your input
@GreggWStone @drjohnm @adamcifu @ADAlthousePhD Food for thought & discussion before our expert updates at #EACTS2019 on saturday!
