1/ Oh no… another tweetorial on #EXCEL

Very interesting to see so many IC colleagues underscore the importance of periprocedure MI in #EXCEL (to the benefit of PCI) and reverse 180° after #ISCHEMIA (as it didn’t benefit PCI)

2/ Let’s go back to the #EXCEL trial paper on PMI, and how the investigators chose a new definition of PMI

Interesting to see that the first 4 authors are all employed by the CRF clinical trials unit. Where does the funding for CRF come from?

3/ Remember, this is the justification that #EXCEL investigators used for developing new definitions of PMI and including it in the primary outcome. So let’s take a deeper dive into this seminal paper

4/ Starting with the rationale and methods: 3rd UDMI require clinical correlates (new EKG changes, new RWMA, etc.) along with biomarker thresholds. Why forgo these clinical correlates in #EXCEL?

5/ Ok, seems reasonable: conflicting data on prognostic significance of PMI, so #EXCEL set out to test thresholds of biomarker values that have a prognostic significance

6/ Uh, wait just a second. The astute reader asks: does this argument justify forgoing clinical correlates and using entirely new PMI definitions?

7/ That’s right folks, the #EXCEL investigators don’t justify at all. They are saying “Let’s just do our thing, and see where that leads us”. Does that seem fair to you?

8/ Ok, back to the manuscript. #EXCEL set out to test biomarker thresholds of prognostic significance… Oh, @RichardWhitloc2 answers this one very astutely https://twitter.com/RichardWhitloc2/status/1178300694492778497?s=20

9/ Ok, so let’s just work with what the manuscript is then: testing a single threshold. How did they do it?
Great, a prespecified analysis! Seems reasonable, right?

10/ Well let’s look at the #EXCEL protocol, conveniently published on @NEJM page of 5yr results

11/ Ok, it must be there somewhere…. only 584 pages to go through the entire PDF

12/ Ah, getting closer... subgroup analyses

Well, nope, still not

13/ Analysis of other endpoints maybe? Still not

14/ How about searching “pre-specified”, “pre-specified analysis” and any other combination… still not

15/ Ok, I give up. #cardiotwitter, to those who are much smarter than me, could you please show me where this analysis was pre-specified? Because I haven’t found a clue to show it. Which makes this statement interesting

16/ Now for methods. #EXCEL routinely collected baseline, 12 and 24h CK-MB. @Ajay had a great thread on how off this was (in ISCHEMIA) from real practice, and the risk of picking up a lot o noise for a small signal

https://twitter.com/ajaykirtane/status/1196795707174244353?s=20

17/ This also defines PMI in #EXCEL: CK-MB >10x URL without any clinical correlates. No justification for this

18/ Now for the statistical analysis: this was a “modified intention-to-treat cohort”. What does this mean? Patients were included only if their revascularization was performed according to the assigned procedure

19/ So let’s go back to the protocole. How was this defined here?

20/ Maybe someone more intelligent can explain me this: isn’t this “modifed ITT” just an “as treated” analysis?

21/ So what’s the problem with using a modified ITT?
1. You lose the power of randomization.
2. You are using jedi mind tricks to make your readers think that your approach doesn’t suffer from 1

22/ Here are the results of this modified ITT: exclusion of 13 PCI and 34 CABG patients.
More importantly, only 90 PMI events for the entire study population

23/ Just 15 patients with CK-MB ≥5x URL + additional criteria (i.e. UDMI).

Hard to write a manuscript around 15 events, don’t you think? So let’s just lump in 75 more, with CK-MB ≥10 URL without clinical criteria, right?

24/ So if this is an “as treated”, this means we loose the advantage of randomization. Table 1 (baseline characteristics) becomes important, no @ADAlthousePhD ?

25/ Well there was 2x as much COPD, but also more insulin-treated DM, less PVD (?!?) and more unstable angina in PMI pts

26/ PMI group had more complex lesions, and received less prasugrel or ticagrelor. Not signfiicant, but with n = 90 in one group, the threshold for a “significant” difference is greatly increased

27/ Anyway, this thread is already too long. Many more items to discuss, and would love to hear your thoughts

28/ Bottom line: I wonder if this manuscript is much better than the infamous ISIS-2 - astrological sign as predictor of survival https://www.thelancet.com/journals/lancet/article/PIIS0140-6736(15)61505-7/fulltext

29/ I'm just trying to discuss a paper that has been heralded as the justification for using the SCAI def of MI in #EXCEL. My read is the evidence isn't that strong that SCAI-definition PMI in CABG has prognostic significance. Would love to hear and learn from #cardiotwitter