Read on Twitter
After almost 1 year, the wait is over - the #EXCEL investigators finally release the UDMI data in @NEJM
https://www.nejm.org/doi/full/10.1056/NEJMc2000645?query=featured_secondary
in a nutshell:
- the EXCEL investigators had the UDMI data, although they said they didn’t
- It shows higher MI risk for PCI
https://www.nejm.org/doi/full/10.1056/NEJMc2000645?query=featured_secondary
in a nutshell:- the EXCEL investigators had the UDMI data, although they said they didn’t
- It shows higher MI risk for PCI
- This is research fraud and the papers should be retracted
@BBCnewsnight have had this data since december, why did it take 7 months more? Gosh @NEJM, what is up?
/2
@BBCnewsnight have had this data since december, why did it take 7 months more? Gosh @NEJM, what is up?
/2
Main finding: PCI has 2x the incidence of PMI (95%CI: 0.5-3.3), and 4.9x the incidence of all MI when using UDMI (95%CI: 2.6-7.2)
https://twitter.com/VictorDayan1/status/1283711323419836416?s=20
/3
https://twitter.com/VictorDayan1/status/1283711323419836416?s=20
/3
This begs a few questions:
- Why not report the p-values?
- Why report as cumulative incidence, and not KM? If you start the PCI curve above the CABG curve, this gives a very different outlook on the primary analysis as below
/4
- Why not report the p-values?
- Why report as cumulative incidence, and not KM? If you start the PCI curve above the CABG curve, this gives a very different outlook on the primary analysis as below
/4
All-cause death, MI and revascularisation all point to PCI inferiority, and superiority for stroke.
Why the event-rate difference between SCAI and UDMI? “no clinical correlation” (required by UDMI)
/5
Why the event-rate difference between SCAI and UDMI? “no clinical correlation” (required by UDMI)
/5
@greggwstone, I thought it was so difficult to do an ECG after CABG? This isn’t very clear: does this mean that in one of the best run trials in PCI vs. CABG (quote from Dr Braunwald, @NEJM 2016 editorial), you couldn’t do a simple post-op ECG and there is missing data?
/6
/6
Or does it mean that these were CLINICALLY INSIGNIFICANT cardiac enzyme bumps, as @ajaykirtane has said about routine enzyme testing after PCI?
/7
/7
The authors look at the prognostic significance of PMI according to definitions. Great! So what does it give?
/8
/8
Maybe a graph will make it easier to understand? Sure - look in the appendix
Wait, what? A letter that needs an online supplementary appendix? Is this for real?
/9
Wait, what? A letter that needs an online supplementary appendix? Is this for real?
/9
This seems much clearer:
- regardless of MI definition, the hazard of CV death is 2-3x if you have an MI after PCI or CABG for LM
- if you have UDMI MI after CABG, the hazard of death is greatly increased.
/10
- regardless of MI definition, the hazard of CV death is 2-3x if you have an MI after PCI or CABG for LM
- if you have UDMI MI after CABG, the hazard of death is greatly increased.
/10
- This last point would seem to say: a UDMI after CABG has a much larger prognostic significance than a SCAI MI. Or am I reading this wrong? This would seem to validate using UDMI rather than SCAI - which is picking up more MIs with little prognostic significance
/11
/11
But wait a second: 56 SCAI MIs in CABG patients? The @NEJM 5yr outcomes stated 57… these numbers don’t add up. Oh, here is why, you have to read the appendix fine print:
/12
/12
This is an as-treated analysis! This excludes 13 PCI pts and 34 CABG pts. Gosh, why the sudden change? We loose any benefit of randomisation, without any justification
/13
/13
Why do we have to dig through the online supplement to see this? The analysis isn’t described in detail in the text, only the results. This is very misleading for the reader.
/14
/14
@NEJM, please fact-check the author’s response. They assert that the SCAI definition of PMI was “previously shown to be associated with mortality after PCI and CABG”. This is factually wrong: the SCAI definitions were published after #EXCEL
/15
/15
The paper looking at the prognostic significance of SCAI definitions of PMI was published in EHJ 2019, 3 years AFTER the @NEJM primary outcome of #EXCEL! How can the @NEJM allow such a false statement?
/16
/16
Our colleagues from Basel ask an important question: as a sensitivity analysis, could the authors report the primary composite outcome using the UDMI definitions? Yet the authors don’t answer this question
/17
/17
Yet the authors stonewall (sorry, pun intended) on this.
We have to refer to @BBCnewsnight, who had access to this data and ran the analysis (as reported in @AATSJournals JTCVS by Mario Gaudino, Nick Freemantle and @dompagano): the 5yr composite shows an HR of 1.4, P = 0.009!
/18
We have to refer to @BBCnewsnight, who had access to this data and ran the analysis (as reported in @AATSJournals JTCVS by Mario Gaudino, Nick Freemantle and @dompagano): the 5yr composite shows an HR of 1.4, P = 0.009!
/18
Gosh, I am at a loss. I can’t fathom why these EXCEL papers remain without a retraction notice.
Leaving the record stand seems worse than the surgisphere debacle of invented data. We have been asking for this data for close a 1 year, and it was all there the whole time.
/19
Leaving the record stand seems worse than the surgisphere debacle of invented data. We have been asking for this data for close a 1 year, and it was all there the whole time.
/19
Why didn’t the authors share it? Because it didn’t show what they wanted to? And why switch from non-inferiority to superiority, and from ITT to as-treated? Without any plausible justification other than p-hacking.
/20
/20
The UDMI data should have been presented in the primary analysis 3 yr paper.
PCI would not have been non-inferior. This data would have been included in the @escardio / @EACTS guidelines.
I don’t take this lightly: this is research fraud which has corrupted our guidelines
/21
PCI would not have been non-inferior. This data would have been included in the @escardio / @EACTS guidelines.
I don’t take this lightly: this is research fraud which has corrupted our guidelines
/21
How can we believe this analysis, when we haven’t seen the data? We need independent teams to analyse all of the data and reach independent conclusions, like @brophyj did with his bayesian re-analysis
/22
/22
Based on these results, I would ask:
- @NEJM, this should have been a retraction notice, and the paper should have been republished with all of the data and conclusions stating the true results. Not a response to LTTE. This is a disgrace.
/23
- @NEJM, this should have been a retraction notice, and the paper should have been republished with all of the data and conclusions stating the true results. Not a response to LTTE. This is a disgrace.
/23
