A Bayesian Approach to Left Main Disease – How has the Excel Trial moved the needle?


A talk that examines the role of Bayesian inference in interpreting the EXCEL trial comparing percutaneouc coronary revascularizations (PCI) to coronary artery bypass grafting (CABG) in patients with left main (LM) disease The 5 year EXCEL study was published in the NEJM and concluded - In patients with left main coronary artery disease of low or intermediate anatomical complexity, there was no significant difference between PCI and CABG with respect to the rate of the composite outcome of death, stroke, or myocardial infarction at 5 years. In summary, there are several problems with this study -

  1. Designed as an non-inferiority trial with a margin of 4.2%. The 3 year results were reported as non-inferior but the 5 year results were reported as a superiority trial (difference of 2.8% against PCI 95% -0.9 - 6.5, p= 0.13). As a superiority trial the null hypothesis is not rejected as p >0.05. However when analysed with the non-inferiority lens, the null hypothesis of a difference between PCI and CABG can’t be rejected as upper limit of 97.5% CI exceeds the pre-specified margin of 4.2% and therefore non-inferiority CAN’t be claimed.

  2. Trial used a unique definition of myocardial infarctions based solely on enzyme elevations with no other collaborating information (symptoms, ECG changes). The more standard MI definition based on the 3rd universal definition of MI (UDMI) was a prespecified secondardy oputcome but not reported in the original publication. The authors claimed the data was unavailable before eventually publishing it here - 9 months after the orginal publication

  3. Home institution of 8 authors (Cardiovascular Research Foundation) received $1 million donation from stent sponsor during study

  4. A Bayesian analysis offers several advantages - i) provides desired answers while avoiding inferential problems with p values, ii) is fundamentally sound, following the rules of probability iii) is intellectually coherent & intuitive -> clear and direct inferences iv) makes use of all available information -> allows flexible, allows complex models v) places emphasis on parameter estimation & uncertainty measures vi) readily computed with modern computers

  5. The Bayesian analysis assists in data interpretation and specifically suggests, whether based on EXCEL results alone (vague prior) or on the totality of available evidence (informative prior), that PCI is associated with inferior long-term results for all events, including mortality, compared with CABG for patients with left main coronary artery disease. The full peer reviewed published article can be found here

My take home messages from this debate were;

  1. While statistical inference is important, it is also important to examine the study for other biases (e.g. protocol deviations, biases, conflicts of interest)
  2. Bayesian perspective may provide additional insights into understanding a study & decision making
  3. CABG remains the preferred revascularization choice for appropriate and eligible patients

Sep 30, 2020 12:00 PM — 1:00 PM
Jewish General Hospital & McGill University Hospital Center
Montreal, QC
Professor of Medicine & Epidemiology

I am a tenured (full) professor with a joint appointment in the Departments of Medicine and Epidemiology and Biostatistics where I work as a clinical cardiologist and do research in cardiovascular epidemiology.