PREDIMED was the trial that moved the Mediterranean diet from observational association to evidence, assigning 7,447 Spanish adults at high cardiovascular risk to one of three diets. Then the authors found : household members enrolled without randomization, participants assigned without randomization at one site, and inconsistent randomization tables at another. They withdrew the 2013 paper. This is the reanalysis. What survived?
The event counts still favor both Mediterranean arms. Among 288 primary end-point events there were 96 in the olive oil group, 3.8 percent, 83 in the nut group, 3.4 percent, and 109 in the control group, 4.4 percent. The revised report rests on analyses that do not assume everybody was randomly assigned, which is a weaker foundation than the original and an honest one.
Observational cohorts and one secondary prevention trial had linked the Mediterranean diet to lower cardiovascular risk. PREDIMED set out to test it properly, assigning 7,447 participants aged 55 to 80, 57 percent of them women, at high cardiovascular risk but with no cardiovascular disease at enrollment, to one of three diets.
Two arms received a Mediterranean diet, one supplemented with and one with mixed nuts, both provided free. The control arm received advice to reduce dietary fat and small nonfood gifts. Everyone got quarterly educational sessions. The was a major cardiovascular event, meaning myocardial infarction, stroke or death from cardiovascular causes. After a median 4.8 years the trial was stopped on a prespecified interim analysis.
Then came the part that makes this card worth writing. After publishing in 2013, the investigators identified protocol deviations: household members enrolled without randomization, some participants assigned without randomization at one of eleven sites, and apparently inconsistent use of randomization tables at another. They withdrew the published report and reanalyzed the data without relying solely on the assumption that all participants were randomly assigned.
A primary end-point event occurred in 288 participants: 96 in the olive oil group at 3.8 percent, 83 in the nut group at 3.4 percent, and 109 in the control group at 4.4 percent. The direction held. What changed is the confidence a reader should attach to it. An analysis adjusted for baseline characteristics is a weaker instrument than a clean randomization, and a trial stopped early at an interim look tends to overstate its own effect.
wat. Events ran at 3.4 and 3.8 percent on the two Mediterranean diets against 4.4 percent on the control, in a trial its authors withdrew and redid.
wat wat
Watching a landmark trial correct itself in public is instructive. The direction of the finding held, and the strength of the inference dropped, which is exactly what a broken randomization should cost a result.
How do we know?
Estruch and colleagues assigned 7,447 participants aged 55 to 80 at high cardiovascular risk to a Mediterranean diet with extra-virgin olive oil, one with mixed nuts, or a control diet advising reduced fat. They then reanalyzed the data without relying solely on random assignment.
What is still unknown
Randomization failed at some sites, so the revised estimates lean on adjustment for baseline characteristics instead. The trial was also stopped early at a prespecified interim analysis, which tends to inflate an effect.
Learn something?
Where to next
Can a mirror help a paralysed arm move again?stroke
Why does a failing heart hurt in the arm?heart
Can swapping the salt in the kitchen prevent a stroke?stroke