Hilda Bastian: A Mediterranean diet trial’s retraction and republication leaves a trail of questions

It started with a paper by John Carlisle in 2017. [1] He analysed baseline data for participants in over 5,000 randomised trials, looking for differences in trial arms that would be improbable if people had been allocated to them randomly. One of the trials he marked as unlikely to be genuinely randomised was the PREDIMED trial—the most influential trial of Mediterranean diets. [2]

It turned out the authors did have some explaining to do. Last week, the 2013 report of the trial was retracted from the New England Journal of Medicine. [3] The paper was republished with new analyses. [4] The authors now report that 21% of the participants were either not randomised at all, or that randomisation was suspect. They maintain that their re-analysed study shows the same benefit of Mediterranean diets, albeit with the evidence status somewhat weaker. I’m not convinced.

Not everyone was convinced about the original incarnation either. It was excluded from two relevant Cochrane reviews because of inadequacies in the control group. [5, 6] This bothered reviewers from NICE, as well. They judged that the trial was at “serious” risk of bias for individual cardiovascular disease outcomes, with “low or very low quality” data, and underpowered for mortality. According to them, PREDIMED “was difficult to interpret because the control group was advised to reduce their fat intake and to follow some of the components of the Mediterranean diet”. [7]

What’s more, people in the control group only received a leaflet initially, while those in the two Mediterranean diet arms—one receiving free olive oil and the other free nuts—had regular group and individual training sessions. A few years in, though, and the training sessions were open to the control group, too.

So PREDIMED 2013 had more than its fair share of issues. The trial was stopped early, which left another question mark hanging over it. [8] Unreported protocol changes covered a sizeable chunk of the randomisation problem. When more than one person was recruited from a household, only the first was randomised—subsequent people were assigned to the same diet as the first person. That unreported change accounted for 425 of the 1,588 participants who were not properly randomised. (The rest were from a centre which randomised by clinic, not person, and a centre which didn’t use the randomisation table properly.)

More data on the trial’s outcomes could have cleared up some questions. The trial’s primary outcome, called “major cardiovascular events”, was a composite endpoint. The three components were myocardial infarction (MI)—which could be “acute, evolving or recent MI” (including a change in biomarkers only which preceded angioplasty or other similar intervention), stroke, and cardiovascular death. Each of these three was a secondary endpoint.

Only one of those component outcomes showed a statistically significant difference—stroke. But it wasn’t significant in all of the analyses, for example, a Kaplan-Meier analysis. We aren’t provided with a detailed breakdown of events, or the secondary endpoints, for the 5,589 randomised participants. There was no protocol provided for the 2018 re-analysis, so we don’t know whether or not analyses were selectively reported.

Where does this leave the evidence for the Mediterranean diet? Although others were convinced that the evidence for health benefits was strong enough, the Cochrane and NICE reviewers were not, even before these revelations about PREDIMED. I’m with them. And if the trials have left a lot of questions in the air, what’s happening outside trials raises even more questions.

We don’t really even know exactly what “the Mediterranean diet” should be. It’s different in each trial. We don’t know what it means for the average person, or outside Mediterranean countries. And the PREDIMED trial looked only at older people with risk factors.

We don’t know how the popularisation of the “Mediterranean diet” has played out, either. It’s one thing to study what a diet could potentially do in an experiment with extensive support and training—and another to see what effect advice has out in the wild. It’s not hard to see why a concept that conjures up oil, pasta, pizza, and red wine would be so attractive to many people—or to see what could go wrong! We may have under-estimated the harm that even small amounts of alcohol can do. [11] If there does turn out to be an adverse effect of the Mediterranean diet phenomenon, it could be in the idea that a healthy diet includes regular red wine.

Hilda Bastian is a scientist, cartoonist and blogger.

Competing interests: None declared.


[1] Carlisle: https://onlinelibrary.wiley.com/doi/full/10.1111/anae.13938

[2] PREDIMED 2013: https://www.nejm.org/doi/10.1056/NEJMoa1200303

[3] Retraction notice: https://www.nejm.org/doi/full/10.1056/NEJMc1806491

[4] PREDIMED 2018: https://www.nejm.org/doi/full/10.1056/NEJMoa1800389

[5] Rees et al: http://cochranelibrary-wiley.com/doi/10.1002/14651858.CD009825.pub2/abstract

[6] Hooper et al: http://cochranelibrary-wiley.com/doi/10.1002/14651858.CD011737/abstract

[7] NICE: https://www.ncbi.nlm.nih.gov/books/NBK248067/

[8] Bassler: https://jamanetwork.com/journals/jama/article-abstract/185591

[9] PREDIMED design: https://academic.oup.com/ije/article/41/2/377/690923

[10] PREDIMED protocol: https://www.nejm.org/doi/suppl/10.1056/NEJMoa1200303/suppl_file/nejmoa1200303_protocol.pdf

[11] Wood: https://www.thelancet.com/journals/lancet/article/PIIS0140-6736(18)30134-X/fulltext