QUESTIONS & ANSWERS
Re-publication of the Predimed trial in N Engl J Med (2018)
WHAT HAPPENED?
On June 13, 2018, the New England Journal of Medicine (NEJM) replaced the 2013 study, “Primary Prevention of Cardiovascular Diseases with a Mediterranean Diet,” [1] as a result of imperfections in randomization procedures affecting a portion of participants in the PREDIMED (Prevención con Dieta Mediterránea) trial. Concurrently, NEJM published a corrected version of the study with reanalyzed data to replace the previous version, “Primary Prevention of Cardiovascular Disease with a Mediterranean Diet Supplemented with Extra-Virgin Olive Oil or Nuts.” [2]
On June 13, 2018, the New England Journal of Medicine (NEJM) replaced the 2013 study, “Primary Prevention of Cardiovascular Diseases with a Mediterranean Diet,” [1] as a result of imperfections in randomization procedures affecting a portion of participants in the PREDIMED (Prevención con Dieta Mediterránea) trial. Concurrently, NEJM published a corrected version of the study with reanalyzed data to replace the previous version, “Primary Prevention of Cardiovascular Disease with a Mediterranean Diet Supplemented with Extra-Virgin Olive Oil or Nuts.” [2]
WHAT IS THE REASON FOR THE NEW PUBLICATION OF PREDIMED?
After July, 2017, the Steering Committee identified 2 small departures from the reporting or application of the trial’s protocol:
a) enrollment of household members (partners of a previous participant) without randomization; members of the household of randomized participants were invited to participate and allocated to the same intervention group as their household member. This was done to avoid assigning members of the same household to different diets. Assigning all participants in a household to the same diet was viewed as the best approach to achieve dietary changes in the household. This procedure was inadvertently omitted in the reporting of the protocol and the original NEJM paper.
b) allocation in clusters (by clinic), instead of individual allocation of some participants at 1 of 11 study sites.
Our aim with the republication is to disclose both facts and to provide reassurance that the results of many new analyses to address these minor issues are consistent with our original conclusions that Mediterranean diet interventions enriched with extra-virgin olive oil or mixed nuts are associated with approximately 30% lower risk of CVD compared to the low-fat control diet. We therefore continue to provide convincing arguments and data that confirm the absence of any substantial impact of such departures from the protocol on our results and conclusions.
According to Dr. Drazen, editor in chief of the NEJM: "The New England Journal’s review did not alter any conclusions and should raise public trust in science, not erode it".
a) enrollment of household members (partners of a previous participant) without randomization; members of the household of randomized participants were invited to participate and allocated to the same intervention group as their household member. This was done to avoid assigning members of the same household to different diets. Assigning all participants in a household to the same diet was viewed as the best approach to achieve dietary changes in the household. This procedure was inadvertently omitted in the reporting of the protocol and the original NEJM paper.
b) allocation in clusters (by clinic), instead of individual allocation of some participants at 1 of 11 study sites.
Our aim with the republication is to disclose both facts and to provide reassurance that the results of many new analyses to address these minor issues are consistent with our original conclusions that Mediterranean diet interventions enriched with extra-virgin olive oil or mixed nuts are associated with approximately 30% lower risk of CVD compared to the low-fat control diet. We therefore continue to provide convincing arguments and data that confirm the absence of any substantial impact of such departures from the protocol on our results and conclusions.
According to Dr. Drazen, editor in chief of the NEJM: "The New England Journal’s review did not alter any conclusions and should raise public trust in science, not erode it".
IN HOW MANY PARTICIPANTS DID YOU FIND EVIDENCE OF DEPARTURE FROM THE PROTOCOL?
These departures may have affected at most 14% of the 7447 participants. They comprised all the participants (652) of one study site (site D) where some were allocated in clusters and 425 other participants (35 of them from site D) who were members of the same household of a previously enrolled participant. These small departures from the protocol are now comprehensively explained in full detail in the new manuscript and in a 100-page supplement published online in The New England Journal of Medicine together with the new article.
WHAT ARE THE OBSERVED CHANGES IN THE EFFECT OF THE PREDIMED’S MEDITERRANEAN-DIET INTERVENTIONS AFTER ACCOUNTING FOR THESE DEPARTURES FROM THE PROTOCOL
Results were essentially unchanged after correcting for intra-cluster correlations (in the small proportion allocated in clusters, namely families or clinics) and after omitting participants whose allocation was known or suspected to have departed from the individual randomization procedure. Sensitivity analyses by adjusting for a multitude of covariates (>30) using propensity score analysis did not change the results either.
WHAT FURTHER ANALYSES DID YOU PERFORM TO CORRECT FOR THESE DEPARTURES?
We conducted three analyses to correct for these irregularities.
First, we conducted propensity score analysis to adjust for a wide range of covariates and we also accounted for the intra-cluster correlation, and we obtained the same results as shown in our original report.
Second, when we omitted subjects who were known or suspected to depart from individual randomization, the results did not change either.
Third, we considered the actual individual degree of adherence to the Mediterranean diet, year after year. We did a comparison using only those participants who truly adhered to the Mediterranean diet and compared them exclusively with those who truly adhered to a low-fat diet. In this causal comparison, using up to date statistical methods for causal inference (link), the results were even better for the Mediterranean diet, with more than 50% relatively lower risk of cardiovascular events.
The meaning of this new finding is that differences in cardiovascular event rates between participants assigned to the Mediterranean diets and controls were greater among participants with better adherence to this food pattern.
First, we conducted propensity score analysis to adjust for a wide range of covariates and we also accounted for the intra-cluster correlation, and we obtained the same results as shown in our original report.
Second, when we omitted subjects who were known or suspected to depart from individual randomization, the results did not change either.
Third, we considered the actual individual degree of adherence to the Mediterranean diet, year after year. We did a comparison using only those participants who truly adhered to the Mediterranean diet and compared them exclusively with those who truly adhered to a low-fat diet. In this causal comparison, using up to date statistical methods for causal inference (link), the results were even better for the Mediterranean diet, with more than 50% relatively lower risk of cardiovascular events.
The meaning of this new finding is that differences in cardiovascular event rates between participants assigned to the Mediterranean diets and controls were greater among participants with better adherence to this food pattern.
WHO DECIDED TO WITHDRAW THE PREVIOUS PUBLICATION AND TO PUBLISH A NEW ARTICLE?
A 2017 report in an Anaesthesia journal prompted the Steering Committee of PREDIMED to take the initiative to contact the editors of The New England Journal of Medicine. The editors took a lot of care and effort to help us to conduct a thorough review of the randomization procedures in each of the 11 PREDIMED recruiting sites. We immensely appreciate the meticulous work of the editors of the NEJM on our publication during the last months. During this review process, the Steering Committee identified some small departures from the protocol for allocation procedures, which we describe in our updated report and an appendix. We also conducted new statistical analyses.
Subsequently, we proposed the withdrawal of our previous publication and the publication of a new article describing the issues that we identified and showing updated analyses and results.
Our updated analyses confirm the findings and conclusions of our original NEJM paper.
Subsequently, we proposed the withdrawal of our previous publication and the publication of a new article describing the issues that we identified and showing updated analyses and results.
Our updated analyses confirm the findings and conclusions of our original NEJM paper.
WHAT WAS THE ORIGIN OF SOME IMBALANCES IN BASELINE VARIABLES DETECTED IN PREDIMED?
Only in one of the PREDIMED recruitment sites (site D) there were some large imbalances in different baseline characteristics among the three trial arms (Mediterranean diet + extra-virgin olive oil, Mediterranean diet + nuts, control group).
In order to understand these imbalances, the Steering Committee conducted several inquiries and analyses and found that the rate of inclusion of participants by intervention group in site D did not follow the expected distribution. Later, it was known that in site D, 185 participants in 3 clinics were allocated individually and 467 participants (6.2% of the total of PREDIMED participants) in 11 clinics were assigned in clusters, with the clinic being the unit of allocation instead of each participant being the unit of allocation.
In order to understand these imbalances, the Steering Committee conducted several inquiries and analyses and found that the rate of inclusion of participants by intervention group in site D did not follow the expected distribution. Later, it was known that in site D, 185 participants in 3 clinics were allocated individually and 467 participants (6.2% of the total of PREDIMED participants) in 11 clinics were assigned in clusters, with the clinic being the unit of allocation instead of each participant being the unit of allocation.
HOW LARGE WERE THE IMBALANCES IN BASELINE VARIABLES IN PREDIMED?
Despite these departures from pure randomization affecting second members of couples and some participants at site D, the distribution of baseline characteristics in the full trial was well balanced for most characteristics, meaning that no important practical consequences could be derived from these departures.
Please check
http://medpreventiva.es/xmPvj
Please check
http://medpreventiva.es/xmPvj
ADDITIONAL QUESTIONS AND ANSWERS, MARCH 11, 2019
The imperfections regarding randomization procedures in a small subset of PREDIMED participants (OTHERWISE PROPERLY ADDRESSED IN THE NEW PAPER) do not affect THE published papers by our group which followed an observational design (that was the case for most of our published papers-link)
According to Dr. Drazen, editor in chief of the NEJM: "The New England Journal’s review did not alter any conclusions and should raise public trust in science, not erode it".
The Nutrition Source explains the replacement of the main PREDIMED paper in NEJM: LINK
