One of my classmates on the eCornell Certificate Programme (Plant-Based Nutrition) runs the excellent blog Simplify Your Health, and has recently written an interesting post about the health benefits of nuts. One of the references she used to support her argument was a media report on the PREDIMED study I wrote about earlier: Primary Prevention of Cardiovascular Disease with a Mediterranean Diet.
I was very interested to read this report by Elaine Watson (Loma Linda, California, 26-Feb-2013):
Mediterranean diet cuts heart disease risk by 30%: ‘Landmark’ study provides compelling evidence that it’s the type, not the level of fat, that counts for cardio health
People following an energy unrestricted plant-based diet supplemented with extra-virgin olive oil or nuts can reduce their risk of a major cardiovascular event by 30% compared with people following a purely low-fat diet, according to a “landmark” new study.
In particular I was fascinated to understand how the study team were able to attribute a cardiovascular risk reduction of 30% to the Mediterranean diet when the study showed an actual reduction in incidence of 0.6% for the Mediterranian diet + olive oil group and 1% for the Mediterranean diet + mixed nuts group. This is about to get a bit mathematical, but stick with me if you can – I find this staggering:
Total participants in the study = 2447
Number of participants in the Mediterranean diet + olive oil group = 2543
Number of participants in the Mediterranean diet + nuts group = 2454
Number of participants in the “low-fat” control group = 2450
Total number of cardiovascular events during follow-up = 288
Number of cardiovascular events in the Mediterranean diet + olive oil group = 96
Number of cardiovascular events in the Mediterranean diet + nuts group = 83
Number of events in the “low-fat” control group = 109
% of those in the Mediterranean diet + olive oil group that experienced and event (incidence proportion*) = 96/2543 x 100 = 3.8%
% of those in the Mediterranean diet + nuts group that experienced an event (incidence proportion*) = 83/2454 x 100 = 3.4%
% of those in the “low-fat” control group that experienced an event (incidence proportion*) = 109/2450 x 100 = 4.4%
So from this we can see that the actual difference in incidence proportion between the Mediterranean diet and olive oil group and the “low-fat” control group was 0.6%. Between the Mediterranean diet and mixed nuts group and the “low-fat” control group it was 1.0%.
How on earth, then, did the article above come to state that this study demonstrated reliably that consuming a Mediterranean diet high in olive oil and nuts resulted in a 30% reduction in cardiovascular risk?
The answer lies in the concept of a hazard ratio (HR)**. The hazard ratio is the incidence in the experimental group as a proportion of the incidence in the control group. For demonstration purposes, if they had used the incidence proportions as given above the hazard ratios would be:
HR for the Mediterranean diet + olive oil group = 3.8/4.4 = 0.86
HR for the Mediterranean diet + mixed nuts group = 0.77
In fact they adjusted their incidence proportions to give figures per thousand person-years, and based on this arrived at hazard ratios of 0.70 for both groups. We don’t need to worry about this difference – it is simply an adjustment for things like loss to follow-up.
What this says is that if you were in either experimental group your risk of suffering a cardiovascular event during the study period was proportionally 30% less than if you were in the control group. All so far so good, but here’s the rub. Let’s just go back to our original hazard ratio of 0.86 for the olive oil group for a minute. Let’s say that the incidence proportion in the olive oil group had been 0.38% instead of 3.8 (tenfold less), and that for the control group had been 0.44%. What is the hazard ratio? Well, it is 0.38/0.44, which is 0.86! And what if the incidence proportions had been 38% and 44% (tenfold more)? Well, the hazard ratio would be 0.86! The same!
The thing is, incidence proportions of 0.38% and 0.44% in this study would correspond to actual numbers of events of 9.6 in the olive oil group and 10.9 in the control group (an actual reduction of 1.3 cases or 0.06% for the olive oil group), whereas incidence proportions of 38% and 44% would correspond to actual numbers of events of 960 in the olive oil group and 1090 in the control group (an actual reduction of 130 cases, or 6%. Either way, the hazard ratio would be the same, indicating a 30% reduction in risk, but the actual difference in incidence would be very different.
You will remember that the actual reduction in incidence for the olive oil group compared to controls was 0.6%, and for the mixed nuts group compared to controls it was 1%. I am not at all sure that this difference would inspire anyone to make the effort to significantly change their diet, but these are not the figures the study team have chosen to present to the media and the public. The 30% figure derived from the hazard ratio is not wrong, so they are not lying, but it could perhaps be viewed as an inflation or misrepresentation of their true findings, relying for its impact on the fact that the majority of people will never read the original paper and have never heard of a hazard ratio.
[N.B. I note that other commentators have raised similar issues, including Dr Dean Ornish: see New England Journal of Medicine, Letters to the Editor, here)
*From Wikipedia: Incidence is a measure of the risk of developing some new condition within a specified period of time. Although sometimes loosely expressed simply as the number of new cases during some time period, it is better expressed as a proportion or a rate with a denominator. Incidence proportion (also known as cumulative incidence) is the number of new cases within a specified time period divided by the size of the population initially at risk. For example, if a population initially contains 1,000 non-diseased persons and 28 develop a condition over two years of observation, the incidence proportion is 28 cases per 1,000 persons, i.e. 2.8%
**From Wikipedia: In its simplest form the hazard ratio can be interpreted as the chance of an event occurring in the treatment arm divided by the chance of the event occurring in the control arm, or vice versa, of a study