The study argues that “regular consumption of egg yolk should be avoided by persons at risk of cardiovascular disease”. It hints at egg yolks being unhealthy in general, possibly even more so than cigarettes. Solid critiques have already been posted on blogs by Mark Sisson, Chris Masterjohn, and Zoe Harcombe (, , ), among others.
These critiques present valid arguments for why the key findings of the study cannot be accepted, especially the finding that eggs are more dangerous to one’s health than cigarettes. This post is a bit different. It uses the data reported in the study to show that it (the data) suggests that egg consumption is actually health-promoting.
I used the numbers in Table 2 of the article to conduct a test that is rarely if ever conducted in health studies – a moderating effect test. I left out the “egg-yolk years” variable used by the authors, and focused on weekly egg consumption (see Chris’s critique). My analysis, using WarpPLS (), had to be done only visually, because using values from Table 2 meant that I had access only to data on a few variables organized in quintiles. That is, my analysis here using aggregate data is an N=5 analysis; a small sample indeed. The full-text article is not available publicly; Zoe was kind enough to include the data from Table 2 in her critique post.
Below is the model that I used for the moderating effect test. It allowed me to look into the effect that the variable EggsWk (number of eggs consumed per week) had on the association between LDL (LDL cholesterol) and Plaque (carotid plaque). This type of effect, namely a moderating effect, is confusing to many people, because it is essentially the effect that a variable has on the effect of another variable on a third. Still, being confusing does not mean being less important. I should note that this type of effect is similar to a type of conditional association tested via Bayesian statistics – if one eats more eggs, what is the association between having a high LDL cholesterol and plaque buildup?
You can see what is happening visually on the graph below. The plot on the left side is for low weekly egg consumption. In it, the association between LDL cholesterol and plaque is positive – eating fewer eggs, plaque and LDL increase together. The plot on the right side is for high weekly egg consumption. In this second plot, the association between LDL cholesterol and plaque is negative – eating more eggs, plaque decreases as LDL increases. And what is the turning point? It is about 2.3 eggs per week.
So the “evil” particle, the LDL, is playing tricks with us; but thankfully the wonderful eggs come to the rescue, right? Well, it looks a bit like it, but maybe other foods would have a similar effect. In part because of the moderating effect discussed above, the multivariate association between LDL cholesterol and plaque was overall negative. This multivariate association was estimated controlling for the moderating effect of weekly egg consumption. You can see this on the plot below.
The highest amount of plaque is at the far left of the plot. It is associated with the lowest LDL cholesterol quintile. (So much for eggs causing plaque via LDL cholesterol eh!?) What is happening here? Maybe egg consumption above a certain level shifts the size of the LDL particles from small to large, making the potentially atherogenic ones harmless. (Saturated fat consumption, in the context of a nutritious diet in lean individuals, seems to have a similar effect.) Maybe eggs contain nutrients that promote overall health, leading LDL particles to "behave" and do what they are supposed to do. Maybe it is a combination of these and other effects.
24 comments:
Ned, great work.
What a nice piece of analysis.
Nothing better serves as Comment but The Bard himself:
"For 'tis the sport to have the engineer
Hoist with his own petard."
Hamlet
(For engineer read epidemiologist.)
Slainte
PS I eat a lot of eggs . . . . !
Great stuff Ned. I can't believe studies like these pass for science nowadays. Not only were the 'egg' data gotten in a very unreliable fashion, but did they also conclude the wrong things out of what they did have. Good thing people like you are setting the record straight! :)
Hi Ned,
Great to meet you at AHS!
It seems like this analysis is pretty limited for the reasons you stated, i.e. n=5 of data from quintiles, and it seems hard to make much of it, especially given that the axis for plaque is positive on the right and negative on the left.
Nevertheless, Maria-Luz Fernandez from UConn has done a few studies showing that eggs do exactly that: they either have no effect on cholesterol, or they increase total-cholesterol without any effect on the HDL/LDL ratio, and increase particle size.
Chris
Hi Leon. Well, every researcher has some preconceptions, so we’ll always have different points of view. The problem with the original analysis is actually one of collinearity.
Hi Nils. The main problem with the original study, as I see it, is collinearity. And, unfortunately, collinearity is rarely tested for in epidemiology studies. Controlling for confounders is nice, but age, egg-yolk years, and even eggs/wk are very strongly correlated, to the point of being collinear.
Whenever you introduce collinearity (a.k.a. multicollinearity) to a model by adding variables as predictors (or independent variables), you can severely distort the results:
http://bit.ly/NoVr91
http://www.youtube.com/watch?v=avPWO324E0g
The authors should have checked for collinearity, which is not hard to do. If they found two or more variables were collinear they could not normally use both variables in their model, because the variables would have been measuring the same “thing.
Hi Chris. Since the values are standardized, the negative plaque values are those below the mean. But, indeed, what you noted is correct. This may well be a reflection of the fact that eggs/wk is not only strongly correlated with age, but also with smoking.
The moderating effect analysis discussed here is an attempt to extricate a possible effect that is submerged in a pool of collinearity. I wish I had access to the original data, as working with the quintiles only is far less than desirable.
A couple of additional posts of interest regarding consumption of cholesterol and saturated fat, and associations with LDL particle size and HDL cholesterol:
Want to improve your cholesterol profile? Replace refined carbs and sugars with saturated fat and cholesterol in your diet
http://bit.ly/NbXxPV
Low fasting triglycerides: A marker for large-buoyant LDL particles
http://bit.ly/PMWTbu
Thanks for this article. Hope it is OK if I crossposted it at www.wmodavis.com. Let me know if not.
@Chris:
>Maria-Luz Fernandez from UConn has done a few studies
Oh yes, nicely sponsered by the "American Egg Board" and the "Egg Nutrition Center".
That's what I would call "independent research"... ;-)
Cheers, guzolany
http://www.canr.uconn.edu/nutsci/nutsci/hpg/mluz.html
Nice work. I can already see the days when papers official publication will be subject to "bloggers review". May be there will be two statuses of publication: pre bloggers review and post bloggers review. Newspapers can than wait for the post bloggers version and save their readers a lot of confusion.
Ned,
> I used the numbers in Table 2 of the article to conduct a test that is rarely if ever conducted in health studies – a moderating effect test... This type of effect, namely a moderating effect, is confusing to many people, because it is essentially the effect that a variable has on the effect of another variable on a third.
This is precisely the "decoupling of coupled systems" that I referred to in my response in your 14% advantage post. Anyone who's gone through undergrad level differential equations class should not be making these types of blunders as a professional scientist.
Still, being confusing does not mean being less important.
As my rocket scientist friend told me when we were in grad school, "People will ignore variables until they understand the problem."
Congrats for exposing the ridiculousness of such studies.
Great analysis. Of course conventional wisdom just wants us to stop eating eggs and chew on fat-free processed crap instead.
Hi PI Bill,
As I'm sure you're aware, hardly any studies are conducted independently of outside grant funding, and even the few that are independent of outside funding are not independent of the researchers' own biases. You may also be aware that public funding, which tends to carry the veneer of "independence" from commercial interests, is limited.
Of course it is desirable to see research replicated by different investigators funded by different sources so that influence of funding source and investigator can be teased out. Sometimes that is not possible because only one group has conducted studies of a certain type.
Are you aware of studies by other investigators that contradict those conducted by Fernandez's group?
Chris
It is indeed very difficult to do research without funding. For example, the vast majority of doctoral students have scholarships, and the money for that must come from somewhere.
Also, a priori beliefs always influence research and interpretation of results. For instance, one main motivation for me (and also others) to look deeper into this egg study was our belief that egg consumption is generally healthy. That was an a priori belief based on past research results.
This is why we need research to be done by different people and groups, with different perspectives. In the end, the preponderance of the evidence should win.
Btw, very interesting research by Dr. Fernandez, including that related to the effect of egg consumption on macular degeneration (mediated by lutein intake). Her web page for those interested:
http://www.canr.uconn.edu/nutsci/nutsci/hpg/mluz.html
"And what is the inflection point? "
You mean the 'turning point', don't you?
As some of the others have said, this is a nice piece of work.
I add my wife's comment: "Pah! How many men know how many eggs they eat a week?"
Hi dearieme. Yes, “turning” point. Thanks, I’ll correct it.
js290: "Anyone who's gone through undergrad level differential equations class should not be making these types of blunders as a professional scientist."
Heck, it doesn't even take calculus. Colinearity is something students are taught to check for in the most elementary stat classes.
happy blogging
Sorry sir i did not understand which type that blogs could send me mail about that Power Prash
Thaks
I hope you are right because I eat 15 to 20 eggs per week. Thanks for a great article.
This is interesting, Ned. However, the fact remains that eggs raise one's LDL, which increases risk for heart disease. I don't think you can quantify this and somehow "refute" what most of the scientific experts have a consensus on.
In other words, you can't say this reduces the risk. Perhaps the higher LDL ends up increasing your risk more than the plaque-removal lowers it.
Post a Comment