Just Read: Dairy fat, not associated with cardiovascular disease – How to make it look like it is

2018.05 Low Carb and Low Carbon – Ted Eytan MD-1001 1115 – (View on Flickr.com)

I’ve recently put out a solicitation for physicians to send me evidence that low carbohydrate or animal source foods are harmful for our health. This is another breakdown of one paper (see my previous one: Just Read: Saturated Fat in Meat is not associated with Cardiovascular Disease – How to make it look like it is ) sent to me by a physician colleague purporting to show that there’s an association between dairy fat and cardiovascular disease, with a tangential connection to meat and plant foods. Th The quick answer on this paper is – no evidence.

My method

I usually look at these aspects of any study I review

  • Is there conflict of interest on the part of the authors / study?
    • This includes financial as well as ideological
  • Is the methodology fit for purpose for the question being asked?
  • What are the results, and are they robust?
  • Is there newer/better research out there on this topic?


  • I read the whole paper
  • I grab the supplemental materials – often the real conclusions are stuffed there
  • I look up any tools used, including food frequency questionnaires

Let’s go….

Conflict of interest level – concerning (and expected)

This particular paper’s conflict of interest section says

None of the authors reported a conflict of interest related to the study.Chen, Mu, Yanping Li, Qi Sun, An Pan, JoAnn E Manson, Kathryn M Rexrode, Walter C Willett, Eric B Rimm, and Frank B Hu. “Dairy Fat and Risk of Cardiovascular Disease in 3 Cohorts of US Adults.” The American Journal of Clinical Nutrition 104, no. 5 (November 1, 2016): 1209–17. https://doi.org/10.3945/ajcn.116.134460.

There is more to this story. I mentioned ideological biases are important as well. Before we get into that, I’m lifting part of the conflicts from the last paper I reviewed because it’s material to this research group and its home institution, the Harvard School of Public Health:

Harvard University has filed a provisional patent application that has been assigned to Harvard University and lists DM as a coinventor to the US Patent and Trademark Office for use of trans palmitoleic acid to prevent and treat insulin resistance, type 2 diabetes, and related conditions.https://doi.org/10.3945/ajcn.112.037770.

This research group and institution are strong proponents of polyunsaturated oils for health. Dr. Willett, one of the principle authors and a world renown nutrition researcher, does not write articles supporting meat or saturated fat as healthy. I’ll put out a call to anyone reading this to find an article that refutes this statement. One could say, “that’s because the association between saturated fat and health is so strong,” which could be true. This would be the case if every piece of published research supported this idea; however, it does not.

I’m linking to an article by the Harvard School of Public Health, extolling the virtues of industrial vegetable oils, complete with industrial vegetable oil stock photography. It’s a strong statement of support.

There’s actually a dossier of Dr. Willett’s alleged conflicts available here: Walter Willett, Potential Conflicts of Interest | Vegetarianism | Diet (Nutrition).

In it, there are details of financial contributions from entities that support plant-based nutrition, as well as the allegation that Dr. Willett does not disclose these conflicts on disclosure forms.

Either way,it’s accurate in this case to look at the body of published literature which is very supportive of plant-based nutrition and polyunsaturated oils. As just one example, in a recent review of Milk and Health, Dr. Willett referenced this paper and reiterated his support for the strange methodology used in it…we’ll get to that.

Methodology – (another) Food Frequency Questionnaire observational study

Again, these types of analyses are best used for exploration rather than finding out if X causes Y, and suffer from a myriad of problems. Unlike the previous study I reviewed, this one relies on many many more food frequency questionnaires, given over several years. That’s good in that the questionnaires are standardized, and there’s much more power in them. It doesn’t change the inherent limits of food frequency questionnaire.

As I recommended, because we have to, I pulled and reviewed the food frequency questionnaires used here. The Harvard School of Public Health has done a nice job organizing them over the years:

The advantage of an analysis in this case is the category of foods being reviewed is more narrow. The problem with this and all other food frequency questionnaire studies is we can never know what all the sources of a food substance are – is dairy fat in the cheese puffs? Was it consumed in organic cheese or ultra-processed cheese? Researchers looked at biomarkers for dairy intake as well, however this is not very robust. They did their best with the data they have, and again, they have a lot of data.

National Nutrition Month - 19961996’s National Nutrition Month, when eating fat was a really really bad thing to do. Source: National Nutrition Month Overview

Pay attention to the years of the food frequency questionnaire and what the secular food trends were. This will help, it’s a poster from National Nutrition Month in 1996. What you can see is that a higher dairy fat intake is probably going against conventional wisdom of the time, and especially so for this well-educated, health professional cohort.

The descriptive statistics of the cohorts are fairly brief but also give a healthy user bias picture – the higher dairy fat consumers smoked more and exercised less. Their blood pressure was lower, though.

It’s also important to note that the source of the dairy fat is not categorized – this is not a study of food eaten, it’s a study of food substance. This turns out to be a big issue in later studies, as the source of the dairy fat (the “food matrix”) is everything, so even the research question is not relevant to an eater today.

Because the authors decided to go even farther than this analysis, we have to look at the way animal fat is assessed, and this is where things become really messy. The authors don’t describe how they do this in the paper at all, so it’s a black box. You can look at examples from my previous post to see how fraught with error this is. In a nutshell, it’s very easy to blame the hamburger for what the bun did – the questionnaires aren’t specific enough. If they were, the form would have 1,000 questions on it instead of 131, I’m sure.

The results – problematic

Using statistical manipulation, the researchers attempted to separate the impact of dairy fat in the diet on cardiovascular disease. And, their analysis turned up an answer of “no impact”

In 3 prospective cohorts of US men and women, we showed that, compared with an equivalent amount of energy from carbohydrates (excluding fruit and vegetables), intake of dairy fat was not associated with risk of CVDChen, Mu, Yanping Li, Qi Sun, An Pan, JoAnn E Manson, Kathryn M Rexrode, Walter C Willett, Eric B Rimm, and Frank B Hu. “Dairy Fat and Risk of Cardiovascular Disease in 3 Cohorts of US Adults.” The American Journal of Clinical Nutrition 104, no. 5 (November 1, 2016): 1209–17. https://doi.org/10.3945/ajcn.116.134460.

One would think that would be it. However, it’s not.

What the researchers then did is a mathematical manipulation of removing diary fat in the diet and replacing it with other substances, like, vegetable fat, animal fat, poly-unsaturated fats, etc etc etc. This analysis earned an actual chart in the paper. The base analysis did not. I created that, above.

What’s wrong with doing this?

  1. Most people don’t understand it (I’m not actually sure I do)
  2. The researchers never explain how they did it, nor did they show the numbers behind it.
  3. The papers they reference on how they designed this method don’t explain it either, and they are references to their own work.
  4. They didn’t actually substitute foods in people’s diets, they just modeled doing that. There’s no basis for this being accurate.

I pinged a few colleagues and elite scientists on this technique and the feeling was one big head-scratch. If elite scientists/physicians cannot make sense of this, how should physicians who don’t do this for a living?

If we take this to a practical level, we’re talking about taking a human being and doing a magical swap for all the diary fat in their diet for something else, like animal fat, or plant fat, or whole grain carbohydrates, which is what they modeled in this paper. That’s not changing a diet, that’s changing a person.

In addition, there are significant nutritional consequences for making a wholesale change of foods containing dairy fat to foods like, “carbohydrate from whole grains” both from a micronutrient and macronutrient perspective. This is not a methodology fit for purpose of the question and it can’t be used to guide a dialogue about causation or dietary recommendations.

But guess what, health professionals all over the internet interpreted them as such:

Nowhere in this paper is there evidence that a person should do a wholesale switch of foods containing dairy fat with whole grain carbohydrates and have less CVD as a result.

The authors included the coded language that all physicians know quite well:

Therefore, whether the food matrix may modify the effect of dairy fat on health outcomes warrants further investigation. Last, because of the observational nature of our cohorts, although we adjusted for established and potential risk factors for CVD, unmeasured and residual confounding was still possible.Chen, Mu, Yanping Li, Qi Sun, An Pan, JoAnn E Manson, Kathryn M Rexrode, Walter C Willett, Eric B Rimm, and Frank B Hu. “Dairy Fat and Risk of Cardiovascular Disease in 3 Cohorts of US Adults.” The American Journal of Clinical Nutrition 104, no. 5 (November 1, 2016): 1209–17. https://doi.org/10.3945/ajcn.116.134460.

This is what’s known as the “eating a waffle under a hedge” statement.

They should have stopped at their original analysis, which flawed as it might be as an epidemiological observation study, has some robustness (but yet still lacking in proof of causality regardless).

Is there newer, better research out there?

This paper is now 5 years old and the swath of research done on the health impacts of dairy foods is dizzying. The way I can avoid summarizing all of that research and stay in scope here is to say that I believe this paper was sent to me as a proof point for the health impacts of meat in the diet.

For that question there are newer analyses as I touched on previously.

  1. JACC State of the Art Review on Saturated Fat – See: Saturated fat does not clog the arteries, increase CVD risk or increase diabetes risk. The end.
  2. NUTRIrecs series – See: Most important part of this week’s published nutritional guideline recommendations – personal preferences and values are respected
  3. The PURE study – See: Just Read: More Validation of Low(er) Carb High(er) Fat Diets: The PURE Study and Lipids

For the question about dairy, the PURE team published their findings which contradict the modeling done here, and they didn’t do mathematical swapping. Dr. Willett collaborated with Dr. David Ludwig more recently in the Milk and Health article and there he repeats the assertions and validity of this substitution technique, again without any explanation. A group of researchers (funded by the dairy industry) duplicated this approach using NHANES data (see below) and found problems with this approach, most notably in impact on micronutrient intake and diet quality. Feel free to review on your own.

What I have to say about dairy is – conflicts of interest are every where – it feels that the truth is out there somewhere. Read with a critical eye as I’ve done here.

  1. Hirahatake, Kristin M, Arne Astrup, James O Hill, Joanne L Slavin, David B Allison, and Kevin C Maki. “Potential Cardiometabolic Health Benefits of Full-Fat Dairy: The Evidence Base.” Advances in Nutrition, 2020. https://doi.org/10.1093/ADVANCES/NMZ132.
  2. Willett, Walter C., and David S. Ludwig. “Milk and Health.” Https://Doi.Org/10.1056/NEJMra1903547 382, no. 7 (2020): 644–54. https://doi.org/10.1056/NEJMRA1903547.
  3. Rehm, Colin D., and Adam Drewnowski. “Replacing Dairy Fat With Polyunsaturated and Monounsaturated Fatty Acids: A Food-Level Modeling Study of Dietary Nutrient Density and Diet Quality Using the 2013–16 National Health and Nutrition Examination Survey.” Frontiers in Nutrition 6 (August 6, 2019): 113. https://doi.org/10.3389/fnut.2019.00113.

My conclusion (again) – try again

For the reasons that this paper did not directly assess animal sourced food intake, the base analysis showed no impact from dairy fat, and has been superseded by newer research that looks at the food matrix, this paper fails the test of usefulness in guiding health decisions. On to the next!


  1. Bernstein, Adam M., Qi Sun, Frank B. Hu, Meir J. Stampfer, JoAnn E. Manson, and Walter C. Willett. “Major Dietary Protein Sources and Risk of Coronary Heart Disease in Women.” Circulation 122, no. 9 (August 31, 2010): 876–83. https://doi.org/10.1161/CIRCULATIONAHA.109.915165.
  2. Chen, Mu, Yanping Li, Qi Sun, An Pan, JoAnn E Manson, Kathryn M Rexrode, Walter C Willett, Eric B Rimm, and Frank B Hu. “Dairy Fat and Risk of Cardiovascular Disease in 3 Cohorts of US Adults.” The American Journal of Clinical Nutrition 104, no. 5 (November 1, 2016): 1209–17. https://doi.org/10.3945/ajcn.116.134460.
  3. Halton, Thomas L, Walter C Willett, Simin Liu, JoAnn E Manson, Meir J Stampfer, and Frank B Hu. “Potato and French Fry Consumption and Risk of Type 2 Diabetes in Women.” The American Journal of Clinical Nutrition 83, no. 2 (February 1, 2006): 284–90. https://doi.org/10.1093/ajcn/83.2.284.

My disclosures

I have none, including that I am not an advocate or activist for any specific dietary pattern. I am an advocate for the desire to know what’s accurate when it comes to dietary patterns. My disclosures page is available here.

Ted Eytan, MD