Milk: Source of Nutrients or “White Poison”?
Milk and Health is an article written by Walter Willett, MD, DrPH and David Ludwig, MD, PhD and published in the New England Journal of Medicine (February 2020). Drs. Willett and Ludwig hail from the Harvard T. H. Chan School of Public Health. Both are respected physicians and nutrition researchers. I wouldn’t be surprised if you’re already familiar with their work.
Willett and Ludwig attempt to cover an immense amount of material in a single journal article with a 3,000 word limit. This results in a series of thinly-referenced general assertions that leaves the reader with the impression that the authors feel you should trust their authoritative knowledge. My curiosity led me to investigate some of their conclusions and my findings are below.
Statement #1: “The United States nutritional recommendations for calcium intake are high.” The article does not define a “high” intake, except in terms of the ratio of recommended daily intake compared to current average intake by U.S. citizens. By that logic, the nutritional guidelines for sodium intake are too low, because Americans consume a lot more than recommended every day.
Statement #2: “Cows have been bred to produce higher levels of insulin-like growth factor I (IGF-I) and are now pregnant most of the time they are milked, which greatly increases the amount of estrogen and other hormones in milk.” No other details are given beyond that remark. The effect of modern agricultural methods on the composition of milk is an important question. For brevity I’ll focus on estrogen. When I read a stand-alone statement like that, I wonder, “from what level before to what level now? Does this have a biological effect? There are various forms of estrogen, some biologically active and some not. How much of the estrogen is absorbed by the gut and gets into the bloodstream?” and some other questions. The authors provide one reference to support their estrogen statement – an article published in 2005 in the journal Medical Hypotheses, an ecologic study correlating average intake of various dietary components with breast, ovarian and uterine cancer rates. In 2005 Medical Hypotheses was a non-peer-reviewed journal that specialized in “radical, speculative and non-mainstream scientific ideas.” A forum for stimulating innovative ideas is useful, but if you want to bolster an assertion in a review article it would be better to provide references reporting original research investigating estrogen levels in consumer milk, types and levels of estrogen, now vs. then, etc. It took me five minutes to find a thoughtful 2018 review of estrogens in milk and their potential biological effects. This review walks through the causal pathway from cow to human disease, including how much and what types of estrogen have been found in commercial milk, and essentially concludes that it is unlikely that estrogens in consumer milk affect adult health (the authors note that very little research is available with respect to the effect of milk on fetal development). This review also notes that the estrogens are stored in milk fat, so if you wanted to avoid them you could drink skim milk. Why didn’t Willett and Ludwig consider (or refute) this source when evaluating levels of estrogen in modern consumer milk?
Statement #3: “Countries with the highest intakes of milk and calcium have the highest rates of hip fractures.” Willett and Ludwig provide a graph showing the average proportion of total energy intake from milk vs. age-standardized osteoporotic hip fracture rates by country. “Age standardized” means the hip fracture rates were adjusted so that the fact that populations of different countries tend to be older or younger has been removed from the comparison. We know that the older you are, the higher your risk of hip fracture. I used their data sources to create this graph. Regarding milk intake, I couldn’t find “proportion of total energy intake from milk” from their source (Food and Agriculture Organization of the United Nations), but that could be because I didn’t know how to search the site as well as they can. However, I did find total milk supply per capita per day and it produces a very similar relationship to what they show. I also added countries that for whatever reason they left out of their graph.
That is an interesting relationship but doesn’t stand on its own due to the ecologic fallacy and the fact that various other phenomena vary among countries. Willett and Ludwig acknowledge that the relationship above may not prove that increased milk consumption predisposes someone to hip fracture, but they state it shows that a low calcium intake is “clearly compatible with low rates of hip fracture.” What they consider “low” is undefined. For levels of milk supply below 200, a country’s hip fracture rate can range from <50 to about 350 per 100,000 population. And if they’re not sure that a high level of milk consumption is associated with increased risk of hip fracture, why do they believe that a low level is “clearly compatible” with a low hip fracture risk?
Graphs examining ecologic (summary, aggregate) relationships between an exposure and an outcome can lead to useful hypotheses about disease causation. However, they rarely if ever provide convincing evidence of a causal relationship. One can develop all manner of ecologic associations between variables. An example is below (remember that the influence of age has been removed from the hip fracture rates). Should we throw “hip fracture parties” to increase life expectancy?
And how about this one? I could add 7-10 years to my life just by drinking lots of milk! Of course, my quality of life will be low because I will have a broken hip.
Statement #4: “The U.S. recommendations for milk consumption are based on studies involving a small number of adults and of inadequate duration.” Willett and Ludwig note that “just 155 adults” form the evidence base for the U.S. recommendations of 1000mg calcium daily for adults, 1200mg for postmenopausal women and 1300mg for adolescents. They counter these recommendations with results from another study that indicated a level of 200mg represents sufficient intake. Thankfully my institution’s health sciences library located this article for me. This reference from 1952 involved ten male inmates of a Peruvian penitentiary. If 155 adults are inadequate, why are ten males acceptable? Are they representative of the current U.S. population? The investigators used a linear regression method (calculation of best straight line fit to their data) which was common at the time but has since been superseded by another method. The estimated calcium balance (and therefore required intake) among the subjects ranged from -310mg to 597mg. Three subjects had negative calcium balances. The authors’ response: “We’ll just call it zero.” So 30% of adults don’t need to ingest any calcium at all? Given the importance of calcium in many physiological functions, this seems implausible. A thorough discussion of calcium balance and needs can be found here. This source noted the variability in estimated calcium requirement from different studies using the linear regression calculation, from the 200mg of the Peruvian prisoners’ study to 990mg for premenopausal women. It’s a somewhat older reference but has a thorough discussion about biological processes that appear to be well delineated. It notes the effect of vitamin D [1,25(OH)2D], protein and sodium intake on calcium absorption which suggests that studying the effects calcium intake alone on health outcomes isn’t very useful.
Dr. Willett is known for promoting a plant-based diet. My theory, and this is just a theory, is that he needs to reduce the recommended daily calcium requirement from 1,000mg because in order to obtain that amount from plants, you will essentially be chewing and pooping all day long eating 10 cups of kale every day or getting fat and going broke eating three cups of almonds every day.
Statement #5: “In international comparisons, dairy product consumption is strongly correlated with rates of breast, prostate and other cancer.” This first sentence concerns findings of ecologic studies, which as previously discussed serve only as starting points in causal inference. The authors feel milk may promote initiation or progression of breast and prostate cancer by increasing levels of insulin-like growth factor I (IGF-I), which is involved in cell growth, differentiation and programmed death (apoptosis). As children our IGF-I levels increase until our late teen years, since we need to grow, and then continually decrease after that. If you’re not familiar with apoptosis, roughly it involves the fact that cells are “scheduled” to die at some point. Cancer cells evade this apoptosis mechanism causing a tumor to grow “out of control.” So the idea with IGF-I is that if blood levels are too high, it may encourage the promotion of cancer. In perusing reference #92 from the Milk and Health paper, there are multiple papers that show higher levels of IGF-I in the blood of breast cancer patients than in age-matched controls (anywhere from 2 – 20 ng/mL difference). However, the question is not only whether these differences are clinically important, but were the elevated IGF-I levels present prior to the initiation or growth of their breast cancers, or possibly have they resulted from production by the cancer tissue? And is this biologically active or inactive IGF-I? The Women’s Health Initiative-Observational Study examined the effects of milk consumption on IGF-I levels in 747 women. What they found was that milk consumption increased free (biologically active) IGF-I levels by an average of 18.6%, but did not affect total IGF-I levels. This sounds like a fairly large effect, but free IGF-I in blood is only about 1% of all blood IGF-I. So the free IGF-I increased from 1% to 1.2% of total IGF-I. Is this biologically meaningful?
The authors also state that “Total dairy intake has been associated with a greater risk of endometrial cancer, particularly among postmenopausal women who are not receiving hormone therapy, a finding possibly related to the sex-hormone content of dairy products.” We are back to the effects of estrogen in milk again. This statement is referenced by a paper which included Willett as an author (#94). The connotation is that milk increases the risk of endometrial cancer for all women, but especially pre-menopausal women. However, the manuscript for reference #94, a Nurses’ Health Study analysis, presents data that for all women, neither total dairy, low fat milk, whole milk, yogurt or hard cheese were associated with risk of endometrial cancer after adjusting for important confounders such as hormone use and body mass index (BMI – obesity is a known risk factor for endometrial cancer). They did find in subgroup analyses that at least three servings of total dairy per day was associated with increased endometrial cancer risk in postmenopausal women who had not used hormone therapy. So only the last part of their statement in Milk and Health is consistent with the findings of their own study. In fairness, the authors do acknowledge that high milk intake is associated with a reduced risk of colorectal cancer – this finding is fairly consistent across studies.
Statement #6: “Large meta-analyses of cohort studies show only weak or no association between dairy intake and incidence of type 2 diabetes.” That’s interesting, since the reference provided to indicate “no association” was a paper which included Willett as an author, and which showed that yogurt consumption was linked with a reduced risk of type 2 diabetes with a risk ratio of 0.82 (95% confidence interval: 0.70-0.96). The last statement in the diabetes section says that the risk of diabetes with milk consumption is increased when it replaces coffee drinking and is referenced by another Willett-authored paper (#89). I couldn’t find this statement in that paper. In table 4 of the supplementary material, the risk ratio for coffee drinking shows a stronger reduction in type 2 diabetes risk than that for milk (which also shows reduced risk). I’m fine with that since I’m a coffee drinker. However, milk and coffee are not directly compared in that analysis. It’s an inappropriate extrapolation to say that coffee and milk have differential effects on type 2 diabetes risk based on table 4 alone.
Some general thoughts. There are other assertions in Milk and Health that I could critique, but this post is already a monster. I have focused on statements that I feel are misleading, but the authors do make some reasonable statements and ask some important questions. I don’t closely follow recommendations for milk drinking but can imagine it may be promoted for certain health benefits based on slim evidence. Does milk prevent cancer, type 2 diabetes, obesity, hip fractures? Given the complexity of our bodies, does any single food item? My impression is that Willett and Ludwig have pivoted too far to the other side and painted milk almost as a “white poison.” This is hinted at in some of their language. They talk of the “implications” and “consequences” of drinking milk, rather than use more neutral terms like “effects.” They bring up the environmental effects of agricultural practices in milk production (e.g., increases in greenhouse gas and water consumption). This is concerning to me too, but is not directly related to the health effects of milk consumption. It’s just another sound byte to convince us to stop drinking milk. This paper might be more accurately titled, “Milk and Death.”
In doing some background work for this article review, I searched PubMed for any article authored by Dr. Willett, and which involved “milk” or “dairy.” That identified 97 articles. Some of these were on topics not related to anything covered in Milk and Health, were only incidentally related to milk and dairy, or were reviews. I focused on 38 articles which addressed at least one of the Milk and Health topics. This is not perfect but should not suffer from too much selection bias.
Six articles found harmful effects of milk or dairy on health, 16 showed no relationship, and 14 identified a beneficial effect of milk on health. Of the six harmful effects, three were cited in Milk and Health. These harmful effects were generally fairly small or applied to a small segment of the population under study. For example, reference #106 suggests that consuming more than four servings of dairy every day increases all-cause mortality. Another non-referenced article found that consuming >1,500mg calcium daily increased the likelihood of advanced or fatal prostate cancer. These intakes are on the extremes of the distribution for the U.S. population. Seven of the 16 neutral articles were cited in Milk and Health, while four of the 14 articles suggesting a beneficial effect of milk were. Willett and Ludwig added eight additional references showing harmful effects of milk or dairy, nine additional neutral references, and five references suggesting beneficial effects (three related to colorectal cancer). So, they overweighted the Milk and Health references in favor of those which identified harmful associations and underweighted those identifying beneficial associations, at least with respect to Willett’s total research on the subject.
In a previous job I worked with masters-prepared dietitians. They told me that for the vast majority of us there is no need to delete any major food group from our diet, and that there is no “one right” diet for everyone. Milk appears to be a reasonable nutrient-dense food choice – even Harvard thinks so. It is also not clear from Milk and Health that three eight-ounce servings daily is “too much.” I would not bank on any form of milk or dairy keeping me from getting cancer, heart disease, diabetes or any other disease.
When I was younger and heard people with respected credentials make claims I wasn’t sure of, I would think they knew better than me. I don’t think that anymore.