From Can Drinking Too Much Milk Make Your Bones More Brittle?

The study, conducted with over 60,000 women (age 39-74) and 45,000 men (age 45-79) found that too much milk – three or more glasses a day – was not only associated with mortality but also an increased risk of fracture and hip fracture. Researchers found this surprising association after following the men and women in this study for 22 and 13 years respectively. Over this time, study participants completed questionnaires about their milk-drinking habits.

After adjusting for a other variables, they found that women who reported drinking three or more glasses of milk each day nearly doubled their risk of death in relation to women who drank less than one glass each day. Men were not as affected as women, but those who drank three or more glasses of milk each day still showed a significant increase in mortality.

This just seems too incredible to believe. Doubled the risk of death!

Should this study be trusted? Are there contradicting studies? Assuming the findings are accurate, is this effect due to the hormones in modern milk, the extra calories, excessive calcium or something else?

Update: I found this apparently contradictory article among the 200 citing the original:

Finally, consumption of milk and dairy products was not associated with all-cause mortality.

It would certainly be nice to have more clarity here.

Did this study control for obesity? (Obese people are probably more likely to consume a lot of food including milk)

  • 24
    Don't we all have a 100% chance of death?
    – RedSonja
    Commented Sep 17, 2018 at 11:16
  • 1
    Any idea if any significant number of the people at that age, who are drinking that much milk, are doing so in response to finding out they have bone density issues? So, correlation instead of cause? Commented Sep 17, 2018 at 20:27
  • @RedSonja - Yes, and these people even had a 200% chance of death!
    – Battle
    Commented Sep 20, 2018 at 12:44
  • @MaxB it does, as my answer explains, under "does the study address potential confounders"
    – De Novo
    Commented Sep 22, 2018 at 2:22
  • 2
    Possible duplicate of Does drinking milk increase risk of death?
    – paradisi
    Commented Oct 18, 2018 at 8:06

2 Answers 2


Does drinking more than 3 glasses of milk a day double your risk of death?

The study was not designed to answer this question

The claim in your question title and in the headline of the Cleveland Clinic blog you linked ("Can Drinking Too Much Milk Make Your Bones More Brittle?") are both about causality. The study your question and this blog post are about is a cohort study, a prospective observational study that looks at baseline characteristics, or risk factors, and investigates their association with an outcome. In a traditional model (see Hulley's Designing Clinical Research) cohort studies identify potential risk factors. The impacts of modifying potential risk factors are then investigated in a different type of study, an interventional study, designed for causal inference. While a cohort study can use a variety of strategies to help enhance causal inference (see Chapter 9 in Hulley) there is always a risk of confounding by variables that aren't measured. There are a few specific problems with the study, but the study design itself is the main problem with the causal claim stated in your question. Though the study authors do advocate for a change in recommendations (which they shouldn't), the blog post you linked to does say:

while the study raises interesting questions, there is not strong enough evidence to warrant a restriction on milk

Does the study address potential confounders?

A confounding variable is a variable that is both associated with a causal variable and an outcome variable, but is not on the causal pathway. A classic example is the association between drinking coffee and developing pancreatic cancer. Drinking coffee is associated with developing pancreatic cancer, but this is only because people who smoke are more likely to drink coffee than people who don't smoke. Whether or not a smoker drinks coffee, they are more likely to develop pancreatic cancer than a nonsmoker.

The study addresses some confounders, but cannot dispense with the problem of potential confounding. You asked in an aside whether the study controlled for obesity. It did, by including BMI in the multivariate model, which you can see in the Statistical analysis subsection under methods. In general, the methods used in the statistical analysis were appropriate and can decrease the concern for confounding by the covariates in the multivariate model (see the second paragraph), but the problem remains: you cannot account for confounders not in the model. You can only do that by random assignment of a potential risk factor, which is an entirely different research design.

Other studies show the opposite association

Are there contradicting studies?

Yes. In addition to the meta-analysis mentioned in @Laurel's answer and the article you linked in the edit to your question, the largest and most comprehensive observational study of the impact of diary consumption on mortality, published online in the Lancet just a few days ago, shows the opposite effect. I would note that each of these studies have the same problem as the study making the opposite claim. They are not designed to tell you what will happen if you change your milk drinking habits, or, what is more relevant to the medical community, what will happen if a doctor or public health agency suggests you change your milk drinking habits.

  • 2
    +1 for pointing out the importance of study design. I'd argue that not only unknown/unaccounted for confounders are a problem, but there's also a hen and egg problem: correlation does not imply causation, and if there is causation, we still have to think about its direction. The study in fact says "Theoretically, the findings on fractures might be explained by a reverse causation phenomenon, where people with a higher predisposition for osteoporosis may have deliberately increased their milk intake." + give some points against this reverse causation (IMHO some more, some less compelling). Commented Sep 21, 2018 at 22:13
  • @anonymized thank you! Though it's hard to consider the reverse causation phenomenon for the association they found between death and milk drinking, there's certainly a possibility for fractures, as you say :) They did address this in part with a sensitivity analysis, if I recall, and a prospective cohort makes it easier to deal with reverse causality than a cross sectional study (you can identify the temporal relationship between exposure and event), but it's still always something to think about when interpreting an association.
    – De Novo
    Commented Sep 21, 2018 at 22:19
  • To me it isn't hard to come up with plausible stories for reverse causation of mortality: all we need is a population with higher mortality who believes that it is healthy to drink milk and acts accordingly. E.g. for the CVD deaths having high blood pressure and cutting out coffee replacing it by milk. Fermented (with lower risk) vs. non-fermented milk consumption may have a correlation as well to socioeconomic status (imagine 70 year old women who drink Kefir vs. 70 year old women who drink milk). Ahh, far to easy to invent plausible stories... Commented Sep 21, 2018 at 22:57
  • 1
    @MaxB that's a misinterpretation of the problem. It's not a statistical flaw, it's a relationship the study simply cannot measure. If that's unconvincing, consider that the other studies show the opposite association (drinking milk is associated with a lower risk of death).
    – De Novo
    Commented Sep 22, 2018 at 5:41
  • 1
    DeNovo, you are right, of course. Too late yesterday evening for clear reasoning. @MaxB: not only did they control for obesity, but they also write that they did not observe BMI differences between the milk-drinking groups - even though the milk drinkers had higher energy intake. Commented Sep 22, 2018 at 11:58

The study that article is referencing is Milk intake and risk of mortality and fractures in women and men: cohort studies, which found:

A higher consumption of milk in women and men is not accompanied by a lower risk of fracture and instead may be associated with a higher rate of death. Consequently, there may be a link between the lactose and galactose content of milk and risk as suggested in our hypothesis, although causality needs be tested using experimental study designs. Our results may question the validity of recommendations to consume high amounts of milk to prevent fragility fractures. The results should, however, be interpreted cautiously given the observational design of our study. The findings merit independent replication before they can be used for dietary recommendations.

Specifically, "nearly doubled" refers to:

In women the adjusted mortality hazard ratio for three or more glasses of milk a day compared with less than one glass a day was 1.93 (95% confidence interval 1.80 to 2.06)

That is, the proportion of women who consumed the equivalent of 600 mL of milk per day who died during the whole 22 year study was 80-106% higher than the proportion of women who consumed very little dairy products per day.

Overall 15,541 women (of 61,433 women at the start of the study - around 25%) died during the study - about half from cancer and cardiovascular disease.

The potential mechanism given by the paper that might cause this is D-galactose.

This study is mentioned by another paper (Higher milk intake increases fracture risk: confounding or true association?), which says:

This study has raised several issues. The highest relative risks were in women with high milk intake (9% of the population). These women, aged >50 when the study started were followed for 20 years. At age 70, people are likely to die from multiple causes and not due to milk intake per se. These women were at increased risk of comorbidities, prone to dietary changes pursuant to age, loss of spouse, and changes in living arrangement. Furthermore, baseline vitamin D status was missing, and Charlson’s comorbidity index does not account for diabetes, hypertension, or hypercholesterolemia separately. The high risk associations became weaker if only baseline (compared to two repeated measures) milk intake was associated with mortality risk. The FFQ had limited food items and was not validated for milk.

This paragraph continues, referencing the meta study Milk and dairy consumption and risk of cardiovascular diseases and all-cause mortality: dose–response meta-analysis of prospective cohort studies:

The conclusion disagrees with a recent meta-analysis of 29 prospective cohort studies on milk and mortality that found no association.

  • It would be helpful to explain what "adjusted mortality hazard ratio" means. Is a 95% increase something to worry about?
    – Oddthinking
    Commented Sep 15, 2018 at 4:30
  • @Oddthinking "Hazard Ratio (i.e. the ratio of hazards) = Hazard in the intervention group ÷ Hazard in the control group" (src). "Adjusted" means that they adjusted for variables such as "age, body mass index, energy intake, education".
    – Laurel
    Commented Sep 15, 2018 at 4:38
  • @MaxB I'm not sure what you're asking here. The theory the authors of the first paper have about galactose is based off experiments done on animals. The study they did to get the "almost 2x" number was observational and can't be used to determine causality, plus it has all the problems mentioned.
    – Laurel
    Commented Sep 15, 2018 at 4:52
  • 1
    @matt_black: I have edited in some contextual figures for absolute risk - about 25% over the 20+ year life of the study. I confess: higher than I expected.
    – Oddthinking
    Commented Sep 15, 2018 at 15:24
  • 2
    @Oddthinking According to the raw mortality table in the data appendix the absolute mortality in the consumption groups looked like this (expressed as deaths/1000 person years): <1 glass (11.9); 1-2 glasses (12.2); 2-3 glasses (13.5); 3 or more glasses (15.3) which is probably a better way to express the results and less scary. Still looks large (and consistent with your edits) but then those results are contradicted by other studies anyway.
    – matt_black
    Commented Sep 17, 2018 at 16:06

You must log in to answer this question.