Milk intake and risk of mortality and fractures in women and men: cohort studies

The observational study by the Swedish surgeon Karl Michaëlsson and co-workers published in British Medical Journal concluded that high milk intake was associated with increased risk of bone fractures and mortality (Michaëlsson 2014). The message gave rise to global media attention, not least due to the fact that the results of the study seem to go against previous systematic reviews, meta-analyses and recommendations from most authorities in the world.
But are the new findings valid and do they change our view on dairy in any way? Or is the study flawed, or are there reasons simply to look at the result as just an outlier that needs to be viewed in the context of the totality of evidence? We have analysed the publication and the population studies behind the data.

The study is based on analyses of two Swedish population cohorts, the Swedish Mammography Cohort (SMC), and the Cohort of Swedish Men (CSM) that were followed for a mean of 20 years. At baseline in 1987-90, and 1997, respectively, various pieces of information were recorded including self-reported information about dietary intakes, and this information was analysed in relation to subsequent health outcomes. In both cohorts all subjects ranged from middle aged to elderly at baseline.

The positive findings
The intake of cheese and fermented milk products was associated with a reduced mortality and a reduced risk of fractures. For each serving, the rate of mortality and hip fractures was reduced by statistically significant 10-15 %, which is consistent with other studies.

The negative findings
For every glass of milk the women reported to drink at baseline it was found, after various adjustments, that they had a 15 % higher mortality in the subsequent 20 years, and a 3 % higher mortality for males. They also found that for every glass of milk women had a 2 % higher risk of all fractures, and a 9 % higher risk of hip fractures, whereas no such significant risk was found among the males. But how do these results fit with the totality of evidence on cardiovascular risk and mortality?

A single study versus totality of evidence
Two recent meta-analyses have analysed all available observational studies. Based on 26 studies O’Sullivan et al. (2013) concluded that dairy and milk intake had neither any effect on total mortality nor cardiovascular mortality. Soedamah-Muthu et al. (2011) also did a meta-analysis of 17 prospective cohort studies and found a modest inverse association between milk intake and risk of overall cardiovascular disease i.e. a 6 % decrease in risk per 200 mL/day. But milk intake was not associated with risk of CHD, stroke or total mortality. They concluded that their dose-response meta-analysis of prospective studies indicated that milk intake is not associated with total mortality, but may be inversely associated with overall CVD risk. A meta-analysis of milk intake and risk of hip fractures could not find any increased risk (Bischoff-Ferrari et al., 2011), but systematic reviews involving totality of evidence conclude that dairy is an important contributor to bone health (Weaver 2014). It is also worth remembering that osteoporosis is a ʻpediatric disease with geriatric consequencesʼ (Power et al.,1999) with good evidence that low milk and hence low calcium intake during childhood/adolescence is associated with significantly increased risk in middle and older age, particular in women (e.g. Kalkwarf et al., 2003). Thus in middle aged and older women it may not be surprising that that increased milk intake did not reduce the risk fractures although a recent study challenges this (Sahni et al., 2014). They reported that a total of 830 men and women (mean age at baseline 77 years) from the Framingham Original Cohort, completed a food-frequency questionnaire in 1988 to 1989 and were followed for hip fracture rate until 2008. Those with medium (>1 and <7 servings/wk) or higher (≥7 servings/wk) milk intake tended to have lower hip fracture risk than those with low (≤1 serving/wk) intake (high versus low intake HR 0.58, 95% CI 0.31–1.06; medium versus low intake HR 0.61, 95% CI 0.36–1.08; p trend = 0.178). There appeared to be a threshold for milk, with a 40% lower risk of hip fracture among those with medium/high milk intake compared with those with low intake (p = 0.061). A similar threshold effect was observed for milk plus yogurt intake. These results do indeed suggest that greater intakes of milk and milk + yogurt may lower risk for hip fracture in older adults.
These conclusions are supported by several newer observational studies using objective biomarkers for dairy intake, and they have strengthened the evidence from observational and intervention studies to support that dairy reduces risk of type 2 diabetes and CVD (Astrup 2014, AJCN).

Two mysteries
We have identified several shortcomings and limitations that can be found in most, if not all, observational studies. However, we have identified two mysteries that go beyond our scientific skills and capacity to explain:

Mystery one
Since this study is clearly an outlier it is reasonable to consider if there is something special regarding the health effects of dairy in the Swedish population. The answer to this question is interestingly enough found in publications from another research group in Sweden that has published similar results from exactly the same study populations, although of somewhat smaller sizes.
Patterson et al. (2013), who also used the Swedish Mammography Study of ~33,000 women found that total dairy intake at baseline was inversely associated with risk of myocardial infarction (MI) when comparing highest versus lowest quintile during 11.6 years follow up (HR 0.77; 95% CI, 0.63-0.95). Also cheese intake was inversely associated with MI (HR 0.74; 95% CI, 0.60-0.91). Interestingly, butter on bread, but not butter on cooking, was associated with an increased risk (HR 1.34; 95% CI, 1.02-1.75). The latter finding is highly suggestive of a confounding effect of starchy carbohydrates i.e. white bread intake, as Jacobsen et al. (2010) have reported that replacement of saturated fat with refined carbohydrate increases CVD risk. In the study by Patterson et al. (2013) there was not even a trend for an increased MI risk for high intakes of milk. In another analysis from the same research group Larsson et al. (2012) examined dairy intake and risk of the other CVD end-point stroke in the both the SMC (also used by Patterson et al. (2013)) and CSM cohorts with a mean 10.2 years follow up. She found that low-fat dairy was associated with reduced risk for total stroke, whereas there was no effect of other dairy products. Again milk intake was not significantly positively associated with stroke, but with RR values of 0.92-0.94 (P=0.15) there was a suggestion a possible small protective effect, if anything (Larsson et al., 2012). Moreover there was a significant negative association between low-fat dairy and total stroke (RR 0.88; CI 0.80-0.97, P=0.03). It is also very interesting to note that if the mortality data in the paper of Michaëlsson et al. (2014) are expressed as a simple proportion of the population that died; increasing milk intake is associated with a linear decrease in mortality while the reported age-adjusted HR shows a similar linear but increasing mortality with increasing reported milk intake (See commentary by Hellstrand above).
The same group have also examined the potential role of dairy for risk of major cancers with possible protective role of dairy for bladder and colorectal cancers using the same to cohorts. Again, total dairy intake was not significantly associated with risk of bladder cancer, but sour milk and yogurt intakes were associated with a significant 38 % lower risk in men and women combined (Larsson et al., 2008). For milk there was no signal of any increased risk, only a non-significant lower risk. In another paper they reported that total dairy intake was associated with reduced risk of colorectal cancer in men, and more than 7 servings/d of total dairy foods compared with less than 2 servings were associated with a 57 % lower risk of colonic cancer (Larsson et al., 2006). A recent meta-analysis of 19 cohort studies also concluded that increased intake of milk and dairy products was associated with reduced colorectal cancer risk reporting summary RR of 0.83 (CI, 0.78–0.88) per 400 g/day of total dairy products and 0.91 (0.85–0.94) per 200 g/day of milk (Aune et al., 2012)
So studies using both the SMC and CMS cohorts, with a shorter follow-up and smaller cohort size than the recent BMJ study, show significantly lower risk of major CVD and cancers for those who consumed higher amounts of dairy, and with no indication of any possible increased risk for high milk intakes. It is unfortunate that the authors of the BMJ paper did not comment on these previous findings from the same cohorts, and failed to present data on both cancer and CVD incidence (relying only on mortality), including details on site specific cancers and strokes and MI. They actually do not even cite that papers in their discussion of their results, despite the fact that the opposing results from these cohorts relied on the same baseline dietary recordings.
From a scientific point of view it is very unusual that a longer follow up i.e. 10 years versus 20 years follow up completely turns the results upside down, particularly when no new information of dietary intake is added during follow up so the impact of changes in intakes can be addressed. There are concerns however that dietary data collected at baseline will be less representative of habitual diet after 20 years than after 10-11 years. Although the SMC cohort was larger in size in the BMJ analyses, the discrepancy with other results from the same cohorts severely questions the validity of the analyses and statistics, and calls for a re-analyses of all data by independent scientists.

Mystery two
The validity of self-reported information about dietary intake is a major problem in population studies, and failure to address this may lead to false conclusions (Dhurandhar et al., 2014). Some “sensitivity analyses” are normally required to avoid false conclusions being made due to statistical artefacts that occur due to a systematically biased under- and over-reporting. It is well-known that all people are prone to underreport unhealthy foods and drinks, and over-report food and drinks considered as healthy. It is impossible to take misreporting into account and adjust for it unless there are measurements of objective biological markers. This has not been done in the Swedish studies, so we can just do some basic validity assessments.
Are the baseline characteristics of the participants in the different quintiles of dairy food consumption comparable and consistent with basic physiological principles? Across the quintiles of dairy intakes BMI is very similar, and so is Total Physical Activity (MET-h/day), so consequently these groups are supposed to have similar basal metabolic rate, and total energy expenditure, which means that they should have same total energy requirements. However, with increasing intake of dairy intake total energy intake goes significantly up i.e. from 1412 to 1965 kcal/day in lowest to highest quintile of dairy intake, a difference of 553 kcal/day (or 39 %). In a previous publication reporting on the same study (Larsson et al., 2012), also here BMI and physical activity were identical across the quintiles of dairy food intakes, but total energy intake went from 1709 to 3013 kcal/day from lowest to highest intake of dairy (76 %).
The authors of the studies do not comment on this paradox, except saying that it is well-known that a high intake of dairy is not associated with increase in body weight and BMI. However, according to thermodynamic laws, it is obvious that the existence of two weight stable groups of persons with same BMI, physical activity level and energy requirements cannot exist when one group is consuming ~50 % more energy than the other. It is a violation of the laws of thermodynamics. However, there is a very sensible explanation: all information about height and weight in the Swedish cohorts are self-reported, which is well known to produce a severe under-reporting of body weight among the overweight and particularly obese individuals (Kuskowska-Wolk & Rössner, 1989).
It means that the true BMI across the quintiles would be increasing, and that total energy intake, and intakes of most foods would go up. An increased BMI causes an increased mortality due to both cancer and CVD. So the most likely explanation for the findings are that the authors have in reality studied, not the effect of high dairy intakes, but the results of being obese. It would be interesting to invite subsets of the participants of these studies to a measured height and weight to get this hypothesis tested.

Professor Arne Astrup Professor D Ian Givens

References
Astrup A. A changing view on saturated fatty acids and dairy: from enemy to friend. Am J Clin Nutr. 2014 Dec;100(6):1407-8.
Aune D, Lau R, Chan DS M, Vieira R, Greenwood DC, Kampman E and Norat T. (2012) Dairy products and colorectal cancer risk: a systematic review and meta-analysis of cohort studies. Annals of Oncology 23: 37–45.
Bischoff-Ferrari HA, Dawson-Hughes B, Baron JA, Kanis JA, Orav EJ, Staehelin HB, Kiel DP, Burckhardt P, Henschkowski J, Spiegelman D, Li R, Wong JB, Feskanich D, Willett WC (2011) Milk intake and risk of hip fracture in men and women: a meta-analysis of prospective cohort studies. J Bone Miner Res 26:833-839.
Dhurandhar NV, Schoeller D, Brown AW, Heymsfield SB, Thomas D, Sørensen TI, Speakman JR, Jeansonne M, Allison DB (2014) Energy balance measurement: when something is not better than nothing. Int J Obes Epub ahead of print doi: 10.1038/ijo.2014.199.
Jakobsen MU, Dethlefsen C, Joensen AM, Stegger J, Tjønneland A, Schmidt EB, Overvad K. Intake of carbohydrates compared with intake of saturated fatty acids and risk of myocardial infarction: importance of the glycemic index. Am J Clin Nutr. 2010 Jun;91(6):1764-8.
Kalkwarf H J, Khoury JC and Lanphear BP (2003) Milk intake during childhood and adolescence, adult bone density, and osteoporotic fractures in US women. Am J Clin Nutr 77:257–65.
Kuskowska-Wolk A, Rössner S. The "true" prevalence of obesity. A comparison of objective weight and height measures versus self-reported and calibrated data. Scand J Prim Health Care. 1989 Jun;7(2):79-82.
Larsson SC, Bergkvist L, Rutegård J, Giovannucci E, Wolk A (2006) Calcium and dairy food intakes are inversely associated with colorectal cancer risk in the Cohort of Swedish Men. Am J Clin Nutr 83:667-673.
Larsson SC, Andersson S-O, Johansson J-E, Wolk A (2008) Cultured milk, yogurt, and dairy intake in relation to bladder cancer risk in a prospective study in Swedish women and men. Am J Clin Nutr 88:1083-1087.
Larsson SC, Virtamo J, Wolk A (2012) Dairy consumption and risk of stroke in Swedish women and men. Stroke 43:1775-1780.
Michaëlsson K, Wolk A, Langenskiöld S, Basu S, Lemming EW, Melhus H, Byberg L (2014) Milk intake and risk of mortality and fractures in women and men: cohort studies. BMJ 349:g6015.
O'Sullivan TA, Hafekost K, Mitrou F, Lawrence D. Food sources of saturated fat and the association with mortality: a meta-analysis. Am J Public Health. 2013 Sep;103(9):e31-42. doi: 10.2105/AJPH.2013.301492.
Patterson E, Larsson SC, Wolk A, Åkesson A (2013) Association between dairy food consumption and risk of myocardial infarction in women differs by type of dairy food. J Nutr 143:74-79.
Power ML, Heaney RP, Kalkwarf HJ, Pitkin RM, Repke JT, Tsang RC and Schulkin J (1999) The role of calcium in health and disease. Am J Obstet Gynecol 181:1560–1569.
Sahni, S., Mangano, K. M., Tucker, K. L., Kiel, D. P., Casey, V. A. and Hannan, M. T. (2014), Protective Association of Milk Intake on the Risk of Hip Fracture: Results from the Framingham Original Cohort. J Bone Miner Res, 29: 1756–1762. doi: 10.1002/jbmr.2219
Soedamah-Muthu SS, Ding EL, Al-Delaimy WK, Hu FB, Engberink MF, Willett WC, Geleijnse JM. Milk and dairy consumption and incidence of cardiovascular diseases and all-cause mortality: dose-response meta-analysis of prospective cohort studies. Am J Clin Nutr 2011;93:158-71.
Weaver CM. How sound is the science behind the dietary recommendations for dairy? Am J Clin Nutr. 2014 May;99(5 Suppl):1217S-22S.