Intended for healthcare professionals

CCBYNC Open access

Rapid response to:

Research

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

BMJ 2014; 349 doi: https://doi.org/10.1136/bmj.g6015 (Published 28 October 2014) Cite this as: BMJ 2014;349:g6015

Rapid Response:

Not only do Helen M Macdonald, Alexandra Mavroeidi and Alison Avenell dislike our results in general, they furthermore criticize our findings as not based on sound science from mainly three aspects: misclassification of milk intake by a food frequency questionnaire, residual confounding and the fact that galactose is existent in fermented dairy products.

We fully agree that an accurate dietary assessment is difficult to accomplish. In our cohort studies, we used comprehensive food frequency questionnaires. These questionnaires have been validated as described in our Methods. According to these validation studies of milk intake, the correlation between the FFQ and four gold standard reference weighed 7-day food records every third month have been approximately 0.7 (1). Further, in both women and men we have previously found a positive association between reported intake of milk and the fat tissue content of pentadecanoic acid, a biologic marker reflecting average long-term dairy fat intake (2, 3). Nevertheless, the consequence of non-differential misclassification when measuring dietary intake in cohort studies can partially be compensated by an increase in study size and in number of outcomes, and additionally by repeat dietary assessments (4) - as was the case in our female cohort study. Indeed, the associations were stronger among women where we used time-dependent exposure analysis taking advantage of the repeat FFQs and where the analysis was based on a larger number of outcomes compared with the male cohort analysis (5), and also compared with previous fracture analysis of the female cohort (6). Non-differential misclassification of the exposure normally leads to conservatively biased estimates.

The possibility of residual confounding has been well acknowledged by us in the article (5). Helen M Macdonald et al seem, however, not to have carefully read our text where it is clearly described that we did take into account fat intake, including saturated fat intake, and retinol in our analyses (p 3, Statistical analysis, and p 4, Sensitivity analysis). Our estimates were not attenuated by these adjustments (Supplementary table E). Coffee intake was related to a somewhat higher risk of fracture in one previous analysis (7) but not in later more extensive analyses, neither in women (8) nor in men (9). Moreover, coffee intake has in recent large observational studies (10-12) been suggested to lower the rate of mortality, not the opposite. Nonetheless, with the addition of coffee intake to our multivariable model and with <1 glass of milk per day as the reference, the HRs of hip fracture for the higher categories were 1.19 (95% CI 1.11.1.28; 1 to <2 glasses/day), 1.54 (95% CI 1.41-1.68; 2 to <3 glasses/day), and 1.58 (95% CI 1.37-1.82; ≥3 glasses/day). Similarly and expectedly, our estimates for mortality were only marginally affected after adjustment for coffee intake.

Finally, we have never claimed that soured milk and yogurt are free of galactose. What we in the Introduction stated was: “Separating milk intake from the consumption of other dairy products may be of importance since a less pronounced induction of oxidative stress and inflammation in humans is expected with cheese and fermented dairy products (for example, soured milk and yogurt) because of their lower or non-existent lactose and galactose content,14 15 possible probiotic antioxidant and anti-inflammatory effects,16 17 18 and effects on gut microbiota.19 20 21 Indeed, a high intake of fermented milk products has been associated with a decreased risk of cardiovascular diseases,18 22 23 24 whereas a high milk intake is related to a tendency of an unfavourable risk profile for the development of diabetes and cardiovascular disease.18 23 24”
Detailed analysis of our article’s reference 14 (by Livia Alm), an old but highly relevant reference regarding lactose and galactose in Swedish dairy products, revealed 10-50% lower galactose content in different fermented milk products compared with non-fermented milk. By use of more contemporary analytical techniques, there is an ongoing examination at the Swedish University of Agricultural Sciences of the sugar components in present Swedish liquid dairy products. Hard cheeses are nowadays recommended for people with galactosaemia since they have undetectable levels of galactose (13). Given these facts and our hypothesis, it is for us surprising to find a suggestion from Macdonald et al that all milk products should in an etiologic analysis have been analyzed together. In an effort to gain further knowledge, such an approach would not be recommended from a scientific point of view.

Arne Astrup and Ian Givens have made a rhetorically elegant attempt to criticize our publication. Had they, however, read our article carefully, including our supplementary material, our previous responses to commentaries and their own references they would have saved words and time.

They begin by citing some recent meta-analyses concluding a null association between milk intake and all-cause mortality as well as cardiovascular mortality. One problem, if you carefully scrutinize the original studies, is that several of them have not separated non-fermented from fermented milk intake in the exposure assessment, with the risk of mixing apples and oranges. There are also formally tested heterogeneities in the study results with some studies displaying a tendency of lower and some tendency towards higher mortality rates with increasing reported consumption of milk, which is an indication of methodological differences in study designs and analyses. Indeed, meta-analysis of published data from dietary studies is problematic because of the wide variety of ways that data have been analyzed and presented. Valid dose-response examinations in meta-analyses are also difficult to accomplish without individual data. The use of pooled analyses based on primary data is a more attractive methodology that allows a coherent analytic approach in included studies (14). Let us remind the reader of the fact that our female cohort included a large number of outcomes compared with earlier studies, that we have specific analysis of different dairy products, a wide exposure range and have applied time-dependent exposure analysis – all highly relevant when interpreting the study findings. We are fully aware that an accurate dietary assessment is difficult to accomplish in large scale studies, as also discussed above. However, the total error in the exposure assessment can be reduced by repeated examinations and by a larger study size, which improves the possibility to detect not only an association but also to determine the strength of the association. Meta-analysis of smaller studies will only render higher precision of the composite estimate and will not modify the magnitude of the estimate.

We note with interest the claim that osteoporosis is a paediatric disease with geriatric consequences. This claim is of especial interest given the recent study results by Feskanich et al based on the Nurses' Health Study and the Health Professionals Follow-up Study in the United States, displaying higher – not lower - risk of hip fracture in men with a high teenage consumption of milk and a null finding in women (15). Arne Astrup and Ian Givens forgot to mention these recent results.

Furthermore, in support of their view that high milk consumption will reduce fragility fractures Astrup and Givens mention the study by Sahni et al (16). This cohort analysis, based on a single FFQ, included 764 (not 830) men and women with an age range between 68 and 96 years. During the 12 year follow-up, a total of 97 hip fracture cases were identified. These hip fractures were self-reported, which is of potential importance for the interpretation of the study results since some fractures might not have been identified, especially since there are high mortality rates in the first months after the hip fracture event (17). No associations between dairy intake (milk or other products) were found when the exposures were examined as continuous variables. No spline curves are presented. The tendencies of statistically significant results from the categorical analyses emerged based on different categorizations depending on type of dairy analyzed. Choosing a reference category (e.g., 1 serving/week or less of milk) with the lowest number of outcomes and participants is not a recommended epidemiological approach. Only multivariable adjusted results are presented, which is also not recommended by the STROBE guidelines (18), but the attempts for adjustment did not include comorbid conditions. Moreover, the change in estimate approach is nowadays not recommended for covariate selection (4). Given these facts, it seems somewhat remarkable that Arne Astrup and Ian Givens consider the Framingham study to offer good scientific support for the notion that greater intakes of milk lower the risk for hip fractures in older adults.

Regarding biomarkers, milk intake estimated by our FFQ has been validated against the fat tissue content of pentadecanoic acid, a biologic marker reflecting average long-term milk fat intake, i.e., existent in both milk and fermented milk products (2, 3). Such biomarkers cannot distinguish between effects on health from non-fermented and fermented milk products.

Mystery one
We disagree with the authors’ opinion regarding our study as an outlier. As already explained in regard to the previous meta-analytical approaches made, these results stem from heterogenetic study results with some individual studies displaying tendencies of higher risk of mortality and fracture with high milk consumption while other researchers find associations in the opposite direction. Although we can’t rule out the possibility of biased results with the four observational study designs included in our article, a more likely explanation for our more obvious study findings compared with earlier publications is different methodology (study size, number of outcomes, no loss to follow-up, repeat dietary assessment in women, distinguishing between different types of dairy products, etc) as described above. Moreover, in a previous response (http://www.bmj.com/content/349/bmj.g6015/rr/779932), we have already described the old intriguing results from reducing gastric ulcer disease pain by drinking milk. Fifty years ago, such a diet to reduce the symptoms from peptic ulcers was reported to increase the risk of myocardial infarction (19). Thus, based on autopsy data both from the United States and Great Britain the relative risk of myocardial infarction was more than doubled in ulcer patients prescribed a milk diet compared with ulcer patients without such a diet (19). By use of data provided in the original publication (19), the RRs can be calculated to 2.3 (95% CI 1.4-4.0) and 4.7 (1.4-15.3), respectively. The pooled RR estimate is 2.6 (1.6-4.3). In addition, although ecological studies should be interpreted cautiously, both cardiovascular disease mortality and hip fracture rates have additionally displayed positive associations with lactose and milk intake (20-23). Therefore, clear indications that high milk consumption could negatively affect our health have long existed. Our study can, however, not prove causality and it should not be used for dietary advice but it is one piece in a scientific puzzle.

Regarding previous publications using the Swedish Mammography Cohort, Patterson et al (24) used the second FFQ in late 1997 as the baseline investigation. We don’t find these previous study findings conflicting with our new results. A high total dairy consumption conferred a lower risk of myocardial infarction (1392 cases of MI) during 12 years of follow-up. On average, only one fourth of the total dairy servings consumed was non-fermented milk. Since we also found a reduced risk of cardiovascular mortality with high cheese and fermented milk product intake, the results are not incompatible. However, highest (median 3.1 serving per day) compared with lowest category of milk intake (median 0.1 serving per day) in the Patterson study was associated with an age-adjusted HR of 1.17 (95% CI 0.99-1.39), with a p for trend across categories of 0.02. When other dairy products were considered in the model, the estimate remained essentially similar (HR 1.15; 95% CI 0.96-1.38). Although Patterson examined MI and we mortality (including cardiovascular mortality), the results from the two studies are definitely compatible given the single FFQ used in the Patterson analysis, a lower number of outcomes compared with our mortality analysis and a somewhat more heterogenic outcome (myocardial infarction compared with death). Accordingly, the claim by Arne Astrup and Ian Givens that “In the study by Patterson et al. (2013) there was not even a trend for an increased MI risk for high intakes of milk”, is untrue. We note with interest that these authors also reference the incorrect calculations of crude mortality made by Staffan Hellstrand, who is consultant for the Federation of Swedish Farmers, an organization who is owner of milk industries (e.g., Arla Foods). In a previous response (http://www.bmj.com/content/349/bmj.g6015/rr/779932), we have corrected Staffan Hellstrand and guided him to the correct data for use. Despite erroneous figures, Astrup and Givens have chosen to repeat the presentation of Staffan Hellstrand’s incorrect calculations – an interesting action given the fact that Arne Astrup should be an objective key opinion leader in field of nutritional research by his role as editor of the American Journal of Clinical Nutrition. Hopefully, his action to criticize our study by distorted arguments is not related to his financial conflicts of interest, for example research support from Arla Foods (25). In addition, we focused on deaths and fractures as outcomes in our study. Presentation of all possible outcomes and references, including bladder cancer, stroke and colorectal cancer, would not have been feasible given both relevance and space limitations. By the second last sentence under “Mystery one”, Astrup and Givens expose their lack of knowledge in epidemiology since they obviously have not understood the time-updated Cox’s regression analysis in our use of the Swedish Mammography Cohort data.

Mystery two
This part of Astrup’s and Givens criticism deals with the apparent paradox of similar average BMI levels of the different milk categories despite higher calculated energy intake in high consumption categories and without higher estimated metabolic equivalents (MET-hours/days). Astrup and Givens claim that we did not comment on this paradox but our comment can be found, including four references to meta-analyses of both cohort and randomized clinical trials, in the first 9-line long paragraph of the Discussion under the subheading “Comparison with other studies”. On average, in both the cohort studies and in the randomized clinical trials - despite a higher energy intake - no weight gain has been observed with high dairy consumption. Interestingly, the authors themselves have been involved in studies where they have found higher energy intakes with high consumption of dairy products accompanied by lower body mass index (26, 27). This fact was not discussed and not even mentioned in the cohort study by Ian Givens et al (26). Nonetheless, the mechanism behind this paradox is not at present known but may involve fecal excretion of fat by soap formation with calcium (28), as also discussed by Ian Givens and Arne Astrup previously (27, 29).

We would also draw the attention to the results for cheese and fermented milk products. High consumption of these products was also associated with a higher total energy intake (Supplementary table A and B; http://www.bmj.com/content/bmj/suppl/2014/10/30/bmj.g6015.DC1/mick020390...) whereas the mortality and fracture rates were lower compared with a low consumption of these products (Supplementary Table C and D). Moreover, if selective underreporting of body weight was the case in the high consumption category, a higher BMI should not be related to a higher hip fracture risk but to a lower risk (30, 31).

Selective underreporting of body weight in the high consumption categories of dairy products is, nevertheless, unlikely. Indeed, since 5022 of the women in the Swedish Mammography Cohort during the period 2003-2009 have been involved in a clinical examination including measurement of body height and weight as well as determination of body composition by DXA (32), we have had the possibility to accurately examine BMI levels by categories of milk intake. Measured average BMI by categories of milk intake was 25.8 (SD 4.2) kg/m2 in the first category of milk intake (<1 glass per day), 26.2 (SD 4.3) kg/m2 in the second category (1 to <2 glasses/day), 26.3 (SD 4.4) kg/m2 in the third category (2 to <3 glasses/day), and 26.0 (SD 4.1) kg/m2 in the highest category of milk intake (3 glasses or more per day). Therefore, the assumption made by Astrup and Givens that women in the highest category actually have a higher BMI is not correct. Moreover, Astrup and Givens again make a gross exaggeration by stating that self-reported weight “produce a severe under-reporting of weight” with reference to the study by Kuskowska-Wolk et al (33). Based on reports from residents in a health care centre, body weight was underreported in that study but the extent of underreporting and the degree of misclassification even by the selective recruitment of the participants is highly unlikely to explain more than small differences in mortality rates (34). A modest underreporting of actual body weight (<2 kg) is also well-known in population-based settings (35) but, again, the magnitude is not able to explain our higher risk of mortality and fractures. In conclusion, the non-differential underreporting in reported body weight by dairy categories is therefore not a probable explanation for our findings.

Misreporting of dietary items was also discussed in our previous response (http://www.bmj.com/content/349/bmj.g6015/rr/779932) and we used the established method as suggested by Walter Willett to compensate for overall under- or overreporting (36, 37).

Finally, we find it remarkable that two researchers with a focus on nutrition research do not find more interest and curiosity in our findings and instead devote themselves to distort the truth. Distorting evidence will not in the long run benefit anyone; neither will hiding one’s head in the sand ignoring potentially important results.

Many of the Rapid responses include comments that disqualify the thorough work effort made by the BMJ reviewers (including a statistical reviewer) and the editorial team who have scrutinised our manuscript before allowing it to be published. We look forward to a scientific debate based not on opinion but rather on peer-reviewed scientific results and encourage researchers with information on dairy data to perform additional analyses where intake from non-fermented milk, fermented milk and cheese is analysed separately.

Karl Michaëlsson, Professor of Medical Epidemiology
Alicja Wolk, Professor of Nutritional Epidemiology
Liisa Byberg, Associate Professor of Medical Epidemiology

References
1. Larsson SC, Andersson SO, Johansson JE, Wolk A. Cultured milk, yoghurt, and dairy intake in relation to bladder cancer risk in a prospective study of Swedish women and men. Am J Clin Nutr. 2008;88:1083-7.
2. Jiang J, Wolk A, Vessby B. Relation between the intake of milk fat and the occurrence of conjugated linoleic acid in human adipose tissue. The American journal of clinical nutrition. 1999;70:21-7.
3. Wolk A, Vessby B, Ljung H, P. B. Evaluation of a biological marker of dairy fat intake. Am J Clin Nutr. 1998;68:291-5.
4. Rothman KJ, Greenland S, Lash TL. Modern epidemiology. 3rd ed. Philadelphia: Wolters Kluwer Health/Lippincott Williams & Wilkins; 2008. x, 758 p. p.
5. Michaelsson K, Wolk A, Langenskiold S, Basu S, Warensjo Lemming E, Melhus H, et al. Milk intake and risk of mortality and fractures in women and men: cohort studies. BMJ. 2014;349:g6015.
6. Michaëlsson K, Melhus H, Bellocco R, Wolk A. Dietary calcium and vitamin D intake in relation to osteoporotic fracture risk. Bone. 2003;32(6):694-703.
7. Hallstrom H, Wolk A, Glynn A, Michaelsson K. Coffee, tea and caffeine consumption in relation to osteoporotic fracture risk in a cohort of Swedish women. Osteoporos Int. 2006;17(7):1055-64.
8. Hallstrom H, Byberg L, Glynn A, Lemming EW, Wolk A, Michaelsson K. Long-term Coffee Consumption in Relation to Fracture Risk and Bone Mineral Density in Women. Am J Epidemiol. 2013.
9. Hallstrom H, Wolk A, Glynn A, Michaelsson K, Byberg L. Coffee consumption and risk of fracture in the Cohort of Swedish Men (COSM). PLoS One. 2014;9(5):e97770.
10. Crippa A, Discacciati A, Larsson SC, Wolk A, Orsini N. Coffee consumption and mortality from all causes, cardiovascular disease, and cancer: a dose-response meta-analysis. Am J Epidemiol. 2014;180(8):763-75.
11. Freedman ND, Park Y, Abnet CC, Hollenbeck AR, Sinha R. Association of coffee drinking with total and cause-specific mortality. N Engl J Med. 2012;366(20):1891-904.
12. Malerba S, Turati F, Galeone C, Pelucchi C, Verga F, La Vecchia C, et al. A meta-analysis of prospective studies of coffee consumption and mortality for all causes, cancers and cardiovascular diseases. Eur J Epidemiol. 2013;28(7):527-39.
13. Portnoi PA, MacDonald A. Determination of the lactose and galactose content of cheese for use in the galactosaemia diet. Journal of human nutrition and dietetics : the official journal of the British Dietetic Association. 2009;22(5):400-8.
14. Smith-Warner SA, Spiegelman D, Ritz J, Albanes D, Beeson WL, Bernstein L, et al. Methods for pooling results of epidemiologic studies: the Pooling Project of Prospective Studies of Diet and Cancer. Am J Epidemiol. 2006;163(11):1053-64.
15. Feskanich D, Bischoff-Ferrari HA, Frazier AL, Willett WC. Milk consumption during teenage years and risk of hip fractures in older adults. JAMA pediatrics. 2014;168(1):54-60.
16. Sahni S, Mangano KM, Tucker KL, Kiel DP, Casey VA, Hannan MT. Protective association of milk intake on the risk of hip fracture: results from the Framingham Original Cohort. J Bone Miner Res. 2014;29(8):1756-62.
17. Michaelsson K, Nordstrom P, Nordstrom A, Garmo H, Byberg L, Pedersen NL, et al. Impact of hip fracture on mortality: a cohort study in hip fracture discordant identical twins. J Bone Miner Res. 2014;29(2):424-31.
18. von Elm E, Altman DG, Egger M, Pocock SJ, Gotzsche PC, Vandenbroucke JP, et al. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. PLoS Med. 2007;4(10):e296.
19. Hartroft WS. The Incidence of Coronary Artery Disease in Patients Treated with Sippy Diet. Am J Clin Nutr. 1964;15:205-10.
20. Hegsted DM. Calcium and osteoporosis. J Nutr. 1986;116:2316-9.
21. Hegsted DM. Calcium and osteoporosis? Adv Nutr Res. 1994;9:119-28.
22. Hegsted DM. Fractures, calcium, and the modern diet. Am J Clin Nutr. 2001;74(5):571-3.
23. Segall JJ. Hypothesis: is lactose a dietary risk factor for ischaemic heart disease? Int J Epidemiol. 2008;37(6):1204-8.
24. Patterson E, Larsson SC, Wolk A, Akesson A. Association between dairy food consumption and risk of myocardial infarction in women differs by type of dairy food. J Nutr. 2013;143(1):74-9.
25. Astrup A. A changing view on saturated fatty acids and dairy: from enemy to friend. Am J Clin Nutr. 2014;100(6):1407-8.
26. Livingstone KM, Lovegrove JA, Cockcroft JR, Elwood PC, Pickering JE, Givens DI. Does dairy food intake predict arterial stiffness and blood pressure in men?: Evidence from the Caerphilly Prospective Study. Hypertension. 2013;61(1):42-7.
27. Astrup A, Chaput JP, Gilbert JA, Lorenzen JK. Dairy beverages and energy balance. Physiology & behavior. 2010;100(1):67-75.
28. Wyshak G. Dietary animal fat intake, calcium intake, and bone fractures in women 50 years and older. J Women's Health. 1993;2:329-34.
29. Dougkas A, Reynolds CK, Givens ID, Elwood PC, Minihane AM. Associations between dairy consumption and body weight: a review of the evidence and underlying mechanisms. Nutrition research reviews. 2011;24(1):72-95.
30. Farahmand BY, Michaelsson K, Baron JA, Persson PG, Ljunghall S. Body size and hip fracture risk. Swedish Hip Fracture Study Group. Epidemiology. 2000;11(2):214-9.
31. De Laet C, Kanis JA, Oden A, Johanson H, Johnell O, Delmas P, et al. Body mass index as a predictor of fracture risk: a meta-analysis. Osteoporos Int. 2005;16(11):1330-8.
32. Michaëlsson KW, A. Byberg, L. Ärnlöv, J. Melhus, H. Intake and serum levels of α-tocopherol in relation to fractures in elderly women and men: two cohort studies. Am J Clin Nutr. 2013;accepted.
33. Kuskowska-Wolk A, Karlsson P, Stolt M, R”ssner S. The predictive validity of body mass index based on self-reported weight and height. Int J Obes. 1989;13:441-53.
34. Flegal KM, Kit BK, Orpana H, Graubard BI. Association of all-cause mortality with overweight and obesity using standard body mass index categories: a systematic review and meta-analysis. JAMA. 2013;309(1):71-82.
35. Nyholm M, Gullberg B, Merlo J, Lundqvist-Persson C, Rastam L, Lindblad U. The validity of obesity based on self-reported weight and height: Implications for population studies. Obesity (Silver Spring). 2007;15(1):197-208.
36. Willett WC, Howe GR, Kushi LH. Adjustment for total energy intake in epidemiologic studies. Am J Clin Nutr. 1997;65(4 Suppl):1220S-8S; discussion 9S-31S.
37. Willett W. Nutritional epidemiology. 3rd ed. Oxford University Press: Oxford; 2013.

Competing interests: No competing interests

04 December 2014
Karl Michaëlsson
Professor of Medical Epidemiology
Professor Alicja Wolk; Associate professor Liisa Byberg
Department of Surgical Sciences, Uppsala University
UCR/MTC, Dag Hammarskjölds väg 14b, 752 37 Uppsala, Sweden