Associations of dairy intake with risk of mortality in women and men: three prospective cohort studies

We appreciate Ding et al for the meticulous work done on the study “Associations of dairy intake with risk of mortality in women and men: three prospective cohort studies”1 recently published in The BMJ. However, we wish to draw attention to certain factors of public health significance which we believe could have impacted the results.

It is an undeniable fact that diet is a major contributor to overall health and mortality and certainly the role of milk cannot be overlooked. However, it is also worth noting that foods don’t work in isolation, requiring that certain influencing factors are taking into account in relating any particular food or diet to overall health and mortality. Accumulating evidence demonstrate that breastfeeding in early life is a critical “window of opportunity” that allows for long term impact on the gut microbiota which is now a “hot cake” subject linked to many diseases and mortality including cancer, cardiovascular and many other chronic diseases.2 Interestingly, findings also indicate that the sterile in utero gastrointestinal tract gets its first microbiota colonization through vaginal delivery but not caesarian section (CS) and the colonized bacteria species persist even into aduldhood. Therefore the mode of delivery (vaginal or CS) and subsequent breastfeeding status (breastfed or not breastfed) are important in shaping the gut microbiota. Consequently, it is imperative that studies relating any factors to disease and mortality even in later life account for the mode of delivery and breastfeeding status since these are critical factors that could impact overall health and wellbeing even later in life via gut microbiota programming. We realized that, in the report by Ding et al, like many similar studies already published, this important public health factor was not reported and/or accounted for in the study population and we believe this could have significantly impacted the overall results. We therefore suggest that future study designs account for these.

In addition, the authors grouped dairy products into milk, cheese, yogurt, and other types of dairy foods including ice cream, cream, and sherbert. Even though this model seems common in many epidemiological studies, we are of the view that it does not capture the important factor of the effect of processing technology on milk (especially milk proteins) and health. It is known that treatment technologies like chemical, biochemical, thermal and physical treatments of milk all have varying effects on the quality of milk with attendant changes in antigenicity and health in general. For example, the type and level of heat treatment could significantly impact the digestibility and digestion kinetics of the milk proteins as well as the bioavailability of essential amino acids with potential consequences for immunity. Processing levels could also induce reactions such as Maillard reactions which could generate intermediate products that could be toxic and/or antigenic upon consumption at elevated levels.3 4 Moreover, it has been reported that thermal processing of food significantly reshaped the gut microbial diversity and differentially triggered microbial alterations5 with significant consequences for health, disease and mortality. Therefore in evaluating associations between dairy consumption and disease and / or mortality, dairy types must reflect the processing techniques used as these might better reflect physiological impact.

Furthermore, taking yogurt for example, commercially supplied yogurt products comes in plain or with added fruits, flavor enhancers, coloring agents and other food ingredients (including food additives, either natural or artificial). Added agents and enhancers could potentially counter the beneficial effects of yoghurt, if any. Interestingly, yogurt topped with fruits could supply mixed nutrients and exert combined health benefits through potential prebiotic and probiotic effects.6 Therefore, yogurt topped with different fruits and other agents such as coloring might exert varying influences on the conclusion of such studies and so generalizing the effect of yoghurt consumption without accounting for these may confound the outcome.

In conclusion, dairy consumption would definitely impact health and mortality but certain critical factors that could potentially and significantly impact the direction of such relationship directly or indirectly must be accounted for, especially when previous study results have been inconsistent. This would allow for more robust findings and conclusions to be made.

Conflict of interest: We declare that we have no conflict of interest

References
1. Ding M, Li J, Qi L, et al. Associations of dairy intake with risk of mortality in women and men: three prospective cohort studies. bmj 2019;367
2. Van den Elsen L, Garssen J, Burcelin R, et al. Shaping the gut microbiota by breastfeeding: the gateway to allergy prevention? Frontiers in pediatrics 2019;7:47.
3. van Lieshout GA, Lambers TT, Bragt MC, et al. How processing may affect milk protein digestion and overall physiological outcomes: A systematic review. Critical reviews in food science and nutrition 2019:1-24.
4. Borad SG, Kumar A, Singh AK. Effect of processing on nutritive values of milk protein. Critical reviews in food science and nutrition 2017;57(17):3690-702.
5. Zhang Z, Li D. Thermal processing of food reduces gut microbiota diversity of the host and triggers adaptation of the microbiota: evidence from two vertebrates. Microbiome 2018;6(1) doi: 10.1186/s40168-018-0471-y
6. Fernandez MA, Marette A. Potential health benefits of combining yogurt and fruits based on their probiotic and prebiotic properties. Advances in Nutrition 2017;8(1):155S-64S.

Ding et al (1) have provided a thorough investigation on the association between dairy consumption and time to death by use of three large well-characterized cohorts within a setting of health professionals in USA. The study is, as expected with this team of authors, carefully done. For the reader it is, however, not easy to interpret these new results in the context of other prior cohort results. We would therefore take the opportunity and ask the authors if they can provide some additional information that would facilitate interpretation of their results. In addition, we would like to make some own remarks and also compare these new results with other cohort studies undertaken within other settings. Specifically, there are substantial differences in exposure range, reference category used and type of exposure.

The main exposure in the study (1) is total dairy consumption, a mixture of different dairy products and the composition of this mixture will depend on the setting. It would be of value if the authors present the relative contribution of different dairy products for the categories of total dairy intake presented in table 1. What were the relative contributions of e.g., non-fermented low-fat milk, high-fat milk, cheese, yogurt, etc? It would also be of interest to see average fruit and vegetable consumption in the categories of dairy, to facilitate comparison with other studies (2).

Reference and exposure categories vary considerably in the analyses (1). For NHS I, the reference averaged 0.7 serving/day of total dairy and highest category 4.1 servings/day. This is substantially different compared with the more restricted exposure width when low-fat milk was analyzed, with less than 1 serving/week as reference and 1.5 serving/day or more as highest category. For high-fat milk/whole milk, the reference was less than one serving/month and highest category 2 servings/week or more. Moreover, it is unclear how the authors treated whole milk consumption when low-fat milk was considered as the exposure, and vice versa. Were whole milk consumers retained in for example the reference category of low-fat milk? In a footnote below Table 3 it is stated that “Subtypes of dairy foods were adjusted for each other” and it is unclear whether the authors by this formulation meant dairy food groups described in the Methods, or mutual adjustment for the dairy products used as exposures in Table 3. Nevertheless, an adjustment does not remove the problem with a “contaminated” reference group. At baseline in our previous study among women (3), there are inverse correlations between reported intakes of different non-fermented milk products; the Swedish market provides mainly three types of milk based on fat content: 0.5% fat, 1.5% fat, and 3% fat milk. Thus, low fat (0.5%) milk intake has a correlation of -0.18 (p<0.00001) with 1.5% fat milk and a correlation of -0.22 with 3% fat milk (p<0.00001). More importantly, 45% of women consuming <1 glass reduced fat milk/day (0.5% or 1.5% fat) consumed 1 glass or more of 3% fat milk/day and 64% of women consuming <1 glass 3% fat milk/day consumed 1 glass or more of reduced fat milk. If the pattern is similar in the US cohorts, the reference categories in Table 3 will consist of not only less than one serving of the subtype of milk but a mixture of low consumption of the exposure and higher consumption of other milk subtypes. We now provide non-fermented milk product specific results based on our previous study (3) illustrating the consequences of a mixed reference category (Table: http://admin.webb.uu.se/infoglueDeliverWorking/digitalAssets/823/c_82363...), also highlighting the different exposure widths used in the US studies and in our Swedish setting.

Specifically, the results are different when consumers of other non-fermented milk types are kept in or excluded from the analysis when subtypes of non-fermented milk consumption (based on fat %) are studied in relation to mortality. When the reference category is “contaminated” with exposure to other milk products, the hazard ratios are clearly attenuated. Adjustments for type of milk influence our estimates only modestly. Moreover, our estimates by use of this approach seems to be compatible with the results from the US cohorts (1) when examining similar exposure categories.

To facilitate the comparison with our previous study (3) and interpretation of their own results, we would like to suggest that the authors run if possible additional analyses in the cohorts using less than one serving/day of total non-fermented milk intake (low and high fat milk in combination) as the reference and 3 servings/day of total non-fermented milk intake or more as the highest category. We assume that few women and men in the cohorts are high consumers of non-fermented milk (more than 3 glasses/servings per day) but we might be wrong.

As mentioned by the authors, a large Japanese cohort study in a population with traditionally low consumption of milk has shown lower mortality rates with “high” consumption of milk (4). Careful evaluation of the exposure reveal that the lowest exposure and reference consisted of never users of milk (zero consumption) while the highest category was consumption of milk almost every day (corresponding to an average of only 100 g/day). This is a considerably different exposure width compared with our Swedish cohorts (3). Of note, non-fermented milk was for long in many families a standard beverage at every meal in Sweden. In total, given the different cohort study results in different populations, there seems to be a U-shaped pattern of mortality with non-fermented milk consumption and we ask the authors (1) to perform the suggested additional analyses to shed further light on this assumption.

An additional concern from our point of view for the interpretation of milk results by use of NHS and HPFS is that no adjustment was made for socioeconomic status, arguably because study participants all initially had a similar high educational level. During 30 years of follow-up considerable changes in family income between study participants may have occurred (5). In contrast to Sweden and Europe (6), milk intake in the US seems to be linked with income and social class (7-11). Can the authors reassure us that our concern is unjustified?

References
1. Ding M, Li J, Qi L, Ellervik C, Zhang X, Manson JE, et al. Associations of dairy intake with risk of mortality in women and men: three prospective cohort studies. BMJ. 2019;367:l6204.
2. Michaelsson K, Wolk A, Melhus H, Byberg L. Milk, Fruit and Vegetable, and Total Antioxidant Intakes in Relation to Mortality Rates: Cohort Studies in Women and Men. Am J Epidemiol. 2017;185(5):345-61.
3. 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.
4. Wang C, Yatsuya H, Tamakoshi K, Iso H, Tamakoshi A. Milk drinking and mortality: findings from the Japan collaborative cohort study. J Epidemiol. 2015;25(1):66-73.
5. Lee S, Kawachi I, Berkman LF, Grodstein F. Education, other socioeconomic indicators, and cognitive function. American journal of epidemiology. 2003;157(8):712-20.
6. Darmon N, Drewnowski A. Does social class predict diet quality? The American journal of clinical nutrition. 2008;87(5):1107-17.
7. Kirkpatrick SI, Dodd KW, Reedy J, Krebs-Smith SM. Income and race/ethnicity are associated with adherence to food-based dietary guidance among US adults and children. J Acad Nutr Diet. 2012;112(5):624-35 e6.
8. Ervin RB, Wright JD, Kennedy-Stephenson J. Use of dietary supplements in the United States, 1988-94. Vital Health Stat 11. 1999(244):i-iii, 1-14.
9. Davis CG, Yen ST, Dong D, Blayney DP. Assessing economic and demographic factors that influence United States dairy demand. J Dairy Sci. 2011;94(7):3715-23.
10. Stewart HD, D. Carlson, A. Is Generational Change Contributing to the Decline in Fluid Milk Consumption? Journal of Agricultural and Resource Economics, Western Agricultural Economics Association. 2012;37(3):1-20.
11. Dubois L, Girard M. Social position and nutrition: a gradient relationship in Canada and the USA. Eur J Clin Nutr. 2001;55(5):366-73.