Consumption of Dairy Products in Relation to Changes in Anthropometric Variables in Adult Populations: A Systematic Review and Meta-Analysis of Cohort Studies

Background The current state of knowledge regarding the association of dairy products and weight gain, overweight, and obesity is based on studies reporting contradicting and inconclusive results. The aim of the present study was thus to clarify the link between dairy consumption in relation to changes in anthropometric measures/adiposity by a meta-analytical approach. Methods For the meta-analysis PubMed, EMBASE, Web of Sciences, and google scholar were searched by two independent authors up to May 2016 with no restriction to language or calendar date. Prospective cohort studies reporting about intake of dairy consumption (including milk, yogurt, cheese, butter) and changes in body weight or waist circumference, risk of overweight, obesity, or weight gain were eligible. Pooled effects were calculated using a random effects model, and also a fixed effect model for sensitivity analysis. Due to the heterogeneity of statistical analytical approaches of the studies the analysis were done separately for beta-coefficients of changes in body weight and/or waist circumference per serving of dairy, for differences in weight gain/gain in waist circumference when comparing extreme categories of dairy consumption, and for odds ratios in regard to weight gain, overweight/obesity, or abdominal obesity. Findings 24 studies (27 reports) met the inclusion criteria for the systematic review, and 22 studies provided sufficient data for inclusion in the meta-analysis. The meta-analysis of the five studies on changes in body weight per serving of dairy no significant results could be found for whole fat dairy and low fat dairy. However, there was inverse association between changes in body weight for each serving’s increase of yogurt (beta: -40.99 gram/year, 95% CI, -48.09 to -33.88), whereas each serving’s increase of cheese was positively associated (beta: -10.97 gram/year, 95% CI, 2.86 to 19.07). Furthermore, the highest dairy intake category was associated with a reduced risk of abdominal obesity (OR: 0.85; 95% CI, 0.76 to 0.95), and risk of overweight (OR: 0.87; 95% CI, 0.76 to 1.00) compared to the lowest intake category. No significant association could be observed for risk of weight gain. Conclusion In summary the results of the meta-analysis still reflect that dairy consumption was not positively related to changes in body weight. Yogurt was the only dairy food that showed some evidence for a beneficial effect, where higher intakes were inversely associated a reduced risk of obesity, changes in body weight or waist circumference. Further research is needed, since the overall interpretation of the results is limited by heterogeneous risk estimates.


Methods
For the meta-analysis PubMed, EMBASE, Web of Sciences, and google scholar were searched by two independent authors up to May 2016 with no restriction to language or calendar date. Prospective cohort studies reporting about intake of dairy consumption (including milk, yogurt, cheese, butter) and changes in body weight or waist circumference, risk of overweight, obesity, or weight gain were eligible. Pooled effects were calculated using a random effects model, and also a fixed effect model for sensitivity analysis. Due to the heterogeneity of statistical analytical approaches of the studies the analysis were done separately for beta-coefficients of changes in body weight and/or waist circumference per serving of dairy, for differences in weight gain/gain in waist circumference when comparing extreme categories of dairy consumption, and for odds ratios in regard to weight gain, overweight/obesity, or abdominal obesity.

Findings
24 studies (27 reports) met the inclusion criteria for the systematic review, and 22 studies provided sufficient data for inclusion in the meta-analysis. The meta-analysis of the five studies on changes in body weight per serving of dairy no significant results could be found for whole fat dairy and low fat dairy. However, there was inverse association between changes in body weight for each serving's increase of yogurt (beta: -40.99 gram/year, 95%

Introduction
Most dietary guidelines recommend the consumption of milk and dairy products as important components of a healthy, well-balanced diet. Dairy products are a high-quality protein source providing a substantial proportion of the recommended adult nutrient intake of calcium, iodine, riboflavin, and vitamin B 12 . It is thus well reasoned that for instance public health measures in the USA recommend an increased intake of fat-free or low-fat milk and milk products [1].
Several recent meta-analyses of prospective cohort studies have shown that dairy consumption is associated with a reduced risk of cardiovascular disease, hypertension, stroke, and diabetes [2][3][4][5]. In line with these findings, a recently published meta-analysis of 76 studies including 659,298 participants showed that higher levels of circulating margaric acid (a component of ruminant milk fat) was associated with a 23% risk reduction of coronary heart disease [6].
While some meta-analyses have highlighted more favorable effects of low-fat dairy compared to higher-fat dairy products [4,7], it has been demonstrated that study participants who consumed more dairy fat and/or high-fat dairy foods at baseline, were leaner and/or gained less weight over time compared to participants who consumed less [8]. In contrast, a current meta-analysis of clinical trials could not support the beneficial effect of an increased dairy consumption on body weight and fat loss in long-term studies or studies without energy restriction [9]. While another meta-analysis of randomized controlled trials (RCTs) has shown that increasing both, whole and low fat dairy food consumption increases weight [10], a number of short-term or energy-restricted RCTs report modest benefits of dairy products in facilitating weight loss [9]. A recent summary of RCTs showed that dairy intake resulted in greater fat mass reduction in the presence of energy restriction in the short-term compared to a control group [11].
As a result of the available contradicting and inconclusive results, the association of dairy products and weight gain, overweight, and obesity has been critically questioned [12,13]. This regards the biological nature (e.g. modulation of adipocyte lipid metabolism) as well as the overall quantitative effect size of the relationship in free living populations. Since we could not find data on overall effect size of dairy consumption on anthropometric changes, including changes in body weight and waist circumference (WC), and risk of weight gain/overweight/ obesity we conducted a quantitative meta-analysis of cohort studies with the focus on dairy consumption and its association with adiposity as complement to the already conducted metaanalyses of RCTs.

Materials and Methods
This systematic review was planned and conducted according to the standards of the Metaanalysis of Observational Studies in Epidemiology [14]. Our protocol has been registered in the PROSPERO International Prospective Register of Systematic Reviews. (Registration number: CRD42014013997)

Literature search
A literature search was performed in PubMed, EMBASE, Web of Science, and google scholar (until May 2016) with no restrictions to language, and calendar date using the following search terms: ("dairy" OR " milk" OR " yogurt" OR " cheese") AND (" weight" OR " waist circumference" OR "BMI" OR "body mass index" OR " obesity" OR "hip") AND (" longitudinal" OR " prospective" OR " cohort" OR " change" OR "follow-up"). Moreover, the reference lists from retrieved articles and systematic reviews, were checked to search for further relevant studies as well as citations of those articles. Literature search was conducted independently by two authors (LS and TKU), with disagreements resolved by consensus.

Eligibility criteria
Studies were included in the meta-analysis if they met all of the following criteria: (i) longitudinal study design; (ii) data related to consumption (quantitative amount; i.e. servings per day/ week, gram per day/week intake) of dairy products (including milk, yogurt, cheese, butter); the primary outcomes were: changes in body weight, either measured continuously (g/year) or binary as incident major weight gain (e.g. incidence of gaining a specific amount of weight, risk of overweight, abdominal obesity); the secondary outcomes were: changes in waist circumference; (iii) reported adjusted beta-coefficients with corresponding standard error or data necessary to calculate these (95% confidence intervals, (95% CI), standard deviations, p-values); (iv) mean differences with corresponding standard error in change in measure of adiposity over time between participants with the highest and lowest intake category; (v) adjusted odds/risk/ hazard ratios with corresponding 95% CI or standard error comparing the highest vs. lowest dairy intake category; (vi) when a study appeared to have been published in duplicate, the version containing the longest follow-up was selected; (vii) study participants were free of chronic disease (cardiovascular disease, cancer) at baseline of study; (viii) only adults were included (minimum of 18 years of age).

Data extraction and quality assessment
We extracted the following data from each study: first author's name, publication year, location, cohort name, type of outcome, population (age, sex, and sample size), follow-up duration, dietary assessment method, definition of dairy products (including the unit of consumption), whether dairy was modeled continuously or categorical, outcome definition, outcome assessment, statistical analysis method, and variables adjusted for. Further, we extracted data of the effect sizes with their corresponding uncertainties. When a study reported only separate effect estimates for males and females, they were treated as different studies. When a study provided effect estimates for different degrees of adjustment, those from the most completely adjusted model were chosen. Study quality was assessed in accordance with a recent meta-analysis of sugar intake and adiposity [15], and included risk of bias (selection of exposed and unexposed in cohort studies from different populations; partially flawed measurement of both exposure (i.e. relied only baseline dairy consumption; measurement error) and outcome (i.e. selfreported); Inconsistency: i.e. point estimates vary widely across studies; confidence intervals shows minimal or no overlaps; statistical test for heterogeneity shows a low p-value; I 2 is large [16,17].

Statistical analysis
We classified the prospective studies according to the type of measure of association they used [18]: 1. Adjusted beta-coefficients for the association of dairy intake with subsequent changes in anthropometric outcomes per g intake; 2. Mean differences of change in anthropometric measures over time comparing the highest and lowest category of dairy consumption; 3. Odds/risk/hazard ratios for risk of overweight or abdominal obesity, or for risk of gaining a particular amount of weight.
Study specific results were pooled with random effects models using the DerSimonian-Laird method, which incorporated both within and between study variability [19]. To evaluate the weighting of each study, the standard error of the beta-coefficient, mean differences, and for the logarithm odds/risk/hazard ratios of each study was calculated and regarded as the estimated variance of the beta-coefficients, mean differences, and odds ratio using the DerSimonian-Laird method and the inverse variance method [19]. Mean changes in body weight and waist circumference and odds/risk/hazard ratios of incident weight gain, overweight, abdominal obesity were pooled comparing participants who had the highest intakes of dairy with those who had the lowest intakes.
Separate meta-analyses were performed for the beta-coefficients per g intake, for the mean changes and for the odds/risk/hazard ratios (adiposity measures) taking the different categories of intake of dairy products such as total dairy, whole fat dairy, low-fat dairy, cheese, yogurt, milk). Heterogeneity was tested with the Q statistic and quantified with the I 2 statistic. I 2 values >50% were indicative for substantial heterogeneity across studies [20]. Funnel plots, in which the effect sizes are plotted against their corresponding uncertainty, were used to assess potential publication bias if at least 10 studies were available [21]. In order to check sensitivity of the results we also conducted all analysis with a fixed effects model using an inverse variance method [19]. We used the "metan" command in Stata 12.0 (Stata Corp. 2011, Texas, USA) for all meta-analyses.

Changes in body weight (beta coefficients and mean differences)
In studies that analyzed the association between whole fat dairy intake and changes in body weight, one study found a significant positive association. In this study by Mozaffarian et al. [28] (Fig 1). Surprisingly, the fixed effect models showed significant results in both cases when the sensitivity analysis was performed. On the other side, an inverse association between changes in body weight for each serving's increase of yogurt

Changes in waist circumference (beta coefficients and mean differences)
Concerning changes in waist circumference, the study by Romaguera [34] observed that every 100 kcal increase in milk, yogurt, or cheese was associated with a decrease in waist circumference in both men and women, whereas no significant association between a 100 kcal increase in whole, and skim/low fat milk could be observed. In the random effects model the value for milk was -0.02 cm/year (95% CI, -0.02 to -0.01; I 2 = 47.9%) annual waist circumference change per serving increase, for yogurt the value was -0.02 cm/year (95% CI, -0.04 to -0.01; I 2 = 0%), and for cheese -0.02 cm/year (95% CI, -0.04 to -0.00; I 2 = 0%) (S3 Fig).
When comparing the highest versus lowest categories of dairy consumption and changes in waist circumference, it could be shown that dairy was associated with a non-significant (-0.07 cm/year, 95% CI, -0.09 to 0.01; I 2 = 54%) annual decrease in the random effects model (S4 Fig). No significant associations could be observed for milk, low-fat dairy, high-fat dairy, yogurt and cheese consumption.

Risk of overweight, abdominal obesity/weight gain (odds/risk/hazard ratio)
Pooling three studies [27,31,48] showed a reduced risk between dairy consumption and risk of overweight (OR/RR/HR: 0.87; 95% CI, 0.76 to 1.00; I 2 = 0%), one of which focused on yogurt consumption (S5 Fig). A significant inverse association could be also observed between dairy products and risk of abdominal obesity (OR/RR/HR: 0.85; 95% CI, 0.76 to 0.95; I 2 = 81.1%) (Fig 2). No significant associations were observed for overall dairy intake and weight gain, and the intake of milk and risk of weight gain and overweight. However, sensitivity analysis comparing low-fat vs. wholefat dairy yielded in a significant reduction for adiposity only for whole fat dairy products (S6 Fig).

Study quality, methodological issues
The present systematic review included 24 longitudinal studies (27 reports) lasting at least 9 months, and in which data relating to an association between dairy products and a measure of adiposity could be extracted. Of these 24 studies, five used self-reported estimates of body weight or waist circumference, validated in a subsample [27,28,30,33,36,48] and three used self-assessed anthropometric parameters only in the follow-up [34,37]. Regarding the dietary assessment methods, all but four applied food frequency questionnaires, two of which used dietary records [22,32], one repeated 24-h dietary records [41], and one study a 15-item questionnaire [26]. Collected exposure data from questionnaires where the validity for assessing dairy intakes was not stated or not assessed included five studies [26,32,38,40,45], and 13 (14 reports) out of 23 studies provided estimates that were adjusted for total energy intake. The wide range of methods of assessing dairy exposures and adiposity outcomes as well as the inconsistency in the covariates used to adjust analyses made pooling studies difficult.

Discussion
Overall the results of the present meta-analysis showed that higher intakes of dairy products were not associated with increased body weight gain. In the fixed effects model whole fat dairy products were associated with a small but significant increase in body weight, whereas low-fat dairy was inversely related to changes in body weight. An increase in dairy intake was inversely related to changes in waist circumference. The pooled results from the Harvard cohorts suggest that higher yogurt intake is the only dairy food inversely related with changes in body weight [28], whereas each serving's increase of cheese was positively associated with body weight. Furthermore the comparing the highest category of dairy consumption was associated with a reduced risk of overweight, and abdominal obesity (mainly contributed throughout whole-fat dairy products).
As stated by Prentice [49], there are only two foods consumed by humans that have been designed to meet the entire nutrient needs of a complex organism: milk and eggs. Diets containing high proportions of both foods are thus unsurprisingly beneficial for healthy growth and development.
To summarize the relationship between dietary calcium intake and BMI after correction for trials effect, a mixed model regression analysis was conducted using mean data reported in the analyzed studies. While dairy products remain the major source for calcium [50], the role of dairy components in the regulation of energy metabolism, weight and body fat regulation, has not been clearly elucidated. However, the most reasonable mechanism refers to a modulation of adipocyte lipid metabolism and fatty acid absorption from the gastrointestinal tract by the effects of dietary calcium on intracellular calcium [51]. Additionally, research indicates that high calcium intake may reduce lipogenesis and increase lipolysis by hormone regulation [52]. A meta-analysis of clinical trials showed that calcium supplementation was associated with a significant weight and fat mass reduction. However, this meta-analysis is not interpretable and has major methodological flaws, since the authors used the standard error instead of the standard deviation for one study, causing an overestimation of the effect size [53].
In addition, other dairy compounds with beneficial effects on weight and body fat loss have been proposed. As such, whey protein (effects on muscle sparing and lipid metabolism [54][55][56]), conjugated linoleic acid (regulation of adipogenesis, inflammation, and lipid metabolism [57]) are promising constituents. Additional to their effect on satiety [58], milk proteins are insulinotropic. Taken together, milk proteins may play an important role in explaining the association between dairy consumption and body weights [59].
Evidence from the present meta-analysis indicates an inverse association between yogurt consumption and weight gain. Nutrients from yogurt have a higher bioavailability compared to other forms of dairy [60]. Recent research indicates that gut microbiota plays a decisive role in weight control [61][62][63]. It is hypothesized that probiotic yogurt enhances the growth of beneficial intestinal microbiota and modulates gut function through regulation of the immune system. These effects have been discussed to facilitate weight loss or weight maintenance [64], which is also the crucial factor in the prevention of type 2 diabetes mellitus [65]. In line with our results, meta-analyses of prospective studies have shown that fermented dairy products were associated with a 12-17% reduction in type 2 diabetes mellitus [4,5].
To our knowledge, no high quality long-term RCTs analyzing weight outcome were conducted, so the level of evidence is limited.
When analyzing dairy products as part of a healthy diet, the unfavourable effects cannot be left disregarded. First, according to data from the NHANES III certain dairy products (regular cheese, dairy desserts, low-fat milk, whole-fat milk, and butter) contribute to 25% of the saturated fat intake in the US diet [66]. In European countries, these numbers are even higher with a proportion of 41% [67]. Second, reducing saturated fat intake is pivotal of most dietary recommendations to reduce the risk of coronary heart disease [68,69].
A number of prospective epidemiological studies have investigated the dairy-weight loss hypothesis . It should be noted that US and European prospective cohort studies are susceptible to unmeasured confounding by individual dietary components. Diets arise from a complex network of foods consumed within and between meals where clustering of "healthy" or "unhealthy" behaviors is common. Specifically, yogurt and milk consumption have often been associated with a better overall dietary profile and inversely associated with the consumption of sugar-sweetened beverages, especially soda and fruit juices [70][71][72].
Regarding anthropometric outcomes, a longitudinal study showed that a healthy dietary pattern high in reduced-fat dairy was inversely associated with BMI and with waist circumference [73]. In the case of dietary patterns as exposures, the individual contribution of specific food groups to the effect cannot be quantified, but rather the results are seen as the effect of the interaction of this complex network of foods over the outcome.
Dairy consumption patterns may differ across countries. As such, in the US, a large proportion of dairy fat is consumed via commercial foods such as ice cream and pizza [74]. Although this holds true throughout the industrialized world, there is a stronger tradition in European countries to consume whole-fat dairy products (plain cheeses, unsweetened yogurt and plain butter).

Limitations
The present systematic review has several limitations. First, baseline and follow-up body weight assessment were measured differently across cohorts (data based on self-report and exact weight measure method). For some included studies, only baseline dairy consumption was used (assuming a stable consumption over time). Common problems in prospective cohort studies, like measurement errors, have to be noted. In most studies, dietary intake is calculated from self-assessment tools such as FFQs [75]. As it is difficult to disentangle the effects of dairy fat from the food within which it was consumed using observational methods, this may also be considered as an additional confounding factor [8].
Second, the included studies showed substantial heterogeneity with respect to the analyzed population size, follow-up length, baseline age, dairy categories, adjustment factors, and outcome estimates.
Thirdly, since the power of each meta-analysis was limited by the low number of studies, potential confounding explored by stratified analysis could not be conducted.

Conclusion
In conclusion, the results of the present meta-analysis indicate that dairy consumption was not related to changes in body weight, but inversely associated to changes in waist circumference, risk of overweight and abdominal obesity. Yogurt was the only dairy food that showed some evidence for a beneficial effect, where higher intakes were inversely associated with a reduced risk of obesity, changes in body weight and waist circumference. Further research is needed, since the overall interpretation of the results is limited by heterogeneous risk estimates.