Milk Intakes Are Not Associated with Percent Body Fat in Children from Ages 10 to 13 Years

Abstract

Epidemiologic studies report conflicting results for the relationship between milk intake and adiposity in children. We examined prospective and cross-sectional associations between milk intake and percent body fat among 2245 children from the Avon Longitudinal Study of Parents and Children. Cross-sectional analyses were performed at age 13 y between total, full-fat, and reduced-fat milk intake assessed using 3-d dietary records and body fat from DXA. Prospective analyses were conducted between milk intakes at age 10 y and body fat at 11 and 13 y. Models were adjusted for age, sex, height, physical activity, pubertal status, maternal BMI, maternal education, and intakes of total fat, sugar-sweetened beverages, 100% fruit juice, and ready-to-eat cereals; baseline BMI was added to prospective models. Subset analyses were performed for those with plausible dietary intakes. Mean milk consumption at 10 and 13 y was (mean ± SD) 0.90 ± 0.73 and 0.85 ± 0.78 servings/d [1 serving = 8 oz of milk (244 g of plain and 250 g flavored milk)], respectively. Cross-sectional results indicated an inverse association between full-fat milk intake and body fat [β = −0.47 (95% CI = −0.76, −0.19); P = 0.001]. Milk intake at age 10 y was inversely associated with body fat at 11 y [β = −0.16 g/d (95%CI = −0.28, −0.04); P = 0.01], but not among those with plausible dietary intakes, suggesting that this association was influenced by dietary measurement errors. Milk intake was not associated with body fat at age 13 y after adjustment. Although our prospective results corroborate other findings of a null associations between milk intake and adiposity, our inconsistent findings across analyses suggest further investigation is needed to clarify the relation, and accounting for dietary reporting errors is an important consideration.

Introduction

In 2007–2008, 14.8% of U.S. children aged 2–19 y were overweight and 16.9% were obese (1). Although the prevalence of childhood overweight and obesity remains high, a recent study indicates that the obesity epidemic may have slowed (1). In the UK, 29% of children aged 5–17 y were overweight or obese in 2004 (2) and, more recently, 4.8 and 6.1% of boys and girls aged 11–18 y, respectively, were obese in 2007 (3). Although many factors likely contributed to the increased obesity rates seen over the past decades, milk intake significantly decreased and sugar-sweetened beverage intake increased concurrently during that time (4–6). Mean milk intake was <244 g/d for U.S. children aged 12–19 y in 2005–2006 (7) and ~107 g/d for UK girls and ~172 g/d for UK boys aged 11–18 y in 2008–2009 (8). Milk is a good source of several nutrients, such as calcium and phosphorus (9–11) and bioactive compounds (12). High dietary calcium may play a role in preventing fat accumulation by suppressing calcitrol, which prevents the influx of intracellular calcium into the adipocyte, and through genetic and nongenetic mechanisms that increase energy metabolism (13). Dietary calcium may also prevent adiposity through increased fecal fat loss and increased thermogenesis (14).

Several studies have suggested an inverse relationship between milk and/or dairy intakes and adiposity in children (15–17). Others, however, reported null (18, 19) or positive associations (20) with body weight and fatness. The relationship between milk intake and body composition is inconsistent (15, 17, 19–21). This may be due to differences in study design, methods for assessing diet and body composition, and/or adjustment of potential confounding factors in the analysis (22). The inclusion of individuals with implausible dietary intakes in diet-obesity studies may also lead to inconsistent findings (23, 24), as shown in a recent study, where associations were attenuated when only individuals with plausible dietary intakes were considered (25). Dietary reporting errors are common in dietary studies (23, 26, 27), especially among individuals who are overweight or obese (28, 29). Measurement error that occurs due to the inclusion of individuals with implausible dietary intakes can obfuscate associations between diet and obesity. Thus, it is important to examine associations that account for dietary reporting errors.

The objective of this study was to move forward the literature on the milk-body fat hypothesis by rigorously examining associations between total, full-fat, and reduced-fat milk intakes measured using diet records and percent body fat measured using DXA in children aged 10–13 y. We examined the influence of dietary reporting errors on associations between milk intake and adiposity by repeating analyses in a subsample of individuals with plausible dietary intakes only. To better understand the impact of study design on these associations, we conducted a set of cross-sectional and prospective analyses. Milk intake and body composition were not concurrently measured at baseline; thus, cross-sectional associations between milk intake and body fat were evaluated only at age 13 y. Prospective associations were investigated between baseline milk intake and body fat at ages 11 and 13 y. In secondary analyses, we examined the relationship between changes in milk intake from 10 to 13 y and changes in body fat from ages 11 to 13 y.

Materials and Methods

Study population.

The study sample included children aged 10–13 y from the Avon Longitudinal Study of Parents and Children (ALSPAC), a prospective cohort study investigating the relationship among health, growth, and development in children living around Bristol, UK (30). Pregnant women from this area of southwest England with an expected delivery date between April 1991 and December 1992 were recruited; 13,988 babies were alive at 1 y with 548 new participants recruited at 7 y. Data collection included self-reported questionnaires, health and education records, in-home observations, and biological samples. From the age of 7 y, participants were invited to regularly attend research clinics to undergo diet record collection and anthropometric and DXA measurements. There were 2245 children available for prospective analyses with complete dietary data at age 10 y and percent body fat at ages 11 and 13 y and 2270 children with dietary and percent body fat data available at age 13 y. This study was approved by the ALSPAC Law and Ethics Committee and the Local Research Ethics Committees and the Institutional Review Board of Boston University Medical Center. Informed consent was obtained from parents upon enrollment and for any additional measures.

Dietary assessment.

At ages 10 and 13 y, 3-d diet records were collected prior to the regular clinic visit. Children were instructed to record all foods and beverages consumed in household measures for 2 weekdays and 1 weekend day. Parental help was enlisted as needed. Records were reviewed for completeness by a nutrition fieldworker. Reported foods and beverages were allocated food codes and weights using Diet in Data Out, developed by the MRC Human Nutrition Research Unit in Cambridge, UK (31). Mean daily nutrient intakes were calculated using an in-house nutrient analysis program, BRIGADE, based on a nutrient databank that included the 5th edition of McCance and Widdowson’s (32) food tables and supplements. Additional nutrient databases, manufacturers’ information, and recipe calculations provided nutrient information for foods not found in the databank (33).

Reported beverages from dietary records were used to create milk groups. Total milk intake included full-fat, reduced-fat, and skim (nonfat) plain and flavored cow's milk. Milk groups were further categorized into full and reduced fat based on fat content. A small number of children consumed skim milk (n = 200 at age 10 y and n = 120 at 13 y); thus, we did not separately examine intakes for this group. Milk intakes were quantified in g/d and servings/d, where an 8-oz serving is 244 g of regular milk and 250 g for flavored milk (21).

Anthropometry.

A Lunar Prodigy DXA scanner (GE Medical Systems Lunar) was used to measure body composition of children at ages 11 and 13 y and provided estimates of total fat mass, lean body mass, and bone mass. Height and weight were measured at ages 10, 11, and 13 y. Height was measured using a Harpenden stadiometer (Holtain) and weight was obtained using a weighing scale (Tanita). BMI was calculated as weight (kg) divided by height (m²).

Potential covariates.

Age at attendance was calculated for each research clinic visit. Maternal educational attainment was classified as none/Certificate of Secondary Education (nonacademic qualification at age 16 y), vocational, O level (national academic qualification at age 16 y, higher than CSE (Certificate of Secondary Education), or A level (national academic qualification at age 18 y), or degree, following a standard protocol. A uniaxial accelerometer (Actigraph) was used to assess physical activity at ages 11 and 13 y. This instrument assesses the frequency and intensity of movement in the vertical plane and provides mean counts over a specified time period (1 min in this study) (34). Children were asked to wear the accelerometer for 7 d during waking hours and to remove it only for bathing/showering or participation in contact or water sports (34). Parents and/or children noted stage of sexual maturity on diagrams based on the Tanner charts for children at age 11 y. Children were classified as prepubertal (stage 1), early pubertal (stages 2–3), or late pubertal (stages 4–5) based on pubic hair development. Maternal BMI [weight (kg)/height (m²)] was calculated from self-reported height and weight when their child was age 9 y. Potential dietary confounding factors, including consumption of total fat, ready-to-eat breakfast cereal, sugar-sweetened beverages, and 100% fruit juice, were assessed from 3-d diet records as g/d. Frequency of dieting during the past year was assessed at age 13 y; a dichotomous dieting variable was created (yes/no).

Dietary reporting error assessment.

Dietary reporting errors were characterized using methods described elsewhere by Noel et al. (35). Briefly, accelerometer data at age 13 y were applied to methods developed by Huang et al. (23) for quantifying reporting errors. This method creates cutoffs for the ratio of reported dietary intakes:predicted energy requirements. Predicted energy requirements were estimated using equations from the US DRI (36). These methods were repeated using accelerometer data at age 11 y and body composition data at 10 y to capture those with plausible and implausible dietary intakes at age 10 y. We examined associations between milk intake and percent body fat in those with plausible dietary intakes at ages 10 y (n = 907, 42%) and 13 y (n = 882, 44%).

Statistical analysis.

All statistical analyses were completed using SAS (version 9.1, SAS Institute). Data were examined for outliers. Assumptions and model fit were tested using residuals and residual plots. Means, SD, and proportions were calculated for anthropometric and sociodemographic characteristics. Separate multivariable linear regression models were used to examine associations between milk intakes (assessed as per 100 g and as servings/d) and percent body fat. Associations between total milk and plain milk at ages 10 and 13 y were examined using Spearman correlation coefficients.

We built 5 multivariable adjusted regression models. Model 1 was adjusted for age, sex, height, and height squared, because previous work in this study population indicated quadratic relationships between height and fat mass (37). Baseline BMI was added to all prospective models to account for baseline adiposity, which is a major factor related to subsequent body fat gains. (Note that we used BMI rather than baseline body fat, because DXA was not performed at age 10 y, as previously noted.) Model 2 also included physical activity (counts per minute), pubertal status, maternal BMI, maternal education, and dietary intakes of total fat, ready-to-eat breakfast cereal, 100% fruit juice, and sugar-sweetened beverage intake. These variables were included as potential confounding factors, as shown elsewhere (38–40), because they are associated with both dairy intake and adiposity. Model 3 was further adjusted for calcium intake to test whether associations between milk and adiposity were independent of calcium, as done elsewhere (20, 40). In model 4, total energy was included, because it can account for extraneous variation in body size, metabolic rate, and physical activity (41). However, because total energy may be in the causal pathway between milk and body fat (i.e. higher milk consumption leads to higher energy intake, which in turn would be related to higher body fat), this term was not included in our main model (model 2). In our final model, model 5, we restricted analyses to those with plausible dietary intakes only, adjusting for all of the main covariates and potential confounding factors, as described in model 2. In secondary analyses, we examined the influence of dieting on associations between milk intake and body fat by adjusting for dieting behavior at age 13 y.

Cross-sectional analyses modeled intakes at age 13 y with percent body fat at age 13 y. Our 2 main prospective analyses modeled baseline intakes (age 10 y) with percent body fat at both ages 11 and 13 y. We had >80% power to detect associations between milk intakes and percent body fat as small as 0.003 and 0.003 at ages 11 and 13 y, respectively, at α = 0.05. In secondary analyses, we also built models examining changes in milk intake from ages 10 to 13 y with changes in percent body fat from 11 to 13 y. All models were tested for effect modification of milk on adiposity by sex and baseline overweight/obesity using separate 2-way interaction terms. Interactions with sex were observed between changes in total milk and body fat (P-interaction = 0.09) and changes in reduced-fat milk and percent body fat (P-interaction = 0.08). Interactions with baseline overweight/obesity were observed for associations between high-fat milk intake at age 10 y and body fat at age 11 y (P-interaction = 0.09) and between reduced-fat milk at age 13 y and body fat at age 13 y (P-interaction = 0.09). These P values were above our prespecified criteria of P-interaction = 0.05 for detecting effect modification. Also, stratified results were inconsistent and did not produce a clear picture of the relationship between milk and percent body fat. Observed interactions may have occurred by chance, because many interaction terms and models were tested. Thus, we present results for only the whole sample in this manuscript.

Results

At baseline, the mean age was 10.6 ± 0.22 y (Table 1). Girls had a greater percentage of body fat at ages 11 and 13 y compared with boys (P < 0.001). Girls were less physically active than boys at ages 11 and 13 y (P < 0.001). At age 13 y, a greater proportion of girls were dieting compared with boys (P < 0.001). Maternal BMI and educational attainment were similar for boys (P = 0.36) and girls (P = 0.31). Total milk intakes were similar at ages 10 and 13 y for boys (1.04 ± 0.78 and 1.03 ± 0.85 servings/d, respectively; P = 0.68) and small differences were seen for girls (0.79 ± 0.67 vs. 0.70 ± 0.69 servings/d, respectively; P = 0.001). Boys consumed more servings of total, full-fat, and reduced-fat milk compared with girls. Full-fat milk intake decreased for boys (0.36 ± 0.67 vs. 0.24 ± 0.60 servings/d; P < 0.001) and girls (0.26 ± 0.52 vs. 0.14 ± 0.42 servings/d; P < 0.001) from ages 10 to 13 y (data not shown). Reduced-fat milk intake was slightly lower at age 10 y compared with 13 y for boys (0.65 ± 0.75 vs. 0.75 ± 0.83 servings/d, respectively; P = 0.004) and girls (0.50 ± 0.63 vs. 0.53 ± 0.64 servings/d, respectively; P = 0.30).

TABLE 1.

Sociodemographic and anthropometric characteristics for boys and girls at ages 10 and 13 y from ALSPAC¹

	Boys	Girls
Sample characteristics	(n = 1030, 45%)2	(n = 1240, 55%)2	P value
Age at clinic visit, y
10 y	10.6 ± 0.22	10.6 ± 0.22	0.74
11 y	11.7 ± 0.19	11.7 ± 0.20	0.15
13 y	13.8 ± 0.18	13.8 ± 0.17	0.25
Body fat, %
11 y	22.7 ± 9.3	27.8 ± 8.4	<0.001
13 y	18.9 ± 9.6	29.1 ± 8.2	<0.001
Change	−3.75 ± 5.8	1.33 ± 4.6	<0.001
Body weight, kg
10 y	37.0 ± 7.5	38.0 ± 8.3	0.004
11 y	42.0 ± 9.0	44.0 ± 9.7	<0.001
13 y	53.9 ± 11.0	54.3 ± 10.2	0.40
Height, cm
10 y	144 ± 6.4	144 ± 6.8	0.46
11y	150 ± 7.0	151 ± 7.2	<0.001
13y	165 ± 8.7	162 ± 6.4	<0.001
Total milk intake,3servings/d
10 y	1.04 ± 0.78	0.79 ± 0.67	<0.001
13 y	1.03 ± 0.85	0.70 ± 0.69	<0.001
Physical activity, counts/min
11 y	644 ± 179	540 ± 148	<0.001
13 y	602 ± 202	480 ± 161	<0.001
Pubertal status 11 y, %			<0.001
Prepubertal	37.2	30.0
Early pubertal	58.6	51.9
Late pubertal	4.2	18.1
Dieting at age 13 y, %	10.7	35.7	<0.001
Maternal BMI, kg/m2	24.6 ± 4.6	24.4 ± 4.4	0.36
Mother's educational attainment, %			0.31
CSE/vocational	16.8	16.1
Ordinary level	34.2	37.4
Advanced level/degree	49.0	46.5

¹Values are means ± SD or proportions. Significant differences between boys and girls were assessed using tests for continuous variables and chi-square analysis for proportions. ALSPAC, Avon Longitudinal Study of Parents and Children; CSE, Certificate of Secondary Education.

²Sample sizes differed due to missing data as follows: baseline BMI ( = 1023 for boys, n = 1227 for girls); weight, age 10 y (n = 1029 for boys and n = 1035 for girls), 11 y (n = 1239 for girls), and 13 y (n = 1239 for girls); height, age 10 y (n = 1024 for boys and n = 1230 for girls) and 11 y (n = 1027 for boys and n = 1238 for girls); pubertal status (n = 838 for boys and n = 1048 for girls); dieting (n = 871 for boys and n = 1097 for girls); maternal BMI (n = 801 for boys, n = 945 for girls); and maternal educational attainment (n = 968 for boys, n = 1152 for girls).

³1 serving = 8 oz of milk (244 g of regular milk and 250 g of flavored milk).

Cross-sectional findings showed an inverse association between milk intake and percent body fat at age 13 y in the simple adjusted model (P < 0.001), but associations were attenuated in multivariable adjusted models (Table 2). Higher full-fat milk intake at age 13 y was inversely associated with percent body fat at 13 y in all models and in those with plausible dietary intakes (P ≤ 0.01 for all); however, reduced-fat milk was not associated with percent body fat.

TABLE 2.

Cross-sectional associations between total, full-fat, and reduced-fat milk intakes and percent body fat at age 13 y^1–3

	β (95% CI)
Daily milk intakes	Per 100 g milk consumed	Per milk servings/d4	P value
Total milk
Model 1	−0.46 (−0.66, −0.27)	−1.13 (−1.60, −0.65)	<0.001
Model 2	−0.16 (−0.36, 0.04)	−0.39 (−0.87, 0.09)	0.11
Model 3	−0.05 (−0.33, 0.24)	−0.11 (−0.80, 0.58)	0.75
Model 4	−0.003 (−0.21, 0.20)	−0.006 (−0.50, 0.49)	0.98
Model 5	−0.02 (−0.32, 0.28)	−0.05 (−0.78, 0.68)	0.90
Full-fat milk
Model 1	−0.76 (−1.06, −0.47)	−1.86 (−2.58, −1.15)	<0.001
Model 2	−0.47 (−0.76, −0.19)	−1.15 (−1.84, −0.45)	0.001
Model 3	−0.44 (−0.72, −0.15)	−1.06 (−1.77, −0.36)	0.003
Model 4	−0.46 (−0.74, −0.18)	−1.12 (−1.81, −0.43)	0.002
Model 5	−0.54 (−0.97, −0.11)	−1.32 (−2.36, −0.27)	0.01
Reduced-fat milk
Model 1	−0.20 (−0.40, 0.008)	−0.48 (−0.98, 0.02)	0.06
Model 2	0.01 (−0.19, 0.22)	0.04 (−0.46, 0.53)	0.89
Model 3	0.19 (−0.06, 0.44)	0.47 (−0.13, 1.08)	0.13
Model 4	0.18 (−0.03, 0.38)	0.43 (−0.08, 0.93)	0.10
Model 5	0.24 (−0.04, 0.53)	0.60 (−0.11, 1.30)	0.10

¹ = 2270, regression estimates (95% CI) and P values were calculated using multivariable linear regression adjusted for age, sex, height, and height squared (model 1).

²Model 2 was adjusted for all variables in model 1 and counts per minute, pubertal status, and maternal BMI and educational attainment, total fat intake and ready-to-eat cereal, sugar-sweetened beverage, and 100% fruit juice intake. Model 3 was additionally adjusted for calcium intake and model 4 for total energy intake.

³Model 5 examined associations among only plausible reporters ( = 882) adjusted for age, sex, height, height squared, pubertal status, and maternal BMI and educational attainment, total fat intake and ready-to-eat cereal intake, sugar-sweetened beverage, and 100% fruit juice intake.

⁴1 serving = 8 oz of milk (244 g of regular milk and 250 g of flavored milk).

In prospective analyses, total milk at age 10 y was associated with body fat at age 11 y in multivariable adjusted models (P = 0.01); the association remained after additional adjustment for total energy (P = 0.03) but was attenuated in the analysis among those with plausible dietary intakes (P = 0.16) (Table 3). Total milk intake at age 10 y was not associated with body fat at 13 y in multivariable adjusted models (Table 4). Full-fat and reduced-fat milk at age 10 y was not related to percent body fat at ages 11 or 13 y.

TABLE 3.

Prospective associations between total, full-fat, and reduced-fat milk intakes at age 10 y and percent body fat at age 11 y^1,2,3

	β (95% CI)
Daily milk intakes	Per 100 g milk consumed	Per milk servings/d4	P value
Total milk
Model 1	−0.16 (−0.28, −0.04)	−0.39 (−0.68, −0.09)	0.01
Model 2	−0.16 (−0.28, −0.04)	−0.39 (−0.69, −0.10)	0.01
Model 3	0.01 (−0.18, 0.20)	0.03 (−0.44, 0.50)	0.89
Model 4	−0.14 (−0.26, −0.01)	−0.34 (−0.64, −0.03)	0.03
Model 5	−0.16 (−0.37, 0.06)	−0.38 (−0.91, 0.15)	0.16
Full-fat milk
Model 1	−0.06 (−0.20, 0.09)	−0.13 (−0.49, 0.22)	0.46
Model 2	−0.08 (−0.23, 0.06)	−0.20 (−0.56, 0.15)	0.26
Model 3	−0.03 (−0.18, 0.12)	−0.07 (−0.44, 0.29)	0.70
Model 4	−0.10 (−0.24, 0.05)	−0.24 (−0.60, 0.12)	0.19
Model 5	−0.04 (−0.32, 0.24)	−0.09 (−0.77, 0.59)	0.80
Reduced-fat milk
Model 1	−0.11 (−0.24, 0.01)	−0.27 (−0.58, 0.04)	0.08
Model 2	−0.08 (−0.20, 0.04)	−0.19 (−0.49, 0.11)	0.21
Model 3	0.07 (−0.08, 0.22)	0.17 (−0.19, 0.54)	0.36
Model 4	−0.05 (−0.17, 0.08)	−0.11 (−0.41, 0.19)	0.47
Model 5	−0.09 (−0.30, 0.13)	−0.21 (−0.73, 0.32)	0.44

¹ = 2245, regression estimates (95% CI) and P values were calculated using multivariable linear regression adjusted for age, sex, baseline BMI, height, and height squared (model 1).

²Model 2 was adjusted for all variables in model 1 and counts per minute, pubertal status, and maternal BMI and educational attainment, total fat intake, and ready-to-eat cereal, sugar-sweetened beverage, and 100% fruit juice intake. Model 3 was additionally adjusted for calcium intake and model 4 for total energy intake.

³Model 5 examined associations among only plausible reporters ( = 907) adjusted for age, sex, baseline BMI, height, height squared, pubertal status, maternal BMI and education attainment, total fat intake and ready-to-eat cereal, sugar-sweetened beverage, and 100% fruit juice intake.

⁴1 serving = 8 oz of milk (244 g of regular milk and 250 g of flavored milk).

TABLE 4.

Prospective associations between total, full-fat, and reduced-fat milk intakes at age 10 y and percent body fat at age 13 y^1–3

	β (95% CI)
Daily milk intakes	Per 100 g milk consumed	Per milk servings/d4	P value
Total milk
Model 1	−0.48 (−0.65, −0.30)	−1.16 (−1.58, −0.74)	<0.001
Model 2	−0.09 (−0.24, 0.07)	−0.21 (−0.59, 0.16)	0.26
Model 3	0.03 (−0.22, 0.27)	0.06 (−0.53, 0.65)	0.83
Model 4	−0.06 (−0.21, 0.09)	−0.15 (−0.52, 0.23)	0.45
Model 5	−0.20 (−0.43, 0.03)	−0.49 (−1.06, 0.09)	0.10
Full-fat milk
Model 1	−0.09 (−0.26, 0.09)	−0.21 (−0.65, 0.22)	0.34
Model 2	−0.09 (−0.27, 0.10)	−0.21 (−0.65, 0.24)	0.36
Model 3	−0.06 (−0.24, 0.13)	−0.14 (−0.59, 0.32)	0.56
Model 4	−0.10 (−0.28, 0.08)	−0.24 (−0.69, 0.20)	0.28
Model 5	−0.21 (−0.48, 0.07)	−0.50 (−1.17, 0.17)	0.14
Reduced-fat milk
Model 1	−0.05 (−0.21, 0.10)	−0.13 (−0.50, 0.25)	0.51
Model 2	−0.01 (−0.17, 0.14)	−0.03 (−0.41, 0.35)	0.87
Model 3	0.09 (−0.10, 0.28)	0.22 (−0.24, 0.68)	0.34
Model 4	0.03 (−0.13, 0.18)	0.06 (−0.32, 0.44)	0.74
Model 5	−0.03 (−0.27, 0.20)	−0.08 (−0.66, 0.50)	0.78

¹ = 2245, regression estimates (95% CI), and P values were calculated using multivariable linear regression adjusted for age, sex, baseline BMI, height, and height squared (model 1).

²Model 2 was adjusted for all variables in model 1 and counts per minute, pubertal status, maternal BMI and educational attainment, total fat intake, and ready-to-eat cereal, sugar-sweetened beverage and 100% fruit juice intake. Model 3 was additionally adjusted for calcium intake and model 4 for total energy intake.

³Model 5 examined associations among only plausible reporters ( = 876) adjusted for age, sex, baseline BMI, height, height squared, pubertal status, and maternal BMI and education attainment, total fat intake and ready-to-eat cereal, sugar-sweetened beverage, and 100% fruit juice intake.

⁴1 serving = 8 oz of milk (244 g of regular milk and 250 g of flavored milk).

In secondary analyses, there were no associations between changes in milk intake from ages 10 to 13 y and changes in percent body fat from ages 11 to 13 y (data not shown). Results were unchanged after additional adjustment for calcium or total energy and among those with plausible dietary intakes (P ≥ 0.31 for all). There was no relationship between change in full-fat or reduced-fat milk intake and change in body fat (P ≥ 0.09). Results for prospective analyses of milk intake at age 10 y and changes in percent body fat from 11 to 13 y were similar to analyses including change in milk intakes (data not shown).

Models with adjustment for dieting at age 13 y produced similar results as those described above (data not shown). Associations between plain milk intake and percent body fat for all models were similar to those observed with total milk intake (data not shown). Plain and total milk intakes were highly correlated (Spearman correlation coefficient: 0.92 and 0.93 at age 10 and 13 y, respectively; P < 0.001).

Discussion

We found an inverse cross-sectional association between full-fat milk intake and percent body fat after adjusting for covariates and potential confounding factors. Our study identified a small association between milk intake at age 10 y and percent body fat at age 11 y in multivariable adjusted models; however, this result was attenuated when analyses were restricted to those with plausible dietary intakes. There was no association between milk intake at age 10 y and body fat at age 13 y in multivariable adjusted models. No relationship was observed between changes in milk intake from 10 to 13 y and changes in body fat from ages 11 to 13 y.

Findings from other cross-sectional studies on associations between milk consumption and body composition in children have been inconsistent (15, 42). One study reported an inverse association between frequency of milk consumption and BMI Z-score in children (mean age ~ 7.5 y) (15). Another found no association between milk intake, regardless of the type (e.g. skim), and weight status (BMI percentiles) (42). We observed a cross-sectional association between full-fat milk intake and percent body fat at age 13 y in all models. A recent randomized controlled trial found that supplementation with CLA, naturally occurring in dairy fat and ruminant animal products, decreased body fat in children (43). A Swedish study in women aged 40–55 y also reported a lower risk of weight gain for whole milk consumption even after adjusting for CLA (44). Due to the potential for reverse causation, whereby individuals with higher BMI may adjust their dietary habits, these findings should be interpreted with caution.

Although there was a small inverse relationship in the prospective analysis between milk intake at age 10 y and body fat at age 11 y, there was no association between baseline milk intake and body fat at age 13 y after multivariable adjustment. Further, our prospective change models revealed no association between changes in milk intake and changes in body fat over 2 y. The relationship between baseline milk intake and body fat was attenuated among those with plausible dietary intakes, indicating that this association may have been a type I error resulting from analyses that included those with implausible dietary intakes. Adjusting for reporting errors will often attenuate an observed spurious association, which may have been the case in our study. Results among plausible dietary reporters only were consistent with our other prospective models, showing no association between milk intake and adiposity. It is also possible that the relationship between milk consumption and adiposity changes as children develop and progress through puberty. Although we adjusted for pubertal status at age 11 y, there remains the potential for residual confounding.

The majority of our longitudinal results support several other well-designed prospective studies that have reported null findings between milk intake and body composition (18, 19, 45). Others, however, have noted inverse associations with adiposity among children (46–48). A recent study reported associations between whole milk intake at age 2 y and BMI Z-score at age 3 y; findings disappeared when analyses were restricted to those with a normal BMI (46). One study found that higher milk intake was associated with gains in BMI, although this appeared to be mediated by higher energy intakes from greater milk consumption (20). In that study, the relationship was stronger for skim and 1% milk intake than for whole or 2% milk. We found no association with body fat regardless of the type of milk consumed.

It is possible that the lack of association in our change models may be due to low milk consumption at baseline (boys: 1.04 ± 0.78 servings/d; girls: 0.79 ± 0.67 servings/d). In one prospective study that observed inverse findings between milk intake at follow-up (adjusted for baseline intake) and change in BMI, over 56% of adolescents consumed ≥7 servings/wk of milk (48). The lack of findings between baseline milk intake and body fat at 13 y could be due to the fact that measured milk intake at age 10 y was too distal to body fat measured at 13 y, although our study was adequately powered to detect associations in our prospective analyses. However, this study was not powered to detect associations between changes in milk intake and changes in body fat, which were quite small and of limited variability (boys: −0.01 ± 0.8 servings/d; girls: −0.09 ± 0.7 servings/d). It is also possible that our null findings may be due to a true lack of association between milk and adiposity, which corroborates several other reports as noted.

There are considerable methodological differences across studies examining milk intakes and adiposity among children, including length of study follow-up, dietary assessment methods (FFQ, recalls), treatment of milk intake (servings per day, consumers vs. nonconsumers), and assessment of body composition (BMI, body fat) as well as confounding factors included for adjustment. In a subsample of children from ALSPAC, milk intake at ages 5 and 7 y was associated with change in fat mass at age 9 y in an adjusted model (47). That study included younger children and milk intakes were higher [median (IQR): 257 g/d (133–388) at age 5 y and 242 g/d (177–376) at age 7 y] compared with our study. Other study populations have reported declines in milk intake with age (4, 5). In nutritional epidemiologic studies, including individuals with implausible dietary intakes in diet-obesity studies may lead to inaccurate results and accounting for this measurement error may clarify associations (23, 24). For example, Huang et al. (23) noted that associations among reported energy intake, meal portion size, and meal energy were more strongly and consistently associated with BMI percentile in children with plausible dietary intakes compared with the full sample. On the other hand, a cross-sectional study of 172 girls aged 11 y reported inverse associations between higher dairy intake and percent body fat; however, there were no differences in a subsample limited to plausible reporters (25). In our study, accounting for reporting errors did not influence results in most analyses conducted. However, our findings between milk intake at age 10 y and percent body fat at age 11 y among plausible reporters were consistent with our other longitudinal models. This does not undermine the importance of considering reporting errors in dietary studies. Perhaps adjustment was less important in our study, which may have been limited due to low intakes of milk in this population.

Similar to our study, Fiorito et al. (19) reported that milk intakes at age 5 y did not predict body fat in girls from age 5–15 y in unadjusted models. Although this study also employed DXA (and skinfold-thickness measures) to assess adiposity, they examined beverage intake at age 5 y, had a longer follow-up period, and examined milk intakes only in models that did not adjust for potential confounding variables. On the other hand, Vanselow et al. (48) reported a nonlinear dose-response relationship between milk intake and change in BMI over 5 y in 2294 adolescents. That study used a FFQ to assess intakes and self-reported heights and weights to calculate BMI and did not assess pubertal status. The authors noted that although the FFQ was a validated dietary assessment tool, portion size was not accurately assessed and intakes may have been underestimated. A recent, 16-wk, randomized controlled trial did not find improved weight loss in overweight children consuming 4 servings/d compared to 1 serving/d of milk as part of a healthy diet (49). These studies raise the question of whether diversity in study methods and designs contribute to the equivocal findings on the milk-body fat hypothesis.

Our study has several strengths and limitations. We used rigorous methods to assess dietary intake and body composition, including 3-d dietary records and DXA. We also included a wide range of maternal and child characteristics and other dietary variables as potential confounding factors in our regression models. Also, to evaluate the potential impact of study design on results, we tested associations using both cross-sectional and prospective analyses. Importantly, we examined the potential impact of reporting errors on our results by building models limited to those with plausible dietary intakes. Despite our large sample size, ALSPAC is not nationally representative and has limited variability in race/ethnicity. Although a range of confounding factors was considered, there are limitations to using a traditional “single food” approach and residual confounding by other dietary factors may confound associations (50). The current work focused on milk intakes in relation to body composition; however, other manuscripts are currently in progress or have examined associations with total dietary patterns or other dietary factors such as beverages consumed in this study population (47, 51, 52). There may have been misclassification of reporting status, because we used prediction equations to assess reporting errors. Additionally, although accelerometers are currently one of the most accurate methods used to assess physical activity, there are some limitations to using this method (34, 53). For example, this instrument cannot be worn during certain activities such as water sports and may not reflect the exercise intensity of come activities such as cycling (34). In some cases, this may have resulted in an underestimation of predicted energy requirements, because total physical activity may have been underestimated. Lastly, baseline dietary intake at age 10 y was not collected concurrently with the baseline DXA measure, which was first measured at age 11 y in this study.

In summary, we performed a set of rigorous analyses that employed a number of different study designs and analyses to examine the milk-body fat hypothesis in a large sample of children. Associations between full-fat milk and adiposity were noted only in cross-sectional models. Results between total milk at age 10 y and percent body fat at age 11 y were attenuated after accounting for dietary reporting errors, suggesting that this may have been a spurious finding that was attenuated after accounting for individuals with implausible dietary intakes. Findings from the majority of our longitudinal models support findings from other studies that showed no relation between milk intakes and adiposity in children after adjusting for confounding factors such as age, sex, sedentary behaviors, sociodemographic variables, total energy, and consumption of nondairy beverages. Also, our prospective analyses were powered to detect associations. However, due to the inconsistent findings across analyses in the present study, further investigation is warranted to clarify the relationship between milk and adiposity, and accounting for dietary reporting errors is an important consideration.