Association of cow's milk intake in early childhood with adiposity and cardiometabolic risk in early adolescence

ABSTRACT

Background

Prior studies have provided conflicting evidence regarding associations of pediatric milk consumption with subsequent adiposity.

Objectives

We aimed to estimate associations of the consumption frequency and fat content of early childhood milk intake with early adolescent adiposity and cardiometabolic risk.

Methods

We analyzed data collected prospectively from 796 children in Project Viva, a Boston-area prebirth cohort. Parents reported the frequency (times/day) and fat content [higher-fat: whole (3.25%) or 2% milk; lower-fat: 1% or skim milk] of cow's milk consumed in early childhood (mean, 3.2 years) via food-frequency questionnaires. We measured adiposity and cardiometabolic markers in early adolescence (mean, 13.2 years) and conducted multivariable regressions to assess associations adjusted for baseline parental and child sociodemographic, anthropometric, and dietary factors.

Results

In early childhood, mean milk intake was 2.3 times/day (SD, 1.2 times/day), and 63% of children drank primarily higher-fat milk. The early childhood BMI z-score (BMIz) was inversely associated with the fat content of milk consumed in early childhood. After adjustment for baseline parent and child factors, early childhood intake of higher-fat compared with lower-fat milk was associated with lower adiposity; however, the 95% CIs for most adiposity outcomes—except for the odds of overweight or obesity (OR, 0.60; 95% CI, 0.38–0.93)—crossed the null after adjustment for the baseline child BMIz and BMIz change between ages 2 and 3 years. Early childhood consumption of higher-fat milk (compared with lower-fat milk) was not associated with adverse cardiometabolic outcomes. The frequency of cow's milk consumption in early childhood was not associated with adiposity or cardiometabolic risk in early adolescence.

Conclusions

Consumption of higher-fat cow's milk in early childhood was not associated with increased adiposity or adverse cardiometabolic health over a decade later. Our findings do not support current recommendations to consume lower-fat milk to reduce the risk of later obesity and adverse cardiometabolic outcomes. This trial was registered at clinicaltrials.gov as NCT02820402.

Introduction

Dairy products provide a rich source of nutrients for growth and bone health during childhood (1); however, concern exists regarding the extent to which dairy fat might contribute to excess weight gain. The 2020–2025 Dietary Guidelines for Americans (2), as well as guidelines in numerous other countries (3) and from multiple professional organizations (4, 5), recommend transitioning from whole milk (3.25% milk fat) to lower-fat milk (0%–1% milk fat) starting at age 2 years (6). These recommendations are based on meeting estimated calcium requirements for bone health while minimizing energy and saturated fat intake, assuming that the latter strategy will lessen the risk of future overweight or obesity and promote cardiometabolic health.

Prior studies examining the association of the frequency and fat content of cow's milk consumed in early life with later adiposity and cardiometabolic risk have yielded conflicting results with significant limitations, such as cross-sectional study designs, small sample sizes, lack of adjustments for critical confounders, and short study durations (3, 7–20). The possibility of confounding by indication is particularly salient in studies assessing the association of milk fat content with adiposity, because parents of children with higher BMIs are more likely to select lower-fat milk for their children based on current dietary guidelines, and parents of children with lower BMIs are more likely to select higher-fat milk (7). Existing prospective cohort studies that have attempted to mitigate confounding by indication by adjusting for baseline anthropometrics have shown a potential protective effect of higher-fat milk against later adiposity and have thus challenged the conventional wisdom that lower-fat milk protects against future excess weight gain (3, 8, 21). As longstanding dietary preferences and habits may be formed early in life at a time when adults dictate child consumption patterns (7, 22, 23), evidence-based nutritional guidelines regarding childhood milk consumption may have profound implications for population health, particularly in the United States and other settings in which cow's milk is a significant component of the childhood diet (24).

To address these knowledge gaps and inform dietary guidelines, we examined associations of fat content and frequency of cow's milk intake in early childhood with body composition and cardiometabolic risk in early adolescence. We hypothesized that higher-fat (whole or 2%) relative to lower-fat (1% or skim) cow's milk consumption would be associated with lower measures of adiposity and cardiometabolic risk and that the frequency of cow's milk intake (times/day) would not be associated with adiposity or cardiometabolic risk. We compared higher-fat with lower-fat milk intake to align our exposure with current national guidelines in this age group. In a sensitivity analysis, we also compared whole milk with the remaining types of milk fat contents (2%, 1%, and skim) consumed in early childhood.

Methods

Study population

We studied participants in Project Viva, an ongoing prebirth, Boston-area prospective cohort established to examine associations of pre- and postnatal factors with maternal and child health. Full details of the cohort, including recruitment process, inclusion and exclusion criteria, and follow-ups, have been previously published (25), and the study is registered at clinicaltrials.gov (NCT02820402). Briefly, we recruited pregnant individuals in 1999–2002 at prenatal visits at Atrius Harvard Vanguard Medical Associates, a large, group practice in eastern Massachusetts. Study protocols and instruments are available at https://www.hms.harvard.edu/viva/. The Institutional Review Board of Harvard Pilgrim Health Care approved the project. All mothers provided written informed consent at enrollment and follow-up visits, and all children provided assent at the midchildhood and early adolescent visits.

Of 2128 live, singleton births, we excluded 1089 without early adolescent outcomes, 203 with missing or invalid exposure in early childhood, 17 who drank primarily soy milk, 10 who drank primarily “other” milk (e.g., rice milk and chocolate cow's milk), and 13 who did not drink milk (Supplementary Figure 1). There were notable differences between the 796 included participants and the 1332 excluded participants (Supplementary Table 1). Compared to those excluded, included participants had parents with higher education levels and mothers with higher ages, rates of being married or cohabitating, household incomes, and median neighborhood household incomes based on the census tract at the child's birth, and with lower rates of smoking during pregnancy. Included participants were younger at the early childhood visit, were more likely to be identified as being of White race and ethnicity, were born at a higher gestational age, were breastfed for a longer duration, and consumed more protein and less dietary fiber than excluded participants. As expected, since we excluded children who did not drink milk from this analysis, included participants also had higher dietary intakes of calcium and vitamin D and a higher measured serum vitamin D in early childhood. We observed no differences in other baseline characteristics; notably, the mean milk intake frequency [2.2 times/day (SD, 1.2 times/day) compared with 2.3 times/day (SD, 1.2 times/day), respectively; P = 0.11) and percentage of children drinking higher-fat milk (66.2% versus 63.2%, respectively; P = 0.68) were both similar among those excluded children with valid, non-zero cow's milk consumption data (n = 355; 25.2%) compared with included participants.

Exposures: cow's milk intake frequency and fat content in early childhood

At the early childhood visit (mean, 3.2 years; SD, 0.2 years), mothers responded to an 87-item, semiquantitative FFQ previously validated in this age group (26). Our 2 exposures in this analysis were cow's milk fat content and cow's milk intake frequency over the last month, as assessed in the FFQ. We based the assessment of cow's milk fat content on the response to the FFQ item: “what kind of milk does your child usually drink?,” with answer choices of: whole milk, 2% milk, 1% milk, skim milk, soy milk, “other (please specify),” and “my child does not drink milk.” The answers listed under “other (please specify)” were examined and then recoded as a type of cow's milk according to the milk fat content if appropriate, or removed if not a type of cow's milk (i.e., oat milk, rice milk). When mothers specified more than 1 milk type, we used the cow's milk option with higher fat content. We excluded all participants who selected soy milk, and we excluded those who did not drink milk (n = 13) to enable adjustment for the milk fat content consumed among all included participants.

We based our assessment of the cow's milk intake frequency on the response to the question of how often the child drank milk, including chocolate milk, on average over the past month, with response choices of: never, less than once per week, once per week, 2–4 times per week, nearly daily or daily, 2–4 times per day, and 5 or more times per day. As above, we excluded the participants who selected never from the analysis to allow for adjustment for milk fat content among all participants. We transformed frequency into a continuous variable for our multivariable analysis by reporting the average within each category and recoding weekly measures as daily measures by dividing by Huh et al. (7). As a result, 2–4 times per week was averaged to 3 times/week and recoded as 0.429 times/day.

Outcomes in early adolescence

Body composition outcomes

At the early adolescent visit (mean, 13.1 years; SD, 0.9 years), trained research assistants measured weight to the nearest 0.1 kg with a calibrated scale (Tanita model TBF-300A; Tanita Corporation of America, Inc.), height to the nearest 0.1 cm using a calibrated stadiometer (Shorr Productions), waist circumference to the nearest 0.1 cm immediately above the iliac crest while standing using a Gulick II measuring tape (Performance Health), and subscapular and triceps skinfold thicknesses to the nearest 0.1 mm using Holtain calipers, from which we calculated sums of subscapular and triceps skinfold thicknesses. We ascertained the percentage of body fat using foot-to-foot bio-impedance (Tanita TBF-300A), and we obtained measures of the lean, total fat, and trunk fat mass using DXA (Hologic model Discovery A). We calculated the following indices (all in kilograms of mass/height in meters squared): BMI, lean mass index, total fat mass index, and trunk fat mass index. We derived age- and sex-specific BMI z-scores (BMIz) using the CDC growth references, and defined overweight or obesity as a BMI ≥ 85^th percentile for age and sex (27). We obtained all measurements using standardized protocols, with high intra- and inter-rater reliability (28). We considered continuous BMIz and waist circumference measurements and the odds of overweight or obesity in early adolescence to be primary outcomes, and the remainder of the adiposity and cardiometabolic measures to be secondary outcomes.

Cardiometabolic health outcomes

We assessed systolic blood pressure (SBP), calculated as the average of 5 measurements taken 1 minute apart, using Dinamap Pro-100 oscillometric automated monitors (GE Medical Services), from which we derived SBP z-scores based on national data (29). Phlebotomists obtained fasting venous blood specimens, which were centrifuged within 24 hours and stored at −80 °C. We measured fasting glucose, insulin, HDL cholesterol, triglycerides, leptin, and adiponectin in the stored plasma. We calculated the HOMA-IR using the standard definition of [fasting glucose (mg/dL) × fasting insulin (µU/L)]/405. We calculated the cardiometabolic risk z-score (MetRiskz) using the mean of 5 sex- and age-specific z-scores referenced to national data for waist circumference (30), SBP (29), inverted HDL cholesterol (31), log-transformed triglycerides (31), and log-transformed HOMA-IR (32). We log-transformed non-normally distributed outcome variables to improve normality. While lower values for most cardiometabolic outcomes in this analysis are more healthful, higher values of log adiponectin and log HDL cholesterol are considered more favorable.

Covariates

At enrollment, mothers reported their pre-pregnancy weight and height, as well as the weight and height of the child's biological father, from which we calculated maternal and paternal BMIs. They also reported their age, smoking status, marital status, and household income, and the education levels of both parents. We ascertained the median neighborhood household income based on the census tract at the time of the child's birth (33). We abstracted information on the delivery date, infant sex, and birth weight from medical records at birth. We calculated gestational age from the last menstrual period or the second-trimester ultrasound if estimates differed by >10 days. We derived sex-specific birth weights for gestational age z-scores based on national US data (34). We defined gestational weight gain as the difference between the last weight recorded before delivery and the self-reported pre-pregnancy weight. During the first and second trimesters, mothers completed FFQs, from which we calculated Alternative Healthy Eating Index for Pregnancy (AHEI-P) scores. We then averaged these measures across both the first and second trimesters as a marker of maternal dietary quality (35).

With the 6-month and 12-month questionnaires, mothers reported whether they were breastfeeding their child and the age at which they stopped breastfeeding, from which we derived the breastfeeding duration in months. At the early childhood visit, we asked mothers, “which of the following best describes your child's race or ethnicity?” Mothers had a choice of ≥1 of the following racial and ethnic groups: Hispanic or Latina, White or Caucasian, Black or African American, Asian or Pacific Islander, American Indian or Alaskan Native, and Other (please specify). If the child's race and ethnicity was missing, we used the mother's self-reported race and ethnicity (36). In this analysis, we combined race and ethnicity categories other than Black or White due to small sample sizes. We included race and ethnicity as a covariate as a marker of social experience rather than biological difference. Mothers also reported the average number of hours/day over the past month that their child spent in active play, television viewing, and sleeping using questionnaires at the early childhood visit.

Based on responses to the early childhood FFQ, we estimated daily energy and individual macronutrient and micronutrient intakes using the Harvard nutrition database. We adjusted all nutrients for energy intake using the nutrient residual method. We computed components of the Youth Healthy Eating Index (YHEI) as a marker of overall dietary quality, based on responses to the FFQ regarding consumption of whole grains, vegetables, fruits, total dairy, meats, snack food (such as chips, cookies, brownies, cake, cupcakes, pie, crackers, Jell-O, donuts, muffins, chocolate candy, and other candy), soda and other drinks [including fruit drinks (Hi-C, Kool-Aid, lemonade); hot chocolate; sugar-free soda], margarine or butter, multivitamins, and fast foods outside of the home [the latter was asked using the question: “in the past month, on average, how often did your child eat something from a fast-food restaurant (e.g., McDonald's, Burger King, Taco Bell, etc.)?”]. Three components of the YHEI were unavailable (visible animal fat, eating breakfast, eating dinner with family); however, as described by Switkowski et al. (37), these items contributed minimally to variability in the total YHEI score in an external sample used for the development of the score (38). Each score was calculated based on the assumption that 1 “time” is equal to 1 serving and that a response of “daily or near-daily” in response to a question of how often a portion of food is consumed can be approximated as 1 serving per day. We excluded the YHEI component derived from dairy consumption due to its correlation with our exposure. We separately assessed “other dairy” consumption, defined as the sum of times/day of cheese, cream cheese, cottage cheese, yogurt, ice cream, and pudding consumption. As prior studies have hypothesized that increasing rates of overweight and obesity among children are related to the concurrent observed rise in consumption of sugar-sweetened beverages (SSBs) and decline in consumption of cow's milk (7, 39), which may indicate that children are substituting cow's milk for SSBs, we did not include the “soda or drinks” component of the YHEI in the aggregate measure. Instead, we assessed SSBs as a separate covariate to isolate their effect. We defined SSBs as regular soda and fruit drinks, but not 100% fruit juice, which we included as a separate covariate; as above, we could not include flavored milk, as we did not assess its consumption frequency in this age group. Consequently, a total of 8 components of the YHEI were summed (modified YHEI) and included as a covariate in our analysis.

We obtained child weight and height data around age 2 years from medical records from well-child visits. We also assessed child weight and height at the early childhood visit around age 3 years. For age 2 years, we derived age- and sex-specific BMIz values using WHO growth standards, as not all children had reached the age of 24 months at the 2-year visit (range, 18–30 months), and CDC BMI reference data are not available for children aged < 24 months. For early childhood BMIz, we calculated both the WHO BMIz and CDC BMIz. We assessed the rate of change in the BMIz from age 2 to 3 years using the WHO BMIz. We used the CDC BMIz as a covariate based on its standard use in the United States in this age group.

Statistical analysis

We presented participant characteristics overall and stratified by cow's milk fat content and intake frequency (Table 1; Supplementary Table 2). We reported P values from the Pearson's chi-squared test or Fisher's exact test if any cell values were less than 5 (with the additional use of simulation by the Monte Carlo method if needed) for categorical variables and from the Wilcoxon or Kruskal-Wallis rank-sum tests for continuous variables.

Table 1.

Baseline characteristics of 796 Project Viva participants overall and by cow's milk fat content consumed in early childhood¹

Characteristic, mean (SD); n (%)	Overall N = 796	Higher-fat milk (whole or 2%), n = 503 (63.2%)	Lower-fat milk (1% or skim), n = 293 (36.8%)	P value2
Parental and prenatal factors
Maternal age at enrollment, years	32.5 (4.9)	31.9 (5.2)	33.4 (4.2)	<0.001
Maternal education, % with college degree	588 (74.0%)	344 (68.5%)	244 (83.3%)	<0.001
Maternal marital status, % married or cohabitating	746 (94.0%)	457 (91.0%)	289 (99.0%)	<0.001
Maternal smoking status, %	—	—	—	0.01
During pregnancy	74 (9.3%)	58 (11.6%)	16 (5.5%)
Former	162 (20.4%)	95 (19.0%)	67 (22.9%)
Never	558 (70.3%)	348 (69.5%)	210 (71.7%)
Maternal pre-pregnancy BMI, kg/m2	24.6 (5.1)	24.7 (5.2)	24.5 (4.8)	0.91
Gestational weight gain, kg	15.6 (5.2)	15.3 (5.2)	16.0 (5.3)	0.08
Average AHEI-P, score	60.8 (9.5)	60.1 (9.5)	61.8 (9.4)	0.02
Maternal parity, % nulliparous	383 (48.1%)	267 (53.1%)	116 (39.6%)	<0.001
Paternal BMI, kg/m2	26.4 (3.9)	26.3 (4.1)	26.7 (3.6)	0.07
Paternal education, % with college degree	527 (70.8%)	300 (65.6%)	227 (79.1%)	<0.001
Median neighborhood household income, 1000	58.3 (21.1)	55.1 (20.7)	63.7 (20.7)	<0.001
Household income > 70,000, %	492 (65.4%)	286 (61.2%)	206 (72.3%)	0.002
Child sociodemographic and early life factors
Child sex, % female	404 (50.8%)	244 (48.5%)	160 (54.6%)	0.10
Child age at early childhood visit, years	3.2 (0.2)	3.2 (0.2)	3.2 (0.2)	0.11s
Child race and ethnicity, %	—	—	—	<0.001
Black	101 (12.7%)	92 (18.3%)	9 (3.1%)
Other	143 (18.0%)	118 (23.5%)	25 (8.5%)
White	552 (69.3%)	293 (58.3%)	259 (88.4%)
Gestational age at delivery, weeks	39.6 (1.6)	39.5 (1.6)	39.8 (1.7)	0.03
Birth weight for gestational age z-score, units	0.21 (0.95)	0.09 (0.96)	0.41 (0.89)	<0.001
Breastfeeding duration (months)	6.7 (4.5)	6.5 (4.5)	7.2 (4.5)	0.06
Change in BMIz from 2–3 years, WHO, SD units	0.27 (0.88)	0.28 (0.88)	0.26 (0.87)	0.81
Early childhood factors (age 3 years)
BMIz at age 3, CDC, SD units	0.45 (0.99)	0.35 (0.99)	0.62 (0.96)	<0.001
Overweight or obesity at age 3, BMI ≥ 85th percentile	189 (25.8%)	106 (22.9%)	83 (30.7%)	0.02
Energy intake, kcal/day	1855.6 (518.9)	1888.9 (541.5)	1800.7 (475.1)	0.04
Protein intake, g/day	58.7 (8.1)	57.2 (8.1)	61.4 (7.5)	<0.001
Dietary calcium intake, mg/day	934.3 (232.3)	882.2 (214.8)	1020.6 (235.0)	<0.001
Dietary vitamin D intake, IU/day	238.3 (97.6)	214.3 (82.8)	278.1 (106.9)	<0.001
Serum vitamin D level, nmol/day	89.7 (26.4)	88.4 (26.8)	91.6 (25.8)	0.21
Sugar-sweetened beverage intake, times/day	0.27 (0.67)	0.34 (0.80)	0.14 (0.30)	<0.001
Juice intake, times/day	1.8 (1.4)	1.9 (1.5)	1.6 (1.3)	0.04
Modified Youth Healthy Eating Index, score	37.9 (8.4)	37.1 (8.4)	39.1 (8.3)	0.001
Sucrose intake, g/day	29.8 (9.0)	29.7 (9.3)	29.9 (8.6)	0.79
Fiber intake, g/day	13.4 (3.4)	12.9 (3.2)	14.1 (3.6)	<0.001
Cow's milk intake, times/day	2.3 (1.2)	2.2 (1.2)	2.4 (1.1)	0.01
Other dairy intake, times/day	1.8 (0.9)	1.7 (0.9)	1.8 (0.9)	0.04
Active play, hours/day	2.9 (1.6)	2.9 (1.6)	2.9 (1.6)	0.59
Sleep, hours/day	11.2 (1.2)	11.1 (1.2)	11.3 (1.0)	0.08
Television viewing, hours/day	1.6 (1.0)	1.7 (1.1)	1.5 (0.9)	0.003

¹AHEI-P, Alternative Healthy Eating Index for Pregnancy; BMIz, BMI z-score; IU, international units.

²Wilcoxon rank-sum test or Pearson's chi-squared test.

We used multivariable linear and logistic regression models to examine the associations of cow's milk fat content and intake frequency in early childhood with adiposity and cardiometabolic risk in early adolescence. We considered covariates for inclusion in the multivariable model based on a priori knowledge and a literature review [such as factors previously shown to be associated with adiposity in this cohort (7, 40–49)] or by their association in the univariate analysis, defined as P value < 0.1, with either the milk intake frequency or milk fat content. We also empirically included sex (for outcomes not adjusted by sex), age at exposure ascertainment, and age at outcome ascertainment (for outcomes not adjusted by age) to improve precision in all models. We did not include calcium intake, vitamin D intake, protein intake, or serum vitamin D levels as covariates due to their potential roles on the causal pathway between milk consumption and adiposity, as well as their presumed multicollinearity with milk intake frequency.

For all analyses, we applied 4 models. In model 1, we adjusted for the other exposure of interest (i.e., we adjusted milk fat content for milk intake frequency, and we adjusted milk intake frequency for milk fat content), as well as for child age at the time of the exposure and for child sex and age at the time of the outcome measurement (for outcomes other than odds of overweight or obesity and z-score outcomes). In model 2, we additionally adjusted for parental education and BMI (both assessed at enrollment); maternal age, marital status, dietary quality during pregnancy, gestational weight gain, and parity; median neighborhood and individual household incomes at birth; and child age at early childhood visit, gestational age at birth, birth weight for gestational age z-score, breastfeeding duration in months, and race and ethnicity. In model 3, we additionally adjusted for early childhood sleep duration, television viewing, modified YHEI score, and intakes of total energy, sugar-sweetened beverages, fiber, juice, sucrose, and other dairy. In model 4, we additionally adjusted for early childhood BMIz values and the change in BMIz values between ages 2 and 3 years, to account for the possibility of confounding by indication.

Because of the expected correlations among our outcomes, we chose not to adjust for multiple comparisons but instead to assess the overall pattern of associations among outcomes and interpret any isolated associations cautiously (50). We evaluated for multicollinearity using generalized variance inflation factors, and considered values <3 as the absence of multicollinearity according to standard statistical practices. We assessed for effect modification by child sex, child weight status at age 3 years (underweight or normal weight compared with overweight or obesity), child race and ethnicity (White compared with Black compared with Other), and maternal education (college degree or higher compared with less than a college degree) by adding multiplicative interaction terms with the milk intake frequency and fat content; we also assessed for interactions between these 2 exposures of interest themselves. We found no evidence of effect modification among any of the tested covariates.

We employed multiple imputation using chained equations to impute missing data for covariates using all exposures, covariates, and outcomes assessed in this study. We generated and combined 50 imputed datasets for all 2128 Project Viva children. We then restricted our analysis to those with complete exposure data (cow's milk intake frequency and fat content in early childhood) and at least 1 early adolescent outcome. Overall, most covariates had less than 5% missing data, except for household income (5.5%), modified YHEI score (5.5%), paternal education (6.5%), AHEI-P score (14.3%), and change in BMIz from ages 2 to 3 years (24.0%). Missingness among outcomes ranged from 0.0% missing for child waist circumference to 42.7% missing for metabolic risk z-score, resulting in a sample size range of 456–796 across different analyses. To assess the robustness of our findings, we repeated all analyses in subjects without missing exposure, covariate, or outcome data ( n = 237–400 for model 4). We also conducted sensitivity analyses, including restricting our multivariate analyses to those children with a baseline weight in the normal range (n = 529) and then separately to those with a baseline weight in the overweight or obese range (n = 189), to address the concern for confounding by indication. We additionally repeated our analysis with exposure groups that compared whole milk consumption with all other milk fat contents (2%, 1%, and skim) in early childhood. Finally, we adjusted for the modified YHEI score assessed at around age 2, instead of at the early childhood visit, to assess the influence of dietary composition prior to exposure ascertainment. We used R version 1.1.463 (RStudio, Inc.) for all analyses.

Results

Participant characteristics

In early childhood, 63% of children drank whole or 2% milk, and 37% of children drank 1% or skim milk. The mean BMIz values were 0.45 (SD, 0.99) in early childhood and 0.35 (SD, 1.1) in early adolescence, with 26% in early childhood and 27% in early adolescence noted to have a BMI ≥ 85^th percentile (overweight or obesity). As shown in Figure 1, a higher BMIz in early childhood was associated with the consumption of lower-fat milk in early childhood (P = 0.001) but not in early adolescence (P = 0.62); this association underscores the importance of addressing confounding by indication in this analysis. Table 1 describes the baseline characteristics overall and according to the type of milk consumed in early childhood. Mothers of children drinking higher-fat milk were more likely to be younger, not have a college degree, not be married or cohabitating, smoke during pregnancy, have a less healthful diet, have lower parity, have a household income ≤ 70,000/year, and have a lower median neighborhood household income. Fathers of children drinking higher-fat milk were also less likely to be college educated. Children drinking higher-fat milk were more likely to be identified as Black or Other race and ethnicity and had higher total daily energy, SSB, and juice intakes, as well as more television viewing. Children drinking higher-fat milk also had lower gestational ages at birth and birth-weight-for-gestational-age z-scores, as well as lower dietary quality and lower intakes of cow's milk, other dairy, fiber, protein, calcium, and vitamin D, though, notably, there was no difference in serum vitamin D levels. Differences in baseline characteristics according to the milk fat content in the 4 categories (whole, 2%, 1%, and skim) are shown in Supplementary Table 3.

Figure 1.

Mean (SE) BMI z-score in early childhood and early adolescence by fat content of milk consumed in early childhood (n = 796).

In early childhood, the mean cow's milk intake was 2.3 times/day (SD, 1.2 times/day). Mothers of children drinking milk more frequently had higher rates of being married or cohabitating, education levels, and median neighborhood household incomes. Children with a higher frequency of milk consumption were more likely to be identified as White race and ethnicity, have higher serum vitamin D levels and sleep durations, and have higher total energy, protein, calcium, and vitamin D intakes, with lower total sucrose and fiber intakes. The highest (≥5 times/day) and lowest (>0 to 4 times/week) frequency milk drinkers had higher SSB intake and lower juice intake compared with those with milk consumption closer to the average (Supplementary Table 2).

Multivariable regression

After adjustment for all confounders (including the baseline child BMIz value and the BMIz change between ages 2 and 3) in model 4, consumption in early childhood of whole or 2% milk, compared with 1% or skim milk, was associated with lower odds of overweight or obesity in early adolescence (Table 2). In contrast, there was no association between the milk intake frequency in early childhood and the odds of overweight or obesity in early adolescence.

Table 2.

Odds (95% CIs) of overweight or obesity in early adolescence by cow's milk fat content and intake frequency in early childhood using multivariable logistic regression (n = 793)

	Model 11	Model 22	Model 33	Model 44
Milk fat content, whole or 2% vs 1% or skim	0.75 (0.54–1.03)	0.57 (0.38–0.85)	0.55 (0.36–0.83)	0.60 (0.38–0.93)
Milk frequency, per time/day	0.97 (0.85–1.11)	1.03 (0.89–1.19)	1.06 (0.89–1.27)	1.01 (0.83–1.22)

¹Model 1 is adjusted for the other exposure of interest (i.e., milk fat content is adjusted for milk intake frequency, and milk intake frequency is adjusted for milk fat content), as well as for child age at the time of the exposure.

²Model 2 is additionally adjusted for parental education and BMI (both assessed at enrollment); maternal age, marital status, dietary quality during pregnancy, gestational weight gain, and parity; median neighborhood and individual household income at birth; and child age at early childhood visit, gestational age at birth, birth weight for gestational age z-score, breastfeeding duration in months, and race and ethnicity.

³Model 3 is additionally adjusted for early childhood sleep duration, television viewing, modified Youth Healthy Eating Index score, and intakes of total energy, sugar-sweetened beverages, fiber, juice, sucrose, and other dairy.

⁴Model 4 is additionally adjusted for early childhood BMI z-score and the change in BMI z-score between ages 2 and 3 years to account for the possibility of confounding by indication.

Consumption of whole or 2% milk in early childhood, compared with 1% or skim milk, was similarly associated with a trend toward lower adiposity in early adolescence (Figure 2A). These associations were significant for all adiposity outcomes after adjustment for all baseline parent factors and most baseline child factors in model 3; however, all 95% CIs crossed the null after adjustment for baseline child BMIz value and BMIz change between ages 2 and 3 in model 4. Similarly, we noted a trend towards associations of higher-fat milk consumption with more favorable cardiometabolic outcomes, but none of these associations were statistically significant (Figure 2B). The frequency of cow's milk consumption in early childhood was not associated with any particular trend in adiposity or cardiometabolic risk outcomes in early adolescence (Figure 2C and D), with the exception of an association with log leptin in models 2–4.

Figure 2.

Associations of cow's milk fat content (whole or 2% compared with 1% or skim) consumed in early childhood with (A) adiposity and (B) cardiometabolic outcomes in early adolescence using multivariable linear regression. Associations of frequency of cow's milk consumption in early childhood with (C) adiposity and (D) cardiometabolic outcomes in early adolescence using multivariable linear regression. Model 1 is adjusted for the other exposure of interest (i.e., milk fat content is adjusted for milk intake frequency, and milk intake frequency is adjusted for milk fat content), as well as for child age at the time of the exposure and for child sex and age at the time of the outcome measurement (for non-z-score outcomes). Model 2 is additionally adjusted for parental education and BMI (both assessed at enrollment); maternal age, marital status, dietary quality during pregnancy, gestational weight gain, and parity; median neighborhood and individual household incomes at birth; and child age at early childhood visit, gestational age at birth, birth weight for gestational age z-score, breastfeeding duration in months, and race and ethnicity. Model 3 is additionally adjusted for early childhood sleep duration, television viewing, modified Youth Healthy Eating Index score, and intakes of total energy, sugar-sweetened beverages, fiber, juice, sucrose, and other dairy. Model 4 is additionally adjusted for early childhood BMI z-score and the change in BMI z-score between ages 2 and 3 years to account for the possibility of confounding by indication. BIA, bio-impedance.

Sensitivity analyses

In our complete case analysis (Supplementary Figure 2), we found the same direction of effect estimates, with narrower CIs that less frequently crossed the null. In particular, after adjustment for all confounders in the complete case analysis, early childhood consumption of higher-fat milk compared with lower-fat milk was associated with lower early adolescent BMIz, waist circumference, sum of subscapular and triceps skinfold thicknesses, DXA fat mass index, DXA trunk fat mass index, DXA lean mass index, MetRiskz, and log leptin, and with higher log HDL cholesterol and log adiponectin. There was no association of early childhood milk intake frequency with adiposity or cardiometabolic outcomes in early adolescence.

To further explore the possibility of confounding by indication, we stratified our analysis according to the baseline weight status in early childhood: that is, normal weight (≥5^th to <85^th percentile; n = 529; Supplementary Figure 3; Supplementary Table 4) and overweight or obesity (BMI ≥ 85^th percentile; n = 189; Supplementary Figure 4; Supplementary Table 5). In the analysis restricted to those with normal weight, the effect sizes of the association between higher-fat cow's milk and early adolescent adiposity tended to be smaller than in the unstratified analysis, with all 95% CIs crossing the null. In the analysis restricted to those with overweight or obesity, we found a more pronounced association of higher-fat milk consumption in early childhood with lower adiposity, particularly waist circumference (ß, −4.18 cm; 95% CI, −7.73 to −0.62), along with DXA lean mass index (ß, −0.75 kg/m²; 95% CI, −1.36 to −0.14), as well as with more healthful cardiometabolic outcomes, particularly log HOMA-IR (ß, −0.27 units; 95% CI, −0.53 and −0.02) and log adiponectin (ß, 0.24 units; 95% CI, 0.05–0.44), in early adolescence.

To align with several recent studies, we repeated our analysis by comparing the consumption of whole milk with all other fat contents (skim, 1%, and 2%). This sensitivity analysis showed similar trends, albeit with generally smaller effect sizes and broader 95% CIs crossing the null for most early adolescent outcomes (Supplementary Figure 5; Supplementary Table 6). Adjustment for YHEI scores at approximately age 2 instead of in early childhood also yielded similar results (Supplementary Table 7).

Discussion

In this prospective cohort study, after controlling for multiple potential confounders, early childhood consumption of higher-fat cow's milk was associated with lower odds of overweight or obesity over a decade later. This finding may be at least partly driven by confounding by indication, given the strong association between early childhood BMIz and the fat content of milk consumed in early childhood and the significant attenuation in associations between higher-fat milk and decreased adiposity outcomes after adjustments for baseline BMIz and change in BMIz in the year preceding exposure ascertainment. Nevertheless, the direction of associations between higher-fat milk and later adiposity remained consistent across the subgroup analyses of children stratified by baseline weight status.

Importantly, despite the ongoing guidance of many professional organizations, we found no protective associations of consumption of lower-fat milk against future adiposity or cardiometabolic risk. Our findings thus do not support current recommendations to consume skim or 1% milk in early childhood to reduce the risk of overweight or obesity and adverse cardiometabolic outcomes in later life. Given the evidence of confounding by indication, however, our results in themselves are not sufficient to endorse a change in policy towards a recommendation to consume higher-fat milk, and instead support the absence of a guideline regarding milk fat content after age 2.

Recent studies have similarly challenged the recommendation that children transition to skim or 1% milk at age 2. Our results are consistent with a prospective cohort study with a mean duration of follow-up of 2.7 years (SD, 1.7 years) among children ages 9 months to 8 years, which found that each 1% increase in cow's milk fat consumed was associated with a 0.05 unit lower BMIz (95% CI, −0.07 to −0.03 SD units), and that those drinking whole compared to reduced-fat milk (3.25% compared with 0.1%–2% fat content) had lower odds of overweight (OR, 0.82; 95% CI, 0.68–1.00) at follow-up (51). A meta-analysis of 11 cross-sectional and 3 prospective cohort studies also found lower odds (crude OR, 0.61; 95% CI, 0.52–0.72) of overweight or obesity among children consuming whole compared with reduced-fat milk (3.25% compared with 0.1–2% fat content); however, only 1 included study had a low risk of bias, and the longest duration of follow-up was 3 years, which is insufficient to assess long-term effects (3). Our sensitivity analysis comparing the consumption of whole milk with all other milk fat contents (skim, 1%, and 2%) similarly did not show a protective effect of lower-fat milk.

The association of higher-fat cow's milk consumption with known obesogenic behaviors—such as higher SSB intake, fast food intake, and television watching, as observed in our study—underscores the importance of rigorous adjustment for confounders in assessing these associations, as families following dietary guidance for milk may be more likely to adhere to other health advice as well. Although higher-fat milk is more caloric and could plausibly lead to excessive adiposity via increased energy intake, it may also be associated with increased satiety, potentially reducing overall energy consumption or displacing low-nutrient-density food choices in the childhood diet (6). The high saturated-fat content in whole milk would suggest an adverse cardiometabolic effect; however, in their systematic review of 10 pediatric studies, O'Sullivan et al. (8) found no increased cardiometabolic risk associated with increased milk fat content, with 1 exception (52), similar to our results. The complex fatty composition of whole cow's milk may be less metabolically detrimental in vivo than previously believed (8).

Prior research has shown mixed results on the association of the frequency or volume of cow's milk intake with later adiposity (9–14) and cardiometabolic risk (15–20). Increased milk consumption may have differential effects on adiposity and cardiometabolic health depending on what, if anything, it may displace in the childhood diet (15), which may vary across individuals and populations. Although our observed association of milk intake frequency with leptin was statistically significant, we interpreted this result with caution, as there was not an overall trend in associations of higher milk consumption with worsened cardiometabolic outcomes, and the prior literature in children (8) and adults (53) has not previously identified an isolated impact on leptin.

Strengths of our study include the prospective data collection, long-term follow-up, and exposure ascertainment at an early age with a paucity of data in the literature. We also assessed milk exposure during early childhood, a crucial time at which milk represents a higher proportion of the total energy intake compared with later ages (24, 54). Additionally, age 3 years is the earliest period by which most families intending to follow dietary guidelines for milk would likely have transitioned to lower-fat milk; thus, milk consumed at age 3 likely represents milk consumption patterns throughout childhood. Other strengths include the assessment of many potential confounders; measurement of a wide range of body composition and cardiometabolic outcomes using standardized protocols in adolescence, a sensitive time for programming of adult adiposity and lifelong cardiometabolic health (55–57); and concurrent examination of the associations of milk fat content and intake frequency.

Limitations of our study include a well-educated, economically advantaged population from Eastern Massachusetts with access to health care, which may not generalize to the US population. Second, the differences between included and excluded children indicate a degree of selection bias. Relatedly, we noted a large proportion of missing data for DXA and cardiometabolic outcomes, which should be interpreted more cautiously than our anthropometric outcomes. We were thus judicious in using multiple imputation to account for missing data, with complete case and imputed data results reported separately. Third, we were unable to account for the contribution of flavored milk, as we collected data on total milk intake frequency that did not quantify the relative contributions of unflavored compared with flavored milk. We excluded participants self-reporting flavored milk as the predominant milk type consumed, but our data do not reflect that some children likely drank more flavored milk than others. Fourth, the FFQs did not have portion sizes, so we were limited to assessing the milk intake frequency rather than volume. This possible measurement error extends to multiple covariates, including energy consumption. As we were using FFQs to rank individuals relative to each other, however, any random error introduced would likely bias towards the null, such that our estimates would be conservative. We also did not include any measures of nutritional composition prior to age 3, except breastfeeding duration. Notably, however, our sensitivity analysis adjusting for dietary quality (i.e., modified YHEI scores) at age 2 instead of age 3 yielded results that were largely similar to our main findings (Supplementary Table 7). Finally, as the large variety of outcomes assessed increases the risk of false-positive results, we interpreted each result in the context of all associated outcomes.

In conclusion, in this prospective cohort study, early childhood consumption of whole or 2% milk, compared with 1% or skim milk, was associated with lower odds of overweight or obesity in early adolescence; however, this modest protective association may be due to residual confounding by indication. Contrary to current dietary guidelines, we did not find evidence that intake of lower-fat milk compared to higher-fat milk is protective against future adiposity or adverse cardiometabolic outcomes. Given the high frequency of early childhood milk consumption in the United States and the prevalences of overweight or obesity and their associated comorbidities, it is imperative to confirm these findings with future research, ideally with large randomized controlled trials. In the interim, it may be prudent for pediatric providers to focus counseling to prevent future overweight or obesity on more evidence-based strategies than the consumption of lower-fat milk.

Notes

The US NIH grants R01 HD034568 and UG3 OD023286 supported this work. JWB is supported by US NIH NIDDK grant K23DK115682.

This work is solely the responsibility of the authors and does not represent the official views of the NIH or any of the other funders.

Supplemental Figures 1–5 and Supplemental Tables 1–7 are available from the “Supplementary data” link in the online posting of the article and from the same link in the online table of contents at https://academic.oup.com/ajcn/.

Abbreviations used: AHEI-P, Alternative Healthy Eating Index for Pregnancy; BMIz, BMI z-score; SBP, systolic blood pressure; SSB, sugar-sweetened beverage; YHEI, Youth Healthy Eating Index

Data Availability

Pending application and approval, the data described in the manuscript and the associated code are available from Project Viva upon request.