Ultra-Processed Food Consumption and BMI-Z Among Children at Risk for Obesity from Low-Income Households

Abstract

Objective:

To evaluate the association between baseline ultra-processed food consumption in early childhood and child BMI Z-score over 36 months.

Methods:

We conducted a prospective cohort analysis as a secondary data analysis of the Growing Right Onto Wellness randomized trial. Dietary intake was measured via 24-hour diet recalls. The primary outcome was child BMI-Z, measured at baseline, 3-, 9-, 12-, 24-, and 36-month timepoints. Child BMI-Z was modeled using a longitudinal mixed-effects model, adjusting for covariates and stratifying by age.

Results:

Among 595 children, median (Q1-Q3) baseline age was 4.3 (3.6–5.0) years, 52.3% children were female, 65.4% had normal weight, 33.8% had overweight, 0.8% had obesity, and 91.3% of parents identified as Hispanic. Model-based estimates suggest that, compared with low ultra-processed consumption (300 kcals/day), high ultra-processed intake (1300 kcals/day) was associated with a 1.2 higher BMI-Z at 36 months for 3-year-olds (95% CI=0.5,1.9; p<0.001) and a 0.6 higher BMI-Z for 4-year-olds (95% CI=0.2,1.0; p=0.007). The difference was not statistically significant for 5-year-olds or overall.

Conclusions:

In 3- and 4-year-old children, but not 5-year-old children, high ultra-processed food intake at baseline was significantly associated with higher BMI-Z at 36-month follow-up, adjusting for total daily kcals. This suggests that it might not be only the total number of calories in a child’s daily intake that influences child weight status, but also the number of calories from ultra-processed foods.

Background:

Childhood obesity disparities are present when children enter kindergarten in the United States.^1–3 Among preschool-aged children, recent estimates from the National Health and Nutrition Examination Survey (NHANES) indicate that the prevalence of obesity among Hispanic/Latino children is 21.1% compared to 12.1% of non-Hispanic Black children and 10.8% of non-Hispanic white children.⁴ There is a broad understanding of the multi-level determinants of these health disparities, that include a confluence of risk factors at the individual, family, and community levels.^5–7 However, these known risk factors are limited in their ability to accurately predict future obesity risk, suggesting the need for novel ways to characterize modifiable intervention targets in early childhood.⁸ For example, in a recent cohort analysis, existing measures of child diet quality (i.e., Healthy Eating Index)⁹ did not predict the emergence of childhood obesity. Rather, the main predictors were child age, parent BMI, and child overweight.¹⁰ Thus, it is unclear what the main dietary intervention targets for preventing childhood obesity should be.

Recent dietary guidelines have shifted away from recommending specific macronutrients (i.e., how much fat to eat/day) but rather focus on overall dietary patterns.¹¹ As a part of that paradigm, foods can be classified based on the degree to which they are processed, rather than macronutrient content.^12,13 For example, unprocessed foods include foods like eggs, nuts, and natural fruits, vegetables and potatoes whereas ultra-processed foods include foods like packaged snacks, ice cream, sweetened beverages, packaged meats, and foods that include hydrogenated vegetable fat. One relatively novel approach to understanding diet quality that may explain the emergence of childhood obesity is the concept of ultra-processed foods.¹⁴ Among adults, consumption of ultra-processed foods has been associated with increased risk of hyperlipidemia,¹⁵ hypertension,¹⁶ and cancer.¹⁷ In addition, a recent study in an energy balance lab suggests that consumption of ultra-processed foods is correlated with weight gain among adults.¹⁸ Similarly, among children, a recent analysis using nationally representative cross-sectional data identified associations between higher ultra-processed food intake and calorically dense, nutritionally poor diet quality.¹⁹

Despite these recent findings, associations between ultra-processed food consumption and childhood obesity remain poorly characterized, especially among younger children from low-income households and racial and ethnic minority groups.²⁰⁻²² Because communities in poverty often have limited access and economic constraints that prevent regular consumption of whole foods, the higher rates of ultra-processed food consumption among low-income minority populations may partially explain health disparities.²³ In addition, ultra-processed foods make up a disproportionate percentage (67%) of daily caloric intake among children in the United States.²⁴ Because dietary practices are established in early childhood, as children develop lasting taste preferences and habits, strategies to appropriately measure disease risk attributable to ultra-processed foods may have important implications for reducing health disparities.⁵

The purpose of this study was to characterize the association between ultra-processed food consumption and child weight status (i.e., BMI Z-score) among preschool-aged children from predominantly low-income households. We hypothesized that higher daily consumption of calories from ultra-processed foods at the baseline timepoint would be associated with less healthy BMI-Z growth over three years of follow-up and that this association would be stronger for children who were younger at baseline.

Methods:

This study was a cohort analysis of previously collected data from a randomized controlled trial (RCT) to evaluate the association between baseline ultra-processed food consumption and child BMI-Z over time. This secondary analysis was conducted on data from the Growing Right Onto Wellness (GROW) Trial (clinicaltrials.gov: NCT01316653), which tested an obesity prevention intervention.²⁵ The intervention did not achieve its primary outcome of preventing childhood obesity, despite children in the intervention group reportedly consuming an average of 100 kcal/day fewer than the control group at the final timepoint.²⁶ The trial was part of the Childhood Obesity Prevention and Treatment Research Consortium, supported by a coordinating center at the University of North Carolina, Chapel Hill.²⁷ The Vanderbilt University Medical Center Institutional Review Board approved the GROW trial, and written informed consent was obtained by bilingual (Spanish- and English-speaking) data collectors in participants’ language of choice.

For the original GROW trial, 610 parent-child pairs were recruited from multiple local settings, including medical practices and community centers, in Nashville, TN. Parent-child pairs were enrolled between August 2012 and May 2014. The final 36-month timepoint occurred between October 2015 and June 2017. Eligibility criteria included child age 3 to 5 years and high normal weight to overweight (BMI ≥50^th and <95^th percentile based on standardized growth curves from the U.S. Centers for Disease Control and Prevention), though five children did have obesity at baseline.²⁸ Caregiver eligibility included being English or Spanish speaking, consistent telephone access, and a commitment to participate in the study. Households also had to qualify for at least one service for underserved populations (e.g., Medicaid, Special Supplemental Nutrition Program for Women, Infants, and Children). Parent-child pairs were excluded if a medical condition precluded regular physical activity or if they lived/worked outside of a 5-mile radius of participating community centers.

The primary exposure variable for the current analysis was the mean number of daily calories consumed from ultra-processed foods at baseline (i.e., prior to randomization in the original trial). That variable was calculated from 24-hour diet recall data collected using Nutrition Data System for Research (NDSR) software as part of the original GROW trial protocol. Parents were asked to complete three 24-hour recalls at each visit with trained and certified research assistants (two weekdays and one weekend day). The unique individual foods consumed were then coded according to their level of processing,²⁹ based on the Nova classification system.^12,30 A minimum of two recalls were required for analysis. Recalls were collected over the telephone in English or Spanish. Recalls were not conducted on consecutive days, and the third recall was collected more than one week after the first recall. The goal was to collect all recalls within 45 days, but recalls collected outside that timeframe were still included for analysis. Quality assurance checks were conducted on at least 10% of the dietary recalls as specified by the NDSR standard protocol. Absolute calories from ultra-processed foods was chosen as the primary exposure rather than percent of calories because it allowed for the same interpretation for all children as opposed to the percent of calories from UPF which means different things depending on the total number of calories consumed in any day.

The primary outcome variable for this analysis was child BMI Z-score as calculated from objective child height and weight measurements that were prospectively obtained during the original GROW trial. Child height and weight were measured by trained and certified data collectors at six times throughout the study: baseline, 3, 9, 12, 24, and 36 months. Height was measured to the nearest 0.1 cm and weight was measured to the nearest 0.1 kg. All measurements were taken twice, and the mean was used as the final value. If the difference between the two measurements exceeded a predefined threshold (0.3 kg or 0.5 cm), a third measurement was taken and the mean of the two closest measures was used as the final value. Quality control measures were obtained on >10% of height and weight measurements, where a second data collector repeated the same set of measurements and accuracy was compared. Height and weight were also cleaned before analysis using both cross-sectional and longitudinal data checks to flag potential data entry errors for review. Reasons for flagging included exceeding 1.5 times the interquartile range or decreased height between timepoints.

Several demographic, behavioral, and psychosocial variables were reported by parents in the study via surveys that were administered by study staff, bilingual in English and Spanish. In addition, child physical activity patterns were measured by accelerometry. Parents helped their children wear an ActiGraph GT3X+ triaxial accelerometer (Actigraph, Pensacola,FL). Children were asked to wear these accelerometers on the waist for at least seven days. To be included, a minimum wear time of 4 days (3 weekdays and 1 weekend day), each consisting of at least 6 hours between 5 am and midnight, was required, although mean wear time far exceeded this requirement (17 hours out of a 19-hour maximum). Records were collected at 40 Hz and integrated into 15-s epochs. Cut-points to categorize child movement into minutes of moderate-vigorous physical activity were based on previously published algorithms.³¹

Statistical Analysis

Participant characteristics were summarized using descriptive statistics, including frequency (%), and median (1^st quartile–3^rd quartile). We present these summary statistics for the total sample and split at the median of the daily ultra-processed food consumption for illustrative purposes. The distribution of child baseline ultra-processed foods consumption (kcals) was presented graphically by plotting kernel density estimates based on the Epanechnikov kernel function.³² To evaluate how child baseline consumption of ultra-processed foods was associated with child BMI-Z score over 36 months of follow-up, this study used a longitudinal mixed-effects linear regression model with two levels (time nested within child), and a random intercept parameter to allow individual variability. A maximum likelihood procedure was used to handle missing data. Time was analyzed as a discrete timepoint variable. Baseline mean daily calories consumed from ultra-processed foods (kcal) was the primary variable of interest. Other baseline covariates in the model were child mean daily total calories (kcal), age (years), sex (male vs. female), mean daily percent of time spent in moderate-vigorous physical activity, parent ethnicity (non-Hispanic vs. Hispanic), household food security (secure vs. insecure),³³ household participation in federal nutrition assistance programs (none vs. WIC and/or SNAP), and random assignment in the original RCT (control or intervention). Importantly, the model included two three-way interactions: $\left(timepoint\right)\times \left({childage}_{baseline}\right)\times \left({totalcalories}_{baseline}\right)$ and $\left(timepoint\right)\times \left({childage}_{baseline}\right)\times \left({ultra-processedcalories}_{baseline}\right)$ to allow estimates of the magnitude of the relationship between ultra-processed calories and BMI-Z to vary over time and to differ depending on baseline child age, while also ensuring that these estimates were adjusted for mean daily caloric intake. Because the regression coefficients were not intuitive to interpret due to the complexity of the model, key results have also been presented as model-based estimates of BMI-Z at each timepoint with 95% confidence intervals. Model-based results have been presented overall and stratified by child baseline age, and, to facilitate interpretation, values were selected to represent low and high ultra-processed intake (300 kcal/day and 1300 kcal/day, respectively).

All analyses were conducted with Stata version 17.0 (StataCorp).³⁴ Statistical significance was defined by a two-sided p-value less than 0.05.

Results:

Of the 610 participants in the original RCT, 595 children (97.5%) with sufficient baseline data were analyzed. Median (1^st quartile–3^rd quartile) baseline age was 4.3 (3.6–5.0) years, 52.3% children were female, and 91.3% of parents identified as Hispanic. Table 1 shows additional characteristics of the included sample stratified by median ultra-processed foods consumption and overall. Baseline median (1^st quartile–3^rd quartile) total daily calories were 1160 (934–1400) kcals and baseline median daily ultra-processed foods calories were 698 (537–914) kcals. Table 2 shows the summary statistics for mean daily total calories, calories from ultra-processed foods, and percent ultra-processed food consumption by child age group. The distribution of child mean daily percent of calories from ultra-processed foods at the baseline timepoint appeared to be similar across different age groups, as shown in Figure 1.

Table 1:

Baseline characteristics, stratified by the median percent of ultra-processed foods and overall.

	Ultra-Processed Food Consumption ≤ 50th percentile	Ultra-Processed Food Consumption > 50th percentile	Total
	N=298	N=297	N=595
Child characteristics
Age (years)	4.0 (3.4–4.8)	4.4 (3.6–5.2)	4.3 (3.6–5.0)
Sex
Male	131 (44.0%)	153 (51.5%)	284 (47.7%)
Female	167 (56.0%)	144 (48.5%)	311 (52.3%)
Daily percent of time in MVPA	8.4 (6.7–10.4)	8.2 (6.3–10.3)	8.3 (6.4–10.3)
BMI-Z	0.8 (0.5–1.2)	0.8 (0.4–1.2)	0.8 (0.5–1.2)
BMI category
Normal	190 (63.8%)	199 (67.0%)	389 (65.4%)
Overweight	105 (35.2%)	96 (32.3%)	201 (33.8%)
Obese	3 (1.0%)	2 (0.7%)	5 (0.8%)
Family characteristics
Ethnicity
Non-Hispanic	5 (1.7%)	47 (15.8%)	52 (8.7%)
Hispanic	293 (98.3%)	250 (84.2%)	543 (91.3%)
Food security status
Food secure	172 (57.7%)	170 (57.2%)	342 (57.5%)
Food insecure	126 (42.3%)	127 (42.8%)	253 (42.5%)
Use of WIC and/or SNAP
No	37 (12.4%)	36 (12.1%)	73 (12.3%)
Yes	261 (87.6%)	261 (87.9%)	522 (87.7%)
Random assignment
Control	145 (48.7%)	152 (51.2%)	297 (49.9%)
Intervention	153 (51.3%)	145 (48.8%)	298 (50.1%)

¹Results are presented as median (1^st quartile–3^rd quartile), or frequency (%). Ultra-processed food consumption has been stratified by the median for presentation in this table but was analyzed as a continuous variable. MVPA: Moderate-to-vigorous physical activity; WIC: Special Supplemental Nutrition Program for Women, Infants, and Children; SNAP: Supplemental Nutrition Assistance Program.

Table 2:

Baseline ultra-processed food intake by child age group

	Child Age Group1
	≥3 years & <4 years	≥4 years & <5 years	≥5 years	Overall
	N=254	N=191	N=150	N=595
Daily total calories	1100 (896–1321)	1172 (952–1395)	1256 (1041–1474)	1160 (934–1400)
Daily calories from ultra-processed foods	639 (489–822)	728 (550–923)	781 (587–1018)	698 (537–914)
Daily percent of calories from ultra-processed foods	61.4 (50.1–69.8)	62.7 (55.4–71.7)	64.3 (55.2–72.2)	62.5 (53.2–71.0)

¹Child age has been categorized into groups for presentation in this table but was analyzed as a continuous variable. Results are presented as median (1^st quartile–3^rd quartile).

Figure 1: Kernel density estimates of baseline mean daily percent of calories from ultra-processed foods, stratified by baseline age (≥3 years & <4 years; ≥4 years & <5years; ≥5 years).

The kernel density was estimated using the Epanechnikov kernel with a bandwidth of 4.04. Child age has been categorized into groups for plotting in this figure but was analyzed as a continuous variable.

Model-based estimates from the longitudinal mixed-effects linear regression analysis indicated that higher baseline mean daily ultra-processed caloric intake was statistically significantly associated with higher BMI-Z at later timepoints for 3- and 4-year-olds but not 5-year-olds. The full model output is contained in Appendix 1 and, to facilitate interpretation, model-based estimates are shown in Figure 2 using representative values for low and high ultra-processed intake (300 kcal/day and 1300 kcal/day, respectively). These model-based estimates suggest that compared with low ultra-processed intake, high ultra-processed intake was associated with a 1.2 higher BMI-Z at 36 months for 3-year-olds (95% CI=[0.5, 1.9]; p<0.001) and a 0.6 higher BMI-Z for 4-year-olds (95% CI=[0.2, 1.0]; p=0.007). The difference at 36 months was not statistically significant for 5-year-olds (−0.1; 95% CI=[−0.6, 0.4]; p=0.7) or overall (0.4; 95% CI=[−0.02, 0.7]; p=0.07).

Figure 2: Model-based estimates of BMI-Z for low and high ultra-processed intakes, overall and stratified by child age.

Plot represents model-based estimates of child BMI-Z for selected representative low and high values of baseline mean child kcals from ultra-processed foods (300 and 1300 kcals, respectively) and for selected child ages. Estimates are from a longitudinal mixed-effects linear regression model, adjusting for baseline child mean daily kcals from ultra-processed foods, mean daily total kcals, age, sex, and mean daily percent of time spent in moderate-vigorous physical activity; parent ethnicity, household food security, WIC/SNAP participation, and random assignment in the original RCT, and with additional interactions involving timepoint and child age. Estimates for selected child ages have been plotted in this figure, but age was analyzed as a continuous variable.

Discussion

In this prospective cohort analysis of children from predominantly low-income households, higher ultra-processed food consumption was significantly associated with higher child BMI-Z over 3 years of follow-up among 3- and 4-year-olds, but the association was not statistically significant for 5-year-olds, or in the overall model. It is important to note that only five of the children in this study had obesity at baseline, allowing for a clear temporal relationship between the ultra-processed food consumption and the emergence of unhealthy BMI Z-scores. These associations were statistically significant even after controlling for the mean daily total number of calories that children consumed. This suggests that it might not be only the total number of calories in a child’s daily intake that drives child weight status, but also the number of calories of ultra-processed foods. In addition, the mean percentage of child daily calories consumed from ultra-processed foods was >60% in this sample, which is consistent with previous reports.²⁴ Taken together, the observations that consumption of ultra-processed foods is both highly prevalent and associated with weight-related child outcomes suggests that ultra-processed foods may be an important intervention target for future childhood nutrition and obesity interventions.

The current study advances the field of child nutrition research by showing a prospective association between ultra-processed foods using the Nova classification and early childhood BMI-Z over 3 years of follow-up. Previous studies have consistently demonstrated associations between certain types of ultra-processed foods (e.g., sweetened beverages, sweets, etc.) and body fatness in childhood.³⁵⁻³⁷ The current study advances this finding by analyzing a unique dataset where every food children reportedly consumed in 24-hour diet recalls had been comprehensively classified into the Nova scheme, allowing for a high-precision evaluation of the number of calories consumed in a day and their level of processing. In addition, the current study advances the field by focusing on the period of early childhood, when disparities in childhood obesity are known to emerge. This adds to the previous literature that has consistently shown associations between ultra-processed food consumption and a range of adult health outcomes, including adult obesity, cardiometabolic risk, type 2 diabetes and even all-cause mortality.¹³ Some of the limitations of these previous studies have included inadequate control for confounding, which the current study is able to address by adjusting for several key covariates, including an objective measurement of child physical activity.³⁸

One notable finding in the current analysis was a statistically significant interaction between ultra-processed food consumption, time, and child age such that higher baseline ultra-processed food consumption for 3- and 4-year-olds was associated with significantly higher BMI-Z over time. The reason for this is unclear. It may be related to a measurement bias, stemming from parents having less detail about what their children eat as they enter school. Or it may be that the importance of ultra-processed food consumption to child weight status is more relevant in early childhood. This finding should be confirmed in other study populations and with methods that can measure school-based dietary intakes more precisely.

One of the main advantages of using the paradigm of ultra-processed foods for child nutrition interventions is the relative simplicity for families.³⁹ It is easier to teach families how to choose minimally processed whole foods than to teach calorie counting or recognition of specific macronutrient profiles (e.g., low fat, low carbohydrate). This may be especially true for families with limited health literacy or numeracy skills.⁴⁰ In addition, a healthy diet pattern that focuses largely on minimally processed and nutrient-dense (e.g., higher quantities of nutrients and minerals) whole foods is consistent with current US Dietary Guideline recommendations.⁴¹ Therefore, one priority of future research should be to test in a randomized trial whether a whole-foods approach to child nutrition could improve child obesity prevention efforts, many of which have met with limited success in predominantly low-income households.

This study had several limitations. Measurement of the primary exposure variable was obtained through 24-hour diet recall. While this methodology has been validated against more rigorous diet measurements, it can be subject to bias. In the context of this study, parents of children with higher weight status may have experienced a stronger social desirability bias, although this may have been mitigated by the ineligibility of children with very high BMI percentiles during recruitment. One area of future research would be to use more objective measures of dietary assessment, especially in school-aged children. As an observational study design, the results are subject to residual confounding and should not be interpreted as causal. Only the association between baseline measurements of child diet and longitudinal weight follow-up was evaluated. The main dataset included subsequent measures of child diet, but, due to issues related to the timing of those measurements and sample size, ultra-processed food consumption was not evaluated as a time-varying exposure. It will be useful for future investigations to describe how consistent measures of ultra-processed consumption are over time and to discover if any characteristics or contextual variables might be related to increased or decreased consumption. Finally, the population included in this study predominantly identified as Hispanic and/or Latino. While this is an important population with high rates of childhood obesity, the results from the current analysis may not be broadly generalizable, and the findings should be confirmed in other populations.

In conclusion, higher mean daily ultra-processed food consumption at baseline, controlling for mean daily total calories and other sociodemographic characteristics, was associated with higher BMI-Z over 3 years of follow-up among 3- and 4-year-old children. This association was not statistically significant among 5-year-old children or in the overall model. Ultra-processed foods may be an important intervention target for future childhood nutrition and obesity interventions.