A Health-Literacy Intervention for Early Childhood Obesity Prevention: A Cluster-Randomized Controlled Trial

As described previously, we implemented a pragmatic, cluster-randomized controlled trial design to assess the impact of the Greenlight intervention on the prevalence of overweight at age 24 months and on interval weight trajectories.⁵²  We chose resident physicians as primary intervention messengers, because of the potential for training to impact long-term practice and the aim to target low-income families, since at least 1 in 5 low-income children receives care in resident clinics.⁵³  To avoid intrasite contamination, because residents could not be easily cohorted within each site, randomization occurred at the site level. We stratified by population density, such that the 2 sites serving the highest population density were assigned to different groups. A statistician, blinded to site location, conducted each site’s random assignment to intervention or active control status, using a random number generator in Stata 9.0 (Stata Corp, College Park, TX). Two sites (New York University [NYU] and Vanderbilt University Medical Center [VUMC]) were randomly assigned to receive the intervention, which incorporated health-literacy principles and focused on obesity prevention, and 2 sites (University of Miami [UM] and University of North Carolina at Chapel Hill [UNC]) were assigned to receive the active control arm, which did not apply health-literacy principles and focused on injury prevention. The study was approved by the institutional review boards at each of the participating university medical centers. A Data Safety Monitoring Board, which included participants from each institution, provided study oversight of protocol implementation, including monitoring of recruitment, retention, and outcomes. The study is registered with the national Clinical Trials Registry (clinicaltrials.gov), study no. NCT01040897.

The Greenlight intervention consisted of 2 components: (1) a low-literacy parent educational toolkit, including developmentally-tailored booklets at each well-child visit [WCV]; and (2) provider training in health communication, including modules on teach-back and goal setting.⁵²  Booklets were designed with low-literacy principles to target age-specific behavioral goals, with a focus on child nutrition (eg, promoting breastfeeding, avoiding sweetened beverages, recognizing satiety cues, using appropriate portion sizes) and physical activity (eg, promoting tummy time, avoiding screen time).^{34,52,54–56}  Tangible tools, which reinforced these goals, included an infant onesie (reading “I’m Sweet Enough. Please, No Juice!”) at 2-month WCV; a sippy cup (with markings to guide juice dilution) at 9-month WCV; a small snack bowl (for pureed foods) at 12-month WCV; and a placemat (illustrating appropriate portion sizes) at 15- to 18-month WCV. Parents from a variety of racial and ethnic backgrounds were consulted to inform toolkit content and design. (Materials accessible at www.greenlight-program.org.) Provider training modules emphasized clear-health communication techniques, including teach-back, goal-setting, and plain language, with specific examples to support toolkit content (eg, avoiding sweetened beverages, recognizing satiety cues).⁵²  Sites randomized to the control arm received The Injury Prevention Program,^57,58  an educational intervention designed by the American Academy of Pediatrics. In each arm, providers received the same number of training hours, parents received the same number of tangible tools, and all materials were provided in English or Spanish. In line with standard practice, in-person or telephone interpreters were provided for those who preferred to conduct the visit in Spanish. Before study initiation, intervention fidelity was reinforced through administration of a 10-item checklist, adapted from previously validated tools,^59–62  applied to each provider at a WCV encounter by a trained observer. Failure to exceed a preset threshold prompted brief feedback and subsequent observation. Before study initiation, the threshold score was exceeded for 95% of initial observations; of those who required subsequent observation, the threshold score was achieved on the second or third observations; none required a fourth. During the study period, 590 families (68% of those randomizly assigned, 74% of those analyzed) completed at least 6 WCVs.

Between April 2010 and October 2014, we enrolled the child’s primary caregiver (“parent”) at the child’s 2-month WCV, with intervention and assessment at each of 7 American Academy of Pediatrics–recommended preventive care visits (2, 4, 6, 9, 12, 15 and/or 18, and 24 months). Each parent provided informed consent before study enrollment.

Eligibility criteria included a child presenting for a 2-month WCV, parent’s ability to speak Spanish or English, and no specific plans to leave the clinic within 2 years. Infant exclusion criteria included gestational age <34 weeks, birth weight <1500 g, enrollment weight less than third percentile on World Health Organization (WHO) growth curve,^63,64  or chronic medical problem causing potential problems with feeding or weight gain. Parent exclusion criteria included age <18 years, serious mental or neurologic illness, or poor corrected visual acuity.

Trained, bilingual research assistants conducted interviews in English or Spanish, on the basis of parent preference, at baseline and at each successive WCV. Acceptable timing for each WCV was 6 to 15 weeks (for 2-month WCV), 16 to 20 weeks (for 4-month WCV), 21 to 32 weeks (for 6-month WCV), 33 to 51 weeks (for 9-month WCV), 52 to 57 weeks (for 12-month WCV), 58 to 67 weeks (for 15-month WCV), 68 to 91 weeks (for 18-month WCV), and 92 to 125 weeks (for 24-month WCV). For missed WCVs, parents were contacted by telephone. Weight and length measurements were collected by clinic staff, who were trained on the basis of guidelines from the Agency for Healthcare Research and Quality. Chart abstraction was completed after the 24-month WCV.

The primary outcome was the proportion of children overweight at 24 months. We defined “overweight” as a BMI ≥85th percentile (adjusting for age and sex) according to WHO growth curves.⁶³  At each WCV, child weight status was assessed as a continuous variable (BMI z score) and as a dichotomous variable (BMI ≥85th percentile), each by WHO standards. Although we originally intended to use the Centers for Disease Control and Prevention growth curves, we discovered that many children presented to their 24-month WCV before the chronological age of 24 months, when Centers for Disease Control and Prevention z scores are not available. We therefore used WHO BMI z scores, which are available from age 2 months onward.

Parent-reported measures included child race and ethnicity, Supplemental Nutrition Program for Women, Infants, and Children (WIC) status, annual household income, and caregiver age, health literacy, reading language, and height and weight. Parent health-literacy skills were assessed with the Short Test of Functional Health Literacy in Adults^65,66  and the Pediatric Health Literacy Assessment Test.³³ 

We summarized baseline characteristics by site (UM, NYU, UNC, VUMC) and treatment arm (intervention, control), using percentiles for continuous variables and proportions for categorical variables. Participants were included in the analysis if they had at least 1 episode of follow-up outcome data at 1 of the scheduled clinic visits after randomization. Of the 865 patients randomized at baseline, 802 had at least 1 follow-up visit and remained eligible for analyses. Among those included, we had follow-up data from 78% of all possible WCVs.

We fit repeated measures, marginalized logistic regression models^67–69  for the binary outcome (BMI ≥85th percentile), and linear mixed models (random intercepts and slopes) for the continuous outcome (BMI z score) to examine site and intervention effects for children from 4 to 24 months of age. For site-specific comparisons, we chose the larger of the 2 control sites (UNC) as referent. To acknowledge clustering by site and to derive the intervention effects, we included in our regression models fixed effects for site, flexible functions of age at each WCV (restricted cubic splines with 2 degrees of freedom), site-by-age interactions, and potential confounders. The treatment-effect estimates were derived from the site-effect estimates. That is, we first estimated site effects over time, and using contrast matrices, we took the difference between the average trajectory in the intervention sites and the average trajectory in the control sites. We used linear contrasts of the site-by-age interactions to ascertain the intervention effect across all ages by subtracting the adjusted average trajectory of the 2 control sites from the adjusted average trajectory of the 2 intervention sites. To adjust for potential confounding due to the small number of sites, we controlled for the following prespecified covariates in each model: baseline child characteristics (sex, race, ethnicity, and the baseline value of the model outcome; ie, BMI ≥85th percentile or BMI z-score at 2-month WCV), baseline caregiver characteristics (age, primary language, annual household income, WIC status, health literacy, and BMI), and the interaction between child age and baseline value of the outcome. The test for treatment-arm effect was based on a Wald test derived from the linear contrasts by using a 2-sided, 0.05 significance level. All repeated measures analyses were clustered by child, rather than by physician, because response dependence was high within children and there was no evidence in the linear mixed model that adding a physician random effect improved model fit. To address missing follow-up data, we conducted multiple imputation with chained equations using predictive mean matching after transforming the data from long to wide format and fitting flexible imputation models.⁷⁰  We combined estimates from the 25 imputations using Rubin’s rule.⁷¹  As a sensitivity analysis, we performed the same analyses, using “weight-for-length z score” as the outcome, which yielded similar results to the BMI z score analysis. In post hoc analysis, we also added the food-insecurity measure to the model, which yielded no change in the results. All analyses were conducted by using R version 3.3.0.⁶⁷ 

We recruited 865 parent–child dyads, 459 at intervention sites and 406 at control sites. Of the 802 in the final analysis, 540 (67%) completed a 24-month WCV. At baseline, children at intervention sites were more likely to be identified as Hispanic (56% vs 43%). Parents at intervention sites were more likely to report, at baseline, a lower age (average 26 vs 28 years), Spanish language as the primary language (39% vs 31%), less than a high school education (29% vs 23%), annual family income <$10 000 (32% vs 30%), and a lower mean BMI (27 vs 28) (See Table 1). At the 24-month WCV, the prevalence of child overweight (BMI ≥85th percentile) was 49% (49% for intervention, 41% for control), and median BMI z score was 1.05 (1.01 for intervention group, 0.78 for control).

For the primary outcome, we found no significant difference between the intervention and control groups in the proportion of children overweight at the 24-month WCV, with adjusted odds (intervention versus control) of 1.02 (95% confidence interval [CI]: 0.63 to 1.64). Overall, there was a nonsignificant trend toward lower odds of overweight in the intervention group, compared with the control group, after adjusting for baseline covariates (P = .057) (Fig 1, top left panel.).

For the secondary outcome, we found a lower BMI z score in the intervention group compared with the control group through 15 months, marginally lower at 18 months, and no effect at 24 months. Adjusted mean differences (intervention minus control) were −0.04 (95% CI: −0.07 to −0.01), −0.09 (95% CI: −0.14 to −0.03), −0.19 (−0.33 to −0.05), −0.20 (−0.36 to −0.03), −0.16 (95% CI: −0.34 to 0.01), and 0.00 (95% CI −0.21 to 0.21) at 4, 6, 12, 15, 18, and 24 months, respectively. Overall, intervention-site growth trajectories differed from control-site trajectories (P = .002) (Fig 1, bottom left panel.).

To provide deeper insight into the observed intervention effect, we examined child growth trajectories at the site level, with the larger of the 2 control sites (UNC) as referent (Fig 1, bottom right). Overall, children at NYU, VUMC, and UM tended to have lower adjusted BMI z scores when compared with UNC. At 1 intervention site (VUMC), the weight trend decreased over time, with adjusted mean differences of −0.10 (95% CI −0.19 to −0.04) at 6 months, −0.26 (95% CI −0.44 to −0.09) at 12 months, −0.34 (95% CI −0.57 to −0.12) at 18 months, and −0.34 (95% CI −0.60 to −0.07) at 24 months. At the other intervention site (NYU), the weight trend exhibited a “U-shaped” trajectory over time, with adjusted mean differences of −0.16 (95% CI −0.23 to −0.08) at 6 months, −0.35 (95% CI −0.53 to −0.18) at 12 months, −0.34 (95% CI −0.56 to −0.12) at 18 months, and −0.11 (95% CI −0.36 to +0.16) at 24 months. Interestingly, the other control site (UM) exhibited a trend similar to the VUMC site. As a sensitivity analysis, we used “weight-for-length z score” as the outcome and found similar results.

The Greenlight intervention had no impact on the primary outcome, the prevalence of overweight at age 24 months. It did, however, result in differences in earlier childhood weight trajectories that may be clinically meaningful and may inform future interventions. In this study, we introduce new evidence about the effectiveness of clinic-based counseling and literacy-sensitive toolkits during infancy to prevent overweight and obesity.^48,72  In addition to addressing an important early childhood precursor of adult health, the Greenlight intervention also provides a model for a health-literacy-informed approach to health behavior change as part of routine primary care.

The Greenlight intervention resulted in less weight gain in the first 18 months of life, which was no longer present by age 24 months. At 12 months, mean BMI z score difference between the intervention and control groups was clinically significant (−0.19), which represents an approximate mean difference of 0.33 kg/m². One explanation is that the intervention “dose” diminished after 12 months, because of the decreased frequency of WCVs, from 5 during the first year to 2 during the second year. Also, the intervention’s onset at the 2-month WCV may have been “too late,” because many parent behaviors associated with child obesity prevention (eg, breastfeeding, recognizing satiety cues, sleep hygiene, screen time) may be established before the 2-month WCV.⁷³  Other factors may have moderated intervention effect, such as the overall prevalence of child overweight at 24 months (49%), which is higher than most national comparisons. Finally, the Greenlight intervention may have failed to influence some caregiver behaviors,^38,42–51  particularly given limited provider time at each WCV, or failed to target other potentially important behaviors, such as infant sleep.⁵¹ 

These findings are subject to limitations that are common to large, pragmatic clinical trials. First, although we randomized by site and adjusted for individual and site characteristics, the observed treatment effects may have been due to site-level confounding that we could not control for in regression analyses. For example, 1 intervention site (VUMC) is located in Tennessee, one of few US states to see an increase in childhood obesity rates during the study period.⁷⁴  Second, despite adjustment for missing data, a somewhat higher attrition rate in the control sites may limit interpretation of treatment effect.⁷¹  Third, the sample included predominantly low-income families, limiting generalizability to higher-income populations. Finally, the intervention period (2–24 months) limits our ability to judge the intervention’s sustained effect into the preschool years, when adiposity rebound occurs and when the effects of obesity prevention may be more clinically meaningful.¹⁰  To ascertain this long-term impact, we continued to follow study participants through age 5 years.

Still, the Greenlight Intervention Study provides unique insights for child obesity prevention and health promotion, particularly for low-income communities. These findings, along with those from other clinical trials,^75–77  support the potential positive impact of primary care interventions to prevent rapid early childhood weight gain. In addition to the reported findings on infant growth trajectories, the Greenlight Intervention Study has provided important information on the early childhood growth trajectories, as well as health behaviors, in low-income and historically underrepresented minority communities.^48,72  Conducted across 4 different geographic regions, the study enrolled an often underrepresented group: low-income families, with a variety of English-proficiency and literacy skills. Observational studies of this cohort have provided context for the role that family factors (eg, ethnicity, language, acculturation, locus of control, food insecurity) may play in the early adoption of behaviors that may protect infants from rapid weight gain.^73,78,79  Because we found that the overweight prevalence in both groups increased from 12 to 24 months, additional modalities, including asynchronous behavioral supports, may be necessary to achieve public health effectiveness. In ongoing and future analyses, we plan to apply predictive modeling, to explore which behaviors are associated with healthy child weight, and to examine family characteristics that may moderate that relationship.

The Greenlight intervention had no impact on the prevalence of overweight at age 24 months, but it did result in less weight gain during the first 18 months of life. Clinic-based interventions may be beneficial for early weight gain, but greater intervention intensity may be needed to maintain positive weight benefits over time. Pediatric primary care may be an important setting for interventions to prevent childhood overweight, especially given the importance of early rapid weight gain for later child obesity and health risks.^2,9,80–83  Interventions like Greenlight should be designed to fit real-world time and staffing constraints of pediatric primary care, as well as the social determinants of child health, including food insecurity. With the findings from the Greenlight study, we suggest the need for further research, to inform and improve obesity prevention and health promotion for young children.

We thank the Greenlight Study Team: Duke University: Asheley Skinner, for intellectual work related to study design, implementation, and analysis; Joanne Finkle, for intellectual contributions related to study implementation and measurement; NYU School of Medicine/Bellevue Hospital Center: Alan Mendelsohn, for intellectual work related to study design and implementation; Benard Dreyer, for initial intellectual work related to study design and implementation and toolkit design; Linda van Schaick, for intellectual work related to toolkit design; Mary Jo Messito, for intellectual work related to implementation and and toolkit design; UM/Jackson Memorial Medical Center: Anna Maria Patino-Fernandez, PhD, for intellectual contributions to study design, measures, and implementation; UNC: Tamera Coyne-Beasley (now at University of Alabama), for intellectual work related to study design and implementation; Michael Steiner, for intellectual work related to study design and implementation; Sophie Ravanbakht, for intellectual contributions related to study implementation and measurements; Vanderbilt University: Shari Barkin, for initial intellectual work related to study design and implementation; Sunil Kripilani, for initial intellectual work related to study design and implementation; Andrea Bronaugh, for intellectual contributions related to study implementation and measurements. The development and implementation of the Greenlight Intervention Study were made possible by the following: NYU School of Medicine/Bellevue Hospital Center: Cynthia Osman, Steve Paik (now at Columbia University), Maria Cerra, Evelyn Cruzatte, Dana Kaplan (now at Zucker School of Medicine at Hofstra/Northwell), Omar Baker (now at Riverside Medical Group; Columbia University), Maureen Egan (now at Children’s Hospital Colorado), and Leena Shiwbaren (now at Flatiron Pediatrics); Stanford University: Thomas Robinson, Mairead Mahoney, Kasiemobi Udo-okoye, Pablo Uribe, Nathan Shaw, Chinyelu Nwobu, Catherine Clark, Lauren Wegner, and Michelle Smith; Duke University: Charles Wood and Janna Howard; UNC: Brenda Calderon, Elizabeth Throop (no longer at UNC Chapel Hill), Margaret Kihlstrom, Christina Anderson, Carol Runyan (now at University of Colorado, Denver), and Mariana Garrettson (now no longer at UNC Chapel Hill); UM/Jackson Memorial Medical Center: Lourdes Forster, Randi Sperling, Stephanie White (now at Dartmouth University), Lucila Bloise, Adriana Guzman, and Daniela Quesada; Vanderbilt University: Ayumi Shintani, Sventlana Eden, Marina Margolin, Alexandra Arriaga, Sujata Ayala, Barron Paterson, and Seth Scholer. We also acknowledge support from American Academy of Pediatrics for use of The Injury Prevention Program materials; Greenlight Study Development Team: Duke University: Asheley Skinner, for intellectual work related to study design, implementation, and analysis; Joanne Finkle, for intellectual contributions related to study implementation and measurement; NYU School of Medicine/Bellevue Hospital Center: Alan Mendelsohn, for intellectual work related to study design and implementation; Benard Dreyer, for initial intellectual work related to study design and implementation and toolkit design; Linda van Schaick, for intellectual work related to toolkit design; Mary Jo Messito, for intellectual work related to implementation and toolkit design; UM/Jackson Memorial Medical Center: Anna Maria Patino-Fernandez, PhD, for intellectual contributions to study design, measures, and implementation; UNC: Tamera Coyne-Beasley (now at University of Alabama), for intellectual work related to study design and implementation; Michael Steiner, for intellectual work related to study design and implementation; Sophie Ravanbakht, for intellectual contributions related to study implementation and measurements; Vanderbilt University: Shari Barkin, for initial intellectual work related to study design and implementation; Sunil Kripilani, for initial intellectual work related to study design and implementation; Andrea Bronaugh, for intellectual contributions related to study implementation and measurements.