Data came from the Data Mining Project, and included: 1) repeated measures of weight and recumbent length (or standing height) for all WIC-participating children in Los Angeles County at the time of certification (entry into the program) and re-certification (every 6–12 months); 2) sociodemographic information on children and their families; and 3) the type of food package children received from WIC throughout their participation. These data come from administrative records; weight and length (or height) are measured at each of the time points described while sociodemographic information is reported by the caregivers upon enrollment into WIC. Data on the type of food package provided by WIC is available for every month of participation. In this study we included children who: participated in WIC from 2003 until 2016 throughout ages 0–4 years (inclusive), enrolled in WIC within 42 days of birth, had at least one weight and length (height) measurement per year, and at least one measurement after the age of 4 years. We restricted the current analysis to children who received the fully formula feeding package every month during their first year of life; i.e., were never breastfed (N = 74,871). For aim 1, based on the calendar years in which the child participated in WIC (Fig. 1), we categorized children according to the duration of receipt (or dose) of the new child food package: 0 years, > 0 to < 1 years, 1 to < 2 years, 2 to < 3 years, 3 to < 4 years, and 4 years. Dose of the new child package was treated as a categorical variable.
Fully formula fed children participating in WIC pre-2009 received cumulatively more formula in the first year of life compared to children participating in WIC post-2009. Prior to the 2009 food package change, fully formula fed children received 403 fl oz. of formula each month during their first year of life, and 96 fl oz. of juice each month between 4 and 11.9 months. After the 2009 food package change, fully formula fed children received 403 fl oz. formula per month between 0 and 3.9 months, 442 fl oz. per month between 4 and 5.9 months, and 312 fl oz. per month between 6 and 11.9 months of age [3]. Juice was removed from the new food package which added a new benefit, 128 oz. of baby food fruits and vegetables monthly between 6 and 11.9 months.
To isolate the potential effect of the new infant food package on obesity for aim 2, we compared a sub-sample of children who received the oldinfant package followed by the newchild package with those who received the newinfant package and the newchild package (Fig. 1). This analysis required comparison of the cohort born in the year before the food package change to children born in the three years following the food package change.
The outcomes of the study were weight-for-height z-score (WHZ) growth trajectories from birth until 4 years (inclusive) and obesity at age 4 years. WHZ values were estimated using sex-specific CDC growth curves [8]; subjects with implausible WHZ values (8 ≤ WHZ ≤ -4) were excluded from analyses [9]. WHZ was used to assess longitudinal growth trajectories because: 1) the CDC does not have body mass index (BMI)-for-age curves for children < 2 years of age, and 2) BMI is correlated with length (height) and thus may not be appropriate for assessing adiposity trajectories in growing children [10, 11]. Obesity at age 4 was defined as BMI-for-age ≥ 95th percentile. Covariates included children’s age, gender, and race/ethnicity as reported by their caregiver (Asian, African American, Hispanic, White, other); maternal education (less than high school, high school graduate, more than high school) and language preference (English, Spanish, other); and family income (< 50% federal poverty level [FPL], 50–100% FPL, > 100% FPL).
Statistical analyses
Frequencies, means and standard deviations were calculated to characterize the sample. We assessed differences in growth trajectories by comparing mean WHZ slopes and differences in WHZ means between new child food package dosage groups (0 years, > 0 to < 1 years, 1 to < 2 years, 2 to < 3 years, 3 to < 4 years, and 4 years) in 6-month intervals for the first two years and each year thereafter (0–0.5, 0.5–1, 1–1.5, 1.5–2, 2–3, 3–4, and 4–5 years). Slopes were estimated using gender-stratified piecewise linear spline mixed models [12, 13], including a random intercept and random slopes for each age interval for each individual, as well as interactions between dose of the new WIC child package and age, allowing mean slope in each interval to vary by dose.
To adjust for confounding, we created post-stratification weights using child’s initial WHZ (categorized as − 4 to − 1, − 1 to 1, and 1 to 8); maternal education and language preference; and household income. Post-stratification weights were calculated as the ratio of the proportion of children in a joint covariate stratum in a reference population (i.e. children with an initial WHZ of greater than − 1 but less than 1, a mother with less than a high school education, a mother who prefers to speak Spanish, and a family income of < 50% FPL) to the proportion in that joint covariate stratum in each category of duration of the new child package received (0, 0 > and < 1, 1 to < 2, 2 to < 3, 3 to < 4 and 4 years). The reference population for this analysis was all children who received exclusively new infant and new child packages (i.e. first enrolled in WIC after October 1, 2009). Weights were calculated for strata defined by every combination of these covariates (initial WHZ; maternal education and language preference; and family income) in each category of the exposure (0, 0 > and < 1, 1 to < 2, 2 to < 3, 3 to < 4, and 4 years of the new child package). Race/ethnicity was excluded from post-stratification weights because of sparseness in joint covariate strata for non-Hispanic children, which created weight instability. In addition to incorporating these post-stratification weights, the mixed models were adjusted for maternal education and language preference, household income and child race/ethnicity.
We used gender-stratified modified Poisson regression models with robust standard error estimation [14] to assess the association between dose of the new child food package and obesity at age 4. Dose of the new WIC child package was treated as both a categorical and an interval-scaled variable. We used similar models to assess the association between infant food package type (old vs. new) and obesity among the sub-sample of children receiving > 3.5 years of the new child package. All Poisson models incorporated post-stratification weights and were adjusted for child race/ethnicity and initial WHZ; maternal education and language preference; and household income. Analyses were conducted using SAS 9.4 (SAS Institute Inc., Cary, NC). A p-value< 0.05 was considered to be of statistical significance.