Stunting in infancy is associated with decreased risk of high body mass index for age at 8 and 12 years of age

Background: Effects of early-life stunting on adiposity development later in childhood are not well understood, specifically with respect to age in the onset of overweight and obesity. Objectives: We analyzed associations of infant stunting with prevalence of, incidence of, and reversion from high body mass index-for-age z score (BMIZ) later in life. We then estimated whether associations of infant stunting with BMIZ varied by sex, indigenous status, and rural or urban residence. Methods: Data were collected from 1942 Peruvian children in the Young Lives cohort study at ages 1, 5, 8, and 12 y. Multivariable generalized linear models estimated associations of stunting (height-for-age z score < 22) at age 1 y with risk of BMIZ > 1 and BMIZ > 2 prevalence, incidence (moving above a BMIZ threshold between ages), and reversion (moving below a BMIZ threshold between ages) at later ages. Results: After adjustment for covariates, stunting at age 1 y was associated with a lower prevalence of BMIZ > 1 at age 8 y (RR: 0.81; 95% CI: 0.66, 1.00; P = 0.049) and 12 y (RR: 0.75; 95% CI: 0.61, 0.91;P=0.004), aswell as a lower prevalence of BMIZ > 2 at age 8 y. Stunting was not associated with incident risk of BMIZ > 1 or BMIZ > 2. Stunting was positively associated at age 5 y with risk of reversion from BMIZ > 1 (RR: 1.22; 95% CI: 1.05, 1.42; P = 0.008) and BMIZ > 2. We found evidence that the association of stunting with prevalent and incident BMIZ > 1 was stronger for urban children at ages 5 and 8 y, and for nonindigenous children at age 8 y. Conclusions: Stunting predicted a lower risk of prevalent BMIZ > 1 and BMIZ > 2, even after controlling for potential confounders. This finding may be driven in part by a higher risk of reversion from BMIZ > 1 by age 5 y. Our results contribute to an understanding of how nutritional stunting in infancy is associated with BMIZ later in life. We analyzed data from Peruvian children in the prospective Young Lives cohort study (23). In 2002, 2052 children aged ∼6–18 mo were recruited (round 1). Follow-up data were collected in 2006 when children were ∼5 y old (round 2), in 2009 when children were ∼8 y old (round 3), and in 2013 when children were ∼12 y old (round 4). To simplify reference to each of these rounds of data collection, we will refer to them as ages 1, 5, 8, and 12 y. Participants were selected through a multistage sampling process. Ten random draws of 20 sentinel sites were conducted from among the 1818 districts in Peru. Consistent with the study’s pro-poor focus, the wealthiest 5% of districts were excluded. From these random draws, one set of 20 sites was selected that best met the study aims of diverse coverage and logistical feasibility. Within selected districts, an initial community was randomly selected as the starting point for recruitment of age-eligible children. Full details of participant recruitment are available elsewhere (24). Weight and length at age 1 y were measured by 6 supervisors who used calibrated digital balances (Soehnle) with 100-g precision and locally made rigid stadiometers with 1-mm precision. At later ages, measurements were taken by all field staff with the use of similar digital platform balances (with 100-g precision), and standing height was measured with the use of locally made instruments accurate to 1 mm. The staff followed standard WHO procedures for measurement of weight, length, and height. To ensure inter- and intrarater reliability, standard measurement procedures were described in the training manual, and repeat measurements were conducted to ensure accuracy (25). HAZ and BMIZ were calculated according to age-appropriate WHO references (2, 4). Our predictor of interest was stunting at age 1 y, defined as HAZ <−2. In this sample, HAZ during round 1 was inversely correlated with age in months (26), so children recruited at a younger age were less likely to be classified as stunted. We therefore adjusted all round 1 HAZ measurements to their predicted value at age 12 mo. Applying methodology described elsewhere (26), we calculated the difference between each child’s HAZ and the mean HAZ for children within ±1 mo of the child’s age. This value was then added to the mean HAZ for children aged 11–13 mo. Outcomes included the prevalence of, incidence of, and reversion from high BMIZ, with the use of thresholds of BMIZ > 1 and BMIZ > 2. The WHO defines overweight and obesity differently for children <5 y of age and those 5–19 y. For children <5 y of age, overweight (including obesity) is defined as BMIZ > 2 and obesity is defined as BMIZ > 3 (2), whereas for children aged 5–19 y, overweight is defined as BMIZ > 1 and obesity is BMIZ > 2 (4). If we adhered to these definitions, children could be considered to develop overweight or obesity without any change in BMIZ. Therefore, for all ages, we consistently defined overweight as BMIZ > 1 and obesity as BMIZ > 2. To maintain clarity, we refer to the exact cutoffs used, rather than the terms overweight and obesity, when referring to the results from this analysis. If a child was above a given BMIZ threshold (i.e., BMIZ > 1 or BMIZ > 2) for the ith round, they were defined as a prevalent case for that threshold in the ith round. If a child was above the threshold at the ith round but below the threshold in the ith − 1 round, then we defined that child as an incident case for that threshold at the ith round. If a child was below the threshold at the ith round but was above the threshold in the ith − 1 round, then we defined that child as reverted from that threshold at the ith round. These transitions are illustrated graphically for the analyzed sample in Figure 1. Transitions across BMIZ > 1 threshold in Peruvian children in the Young Lives cohort at ages 1, 5, 8, and 12 y (n = 1755). Incidence refers to a transition from BMIZ ≤ 1 at a given age to a BMIZ > 1 at the next age. Reversion refers to a transition from BMIZ > 1 at a given age to a BMIZ ≤ 1 at the next age. BMIZ, body mass index–for-age z score. Covariates were selected for the model on the basis of the causal pathway structure supported by the literature, as well as the data available from the Young Lives study. The statistical significance of a covariate was not a criterion for inclusion in the model, although all covariates were significantly associated with stunting status at age 1 y (Table 1). We adjusted for covariates at the child, mother, and household level. At the child level, we adjusted for sex. Child age was not included because it was already adjusted through the BMIZ measure and the adjustment to HAZ in round 1. There was no association between age and BMIZ in any later round. We did not adjust for breastfeeding status because nearly all children (97.7%) had been breastfed for ≥6 mo. We also did not adjust for birth weight because we were interested in stunting at age 1 y as an indicator of chronic malnutrition. Characteristics of stunted and nonstunted Peruvian children at age 1 y in the Young Lives cohort study1 Maternal covariates included height and BMI in round 1. Maternal BMI was categorized into 3 mutually exclusive categories: normal [BMI (in kg/m2) <25], overweight (BMI ≥25 and <30) and obese (BMI ≥30). There were too few underweight women (BMI <18.5; 1.6%) to include in a separate category, so they were included in the normal BMI category. Mothers whose first language was not Spanish (defined by the language the grandmother spoke to the mother) were classified as indigenous. We also included a binary indicator of whether the mother had completed primary education (≥6 grades of schooling). Household characteristics included indicators of whether households had ≥6 members or were in rural areas, and geographic regions (coastal, jungle, or mountain). Household wealth was measured with the use of the Young Lives wealth index, which is the mean of 3 composite scores for housing quality, consumer durables, and service access. A detailed description of the wealth index is published elsewhere (24). Wealth was split into nominal quintile indicators for the statistical analysis. Of the 2052 children initially recruited, 23 were excluded because their ages at recruitment were outside the target range of 6–17 mo. Twenty children were excluded because of documented deaths after baseline, 45 children because of missing HAZ or BMIZ at age 1 y, 11 children because of improbable anthropometric z scores (HAZ <−5 or HAZ >3 or BMIZ <−4 or BMIZ >5) (27) during any round, and 11 children because of missing covariate data at age 1 y. This resulted in a sample of 1942 children with complete data at baseline. An additional 187 children were missing BMIZ data at age 5, 8, or 12 y, resulting in 1755 cases with complete follow-up data for analysis. Details on baseline characteristics of subjects with and without missing follow-up BMIZ data are found in Supplemental Table 1. We observed that missingness was associated with some observed covariates, indicating that a complete case analysis might result in biased estimates. To account for potential selection bias (under the assumption of missing at random), we conducted multiple imputation with the use of chained equations to impute missing values of BMIZ (28). Thirty imputations for each missing value were performed (28). Linear regression was used in the multiple imputation procedure to impute predicted values for missing BMIZ at ages 5, 8, and 12 y. All covariates from the main analysis, baseline outcomes, and an indicator variable for the sampling cluster were included in the imputation models. We stratified the data on stunted status at age 1 y and calculated descriptive statistics. We tested differences in covariate values between stunted and nonstunted children at age 1 y, and between those lost to follow-up and those not lost to follow-up, with the use of Fisher’s exact test, Pearson’s chi-square test, and Student’s t test. We used generalized linear models with a Poisson distribution, log link, and robust variance (29) to estimate the association between stunting status at age 1 y and the risk of subsequent prevalence of, incidence of, and reversion from BMIZ > 1 or BMIZ > 2. Results from 3 models are reported: 1) bivariable regressions of outcomes on stunting at age 1 y; 2) multivariable regressions with controls for potentially confounding covariates with the use of observations with complete outcome data at all ages; and 3) multivariable regressions adjusted for the same covariates as in the second model, but with imputations for missing outcomes. In regressions of incidence and reversion on stunting status in model 3, the population at risk varied across imputed data sets. To permit analysis, we set the at-risk population across imputed data sets by using mean imputed values for BMIZ at age 5 y and age 8 y to determine whether children were at risk of incidence or reversion at ages 8 and 12 y, respectively. We examined, one interaction at a time, the significance of multiplicative interaction terms between stunting status at age 1 y and sex, indigenous status, and rural or urban status. Statistical significance was considered to be P < 0.05. Statistical analyses were conducted with the use of Stata version 13. Ethics committees at the University of Oxford and the Nutrition Research Institute in Lima approved the Peruvian Young Lives study. Parents provided written informed consent in round 1 and verbal reconsent in each subsequent round.