Associations of Enteric Protein Loss, Vaccine Response, Micronutrient Deficiency, and Maternal Depressive Symptoms with Deviance in Childhood Linear Growth: Results from a Multicountry Birth Cohort Study

We identified the determinants of positive (children who had a birth weight, 2.5 kg and/or maternal height, 145 cm but were nonstunted at 24 months of age) and negative (children who had a birth weight $ 2.5 kg and maternal height $ 145 cm but were stunted at 24 months of age) deviance in childhood linear growth. We found that socioeconomic status (b 5 1.54, P, 0.01), serum retinol (b 5 0.05, P, 0.01), hemoglobin (b 5 0.36, P, 0.01), length-for-age Z-score (LAZ) at birth (b 5 0.47, P, 0.01), and tetanus vaccine titer (b 5 0.182, P, 0.05) were positively and maternal depressive symptom (b 5 -0.05, P, 0.01), serum ferritin (b 5 -0.03, P, 0.01), male sex (b 5 -1.08, P, 0.01), and a1-antitrypsin (b 5 -0.81, P, 0.01) were negatively associated with positive deviance. Further, diarrhea episodes (b 5 0.02, P, 0.01), male sex (b 5 0.72, P, 0.01), and a1-antitrypsin (b 5 0.67, P, 0.01) were positively and hemoglobin (b5 -0.28, P, 0.01), soluble transferrin receptor level (b 5 -0.15, P, 0.01), and LAZ score at birth (b 5 -0.90, P, 0.01) were negatively associated with negative deviance. To summarize, enteric protein loss, micronutrient deficiency, vaccine responses and maternal depressive symptoms were associated with linear growth deviance in early childhood. In such a background, public health approaches aimed at reducing the risk of intestinal inflammation and altered gut permeability could prove fruitful in ensuring desired linear growth in children. In addition, maternal mental health issue should also be considered, especially for promoting better nutritional status in children in the context of linear growth deviance. We gathered data from a multicountry birth cohort study named “Etiology, Risk Factors, and Interactions of Enteric Infections and Malnutrition and the Consequences for Child Health” (MAL-ED) study. This study was conducted at eight sites across three continents. In this analysis, we included data from six MAL-ED sites: Dhaka, Bangladesh, and Vellore, India in Asia; Fortaleza, Brazil, and Loreto, Peru, in the Americas; and Venda, South Africa, and Haydom, Tanzania, in Africa. Children were enrolled from November 2, 2009 to February 28, 2014, within 7 days of their birth and were followed uniformly up to 24 months of age after predefined validated protocols.15–17 The MAL-ED study protocol received ethical approvals from the institutional review boards of the respective sites. Informed written consent was taken from the parents or legal guardians of the enrolled children after informing them about the study objective and related methods. An analysis of the MAL-ED birth cohort study reported two nonmodifiable characteristics—maternal height and birth weight—as contributing significantly in modifying linear growth in children throughout the first 24 months after birth.2 Hence, based on the stunting status at 24 months and the maternal height and birth weight of the children, we divided 1,092 children of MAL-ED birth cohort to four deviant groups (Supplemental File 1; Tables 1 and ​and22): Sociodemographic characteristics of the participants EBF = exclusive breastfeeding; LAZ-score = length-for-age Z-score; ND = negative deviant; NND = nonnegative deviant; NPD = nonpositive deviant; PD = positive deviant; SRQ-20: Self-Reporting Questionnaire–20. Factors associated with the chances of being positive and negative deviants CI = confidence interval; EBF = exclusive breastfeeding; LAZ-score = length-for-age Z-score; L-M ratio = lactulose-mannitol ratio; SRQ-20 score = Self Reporting Questionnaire–20. Here, the length-for-age Z-score (LAZ score) for each child was determined using the WHO 2006 Child Growth Standards and stunting was defined as LAZ score < –2 SD of the WHO Child Growth Standards median.18 Enrollment weight, taken within the first 7 days of birth, was used as the surrogate of birth weight. Data on predictor variables were gathered from eight domains: demographic and socioeconomic indicators, dietary intake, maternal depressive symptoms, morbidity, gut inflammation, gut integrity, serum micronutrient status, and vaccine response status. Socioeconomic data was collected at 6, 12, 18, and 24 months of age of a child. The WAMI index (Water, sanitation, hygiene, Asset, Maternal education, and Income index, ranging from 0 to 1), a socioeconomic status index that includes access to improved water and sanitation, eight selected assets (separate room for a kitchen, household bank account, mattress, TV, refrigerator, people per room, table, chair or bench), maternal education, and household income was used as a representative of socioeconomic status of the households.19 A higher WAMI index means a better socioeconomic status. The statistical analysis to calculate WAMI score was done in two phases. First, the best approach for selecting and weighting household assets as a proxy for wealth was identified. Four approaches for measuring wealth (maternal education, principal components analysis, multidimensional poverty index, and a novel variable selection approach based on the use of random forest algorithm) were compared. Second, the selected wealth measure was combined with other relevant variables to form the index.19 For assessing dietary intake of the children, we collected 24-hour dietary recall data monthly from the ninth month onward using a 24-hour multiple-pass dietary recall approach.20 The interviews were conducted on nonconsecutive days, and one out of every four recalls was conducted on a weekend. The total amount of energy taken from protein intake was measured from the dietary intake data using a locally adapted food composition table. Data collectors asked the mother about the liquids the child consumed during the past 24 hours and whether the response followed the WHO definition of exclusive breastfeeding (EBF; no other foods or drink, not even water, except breast milk, including expressed milk, oral rehydration solution, vitamins, minerals, and medicine syrups); if so, the child was considered as exclusively breastfed. Instead of EBF status (yes versus no), EBF days were used during data analysis because it specifies the extent of EBF to specific number of days. Trained fieldworkers recorded the depressive symptoms of a mother using Self Reporting Questionnaire-20 (SRQ-20).21 The questionnaire, comprising 20 binary (yes versus no) type of questions, is developed by the WHO for use in developing countries and is designed to assess maternal psychological adjustment related to depressive symptoms.22,23 Data on diarrheal episodes were collected twice a week. Diarrhea is defined as having three or more loose stools in 24 hours or at least one loose stool with blood reported by the mother, and a diarrheal episode is defined as being separated from another episode by at least two or more diarrhea-free days.24 The markers of gut inflammation (Alpha-1-antitrypsin; [A1AT; ELISA kit: Bio vendor, Chandler, NC], neopterin [Neo; ELISA kit: GenWay Biotech, San Diego, CA], and myeloperoxidase [MPO; ELISA kit: Alpco, Salem, NH]; measured in stool at 7, 15, and 24 months) and gut integrity (lactulose-mannitol ratio, LM ratio; measured in urine at 3, 6, 9, and 15 months) were measured from all the children following a standard protocol.15 All the biomarkers were measured longitudinally, and the mean of all values were used for data analyses. Blood samples for measuring serum zinc, serum retinol, soluble transferrin receptor (sTfR), ferritin, and hemoglobin (Hb) status were collected at 7, 15, and 24 months of age of the children. For hemoglobin, one drop of capillary blood was collected and measured with the Hemocue device (Hb 201, Ängelholm, Sweden). Plasma zinc and ferritin levels were measured using Atomic Absorption Spectrometry and Chemiluminescence Immunoassay, respectively. sTfR levels were measured using immunoturbidimetry method. Hb, ferritin, and zinc values were adjusted for the presence of inflammation using alpha-1-acid glycoprotein values.25,26 Blood samples collected at 7 and 15 months of age was used for assessment of vaccine responses. Quantitative antimeasles, antitetanus toxoid, and antipertussis toxin IgG ELISAs (Euroimmun, Lubeck, Germany) and antipoliovirus IgG ELISAs (Genway) were performed according to the instructions of the manufacturer.27 During administering quantitative antirotavirus serum IgG and IgA ELISAs, microplates were coated with IgG antirotavirus rabbit, and either cell lysate or virus preparedness was applied to alternating rows after washing. Eight 2-fold dilutions were made, beginning with 1:80 dilutions of the IgA and IgG serum levels. Four 2-fold dilutions were prepared of 1:20 dilutions of known reference IgA and IgG and unknown serum or plasma samples. After washing, the microplates were inserted with the serum standard dilutions and serum sample dilutions. After washing again, biotinylated rabbit antihuman IgA (for the IgA plates) or IgG (for the IgG plates) was added and then avidin–biotin-peroxidase complex was washed and inserted. O-phenylenediamine dihydrochloride substrate was applied to each well after the final wash, and the reaction with sulfuric acid stopped. The plates were read at 492 nm, and a four-parameter fit of the transformed optical density values computed the titers.28 We first tabulated variables related to the household, maternal health, and child nutritional status using descriptive statistics (mean, standard deviation, and percentages, as appropriate). Then the group-wise distributions of biomarkers of enteric inflammation and vaccine titers were compared using box plots according to the outcome variables (PD [yes/no] and ND [yes/no]). Normality of the distribution of specific variables was determined using Shapiro-Wilk test. Student’s t test (for normally distributed variables) and Mann-Whitney U tests (for variables with skewed distribution) were applied to test the null hypothesis of no difference between the groups. Finally, multivariable logistic regression models were fit to identify the predictors of growth deviance. The binary outcome variables for the regression analyses were PD (yes/no) and ND (yes/no). Child-specific mean values of the predictor variables were used for fitting the regression models. The data sets that we used contain unequal clusters (Supplemental File 1; Table 2) of nonindependent observational units—namely, country. Statistically, measurements within a country might be more clustered than the measurements between the countries. To adjust this clustering effect, we used generalized linear mixed-effects models (GLMMs) where the intercept of the variable (i.e., country) was kept random. This approach allowed us to calculate more robust estimates of the variance in the outcome variable, both within and between the clusters.29 We built specific bivariate and multivariable logistic regression models for each of the outcome variables where GLMMs estimated the probability of being positive and negative deviants when a child is exposed to the predictor variables detailed above. The inclusion of children in different groups were done according to the definitions of PN, NPD, ND, and NND stated earlier. Initially, we conducted bivariate regression analyses (termed as unadjusted), and the variables showing statistically significant association (P < 0.05) to the specific outcome variables were selected for fitting the final multivariable models (termed adjusted). We reported the results of four multivariable models in Tables 2 and ​and33 (Table 2: factors associated with the chances of being positive and negative deviance; Table 3: association of vaccine titers to positive and negative deviances). Details of the model building process and the model statistics with random effects can be found in Supplemental File 2. Associations between levels of vaccine titers and the chances of being positive and negative deviants CI = confidence interval. We carried out the data analysis in R (version 3.5.1), and the lme4 package was used for fitting the generalized linear mixed-effects models.30 A P value < 0.05 was considered as the margin of statistical significance for all the analyses.