Characteristics associated with the transition to partial breastfeeding prior to 6 months of age: Data from seven sites in a birth cohort study

The WHO recommends exclusive breastfeeding for the first 6 months of life. However, the transition of the infants’ diet to partial breastfeeding with the addition of animal milks and/or solids typically occurs earlier than this. Here, we explored factors associated with the timing of an early transition to partial breastfeeding across seven sites of a birth cohort study in which twice weekly information on infant feeding practices was collected. Infant (size, sex, illness and temperament), maternal (age, education, parity and depressive symptoms), breastfeeding initiation practices (time of initiation, colostrum and pre-lacteal feeding) and household factors (food security, crowding, assets, income and resources) were considered. Three consecutive caregiver reports of feeding animal milks and/or solids (over a 10-day period) were characterized as a transition to partial breastfeeding, and Cox proportional hazard models with time (in days) to partial breastfeeding were used to evaluate associations with both fixed and time-varying characteristics. Overall, 1470 infants were included in this analysis. Median age of transition to partial breastfeeding ranged from 59 days (South Africa and Tanzania) to 178 days (Bangladesh). Overall, higher weight-for-length z-scores were associated with later transitions to partial breastfeeding, as were food insecurity, and infant cough in the past 30 days. Maternal depressive symptoms (evaluated amongst 1227 infants from six sites) were associated with an earlier transition to partial breastfeeding. Relative thinness or heaviness within each site was related to breastfeeding transitions, as opposed to absolute z-scores. Further research is needed to understand relationships between local perceptions of infant body size and decisions about breastfeeding. The MAL‐ED network included eight sites (Bangladesh [Dhaka: BGD], India [Vellore: INV], Nepal [Bhaktapur: NEB], Pakistan [Naushero Feroze: PKN], Brazil [Fortaleza: BRF], Peru [Loreto: PEL], South Africa [Venda: SAV] and Tanzania [Haydom: TZH]). Each site was to recruit and follow a cohort of 200 children to 24 months, and thus, enrolment varied by site based on projected loss to follow up (The MAL‐ED Network Investigators, 2014). Enrolment was staggered over a 2‐year period (during the overall period from 2009 to 2012) to account for seasonality in morbidity and pathogens. Infants were eligible for inclusion in the study if their birth weight or enrolment weight was ≥1500 g, they did not have a chronic illness at recruitment, they were a singleton birth, the family did not plan to move outside the community within 6 months, and their mother was at least 16 years of age. Caregivers provided written consent, and institutional ethical approval was obtained at each site. Further details are available elsewhere (The MAL‐ED Network Investigators, 2014). Here, we examine data from birth to 6 months of age. At enrolment (within the first 17 days of life, median 7 days, inter‐quartile range 4 to 12 days), an interview was conducted to record child and family factors including the sex of the child, maternal age, parity, education and marital status. At that time, mothers reported specific details regarding the timing of breastfeeding initiation after delivery, whether colostrum was given, and pre‐lacteal feeding (Patil et al., 2015). Thereafter, families were visited twice per week and asked about infant feeding practices in the preceding 24 h; specifically, caregivers were asked about the infant’s consumption of breast milk, animal milk, formula, water, tea, fruit juice and other liquids or semi‐solids (Caulfield et al., 2014). Our primary outcome was the consistent reporting of partial breastfeeding, meaning that animal milks (including formula) and/or solids were added to the infant’s diet alongside breast milk for three consecutive reports from the twice weekly breastfeeding data. We chose this definition based on prior work which described the episodic nature of exclusive breastfeeding across the study sites (Ambikapathi et al., 2016; Lee et al., 2014). During these same twice weekly visits, incidence of illness was recorded for all days since the prior visit. Illnesses included diarrhoea (≥3 loose stools/24 h), vomiting, coughing, acute lower‐respiratory infection (ALRI) and fever (Richard, Barrett, et al., 2014). Infant weight and length were measured at enrolment and then monthly on the same birth day throughout the study (Richard, McCormick, et al., 2014). Age‐ and sex‐standardized z‐scores were derived from length and weight utilizing the WHO standards and methods (WHO, 2006). Personnel were trained on a common protocol prior to study enrolment and quality control measures were put in place (Richard, McCormick, et al., 2014); during the study, irreconcilable issues were found with the length data from PKN; therefore, the site was excluded from these analyses. Food insecurity was assessed through the Household Food Insecurity Access Scale (HFIAS) at enrolment (Psaki et al., 2012; Swindale & Bilinsky, 2006), and households were considered to have food insecurity if they answered yes to any of the questions. At 6 months, households were surveyed about socioeconomic status including information on average household income, assets, crowding, and access to improved water and sanitation (as defined by the WHO) (Psaki et al., 2014). Maternal depressive symptoms were captured using the Self‐Reporting Questionnaire at 1‐ and 6‐months postpartum (Beusenberg et al., 1994). The depressive symptoms data at each time point were subjected to psychometric analyses to ensure comparability across sites (Pendergast et al., 2014); because the 6‐month data as opposed to the 1‐month data were related to the outcome, and because the 6‐month data from BRF were not found comparable with the other sites, two analyses were run, with the maternal depressive symptom data (excluding BRF) or without (including BRF). Child temperament was assessed at 6 months via caregiver report using the Infant Temperament Scale (ITS). Psychometric analyses supported the validity of an approach temperament factor across the eight sites (Pendergast et al., 2018). Approach temperament assesses an infant’s tendency to move toward or engage in pleasurable or rewarding stimuli. Our assessment of this dimension of temperament reflected approach towards both social and physical stimuli. We used this as our best measure of infant behaviours, which might affect infant feeding decision‐making. As noted above, we considered the report of feeding animal milks and/or solids on three consecutive visits prior to 6 months to indicate a transition to partial breastfeeding. The first instance of this transition was considered the ‘event’ for the purposes of a Cox proportional hazard model. Cox proportional hazards models were constructed for individual variables and multivariable models. All models included site as a strata to account for differences in baseline hazards between sites. The home visits in which infant feeding and morbidity data were recorded were matched to the closest prior anthropometric measurement. This data structure allowed for time‐varying incidence of illness and the child’s weight‐for‐length z‐score (WLZ) with monthly intervals (the anthropometry collection schedule) for months in which partial feeding was or was not initiated. Those not transitioning before 180 days (276/1470, or 18.8%) were censored. To put variables on a comparable scale, continuous variables, except for WLZ, were scaled to mean zero and standard deviation one across the MAL‐ED sites: average household income (log10 transformed first), the number of people per room, maternal age, maternal years of education, and depressive symptom scores at 6 months postpartum. Odds ratios for these variables therefore reflect the consequence of a one standard deviation change in the variable. The multivariable model was further subjected to backward stepwise selection [minimizing the Akaike Information Criterion (AIC)] to minimize over‐fitting. The models were globally proportional; however, WLZ was not (χ 2 5.06, p = 0.02), largely due to the increase in WLZ observed (in all sites) over the first 2 months. Because field operations could result in shorter or longer than monthly periods between anthropometric measures, sensitivity analyses were conducted to examine the interval (median of 13 days, IQR 6 to 21) between the preceding anthropometry and the change in feeding; the model coefficients presented were found robust to weighting the regression by the interval even considering different weighting schemes.