Do Early Infant Feeding Practices and Modifiable Household Behaviors Contribute to Age-Specific Interindividual Variations in Infant Linear Growth? Evidence from a Birth Cohort in Dhaka, Bangladesh

ABSTRACT Background Causes of infant linear growth faltering in low-income settings remain poorly understood. Identifying age-specific risk factors in observational studies might be influenced by statistical model selection. Objectives To estimate associations of selected household factors and infant feeding behaviors within discrete age intervals with interval-specific changes in length-for-age z-scores (LAZs) or attained LAZ, using 5 statistical approaches. Methods Data from a birth cohort in Dhaka, Bangladesh (n = 1157) were analyzed. Multivariable-adjusted associations of infant feeding patterns or household factors with conditional LAZ (cLAZ) were estimated for 5 intervals in infancy. Two alternative approaches were used to estimate differences in interval changes in LAZ, and differences in end-interval attained LAZ and RRs of stunting (LAZ < −2) were estimated. Results LAZ was symmetrically distributed with mean ± SD = −0.95 ± 1.02 at birth and −1.00 ± 1.04 at 12 mo. Compared with exclusively breastfed infants, partial breastfeeding (difference in cLAZ: −0.11; 95% CI: −0.20, −0.02) or no breastfeeding (−0.30; 95% CI: −0.54, −0.07) were associated with slower growth from 0 to 3 mo. However, associations were not sustained beyond 6 mo. Modifiable household factors (smoking, water treatment, soap at handwashing station) were not associated with infant growth, attained size, or stunting. Alternative statistical approaches yielded mostly similar results as conditional growth models. Conclusions The entire infant LAZ distribution was shifted down, indicating that length deficits were mostly caused by ubiquitous or community-level factors. Early-infant feeding practices explained minimal variation in early growth, and associations were not sustained to 12 mo of age. Statistical model choice did not substantially alter the conclusions. Modifications of household hygiene, smoking, or early infant feeding practices would be unlikely to improve infant linear growth in Bangladesh or other settings where growth faltering is widespread.

1 Supplemental Methods Method 1. Description of the MDIG trial Eligibility Criteria Pregnant women (n=1300) aged 18 years and above and enrolled at 17-24 weeks gestation were randomized into one of five vitamin D treatment arms: 0 IU/week from prenatal enrolment until delivery and 0 IU/week until 26 weeks; 4200 IU/week prenatal and 0 IU/week postpartum; 16800 IU/week prenatal and 0 IU/week postpartum; 28000 IU/ prenatal and 0 IU/week postpartum; or, 28000 IU/week in both prenatal and postpartum periods.

Anthropometry
The birth measurement was the earliest measurement within the first 45 days of life. The 3-month measurement was that which was closest to day 91 (but between day 46 and 136, inclusive). The 6-month measurement was that which was closest to day 182, but between 137 and 227 days. The 12-month measurement was that which was closest to 365, but between 320 and 410. For any infant, if the time between the 0 and 3 month measurements or the 3 and 6 month measurements were fewer than 60 days apart, that infant was excluded from the corresponding analysis where length in that interval was the outcome. Similarly, infants were excluded from analyses where length in the 6 to 12 month interval was the outcome if the 6 and 12 month measurements were fewer than 152 days or more than 212 days apart. If the birth and 12 month measurements were fewer than 304 days apart, the infant would be excluded from analyses where length in the birth to 12 month interval was the outcome.

Breastfeeding Classification
Each child-week was classified as exclusively breastfed, predominantly breastfed, partially breastfed or not breastfed, per the WHO categories (1). Maternal/caregiver report of the infant's feeding 24 hours/and or 7 days preceding each visit was combined to classify breastfeeding in each week during the first 3 months and the first 6 months of life. In order to reconcile rare discrepancies between the 24-hour and 7-day recall items relating to breast milk intake in the caregiver interview, an infant who was indicated as being breastfed in the last week or the last 24-hour period, with no alternative liquids/foods was classified as exclusively breastfed. This combination of the 24-hour and 7-day recall were similarly applied to predominant breastfeeding pattern but with exposures to any additional 2 liquids and to partial breastfeeding pattern but with exposures to any solid/semi-solid/soft foods. No breastfeeding required the caregiver to indicate no breastmilk was given to the infant in both the 24-hour and 7-day recalls. To aggregate weekly data, for each time period of interest, the assigned breastfeeding category corresponded to the least optimal breastfeeding category among any of the observed child-weeks. If an infant was missing breastfeeding data in the last 2 weeks of observation, the infant was designated 'not able to be classified'. Due to the high prevalence of pre-lacteal feeds in this population, the primary derivation of this variable ignored the breastfeeding pattern in the first week to classify breastfeeding in the 0-3 and 0-6 month time periods.
In sensitivity analyses, we derived breastfeeding pattern within the first 3 and the first 6 months of life: 1) including feeding data from the first week of life, not allowing for any deviations in any week; 2) requiring the caregiver to indicate that the infant was breastfed in both the 24-hour and 7-day recalls (rather than at least one of the 2 recalls) in order to be classified as exclusively breastfed or predominantly breastfed, and requiring the infant to have been breastfed over at least the 7-day period to be classified as partially breastfed; 3) allowing for a single deviation in any observed child-week but still excluding the first week of life (e.g., if in the first 3 months, a child was partially breastfed in a single week but otherwise exclusively breastfed, then the single week deviation would be ignored and the child would be classified as exclusively breastfed in the first 3 months); 4) allowing for missing data in the last 2 weeks of the classification time period among infants who were not exclusively breastfed; 5) using the last 2 observation weeks (i.e., feeding pattern in weeks 12 and 13 or weeks 25 and 26 to derive feeding in the first 3 and 6 months, respectively); 6) using data from infants who had no missing breastfeeding data in the first 3 or 6 months of life; and 7) using a 24-hour recall period rather than a 7 day period preceding the visit.

Duration of EBF
Duration of exclusive breastfeeding (EBF) in the first 3 and 6 months was the number of weeks since birth that an infant was reported to be exclusively breastfed. In any given week, an infant was identified as having been exclusively breastfed if the caregiver indicated that the infant was breastfed in the last week or the last 24-hour period and was not given any alternative liquids or foods. For infants for whom a period of EBF was separated from a period of non-EBF with missing child-weeks of observational data, the duration of EBF was set to be the average between the last known week of EBF and the first week of known non-EBF. For infants for whom a period of exclusive breastfeeding was followed by weeks of missing data with no dietary data available following the missing weeks, the duration of EBF was set to be halfway between the end of the interval and the last known week of EBF. As with breastfeeding pattern, the primary derivation of duration of EBF ignored the first week of life and assumed that the infant was EBF due to the high prevalence of pre-lacteal feeds. Therefore, the minimum duration of EBF was 1 week.
In sensitivity analyses, we used the following alternate derivations of duration of EBF in the first 3 and 6 months: 1) allowing for 1 week of deviation from exclusive breastfeeding, excluding the first week of life; 2) including the first week of life, thereby not allowing for any deviations from EBF in any week; 3) requiring the infant to have been indicated as breastfed by the caregiver in both 3 the 24-hour and 7-day recalls (rather than at least one of the 2 recalls) in order to be considered exclusively breastfed; 4) using the last known week of EBF as the duration of EBF rather than taking the average between the last known week of EBF and the first week of known EBF or between the last known week of EBF and the end of the age interval; and 5) using the end of the time interval as the duration of EBF for infants who were exclusively breastfed until a certain week followed by missing data up until the end of the interval.

Additional Sensitivity Analyses
The primary derivation of the binary variables (ever/never) for formula exposure and animal source food exposure (i.e., any of milk, yoghurt, cheese, meats and organ meats, eggs, or fish) included only those infants for whom there were available feeding data for at least half of the weekly visits during the respective time period. In sensitivity analyses, we derived formula exposure and animal exposure variables that included infants who completed less than half of the weekly visits. We also generated an additional animal exposure variable for sensitivity analyses whereby infants were classified as having consumed animal products using data from weekly visits (i.e. the primary feeding data source) as well as data on animal product consumption collected from mothers retrospectively during an interview conducted later in infancy.
To address the robustness of the inferences based on variations in anthropometric outcome definitions, the following sensitivity analyses were performed: 1) relaxed the interval duration requirements to permit inclusion of measurements a minimum of 45 days apart rather than 60 days apart; or 2) Increased the stringency of the interval duration requirement by narrowing the timeframe of each measurement, such that the birth measurement was required to be between 0-14 days, the second measurement between 77-105 days, the third measurement between 168-196 days, and the fourth measurement between 351-379. For these sensitivity analyses, we used the primary derivation of breastfeeding pattern as described above.
Method 3. Description of post-hoc sensitivity analysis using a mixed effect model As a post-hoc sensitivity analysis, we assessed the association of breastfeeding pattern from 0 to 3 months of age with infant growth from birth to 12 months of age using a mixed effects model with linear splines, infant-specific random intercepts and random slopes. Knots were placed at 91 days and 182 days, resulting in 3 spline terms that captured the birth to 3-month interval, 3 to 6 month interval, and 6 to 12-month interval of growth. An interaction between each of the spline terms and breastfeeding pattern was estimated to assess the association of breastfeeding pattern with the change in growth in the interval. 5 Supplemental Table 2. Association of breastfeeding pattern from 0-3 months of age or 0-6 months of age with linear growth in two age intervals (0-3 months or 3-6 months of age) with or without adjustment for reported maternal concern about infant feeding or weight gain, using the residuals model approach 1 Feeding-related factor (age interval in which exposure was ascertained) Age interval in which outcome (linear growth) was ascertained 2

0-3 Months 3-6 Months
Without adjustment for presence of any maternal concern during the breastfeeding period With adjustment for presence of any maternal concern during the breastfeeding period Without adjustment for presence of any maternal concern during the breastfeeding period With adjustment for presence of any maternal concern during the breastfeeding period All models included the following additional covariates: assigned treatment group in the MDIG trial, maternal height, paternal occupation, maternal and paternal education, asset index, neonatal illness, delivery location, delivery mode, maternal and paternal education, wealth index, number of children, maternal postnatal BMI, maternal age, infant sex, newborn weight-for-length z-score, and gestational age at birth. Table 3. Associations of risk factors measured at baseline (2 nd trimester of pregnancy) or 9-months postpartum with linear growth at the 6-to 12-month age interval, using a residuals model approach 1 1 All models adjusted for the assigned treatment group in the MDIG trial and maternal height, maternal and paternal occupation, maternal and paternal education, and asset index. 7

Covariates measured in the 2 nd trimester of pregnancy
Supplemental Table 4. Multivariable-adjusted associations of early postnatal feeding-related factors and household factors with linear growth in five age intervals of infancy in a birth cohort in Dhaka, Bangladesh, based on the ANCOVA modeling approach 1 1 All models adjusted for the assigned treatment group in the MDIG trial and maternal height. 8 Supplemental Table 5. Multivariable-adjusted associations of early postnatal feeding-related factors and household factors with linear growth in five age intervals of infancy in a birth cohort in Dhaka, Bangladesh, based on the 'change score' modeling approach 1 1 All models adjusted for the assigned treatment group in the MDIG trial and maternal height.  Supplemental Table 6. Comparison of effect estimates for the associations of feeding-related and household factors with linear growth outcomes in different age intervals of infancy, using five different growth or size modeling approaches, restricted to associations for which there were significant associations using at least one modeling approach. 1 1 Red cells represent statistically significant associations and blue represent non-significant associations whereby p<0.05 was considered statistically significant.   Table 7. Multivariable-adjusted associations of early postnatal feeding-related factors and household factors with linear growth at four ages of infancy in a birth cohort in Dhaka, Bangladesh, using the attained size modeling approach 1 1 All models adjusted for the assigned treatment group in the MDIG trial and maternal height. 2 Additionally adjusted for maternal and paternal occupation, maternal and paternal education, and asset index. 3 Additionally adjusted for neonatal illness, delivery location, delivery mode, maternal and paternal education, asset index, number of children, maternal postnatal BMI, maternal age, infant sex, newborn weight-for-length z-score, and gestational age at birth. Mixed effects models were used to model length-for-age z-score as a function of age, with child-specific random intercepts and random slopes. Linear splines were used to capture the age intervals of interest; knots were placed at 91 and 182 days of age (corresponding to 3 and 6 months of age). Interaction terms between breastfeeding pattern from 0-3 months and all spline terms for age were included in the model, and the interaction of breastfeeding pattern with the spline terms are presented in this table. Effect estimates are scaled to reflect the difference in LAZ slope for a 91-day increase in age for the 0-3 month and 3-6 month spline terms and a 182 day increase for the 6-12 month spline term. N= 983 infants contributed at least 1 observation to the model. 2 Models adjusted for assigned treatment group in the MDIG trial, maternal height, neonatal illness (ever/never), delivery location, delivery type, maternal and paternal education, wealth index, number of children, maternal postnatal BMI, maternal age, infant sex, newborn weight for length z-score, and gestational age at birth. 3 Marginal effect of breastfeeding pattern from 0-3 months of age on LAZ at 12 months of age was estimated from the mixed effects model with 95% confidence interval 13

Size at birth Attained size at 3 months Attained size at 6 months
Supplemental Table 10. Sensitivity analyses of associations of early postnatal feeding-related factors with linear growth in four age intervals of infancy using alternative breastfeeding classification schemes, exclusive breastfeeding duration criteria, animal intake criteria and formula intake criteria in the first 6 months of life. 1 Feeding-related factor (0-6 months) Age interval in which outcome (linear growth) was ascertained 2