Early newborn ritual foods correlate with delayed breastfeeding initiation in rural Bangladesh

The timing of initiation of breastfeeding within this population was suboptimal, with a mean delay of 30.6 hours between birth and initiation of breastfeeding. Women who fed an ENF initiated breastfeeding significantly later than women who did not feed ENF, even after controlling for other factors likely to hinder breastfeeding initiation and stratifying on the infant’s ability to breastfeed at birth. These results have been shown elsewhere: a recent meta-analysis indicated that early initiation of breastfeeding is low globally [30], and an additional systematic review identified prelacteal feeding as a major barrier to early initiation of breastfeeding in South Asia [31]. Other studies have shown a delay in breastfeeding associated with prelacteal feeding in Ethiopia [23, 32] and western Nepal [33].

In this population, which is demographically representative of rural Bangladesh and similar populations across the greater Gangetic floodplain [27, 28], early neonatal feeding is a common practice [20]. Mothers feeding ENF, and neonates’ inability to breastfeed after birth, are both associated with suboptimal breastfeeding practices. These factors may be interrelated in their respective associations with delayed breastfeeding initiation. For example, neonates who have trouble breastfeeding after birth or who are perceived to be too weak to breastfeed may be given a non-breastmilk food.

This study has at least four limitations. First, the information used in this analysis was derived from interviews performed both during pregnancy and at three months postpartum, and there is a possibility of recall bias. A three month recall bias has been shown to exist in reporting of breastfeeding behaviors [34]. However, this study focused on sensitivity and specificity of self-reported exclusive breastfeeding. The study results indicate that recall bias may overestimate optimal breastfeeding behaviors. Such a bias would reinforce the findings of this study. However, the interviews contained closed ended questions and were conducted by trained interviewers. To limit the possibility of recall bias, we excluded interviews carried out more than 28 days after or fewer than 28 days before the scheduled three month postpartum visit. Excluding interviews where the interview dates were closer to the dates of birth may have removed a subsection of interviews less subject to recall bias, however, this was done in the interest of maintaining internal validity. Women with interview dates similar to each other may be the most likely to give internally consistent interviews. There is a possibility that there may be differential misclassification of the exposure of interest (ENF feeding), based on recall bias about infant difficulty suckling (i.e., women who recall more difficulty suckling also are more apt to remember feeding ENF). However, we believe this possibility to be small due to the high prevalence of ENF feeding throughout the study population. Second, due to the timing of the study interview and the fact that mothers were not asked why they fed ENF, it is impossible to establish the direction of causality in the relationship between delayed initiation of breastfeeding and ENF feeding. Third, the participant interviews used in this analysis were conducted from 2001–2007, and may not describe more recent trends in early newborn feeding and early initiation of breastfeeding.

Finally, given the very large sample sizes of this analysis, we are powered to detect even small differences in the time to initiation of breastfeeding, including those that may be less clinically relevant. However, because of the ubiquity of ENF behaviors across populations in South Asia, rarely have studies been able to compare the attributes of ENF feeding with a very large group of individuals who did not provide their newborn ENFs. This statistical power provides a unique opportunity to identify both major and minor risk factors for delayed initiation of breastfeeding in this context, and allows us to be more confident about the lack of type II error in our statistical analysis.