Factors associated with delayed initiation of breastfeeding: a cross-sectional study in South Sudan

This study was conducted at the maternity ward in Juba Teaching Hospital (JTH) in South Sudan. This is a national referral and teaching hospital for the College of Medicine, University of Juba. The hospital has a capacity of 500 beds and conducts, on average, 20 births daily.

This was a cross-sectional study among 806 mother-infant pairs conducted from October 2016 to January 2017. Mothers who gave informed written consent, had a term birth and live infant were included. We excluded mothers who had ill infants; and those unable to communicate independently.

This study considered a 50% prevalence of early initiation of breastfeeding in sub-Saharan Africa from a recent descriptive analysis in 57 low-and middle-income counties [21], with a precision of 5%, 95% confidence intervals and non-response of 20%. This gave us sample size of 461 participants. This sample size was determined using Open Epi [22].

We also calculated a sample size for factors associated with delay in initiation of breastfeeding. Initially, the sample sizes of various exposures were computed using Open-Epi [22] epidemiological calculator for detecting differences between proportions of two groups (Kelsey formula). Finally, discarding of colostrum was used to calculate the sample size needed for detecting variations between proportions of delay in initiation of breastfeeding between mothers who discarded colostrum and those who did not. Using a study from the Somali region of Ethiopia where 46.3% of the mothers discarded colostrum, while 53.7% did not [23], we assumed the proportion of delay in initiation of breastfeeding of 59% for mothers who discarded colostrum and 49% for those who have not discarded. This gave a sample size of 806; which was enough to detect differences in the proportions of the exposures of interest with 80% power.

A total of 1723 mothers gave birth in JTH, of these 13 had stillbirths; and 94 mothers were enrolled in another study. The study objectives and procedures were explained to the 1616 mothers eligible to participate in the study. In the design of the study, we planned to recruit 10 of the 20 mothers who gave birth daily. However, consecutive sampling would have favored those mothers who gave birth in the day and left out those who gave birth at night; who may have different characteristics. It was also not practical to randomly sample 10 mothers after all births had occurred in a day, because most mothers leave the hospital few hours after birth. Therefore, we decided to randomly sample one of every two mothers who gave birth. This process spread the data collection throughout the day. So, for every two births, the mothers were requested to pick a concealed piece of paper from a box. Those who picked a “yes” were assigned to “participate” and those picking the “no” option to “not to participate” in the study (Fig. 1). Of the 808 mothers who picked to participate in the study, 806 consented to participate in the study; while two mothers with ill infants declined consent.

We used a questionnaire generated from the World Health Organization (WHO) indicators for assessing infant and young child feeding practices [24] and a study in Uganda [25]. The questionnaire was translated to Bari (the local language) and back translated into English by a different translator for validation. Next, it was piloted and pretested in Al Sabbah Children Hospital in Juba. After the pilot test, irrelevant questions were removed, vague questions made clearer and errors corrected. The questionnaire had three sections: sociodemographic characteristics, birth characteristics, and breastfeeding practices. The data were collected by a pair of trained research assistants in a face to face interview. The collected data were checked daily for completeness, consistency and errors by the principal investigator (PI). The data were cleaned, coded, double entered into Epi Info version 6 and stored safely in a password-protected computer to which the PI had sole access.

The outcome variable was timely or early initiation of breastfeeding. The mothers were asked how long after birth their babies were first put to the breast. Initiating breastfeeding within one hour was categorized as “early”, whereas later than one hour was categorized as “late /delayed”. The independent variables have three sections:

In the present study, we reported the descriptive variables as means and standard deviations. The categorical variables were reported as proportions. Chi square tests were performed to identify the independent variables that were associated with outcome variable of interest and recorded their p - values. Factors from the literature known to predict our outcome variable and those with a p - value ≤0.25 at binary analysis, not in the causal pathway and not strongly collinear with other independent variables were entered into the initial multivariable logistic regression model. We evaluated for collinearity and any factor with variance inflation factor (VIF) > 10 was regarded as strongly collinear. In instances where there was collinearity, the factor with a stronger measure of association with the outcome variable was retained in the model and the other dropped. The variables with a p - value ≤0.25 from bivariate analysis were then entered into a backward stepwise multivariate model. Next interaction terms were formed between the variables retained in the multivariate model. The reduced model was then compared with the model with the interaction terms using chunk test to assess for interaction. We evaluated the confounding effect for the variables dropped from the model on the variables that remained in the multivariate model. Any variable that caused a variation ≥10% between the crude and adjusted degrees of association of any of the variables that remained in the multivariate model was kept in the model as a confounder. In the final model, independent variables with a p - value < 0.05 were regarded key associated factors. Finally, we carried out the multivariable logistic regression analysis without discarding of colostrum. Since this variable could be a consequence and not a cause of the delay in initiation of breastfeeding. For example, discarding of colostrum could result from delay in initiation of breastfeeding due to mother’s engorged breast or infant illness. To control for this situation, we repeated the analysis without this variable. Data analysis for the current study was conducted using STATA version 14 (Stata Corp LLC, Texas, USA).

The mean age of the mothers was 24.9 years (SD 5.3) (Tables 1 and 2). Of all the mothers, more than half 51% lived in urban area, 46% owned houses, 96% were married, 17% were uneducated, and 76% were unemployed. The majority, 91% of the mothers attended antenatal care (ANC), 88% gave birth normally, 55% received breastfeeding counseling, and 88% were not exposed to infant formula advertisement a month before birth.

More than half 52% (418/806) of the mothers delayed initiation of breastfeeding. Place of residence, marital status, no house ownership, ANC, mode of birth, birth order, type of birth, discarding of colostrum, and exposure to infant formula advertisement were associated with delay in initiation of breastfeeding (Table 3). The variables which were not associated at this level were excluded in final multivariable analysis.

In this analysis, birth by Caesarean section (Adjusted odd ratio [AOR] 41; 95% confidence interval [CI] 12.21, 138), discarding of colostrum (AOR 9.89; 95% CI 4.14, 23.62), unmarried mothers (AOR 3.76; 95% CI 1.53, 9.24), exposure to infant formula advertisement (AOR 1.82; 95% CI 1.09, 3.02), and no house ownership (AOR 1.52; 95% CI 1.11, 2.09) were associated with delay in initiation of breastfeeding (Table 3). The repeated model without discarding of colostrum found the same results with similar measure of association.

This study was the first to assess early initiation of breastfeeding in a unique conflict setting in South Sudan. The data were gathered within 24 hrs after birth which minimized recall bias. This study had several limitations; the result of this study cannot be generalized since one hospital was studied. This study omitted other methods of infant formula advertisement such as points of sale, social media promotion, internet, health staff role and public display. In addition, the information on infant formula advertisement could be subject to recall bias. Finally, the mothers’ ability to estimate exact time on initiation of breastfeeding was not assessed.