Data source and study sites
This analysis used data derived from the BADUTA study conducted in 2015-2016 in Sidoarjo and Malang District of East Java, Indonesia. The Ministry of Health, Republic of Indonesia identified Malang and Sidoarjo Districts as the study sites for evaluating the BADUTA program since they represented peri-urban and rural areas of East Java Province. In both districts, we selected six sub-districts to conduct the trial. The sub-districts in Sidoarjo District were Tulangan, Wonoayu, Sidoarjo, Prambon, Taman, and Krian; and in Malang District were Dampit, Turen, Tumpang, Poncokusumo, Gondanglegi, and Jabang.
We have presented detailed information about the BADUTA study protocol elsewhere [20]. We used data for this analysis from two independent cross-sectional surveys conducted in 2015 at the beginning, and in 2017 at the end of the project. To assess breastfeeding self-efficacy amongst mothers, we only used information collected from mothers of children less than six months of age.
Background information on study sites
East Java Province is one of the provinces in Indonesia located in Java Island, and the capital is Surabaya City, the second-largest city in Indonesia. The total population of East Java is approximately 37 million people, the second-most populous province in the country [21]. Malang District, located in the centre-south region of East Java Province, has an estimated total population in 2017 of 2,576,596 people [22]. Most of the people were working as labourers or private employees (37.63%) [22]. Sidoarjo District, located north of Surabaya City, has an estimated total population in 2017 of 2,207,600 people [23].
Study design and samples of the study
We conducted an observational epidemiological study to examine factors associated with low breastfeeding self-efficacy. For our analysis, we combined the data from the baseline and endline cross-sectional surveys, for both the intervention and comparison groups in the BADUTA study.
The sampling design in this trial used a three-stage cluster sampling procedure. At the initial stage, in each of the 12 subdistricts selected, 10 villages were selected using the probability proportionate to size sampling method. Next, we selected two sub-villages from each chosen village using simple random sampling method. Finally, in each of the selected sub-villages, we conducted a mini census to list all children aged <2 years, and using the listing as a sampling frame we selected 12 children aged <2 years and their mothers using simple random sampling.
In the baseline survey of the BADUTA study, the sample size for children under two years old was 2435 children; while in the endline survey, the sample for children under two years old was 2740 [20]. For this analysis, we only used information from 1210 women with children under six months (575 from the baseline and 635 from the endline survey).
Survey instruments and field personnel
The field team carried out house-to-house interviews using pre-tested and structured questionnaires. The information collected in this study included socio-economic and demographic characteristics; infant feeding practices as well as the intention of the mother to breastfeed and self-efficacy for breastfeeding of the mothers; child morbidity, reported by mother/caregiver; as well as contact with the health system and exposure to the interventions. Information about the mothers' self-efficacy for breastfeeding was collected using the Breastfeeding Self-Efficacy Scale-Short Form questionnaire developed by Dennis [15], a 14-item instrument aimed at measuring breastfeeding confidence.
At the baseline, we established eight fieldwork teams in each district. However, in the endline study, we established ten fieldwork teams to shorten the duration of data collection. For interviews, each team consisted of one field coordinator, one assistant field coordinator, and ten enumerators. In total, there were 10 field coordinators, 130 interviewers, and 20 nurses or midwives recruited [20]. The nurses and midwives conducted blood samples collection and took anthropometric measurements.
Before data collection, all field workers attended a one-week training program to standardize the enumerators, and their coordinators, with the questionnaire, sampling methodology, as well as interview techniques. The training covered different aspects of the study, i.e., an overview of the BADUTA study, the use of CommCare application; household listing and data collection procedures; explanations of study instruments (listing forms and questionnaires); quality controls for data collection; as well as a field plan. The training program included a two-day tryout to allow all training participants to practice the household listing and interviews using CommCare application. A pre and post-test was also carried out before and after the training sessions, respectively. Enumerators with low post-test results were monitored closely by field coordinators and supervisors, particularly at the beginning of data collection, to ensure their ability and quality to conduct all fieldwork activities.
Data were collected electronically on hand-held devices using CommCare system developed by Dimagi [24]. Information was recorded on structured, error detecting forms on tablets and then dispatched directly to a server for cleaning and merging. Field supervisors and a data manager monitored the data quality regularly.
Outcome variable
The outcome variable in this analysis was mothers' self-efficacy for breastfeeding as a binary variable (low or high self-efficacy on breastfeeding). We defined breastfeeding self-efficacy as the mothers' beliefs and confidence in their ability to breastfeed their infants successfully. Information about the mothers' self-efficacy for breastfeeding was collected using the Breastfeeding Self-Efficacy Scale-Short Form [15]. For each of the 14 statements, we asked the mothers to give a score from 1 to 5 that offered a range of answer options from "strongly disagree" to "strongly agree," respectively. We added all the scores to calculate the total score. As in other studies, we based the classification of breastfeeding self-efficacy on the median of the total score [25, 26]. Previous studies supported using either the mean and or the median as the cut-off point to categorize low and high breastfeeding self-efficacy [26-28]. We classified mothers scoring less than the median as having a low self-efficacy on breastfeeding, and those scoring equal to or above the median as having a high self-efficacy.
Potential predictors
The potential predictors were selected using the analytical framework shown in Figure 1. In total, there were 17 potential predictors of breastfeeding self-efficacy included in the analyses, categorized into six sub-groups: (1) context/demographic variables; (2) household characteristics; (3) maternal characteristics; (4) child characteristics; (5) breastfeeding characteristics; and (6) antenatal and delivery care characteristics.
In the group of contextual and intervention variables, we constructed a composite variable indicating the total number of interventions from 13 variables representing breastfeeding-related interventions in the BADUTA study. Those 13 interventions were: (1) discussing breastfeeding with cadres on a home visit during pregnancy; (2) discussing exclusive breastfeeding in pregnant women's class during pregnancy; (3) did not receive any free formula milk after delivery (part of the Baby Friendly Hospital Initiative); (4) discussing breastfeeding with a village facilitator during pregnancy; (5) watching a breastfeeding-related video shown by the village facilitator during pregnancy; (6) discussing the topic of breastfeeding in emo-demo sessions; (7) receiving mobile phone messages on early initiation of breastfeeding; (8) receiving mobile phone messages on the benefits of colostrum; (9) receiving mobile phone messages on exclusive breastfeeding; (10) receiving mobile phone messages on exclusive breastfeeding problems and how to handle them; (11) receive breastfeeding counseling by midwives during pregnancy; (12) receive breastfeeding counseling by cadres during pregnancy; and (13) watching TV commercials about breastfeeding. For each question, we scored the answers "1 (one)" if the mothers answered "yes," and scored "0 (zero) if answered otherwise. We then summed all the scores to obtain a total intervention score. We then categorized the total intervention score for each individual into "no intervention" (total score = 0); "one intervention" (total score = 1); "two interventions" (total score = 2), and "three or more intervention" (total score is ³3). Finally we calcualted the total intervention score for all women from both the intervention and comparison groups included in this analysis. Our purpose was to assess the impact of any breastfeeding intervention, whether from the study interventions or routine programs on breastfeeding self-efficacy. We have documented a detailed explanation of the interventions in the BADUTA study elsewhere [20].
In household characteristics, we constructed the household wealth index variable using Principal Component Analysis (PCA) [29] from an inventory of the household's facilities and assets. These items included ownership of electricity, drinking water, toilet facility, type of toilet facility, fecal final disposal and ownership of bicycle, television, water heater, 12kg of LPG, fridge, and car. We ranked households by this index and classified them into five quintiles, i.e., poorest, poor, middle, rich, and richest categories of households.
In the breastfeeding knowledge and experience group, we developed one composite variable to represent mothers' level of knowledge about breastfeeding. We constructed this variable from five questions: (1) the best food or liquid to be provided to children aged < 6 months; (2) the duration for exclusively breastfeeding a child; (3) the duration a child should receive breast milk; (4) the benefits of giving breast milk to children; and (5) the time a child should receive complementary feeding. For each question, score "1 (one)" was assigned to all correct answers, and "0 (zero) for all incorrect answers. We summed all the scores to get the total knowledge score, and we calculated the median value. Two categories of knowledge were developed: (1) high level of knowledge, for those whose total knowledge score was greater or equal to the median, and (2) a low level of knowledge for those whose total knowledge score was less than the median. To test if previous experience with feeding infants influenced breastfeeding self-efficacy, we also used an indicator for previous live births as we did not specifically ask the mothers about their earlier breastfeeding experiences. For the variable of problems with breastfeeding, we categorized mothers into four groups: (1) Mothers who did not experience any problems with breastfeeding; (2) Mothers who had breastfeeding problems not related to illness; (3) Mothers who had breastfeeding problems related to illness or anatomical conditions; and (4) Mothers who had both types of problems. We categorized mothers who mentioned their breastmilk was insufficient, or they could not express it, or the infant refused breastfeeds as "breastfeeding problems not related to illness. We categorized mothers with problems due to swollen breasts/mastitis, sore nipples, or flat/embedded/large nipples as a "problem related to illness/anatomical conditions". We categorized mothers reporting both types of breastfeeding problems as "mothers who had both types of problems.
Data analysis
To examine the characteristics of all variables (outcome variables and potential predictors) used in the analysis we used contingency tables. We then applied logistic regression analyses to determine factors associated with all outcome variables using ORs (odds ratios) as the estimated measures of association. We used Stata survey commands (svyset) to adjust for the clustering from the cluster randomization. All estimates presented in this analysis considered the complex sample design.
In the first step of logistic regression, we used bivariate analyses to assess the relationship between outcome variables and their potential predictors independently. In the second step, we performed multivariate analyses using a backward elimination method to remove all variables not significantly related to the study outcome, with a significance level of 0.05. Two variables selected a priori and retained in the final model regardless of the significance level, were: (1) Period of the survey (baseline or endline) and (2) the fulfilment of the minimum requirement of four antenatal care visits by trimester (met or did not met). In the final model, we obtained the adjusted ORs (aOR) and 95% confidence intervals (95% CIs) for all variables in the model.
In multivariate analysis, we used problems of breastfeeding and the number of breastfeeding interventions as composite variables. After obtaining the final model (Model #1), we developed the second model by replacing problems of breastfeeding with each type of breastfeeding problem (Model #2). We also developed the third model by replacing the breastfeeding intervention variable with all the individual exposure to intervention indicators (Model #3). We then retained the other variables in the final model of Model #1 in Model #2 and Model #3. We used Stata/MP software (version 13.1; Stata Corp) for all analyses.
Collaborating institutions
This study was conducted by an International Research Consortium that comprised of experienced researchers from the University of Sydney (Australia), the London School of Hygiene and Tropical Medicine (LSHTM) (United Kingdom), the Center for Health Research Universitas Indonesia (CHR-UI), the Indonesia Nutrition Foundation for Food Fortification (KFI), and the Southeast Asian Ministers of Education Organization (SEAMEO), Regional Center for Food and Nutrition (RECFON).