This was a hospital-based cross sectional study, with a complex sample to represent all births that occurred in hospitals with more than 500 deliveries/year in Brazil (which correspond to 78.6 % of all hospital births), with field work conducted from February 2011 to October 2012. This study, named “Birth in Brazil: national survey into labor and birth”, was coordinated by the Oswaldo Cruz Foundation and the sample was based on the National Information System [9].
The sample design was selected in three stages: in the first stage, hospitals were stratified according to the five Brazilian Regions (North, Northeast, Southeast, Midwest and South), location (state capital and other cities), and type of hospital funding (public, mixed or private), with a total of 30 strata. In this stage 266 hospitals were selected with probability of selection proportional to the number of deliveries in each strata. In the second stage, the number of days needed to interview 90 puerperal women in each hospital was selected by inverse sampling method. In the third stage, the women eligible on each day of the fieldwork were selected. Sample losses due to refusal to participate or hospital discharge were replaced by selecting other puerperal women at the same hospital.
The inclusion criteria to the “Birth in Brazil” survey was: hospital live births with gestational age of more than 22 weeks recorded in the medical file or weight greater than 500 g. All miscarriages were excluded. The sample size was based on a caesarean delivery rate of 46.6 %, to detect differences of 14 % between hospitals, with an alpha of 5 % and power of 95 %, having a minimum of 341 puerperal mothers in each strata.
In total, interviews were conducted with 23,940 women, among 266 hospitals distributed in 191 municipalities, covering all the 27 Brazilian states. Trained field researchers interviewed mothers with an electronic questionnaire in the first 24 h post partum. The questions were related to individual and gestational characteristics, prenatal and delivery care, neonatal characteristics, and breastfeeding at birth. A different questionnaire was applied to the hospital manager. More details about the sample design and field work can be obtained [10].
This study was approved by the Research Ethics Committee of ENSP/FIOCRUZ, under the report n°. 92/2010. Every care was taken towards ensuring privacy and confidentiality of the information. Before each interview was conducted, the interviewee’s consent was obtained, after reading the free and informed consent statement.
The present study investigated the factors associated with breastfeeding at first hour of life (outcome), also denominated ‘timely breastfeeding initiation’, which has been categorized in a dichotomous way (yes, no) based on questions regarding breastfeeding at delivery room and time to initiate breastfeeding. Based on potential conditions that may impede or obstruct breastfeeding at first hour, we have established the following exclusion criteria: mothers tested positive for HIV (according to medical records) and/or with near missing condition [11]; babies who died in the neonatal period; with APGAR < 7 at 5th minute of life; with birth weight <1500 g; and/or gestational age < 32 weeks. In addition, 924 (around 4 %) mothers did not know/answer the questions about breastfeeding initiation, resulting in a final sample of 22,035 mothers and their respective babies.
The exposure variable of being born in a Baby Friendly Initiative Hospital (divided in three categories: yes, in process, and no) was obtained from an interview with the hospital manager and attributed to each mother that had a delivery in that hospital.
Based on a recent literature review [7] and in a conceptual framework [12], we arranged the confounding variables in a hierarchic model, in three distinct levels based on their proximity to the outcome: distal – mother and family socioeconomic characteristics; intermediate – pregnancy and prenatal characteristics; and proximal – related to delivery conditions and newborn characteristics (Fig. 1).
It is important to state that in Brazil we classify race/ethnicity not based on ancestry taxonomy, but based on self-reported skin color/race, according to the official Brazilian Statistic Institute ‘Instituto Brasileiro de Geografia e Estatística’ definitions, 2010.
All analysis has considered the complex sample design, having the mothers that breastfed their children on the first hour of life as reference, implying to interpret the results as the chance to breastfed in the first hour after birth. Initially we estimated the Chi-square test for each variable and the outcome and obtained the unadjusted Odds Ratio (OR) and the 99 % Confidence Interval (99 % CI). To avoid residual confounding, we selected all variables with p-value ≤ 0.20 to compose the modeling process.
In sequence, we estimated logistic regression models, with 99 % CI, following the hierarchic model (Fig. 1) in three steps: first, adjusting all the distal variables together and removing the ones without statistical significance; second, adjusting the intermediate variables along with the remaining distal variables and removing the intermediate ones that did not achieved p < 0.01; third, adjusting all the proximal variables with the remaining ones from the previous steps – only the variables with p-value <0.01 were retained.
