Ethics statement
We used anonymized survey data that were available for academic use ethical approval for this study was not required. We obtained approval to use the data from ICF International Rockville, Maryland, USA in August 2016.
Data source
We analyzed secondary and cross-sectional data from the 2014 BDHS. This survey was conducted from June to November 2014 by Mitra and Associates. Details of this population-based survey including survey design methodologies, findings, and questionnaires have been described elsewhere [8].
Three types of questionnaires were used in the 2014 BDHS: a household questionnaire a women’s questionnaire, and a community questionnaire. Information from ever-married women aged 15-49 years was collected by the women’s questionnaire. A total of 164 field workers were recruited based on experience, education level, and their willingness. Then, they were trained to conduct the oral interview [8].
Women were asked questions on the following topics: background characteristics (e.g. age, education, religion, media exposure), reproductive history, use and source of family planning methods, antenatal care, delivery care, postnatal care, newborn care, breastfeeding and infant feeding practices, child immunizations and illnesses, marriage, fertility preferences, husband’s background and respondent’s work, and awareness of AIDS and other sexually transmitted infections [8].
Sample design
To represent the demographics of women across the country the sample of the BDHS survey was nationally representative. The survey used a sampling frame from the list of enumeration areas (EAs) of the 2011 Population and Housing Census of the People’s Republic of Bangladesh, provided by the Bangladesh Bureau of Statistics (BBS) [8].
The survey was based on a two-stage stratified sample of households. In the first stage 600 EAs were selected with a probability proportional to the EA size there were 207 EAs in urban areas and 393 in rural areas. A complete household listing operation was then carried out in all of the selected EAs to provide a sampling frame for the second-stage selection of households. In this second stage of sampling a systematic sample of 30 households on average was selected per EA the sample provided statistically reliable estimates of the key demographic and health variables for the country as a whole for urban and rural areas separately, and for each of the seven divisions. With this design, the survey selected 18,000 households with the expectation of approximately 18,000 completed interviews with ever-married women [8].
The weighted distribution of urban-rural households in the survey was based on the urban-rural distribution of the 2011 population census of the country a modified urban-rural household distribution was reflected by adjusting the sample weights and any significant differences in the overall survey indicators were not expected among the population [8].
Coverage of the sample
Initially 17,989 households were selected based on the weighting of the population distribution of rural-urban place and division. Among the selected households, 17,565 were occupied during the time of the interview, and 17,300 (99%) of the households were then interviewed. A total of 18,245 ever-married women of reproductive age (15-49years) were identified in these households, and 17,863 women were interviewed (98% response rate). Rural and urban areas had similar response rates [8].
Participants
We examined a cohort of women among those who participated in the BDHS we minimized recall bias by including the women who had given birth to at least one child within the last five years which resulted in 6,855 deliveries within this period. For our study, we considered one delivery per woman as a single unit of analysis. In addition, as some women delivered more than once in this 5-year period, we considered their latest delivery as the delivery to be included in the analysis. In this way, 4,468 women delivered at least one child within the previous 5-year period and they were included in our analyses.
Outcome
Women reported on their last childbirth and stated whether attendants were present during their childbirth. In our analysis we used the WHO definition to incorporate deliveries attended by SBAs.
Using this definition, SBAs included qualified doctors, nurses, midwives, family welfare visitors, and community skilled birth attendants [4]. We defined traditional birth attendants, unqualified doctors, relatives, neighbors, and others as unskilled attendants.
We then coded the birth attendant variable according to the assistance of delivery (SBA = 1 and unskilled attendant = 0).
Exposure variables
We selected the following individual fertility and contextual factors based on published reports and data structure of the BDHS: age of the women and their husbands, parity, birth interval, previous history of deceased children, education level of the women and their husbands, occupation of the women, wealth quintile, exposure to mass media (i.e., radio, television or newspaper), receiving antenatal care (ANC) during pregnancy, division, place of residence (i.e., urban or rural), and religion.
Statistical analyses
A contingency table was used to describe the selected background characteristics and to compare women according to the selected characteristics. Simple and multiple logistic regression analyses were applied to calculate crude (unadjusted) odds ratios (CORs) and adjusted odds ratios (AORs) respectively. The odds ratios (ORs) with 95% confidence intervals (CIs) and significance levels (p-value) were reported. Only variables with a predetermined significance level (p < 0.2) in the simple logistic regression were kept in the multiple logistic regression. Variance inflation factors (VIFs) were estimated to check collinearity. High collinearity was assumed for a VIF greater than 10.
The discrete variables (i.e. age, parity, birth interval, number of children) were converted into categorical variables. The wealth index groups were based on household assets and facilities. The ownership of assets (i.e., television, radio, fridge, car, bicycle, and motorcycle), and facilities (i.e., source of drinking water, type of toilet, electricity, and type of building materials used in the place of dwelling) were weighted using a principal component analysis (PCA). Stata 13.0 (Stata Corp, College Station, TX) was used for all data analyses.