The study was conducted in informal settlements in two eastern municipal wards in Mumbai (M East and L). Both rank lowest on the UN Human Development Index for the city with a comparatively high concentration of slum residency (78 and 85 % respectively), higher infant mortality, lower life expectancy, and lower female literacy and employment. The majority of residents are of Muslim faith . The two wards were included in a cluster randomised controlled trial of community resource centres. Centres served as a base for the collection and dissemination of health information, home visits, care for malnourished children, referral of individuals and families to appropriate services, meetings of community members and providers, and events and campaigns on health issues . Trial areas comprised 40 informal settlements, each of approximately 600 households, and covered a population of ~120 000.
Study design, participants, and tools
We used a sequential mixed-methods design . First, we analysed data from a baseline census to describe determinants of maternity care, then used grounded theory methods to examine women’s choice and utilisation of provider. We used this approach in order to (1) describe the quantitative patterns and determinants of maternity care utilisation, (2) from the quantitative results, purposively select individual women from social, economic, and demographic characteristics and choice of health care provider, (3) explore possible relationships between the observed quantitative patterns and determinants of care, and women’s narratives of care-seeking, and (4) triangulate quantitative and qualitative data.
The research team comprised a principal investigator (TH), a senior data manager (SD), a senior researcher (DO), an experienced male qualitative researcher (GA), two female junior qualitative researchers (KH and SM), SNEHA’s Executive Director of Programs (SP) and the Program Director for the resource centre trial (NSM).
We used two datasets in the study: the trial baseline census for the quantitative analysis and the intervention database to identify participants for qualitative interview. Census respondents were all residents of trial areas and were married women in the 15–49 age group. The actual ages of respondents included in the census ranged from 17–49. The intervention database allowed us to purposively sample individual women based on their care-seeking behaviour and because we did not expect the trial to impact choice of provider. Selection criteria for qualitative interview included married women aged 18 and over who were currently pregnant or had given birth (at home or in a health facility) in the previous two years.
Quantitative data were collected in a baseline census over 18 months from September 2011 to March 2013. All respondents gave signed consent prior to interview. Interviewers took household GPS coordinates and enumerated household members, their ages, schooling and livelihoods. The interview covered duration of residence, assets and amenities, housing fabric and faith. Women provided brief maternity histories and information on family planning.
Data were collected on smartphones running Open Data Kit (www.opendatakit.org), which included inbuilt skips and validation constraints. After checks for completeness, data were uploaded to a secure database in ODK Aggregate. They were cleaned and analysed in Stata 12 (StataCorp, College Station, Tx: www.stata.com).
We used semi-structured topic guides for qualitative data collection, including sections on the respondent’s background (e.g. place of origin, family structure), experiences of pregnancy and childbirth, maternity care, and choice of provider. Women were explained the purpose of the study and assured of confidentiality before giving verbal consent to participate. KH and SM conducted seven focus groups (alternating between moderating and note-taking) with married women (average, eight per group), 16 in-depth interviews, one group discussion with five SNEHA Community Organisers, and an interview with the mother-in-law of two respondents. In total, 78 women from nine clusters participated. Focus groups took place at the nearest nongovernment outreach centre and most interviews in the participant’s home. They were conducted in Hindi or Marathi and lasted from 30 min to over an hour. We stopped data collection when we felt concepts and themes were sufficiently developed.
Focus groups and interviews were digitally recorded and transferred to two password-protected computers. The interviewers anonymised and transcribed their own interviews verbatim and translated them into English for dissemination among the research team. Translated transcripts were randomly selected and cross-checked for accuracy.
We were interested in examining uptake of prenatal and institutional delivery care, whether it was in the public or private sector, and whether women’s choices favoured tertiary public hospitals. We defined prenatal care as attendance for at least three check-ups (the locally recommended minimum). Public sector facilities providing prenatal care included municipal health posts, urban health centres, maternity homes, general hospitals, and tertiary hospitals. We included established, large state government hospitals in the latter group as they provide free or low-cost services. Delivery was possible at all these types of facility except for health posts. Private sector facilities included single-handed practices without inpatient services, small maternity homes and inpatient centres, and larger hospitals. Delivery was possible at all but single-handed facilities without beds.
We chose variables purposively from the available dataset, to reflect socio-economic position (household asset index, maternal schooling), demography (maternal age, parity), establishment and familiarity with healthcare options (duration of residence), and socio-cultural milieu (faith). Maternal schooling was described in an ordered categorical variable as none, primary, secondary, or higher than secondary. Socio-economic position was described by quintiles of an asset index developed from standardized weights of the first component of a principal components analysis [30, 31]. Assets included home ownership, possession of a ration card, robust housing fabric, private water supply, private toilet, finished floor, and possession of a mattress, pressure cooker, gas cylinder, stove, bed, table, clock, mixer, telephone, refrigerator, or television. Duration of residence was a continuous variable describing the number of years the woman had been living in Mumbai. A continuous variable describing parity included the index pregnancy in the preceding two years. Faith was categorized as a binary variable describing Muslim or other faith.
The analyses included women who had reported a birth in the 2 years preceding the census. We tabulated frequencies and percentages of attendance for prenatal care, its location in the private or public sector, and the use of tertiary hospitals and smaller public sector institutions, against the chosen independent variables. We did the same for institutional delivery.
For each combination of dependent and independent variables, we developed a univariable logistic regression model with a random effect for cluster. For prenatal care, whether the woman had 3 or more visits (denominator: all women who had had a pregnancy in the preceding 2 years), whether the prenatal care was in the public rather than the private sector (denominator: women who had made more than 3 prenatal care visits), and whether it was in a large public hospital rather than a smaller one (denominator: women who had made more than 3 prenatal visits in the public sector). For delivery, whether institutional or at home (denominator: women had had delivered in the preceding 2 years), whether it was in the public rather than the private sector (denominator: women who had had an institutional delivery), and whether it was in a large public hospital rather than a smaller one (denominator: women who had delivered in the public sector).
For each outcome, we created a single multivariable logistic regression model with random effect for cluster. All models included adjustment covariates selected as markers of socio-economic position, demography, establishment and familiarity with healthcare options, and socio-cultural milieu. Age and parity were both included in the models since the Stata collin package did not suggest collinearity. All models satisfied quadrature parameters.
We used grounded theory (GT) methods. GT is an inductive research methodology to generate theory through the development of conceptual categories that are grounded in systematically collected and analysed data [32, 33]. We coded the English transcripts in NVivo version 10 (QSR International: http://www.qsrinternational.com).We began by open coding transcripts individually and analysed them collectively to identify and explore descriptive and higher-level conceptual categories. We tested emerging categories and interpretation through constant comparison and presentations to colleagues.
Ethical approval for the study was granted by the UCL Ethics Committee and the Multi-institutional Ethics Committee of the Anusandhan Trust in Mumbai.