Data were collected as part of the 2017 baseline survey for the “Improving Access to Reproductive, Maternal and Newborn Health in Tanzania” (IMPACT) project. The IMPACT project aims to accelerate the reduction of maternal and newborn mortality by addressing major reproductive, maternal and newborn challenges in eight districts of Mwanza Region. This study was conducted in six of the eight districts: Ukerewe, Nyamagana, Illemela, Magu, Sengerema, and Buchosa districts excluding Kwimba and Misungwi districts that had an ongoing similar project (Figure 1).
The present study used a descriptive, cross-sectional, multi-stage design to select eligible households for participation in the survey. With the help of village executive officers, a listing of all households was obtained for each district. This list was used as the master frame from which the number of households were randomly selected for the survey. In the first stage, 30 of 408 villages were selected across the six districts using probability proportional to the size of villages (number of households in the respective villages). The second stage involved random selection of households in each selected village. The required sample size for households was calculated to detect a change of 10% in skilled birth attendants between baseline and study endline, with a 95% level of significance, 0.05 margin of error, and 5.6 crude birth rate, using 2 as a design effect and a 10% non-response rate. This resulted in a required sample size of 1,476 households, of which 1,312 households’ members were present and consented.
No further sampling was undertaken in the sample households; all women in those households aged 15–49 years were eligible to participate. Women present in the household at the time of the visit that consented to participate were interviewed. Data were collected from August to September 2017. A total of 1,612 women met the eligibility criteria but 1,167 consented, resulting in a 72% response rate. Of the 1,167 women who consented, 409 reported a live birth in the preceding 2 years.
Ethical considerations
The baseline survey was approved by the National Institute for Medical Research in Tanzania (registration certificate: NIMR/HQR/R.8a/Vol.IX/2517) and the Institutional Review Board at Aga-Khan University in Dar-es-salaam, Tanzania. The Regional Medical Officer and district reproductive and child health coordinators authorized the survey. Village administrators granted permission to conduct the survey in households in their village of jurisdiction. All survey participants provided oral and written informed consent after receiving an explanation of the purpose of this study, duration of the interview, and their right to refuse or withdraw from the interviews at any time during the study process.
Data collection
All data were collected through face-to-face interviews in either English or Swahili by the enumerators in a private place in the selected household (or in some cases in the homestead) using a translated electronic tool. Data collection was undertaken by teams of trained enumerators who were fluent in both English and Swahili. All enumerators participated in a 6-day training program covering data collection tools, interviewing skills, research ethics, and use of electronic devices (tablets) for data collection. The last day of the training was used for practical exercises in a nearby village, which was not part of the sample used for the baseline survey. Each district had a team of six enumerators. A team lead oversaw data collection and ensured that all data were uploaded onto the server. Data quality during and after the survey was ensured by setting validation checks in the electronic data collection forms, random spot checks on some households, and daily supervision of the data collection process. Any issues identified were discussed the following morning before the start of that day’s data collection.
Measures
The selection of variables was based on a review of relevant literature. Three outcome measures of MHS use were evaluated: number of ANC visits, delivery in a health facility, and postpartum care. Based on the recommended standard of ANC4+ visits during pregnancy, a woman who attended at least four visits received a score of 1. Women that attended none or fewer than four visits received a score of 0. An ANC visit was defined as a woman that reported having visited a nurse, midwife, clinical officer, or medical doctor during their last pregnancy. Health facility delivery was assessed by a question about the place of delivery of the last pregnancy, and was coded dichotomously as 1 if a woman delivered in a formal private/public medical facility with the help of a health professional, or 0 if she delivered at home or on the way to a medical facility. Postpartum care was assessed by asking each woman if a healthcare worker had checked on her health after delivery regardless of the place of delivery. Postpartum care was coded a 1 if a woman received any form of postpartum care (irrespective of number of days after delivery and place of delivery) and 0 if she did not receive a postpartum check-up.
Explanatory variables included: district of residence, maternal age at time of the survey, marital status, and gestational age at first ANC visit, wealth quintile, maternal education, and whether the last pregnancy was wanted. Age was recorded on a continuous scale and later categorized into four groups (15–19, 20–29, 30–39, and 40–49 years). Marital status was collected as a categorical variable (currently married, in-union, and not in-union). The in-union group was merged with the currently married group and coded as 1; the not in-union group was coded as 0. Gestational age at first ANC visit was collected as a continuous variable in weeks and then categorized as first, second, and third trimester/no ANC visit. This was then dichotomized as a first trimester group and a second/third trimester and no ANC group. This allowed comparison of women who had ANC during their first trimester and those who did not attend/had late ANC visits. Wealth index (generated through principal component analysis based on household assets) was grouped into quintiles: poorest, poor, middle, rich, and richest. Maternal education level was dichotomized as primary education or below and secondary education or above. Assistance during delivery was categorized as skilled delivery if the woman was assisted by a nurse, clinical officer, or medical doctor. Otherwise, the delivery was categorized as unskilled. Finally, women that reported that their last pregnancy was wanted were coded as 1, and pregnancies that were not wanted (e.g., mistimed) were coded as 0.
The community level variable identifying district and village of residence was only used in the descriptive table and was not included in the analysis because there was no variation in outcomes at this level as determined by the intra-class correlation coefficient; therefore, it was included in the model as a fixed effect. District of residence was coded to compare urban, rural, peri-urban, and island locations. Magu and Sengerema were classified as peri-urban districts, Ukerewe as an island district, Nyamagana and Illemela as urban districts, and Buchosa as a rural district.
Data analysis
Data were analyzed in three steps. First, data were explored descriptively using frequency (percentages) and median (interquartile range [IQR]). Second, bivariate analyses were completed using chi-square tests to examine associations between individual factors and different outcome measures. This was also used to select variables to be retained in the multivariate analysis. Third, multivariate generalized estimation equation (GEE) regression analysis which is a form of logistic regression for clustered data was conducted to determine the adjusted effects of all explanatory variables on the three outcome variables. Three models were examined to assess the adjusted effects of predictors on ANC4+, facility delivery, and postpartum care. Variables that had a p-value <0.25 in the bivariate analysis were included in the multivariate models. Stata version 12 (College station, Texas, USA) was used for all analyses. Data were de-identified by deleting all personal identifiers before analysis to ensure participants’ anonymity and confidentiality throughout the study.