Female migrants, family members and community socio-demographic characteristics influence facility delivery in Rufiji, Tanzania

Background: Health professionals and public health experts in maternal and newborn health encourage women to deliver at health facilities in an effort to reduce maternal and newborn mortality. In the existing literature, there is scant information on how migration, family members and community influence facility delivery. This study addresses this knowledge gap using 10 years of longitudinal surveillance data from a rural district of Tanzania.Methods: Multilevel logistic regression was used to quantify the influence of hypothesized migration, family and community-level factors on facility delivery while adjusting for known confounders identified in the literature. We report adjusted odds ratios (AOR). Results: Overall, there has been an increase of 14% in facility delivery over the ten years, from 63% in 2001 to 77% in 2010 (p < .001). Women residing in households with female migrants from outside their community were more likely to give birth in a facility AOR = 1.2 (95% CI 1.11-1.29). Furthermore, the previous facility delivery of sisters and sisters-in-law has a significant influence on women's facility delivery; AOR = 1.29, 95% CI 1.15-1.45 and AOR = 1.7, 95% CI 1.35-2.13 respectively. Community level characteristics play a role as well; women in communities with higher socioeconomic status and older women of reproductive age had increased odds of facility delivery; AOR = 2.37, 95% CI 1.88-2.98 and AOR = 1.17, 95% CI 1.03-1.32 respectively.Conclusion: Although there has been an increase in facility delivery over the last decade in Rufiji, this study underscores the importance of female migrants, family members and community in influencing women's place of delivery. The findings of this study suggest that future interventions designed to increase facility delivery must integrate person-to-person facility delivery promotion, especially through women of the community and within families. Furthermore, the results suggest that investment in formal education of the community and increased community socio-economic status may increase facility delivery. This study utilizes delivery records of women who gave birth from 2001 to 2010 in the Rufiji Health and Demographic Surveillance System (RHDSS). The surveillance area was established in 1998 with an initial census conducted by enumerating the entire population. The surveillance system covers an area of 1,813 square kilometers with 33 communities (villages). The surveillance area, and lies about 178 km south of Dar es Salaam (the largest city in the country), and covers approximately 50% of the population of Rufiji District. The study area is geographically formed by 33 administrative villages which will be referred to as communities and serve as a unit of analysis in this study. Dominant ethnic groups are Ndengereko, Makonde and Sukuma. The surveillance system is managed by Ifakara Health Institute (IHI). Rufiji district was the site of two projects aimed at increasing facility delivery. From 1998 – 2003, the Ministry of Health implemented the Tanzania Essential Health Intervention Project (TEHIP). The goal of TEHIP was to increase district capacity to provide expectant mothers with access to basic obstetric and reproductive care. The project provided management tools such as training in management skills, team building, planning, and introduced management systems to help the district focus resources on areas in greatest need [24]. Following TEHIP, the Empower project (2006 – present) upgraded facility infrastructure and supplied equipment to increase access to comprehensive emergency obstetric care. It has trained staff to improve their service provision and increase retention. The population under surveillance in Rufiji is a dynamic cohort that enrolls individuals through birth and in-migration and who exit through death or out-migration. Each enumerated individual is given a unique identification number that is used to longitudinally track demographic events. Field teams visit each registered household three times annually to record dates of births, deaths, in and out -migrations and marital status changes that have occurred to each individual since the previous visit. Birth registration includes detailed information about mother’s place of delivery and assistance during delivery. More details about the surveillance system can be found elsewhere [25]. This paper uses data on all deliveries to women aged 15–49 between 2001 and 2010. A total of 20,049 deliveries were included in the analysis. Although the RHDSS started in 1998, data collection on place of delivery began in 2001. The dichotomous outcome variable compares delivery in a facility to delivery at home. The annual percent of facility deliveries was calculated by the ratio of facility deliveries with all deliveries recorded in the surveillance area. The change in the percent of facility delivery was calculated as difference in proportion in 2001 and 2010 at 5% significance level. The predictor variables of interest were constructed as follows. Presence of an in-migrant in a household was captured from the dataset documenting in and out-migration in the surveillance area. We construct a binary measure indicating presence of an adult female in-migrant in a household two years prior to delivery. Sisters’ place of delivery was constructed by first identifying sisters through matching parent identification numbers, and then determining place of delivery for sisters’ previous births using the birth registry data. We then construct a count variable for how many times she has given birth in a facility. The measure of sisters-in-law’s place of delivery is similarly constructed; women are linked to their husbands, and we then use husband’s identification number to link him to his sisters through matching parent identification numbers. We included sisters-in-law’s experience with facility delivery because Rufiji is a patrilineal society, and women are likely to reside in their husbands’ home village. To capture the influence of the community, we calculate the mean age of women of reproductive age, mean household wealth, and mean years of education of women of reproductive age by village using all residents in the village at the time of birth. Control measures were constructed as follows. Wealth measures were constructed using an asset index constructed through principal components analysis, based on asset ownership (such as bicycle, radio, bed) and structural characteristics of the home (such as roofing and building material) [26]. Asset data are collected annually. Maternal characteristics of education, work (paid/other) and marital status were captured at baseline or during routine updates of household data. Parity, multiple births, season of birth, place of delivery and outcome of every pregnancy were extracted from birth registration data. Distance to the nearest health facility was calculated from GPS points mapped at each household and health facility in the surveillance area. Mapping of geographic boundaries was conducted to locate all 33 communities and main roads. Proportion of facility delivery for each community was added to the surveillance map to geographically explore the facility delivery in relation to village main road connections. Exploratory analysis was conducted by plotting annual proportions of facility delivery (number of reproductive aged women who delivered at the facility/total number of deliveries) by socio-economic status, education, previous birth outcome and previous place of delivery. Multilevel logistic regression was used to estimate the hypothesized influence of in-migrants, family and community-level factors on facility delivery while adjusting for known confounders including mother’s age, birth order, season of birth, place of delivery of previous pregnancy, and marital status. Multilevel models were specified where children are nested within their mothers and mothers within communities. Mother and community level random effects were included because many individual and community-level factors that may influence facility delivery are not collected or are unobserved, such as husband’s approval of facility delivery and tribe. Maximum likelihood was used to estimate regression model coefficients and results are presented as odds ratios with 95% confidence interval. The study protocol was approved by Ifakara Health Institution Internal Review Board, National Institute of Health Research and Commission of Science and Technology, Tanzania. All participants were explained the demographic surveillance system study protocol and were enrolled through informed consent. Our research had adhered to the STROBE guidelines for observational cohort studies (Additional file 1).