Determinants of postnatal care use at health facilities in rural Tanzania: Multilevel analysis of a household survey

Background: Postnatal care (PNC) for the mother and infant is a neglected area, even for women who give birth in a health facility. Currently, there is very little evidence on the determinants of use of postnatal care from health facilities in Tanzania. Methods: This study examined the role of individual and community-level variables on the use of postnatal health services, defined as a check up from a heath facility within 42 days of delivery, using multilevel logistic regression analysis. We analyzed data of 1931 women, who had delivered in the preceding 2-14 months, from a two-stage household survey in 4 rural districts of Morogoro region, Tanzania. Individual level explanatory variables included i) Socio-demographic factors: age, birth order, education, and wealth, ii) Factors related to pregnancy: frequency of antenatal visits, history of complications, mode of delivery, place of delivery care, and counseling received. Community level variables included community levels of family planning, health service utilization, trust, poverty and education, and distance to health facility. Results: Less than one in four women in Morogoro reported having visited a health facility for postnatal care. Individual-level attributes positively associated with postnatal care use were women’s education of primary level or higher [Odds Ratio (OR) 1.37, 95 % Confidence Interval (CI) 1.04-1.81], having had a caesarean section or forceps delivery (2.95, 1.8-4.81), and being counseled by a community health worker to go for postnatal care at a health facility (2.3, 1.36-3.89). Other positive associations included those recommended HIV testing for baby (1.94, 1.19-3.15), and whose partners tested for HIV (1.41, 1.07-1.86). High community levels of postpartum family planning usage (2.48, 1.15-5.37) and high level of trust in health system (1.77, 1.12-2.79) were two significant community-level predictors. Lower postnatal care use was associated with having delivered at a hospital (0.5, 0.33-0.76), health center (0.57, 0.38-0.85), or dispensary (0.48, 0.33-0.69), and having had severe swelling of face and legs during pregnancy (0.65, 0.43-0.97). Conclusions: In the context of low postnatal care use in a rural setting, programs should direct efforts towards reaching women who do not avail themselves of postnatal care as identified in our study.

The target population for our research was rural women who had a childbirth in the preceding 2–14 months and referred to as “recently delivered women” (RDW). The household survey of women (N = 1968) was carried out from August 2011 to November 2011. The baseline household survey was conducted as part of an evaluation of the Integrated Facility and Community (IFC) program being implemented by the MoHSW with support from Jhpiego since 2008. The principal objective of the survey was to establish baseline household characteristics and MNCH care utilization among RDWs. The study sample for the survey was RDWs living in rural areas of four districts in the Morogoro Region of Tanzania- Morogoro District Council, Mvomero, Kilosa and Ulanga. We performed post-hoc power analysis for our sample size of 1931 women – among the respondents, 37 (2 %) were excluded from the analysis because they had experienced pregnancies that resulted in a miscarriage and did not have reason to access postnatal care. The sample size would have 80 % power to detect a 10 % difference in the proportion of women using postnatal care between two groups with a type 1 error of 0.05, assuming an intra class correlation (ICC) of 0.1. A two-stage sampling strategy was employed to select households from 60 clusters, − a ‘cluster’ being defined as a unit of 1000 population. The first stage was the selection of villages (containing the cluster) through probability proportional to size (PPS) sampling using population estimates from the 2002 Tanzania National Population Census. In each selected village, a list of the population and households was made for the constituent sub-village units (Vitongoji) based on data from the local government authorities which were then grouped to create clusters composed roughly of a population of 1000. The second stage of the sampling process was to choose one cluster in a random manner by lottery. In each cluster, the survey team visited every household to list and interview women who had any pregnancy outcome (live born/stillbirth/abortion) in the preceding 2–14 months. The probability of selection for each household in the sample was equal. If the household had more than one eligible woman, only one was randomly selected for interview. The respondent identified in each household was interviewed using a questionnaire adapted from the model questionnaires developed by the MEASURE Demographic & Health Surveys (DHS) program [18]. The questionnaire was adapted to reflect relevant issues related to the larger ongoing evaluation in the region and collect information on background characteristics, pregnancy history, utilization of health care during pregnancy, delivery, and postnatal period as well as barriers to care seeking. If multiple births were encountered, then it was considered as a single pregnancy event. The adapted questionnaire was translated from English into Kiswahili and pilot tested in a district adjacent to the study site (Pwani Region). The respondents were interviewed after obtaining written informed consent. Two teams of trained interviewers fluent in Swahili and English administered the survey questionnaire. A field editor reviewed questionnaires from all the teams using a checklist for completeness, quality, and consistency at the end of each day while the study investigators made periodic checks to ensure quality of data collection and entry. We framed the use of postnatal care adapting the healthcare-seeking behavior model developed by Anderson and Newman [19]. This model proposes that the use of health care services is a function of three sets of characteristics – predisposing characteristics, enabling characteristics, and need characteristics (Fig. 1). We included (i) predisposing characteristics such as age, parity, marital status, education, wealth index, community poverty and peer usage of services, (ii) enabling characteristics such as distance to facilities, cost of services, trust in health system and community outreach activities, and (iii) need characteristics such as perceived susceptibility, seriousness of complications during antenatal, delivery and postnatal period, mode of delivery and need for contraception, and HIV testing. The outcome variable was ‘use of postnatal care’, which was defined as attending postnatal care for the mother’s care at a dispensary, health center, or hospital (government or private) within 6 weeks of delivery. Individual-level explanatory variables included demographic variables such as woman’s age, birth order, education, marital status, and religion. An index of household wealth, based on household assets was created using principal components analysis (PCA) methods proposed by Filmer and Pritchett and used to group households into wealth quintiles [20]. Frequency of antenatal care visits and location of delivery care were included as measures of health system utilization. The variable ‘HIV testing of baby’ refers to the requirement for HIV counseling and testing of the baby during a postnatal visit and functions as a ‘need characteristic’ variable. The ‘partner test for HIV’ variable is proxy for the involvement of men in the process. If expenses were incurred for a delivery, it was coded as a dichotomous variable ‘money spent on delivery’ which is a proxy barrier to care seeking. To create community-level variables, the individual level responses were averaged at the level of the clusters. The asset score of the households in the cluster was averaged to generate a community poverty score that served as proxy for wealth of the community. The proportion of women in the cluster with primary or higher education was used as proxy for literacy of a community. The proportion of women using contraception and the proportion attending 4 or more antenatal visits were used as proxies for community family planning practices and maternal health service utilization. The community level variables for education, contraceptive prevalence and ANC4+ coverage were generated by assigning each cluster with the value of the prevalence of women for the indicators and dividing them into low, middle and high categories. Communities with more than 80 % of respondents who reported trusting a health provider or CHW for advice on pregnancy related issues were classified as communities with high trust. Survey teams, also, collected information on the presence of the nearest functioning health facility and the distance recorded. The distance variable was categorized as 0 km (facility in the village), less than 5 km and more than 5 km. Frequency distributions of the sample women were explored to describe the characteristics of women included in this study. Bivariate and multivariable analyses were performed for individual-level and community-level variables of interest. Multilevel models take into account the hierarchical structure of the data and clustering of responses at the different levels. The following equation illustrates the multilevel model for utilization of Postnatal Care where i and j are the level 1 (individual) and level 2 (community) units respectively; pij is the probability of the outcome of interest for woman i in the cluster j; the b’s are the fixed coefficients; I and C refer to individual-level and community-level explanatory variables, respectively; and ZjXij is a cross-level interaction term; μ shows the random effects for the jth cluster. The error term, ε, represents unmeasured factors that may influence use of postnatal care at a health facility. A multilevel random intercept logistic regression model without covariates (null model) was used to assess the influence of unobserved community-level characteristics on the overall variation in facility use. Three multilevel random intercept logistic regression models were fitted to estimate associations between the individual and community variables and the likelihood of seeking postnatal care at a health facility. The first model included individual-level characteristics only, the second model included community-level characteristics only and the last model (full) includes individual and community-level variables. The independent variables were retained in each of the models if the the p-value was less than 0.2. Important demographic variables were also retained in the multivariable models, in addition to the variables chosen from the bivariate analysis, based on previous literature on the use of maternal health services. All statistical analyses were carried out with STATA 13.1[21]. The extent of missing data was assessed, and patterns of missingness were explored in order to ensure whether data were missing at random. Approximately 15% of our sample was missing data for at least one variable in the final multilevel model. We used multivariate imputation using chained equations (MICE), which uses a Gibbs-like algorithm to impute multiple variables sequentially using univariate fully conditional specifications. This method is considered more appropriate for the imputation of categorical data because it does not assume normality of imputed variables [22]. Independent variables with more than 2% of values missing were imputed, and 50 concatenated datasets were created to reduce any potential bias caused by rounding. Variables imputed included the outcome variable of use of postnatal care, complications during pregnancy, delivery and postnatal period, counseling from CHW regarding postnatal care, mode of delivery, money spent on delivery, partner test for HIV and HIV testing of baby. The full multilevel model was subjected to sensitivity tests to estimate the impact of missing data using multivariate imputations for the independent variables. The percentage change in the standard error for all independent variables was less than 0.5 % and the estimates derived from the imputed model are used. Ethical and administrative approvals were obtained from the Ethics Review Committee of the Johns Hopkins University and the Muhimbili University of Allied Health Sciences, Dar es Salaam, Tanzania. Written informed consent was obtained from each participant.

N/A

• Dima Health