Menu
Menu
Menu
Menu
Background: Persistent challenges in meeting reproductive health and family planning goals underscore the value in determining what factors can be leveraged to facilitate modern contraceptive use, especially in poor access settings. In Mali, where only 15% of reproductive-aged women use modern contraception, understanding how women’s realities and health system design influence contraceptive use helps to inform strategies to achieve the nation’s target of 30% by 2023. Methods: Using household survey data from the baseline round of a cluster-randomized trial, including precise geolocation data from all households and public sector primary health facilities, we used a multilevel model to assess influences at the individual, household, community, and health system levels on women’s modern contraceptive use. In a three-level, mixed-effects logistic regression, we included measures of women’s decision-making and mobility, as well as socio-economic sources of empowerment (education, paid labor), intrahousehold influences in the form of a co-residing user, and structural factors related to the health system, including distance to facility. Results: Less than 5% of the 14,032 women of reproductive age in our study used a modern method of contraception at the time of the survey. Women who played any role in decision-making, who had any formal education and participated in any paid labor, were more likely to use modern contraception. Women had three times the odds of using modern contraception if they lived in a household with another woman, typically a co-wife, who also used a modern method. Compared to women closest to a primary health center, those who lived between 2 and 5 km were half as likely to use modern contraception, and those between 5 and 10 were a third as likely. Conclusions: Despite chronically poor service availability across our entire study area, some women—even pairings of women in single households—transcended barriers to use modern contraception. When planning and implementing strategies to expand access to contraception, policymakers and practitioners should consider women’s empowerment, social networks, and health system design. Accessible and effective health systems should reconsider the conventional approach to community-based service delivery, including distance as a barrier only beyond 5 km.
We conducted a cross-sectional household survey in the communities of seven health catchment areas of the rural Bankass district, Mali from December 2016 to January 2017. This survey served as the baseline for an ongoing cluster-randomized controlled trial (trial registration number {“type”:”clinical-trial”,”attrs”:{“text”:”NCT02694055″,”term_id”:”NCT02694055″}}NCT02694055; N = 137 village-clusters) to assess the effects of a proactive approach to community-based healthcare delivery on child mortality and access to care over three years, including access to modern and long-acting reversible contraception (secondary trial endpoint) [21]. Here, we analyze baseline survey data from women of reproductive age to assess contraceptive use before the launch of intervention activities. The Bankass health district is part of the Mopti region in central Mali, approximately 600 km northeast of the nation’s capital, Bamako. The district has a population of approximately 300,000 people and is served by a public, secondary referral hospital located in Bankass, the largest town in the district [22]. It was chosen for the trial in collaboration with the Malian Ministry of Health and Social Affairs based on high under-five mortality and low healthcare utilization in the region, similar to other rural Malian settings [23], as well as few concurrent health interventions in the district and interest from local authorities to collaborate. Within the Bankass health district, the study was conducted in seven (of 22) contiguous, rural health catchment areas: Dimbal, Doundé, Ende, Kani Bozon, Koulongon, Lessagou, and Soubala, an area with a population of approximately 100,000 people. Each health catchment area is served by a public sector PHC. In the context of rural Mali, extended families often live together in family compounds comprised of multiple households. Our survey definition of a household within a family compound was a monogamous or polygynous marital arrangement with or without children and additional relatives, or a single mother with or without additional relatives. All women aged 15 to 49 years permanently residing in the study area with no plans to leave during the trial period and who provided consent or assent were eligible to participate in the women’s questionnaire component of the household survey. From the present analysis we excluded all women who reported being pregnant at the time of the survey (N = 2022) or who reported having reached menopause or having had a hysterectomy (N = 299). We adapted our household survey instrument (see Additional File 1) from the Mali DHS and programmed it in Open Data Kit. The survey captured detailed information on household and individual socio-demographic characteristics, utilization of reproductive, maternal and child health services, and recorded household geographic coordinates, among other topics. All surveyors were women who were not members of the villages they surveyed, due to the sensitive nature of questions related to contraception and reproductive health. Respondents participated in French, Bamanankan, Peulh, or the Dogon dialects of Tomokan and Tingu. We evaluated women’s self-reported use of a modern method of contraception at the time of the survey. We defined modern methods according to the World Health Organization (WHO) [24] and Malian guidelines, and included female and male sterilization, female and male condoms, intrauterine device (IUD), implant, injectable contraceptive, oral contraceptive pill (OCP), diaphragm, foam/spermicidal jelly, the lactational amenorrhea method (LAM), and the standard days method (e.g., cycle beads). Traditional methods included the rhythm/calendar method, withdrawal, herbal, and other methods. For women who reported using multiple methods concurrently (N = 5), the more efficacious method was chosen for analysis (i.e., sterilization > implant > IUD > injectable > other modern method > traditional method). We descriptively examined length of use and place and cost of last procurement among all contraceptive users. Length of contraceptive use in months was calculated by subtracting the month and year that the woman reported using the current method without interruption from the month and year of the survey. Initiation month was assigned between one and 12 at random using the runiform function in Stata 15 if it was missing (N = 123/710). The place where the current method was last procured was categorized as within the health sector or outside the health sector. Within the health sector included national, regional, or district hospitals, PHCs, CHWs, and private clinics. Outside the health sector included at home, at boutiques, kiosks, bars, or nightclubs, black market vendors, or personal contacts. We elaborated a list of potential predictors a priori based on existing evidence and contextual knowledge. At the individual level, these included: women’s age (5-year categories); number of living children (none, one or two, three or four, five or six, seven or more); marital status (monogamous, polygynous, not currently married); tolerant attitudes for spousal violence (coded any tolerance versus none, based on whether she believed a husband was justified in hitting or beating a wife under any of the seven circumstances evaluated, including if she used contraception without his consent); education (any formal schooling versus none); participation in paid labor (any versus none); and empowerment measures we adapted from existing scales, including mobility and decision-making power. We coded women’s mobility categorically based on their having ever been to the market place, health center, women’s group, or outside the village (never been to any, been to some or all but never alone, been to at least one alone—with which we capture independent mobility) [25]. Women were coded as having any involvement in decision-making versus none, based on whether they reported making decisions, either independently or jointly with someone else in the family, for any of the three domains asked (i.e., her own healthcare, visiting her relatives, household purchasing) [26]. Household level predictors included: another woman of reproductive age in the household using modern contraception; household wealth quintiles constructed using principal components analysis of asset indicators; [27] household food insecurity (coded as any versus none, based on whether the respondent affirmed that in the last 30 days, there was no food to eat due to a lack of resources, or someone went to sleep hungry because there was not enough food, or someone went a whole day and night without eating because there was not enough food); [28] and household distance to nearest public sector health facility. Orthodromic (great-circle) distance estimates were based on Geographic Information System (GIS) data for the entrance to the family compound, each PHC, and the district referral hospital. When GIS data for the family compound was missing (N = 560), we approximated household distance using GIS collected at the central gathering place in the village. We included a community level factor for the availability of CHW services at the time of the survey (coded as having a CHW posted in the village or hamlet at a fixed community health site, having a CHW provide services in the village or hamlet but not posted there, or not having any CHW services available), based on documentation from the Ministry of Health and Social Affairs. All statistical analyses were performed using Stata version 15 (Stata Corporation, Texas, USA). We first examined descriptively sample characteristics and contraceptive use outcomes. Categorical sample characteristics were calculated as proportions of all women in the sample (i.e., including those with missing characteristics data). The main outcome was calculated among those with non-missing modern contraceptive use data. We compared sample characteristics between those with and without missing data on the main outcome. For continuous variables, we calculated summary statistics appropriate to the variable distribution in the sample population (e.g., mean, median). We georeferenced the concession location data using OpenStreetMap. Kernel Density Estimation was employed to generate a density raster (heatmap) in QGIS v3.4.7 and to visualize the spatial/geographic clustering of women using modern contraception as well as multi-user households and village-clusters with no users, across the study area. The radius was set at 0.01 map unit. Within households that had multiple women of reproductive age where some but not all were using modern methods of contraception, we explored descriptively how, within the same household, users compared with non-users in terms of their socio-demographic characteristics, role in the household, and empowerment measures. We also explored descriptively how household level factors, including household decision-making dynamics, compared between households where there was at least one modern contraceptive user and households where there were none. We conducted a multilevel regression analysis to assess factors at multiple levels influencing modern contraceptive use among women of reproductive age. As the percent missing on outcome data and covariates was small, these observations were dropped from the regression analysis. Due to the clustering of female modern contraceptive users within households, family compounds, village-clusters, and health catchment areas, we employed a multilevel modeling approach. We used a three-level, mixed-effects logistic regression with random effects at the family compound and village-cluster levels, and fixed effects for health catchment area in order to adjust for any time-invariant unobserved heterogeneity across catchment areas, such as availability of contraceptive methods or characteristics of provision at the PHCs. The Level 1 equation represents variation at the individual woman level. ηijk= logπijk1-πijk, πijk denotes the probability that the ith woman in the jth family compound and the kth village-cluster uses modern contraception. Xijk denotes a vector of individual woman-level and household-level (e.g., wealth, food insecurity) variables of interest, and β1 represents the coefficients for this set of covariates. εijk is the woman-level error term, with variance σ(1)2. The Level 2 equation represents variation at the level of the family compound, where α0jk is a function of: Z0jk, which denotes family compound-level covariates (i.e., distance to the nearest public sector health facility); δ00k, which is a systematic component modelled as the compound specific random intercept; and μ0jk, representing the family compound-level random effect with variance σ(2)2. The Level 3 equation represents variation at the level of the village-cluster. C00k denotes cluster-level covariates (i.e., availability of CHW services), θ000 represents the village-cluster specific intercept, and υ00k is the village-cluster level random effect with variance σ(3)2. We estimated adjusted odds ratios (AORs) and reported 95% confidence intervals (CIs). For all categorical variables, we conducted a likelihood ratio test to assess the evidence of an association between the variable and the outcome. To assess the extent to which missing outcome data affected the results, we ran the model assuming women with missing contraceptive use data were all not modern contraceptive users, and again assuming they were all modern contraceptive users.
