Determinants of unmet need for family planning in rural Burkina Faso: a multilevel logistic regression analysis

The study used data from the baseline round of a survey which included multiple tools in order to evaluate the impact of a performance-based financing (PBF) intervention on access to and quality of a wide range of healthcare services. Specifically, this study used data from both the household survey and from the healthcare workers’ survey embedded within the larger set of tools needed for the impact evaluation. Both surveys were applied in the twenty-four (24) districts distributed in six (6) regions of Burkina Faso (Boucle du Mouhoun, Centre-Est, Centre-Nord, Centre-Ouest, Nord and Sud-Ouest) where PBF was to be rolled out starting in April, 2014.

Data were collected from October 2013 to March 2014. The household survey relied on a three-stage cluster sampling technique. First, clusters were defined to reflect the catchment areas of the 448 health facilities in the 24 districts. Second, one village was selected in each of the 448 catchment areas. Third, fifteen (15) households were selected in each village. Households were selected on the basis of whether there was a woman living in the household who was currently pregnant or who had been pregnant in the twenty-four months prior to the survey date. Households were selected using a random route approach [27] until the desired sample size was achieved in each village.

Within a household, information was collected on the overall household socio-demographic and economic profile as well as on individual illness patterns, health care seeking behaviour, and related expenditure (for both adults and children).

Specifically, given our focus on unmet need for family planning among pregnant women, we considered as the effective sample for this study only the 1309 currently pregnant women included in the household survey. Currently pregnant women were asked whether their current pregnancy was intended, or whether they would have rather preferred not to have any more children, or to postpone the current pregnancy by at least 2 years. This allowed us to compute unmet need for family planning, further differentiating between unmet need for limiting and unmet need for spacing.

The healthcare workers’ survey targeted the staff working at all 443 facilities included in the study. Specifically, at each facility, the aim was to interview at least three healthcare workers. Respondents were conveniently selected among the staff present in the facility on the day of the survey. Information was collected through means of a structured, close-ended questionnaire with several modules, covering healthcare worker’s roles and responsibilities, their work environment, their training with specific reference to family planning, and facility assessment on availability of family planning methods.

Data collection was carried out by trained interviewers recruited and supervised by the colleagues at Centre-MURAZ. Both the household and the healthcare workers’ surveys relied on digital data collection, using Personal Digital Assistants (PDAs/mini computers) with data being sent to a central server on a daily basis using mobile phone connection.

Table 1 reports the complete list of variables included in our analysis, which were derived from the household and healthcare workers’ survey, as well as the expected direction of the estimated coefficient. Information from the two surveys was merged into one dataset (matched at the health facility level) to account for the fact that a mixture of demand-side (i.e. pertaining to women, their partners, and their households) and supply-side (i.e. pertaining to health system) factors is expected to influence unmet need for family planning [28]. The outcome was defined as a dichotomous variable, differentiating between pregnant women with unmet need for family planning (coded as 1) and pregnant women without such unmet need (coded as 0). According to available information from WHO and demographic and health surveys (DHS) unmet need is estimated from non-contraceptive users (pregnant women and non-pregnant who are fecund and desire to have a child in at least 2 years’ time) [3, 6, 7]. Given the non-availability of information on the non-pregnant women explained in the methodological considerations section and the fact that evidence suggest that women (pregnant and non-pregnant) have differentiated needs and should be targeted in their different sects when designing family planning intervention, our focus on unmet need was among the pregnant women category [3, 6, 8, 17]. A pregnant woman was defined as having unmet need if she indicated in the questionnaire that her pregnancy was either wanted later (mistimed pregnancy) or she did not want to be pregnant (unwanted pregnancy) but was not using any method of contraception before the pregnancy. Women with mistimed pregnancies were classified as having unmet need for spacing while those with unwanted pregnancies were classified as having unmet need for limiting. These two (2) categories are referred to as total pregnant women with unmet need for family planning also referred to as women with unintended pregnancy consistent with the WHO and World Bank definition of unmet need among the pregnant women category which is often used as proxy for unmet need [28, 29]. Pregnant women who indicated that their current pregnancy was desired did not experience unmet need for family planning (intended pregnancies) [29].

Most of the independent variables included in Table 1 are self-explanatory. Number of living children was categorized into two groups with the classification being consistent with prior studies [30, 31]. We looked at sons living as important in relation to unmet need for family planning. In most patriarchal societies, male children are required to maintain the family lineage and as such, women are expected to give birth to male children. In line with prior research [32], household socio-economic status was assessed by computing a wealth index based on a combination of housing infrastructure and ownership of mobile goods, using multiple correspondence analysis.

Four variables, defined in the literature as proximate variables, were included as measures of a woman and her partner’s attitude and decision making towards family planning. Proximate variables are intermediate variables that focus on attitude and decision making [1, 33, 34]. In our analysis, they were: woman’s approval of contraceptive use; partner’s approval of contraceptive use; couple discussion on family planning; and woman’s desire for fewer children in relation to partner. Their inclusion was motivated by the existence of prior evidence suggesting that partners’ involvement in family planning decisions is a key factor shaping women’s reproductive behaviour. Evidence indicates that most women positively adopt family planning methods when they perceive their partner’s approval of contraceptive use [28, 35].

A set of variables from the health facility assessment and from the healthcare provider survey was included to account for health system factors likely to influence unmet need for family planning. Distance to the referral health facility was assessed around the cut-off point of 5 km, in line with WHO guidelines on accessibility [36, 37]. We included a measure of the contraceptives available at each facility, distinguishing between barrier contraceptives, hormonal contraceptives, and IUD. We included two variables to assess healthcare providers’ training, one looking at general training in family planning and one looking more specifically at logistics (procurement and stocking) concerning family planning products.

Bivariate analysis was carried out to assess non-adjusted associations between the single variables and unmet need for family planning. For each of the independent explanatory variables included in our final analysis, we estimated the crude odd ratio using univariate logistic regression. We used multivariate multilevel logistic regression to identify significant associations between unmet need for family planning and the explanatory variables, while controlling for potential confounders. Specifically, we used the Stata command xtlogit [38, 39]. The application of multilevel (random-effect) modelling was used to account for the fact that women were clustered at the district level. Preliminary analysis had in fact detected important differences in unmet need for family planning across districts (Table 2). We purposely did not account for clustering at the household level, given that we recorded multiple women only in 37 households.

Out of the 1309 currently pregnant women included in the household survey, 1070 (81.74%) had intended to get pregnant, while 239 (18.26%) had not intended to (i.e. unintended pregnancies). This value represents the measure of unmet need for family planning, further distinguished between unmet need for spacing 197 (15.05%) and unmet need for limiting 42 (3.21%).

The second set of results in Table 3 reports unadjusted estimates for the association between unmet need and selected explanatory variables. A positive association was detected between unmet need for family planning and number of living children [OR = 1.72; 95% CI (1.19–2.48)], having a child younger than 1 year [OR = 1.56; 95% CI (1.06–2.32)], woman’s disapproval of contraceptive use [OR = 1.94; 95% CI (1.42–2.65)], partner’s disapproval of contraceptive use [OR = 2.51; 95% CI (1.10–2.07)], couple discussion on contraceptive use [OR = 1.33; 95% CI (1.00–1.77)] and a woman’s desire for children compared to partner [OR = 2.22; 95% CI (1.65–2.99)]. A negative association was detected between unmet need for family planning and distance to health facility [OR = 0.75; 95% CI (0.56–0.99)] and health worker training in contraceptive logistic management [OR = 0.46; 95% CI (0.27–0.77)]. No significant association was detected between unmet need and socio-economic status. Table 2 reports important differences in the level of unmet need for FP across districts with the highest rates in Solenzo and Dano (50.00% each), Sapouy and Batié (31.25% each), Gaoua (30.00%), Koudougou (28.13%) and the lowest in Kongoussi (5.88%), Ouargaye (8.16%), Reo (9.38%) and Barsalgho (11.11%).

The second set of results in Table 3 reports adjusted estimates from the multilevel logistic regression model. The model confirmed a positive association between unmet need for family planning and having four or more children [OR = 1.80; 95% CI (1.1–2.99)], having a child younger than 1 year [OR = 1.75; 95% CI (1.04–2.97)], being Christian [OR = 1.29; 95% CI (1.00–1.76)], a woman’s disapproval of contraceptive use [OR = 1.48; 95% CI (1.00–2.21)], partner’s disapproval of contraceptive use [OR = 1.51; 95% CI (1.03–2.10)], couple discussion on contraceptive use [OR = 1.48; 95% CI (1.09–2.02)] and facing a mismatch in the number of desired children with one’s partner [OR = 1.90; 95% CI (1.36–2.67)]. No consistent pattern was identified between unmet need and socio-economic status, although a significant positive association was identified between unmet need and the least poor income quintile [OR = 1.72; 95% CI (1.04–2.85)].

The model detected a negative association between unmet need for family planning and health workers’ training on contraceptive logistics [OR = 0.46; 95% CI (0.24–0.73)] and distance to health facility [OR = 0.72; 95% CI (0.53–0.98)]. The model did not detect any variance attributable at the district level (rho = 0), providing an indication that the demand-side and supply-side factors included in the analysis were sufficient to account for the heterogeneity in unmet need originally observed across districts.

As we appraise the generalizability of our findings, we need to acknowledge the possibility that the associations observed between unmet need and the selected explanatory variables may be context-dependent and not necessarily applicable to other LMICs, even within SSA. Moreover, we need to report that the research team in selecting the explanatory variables to be included in the analysis, there was the possibility that additional explanatory variables not available in our survey may also be relevant in shaping unmet need. Similarly, we ought to recognize that the analysis relied exclusively on quantitative methods, thus making it impossible for the research team to investigate the role of unmeasurable dimensions, such as cultural beliefs and values. We must also mention the estimation of unmet need which used a population of currently pregnant women only. It would have been desirable to replicate the analysis on the entire sample of non-pregnant women of reproductive age included in our survey, but this was not possible due to the unavailability of relevant data to estimate unmet need for the non-pregnant fertile women. Last, but not least, we need to acknowledge the limitation of our quantitative study in addressing an issue as sensitive as family planning. We recognize that qualitative methods of inquiry may be better suited to explore reasons behind low adoption of contraceptives in low-income settings. Nevertheless, we trust that, by identifying a few key associations, quantitative research can be instrumental in generating impetus and working hypotheses for further qualitative inquiry.