Setting
The self-injection pilot study took place across six high-volume public health facilities in Addis Ababa, Ethiopia, that were selected by the Ministry of Health to pilot rollout of DMPA-SC self-injection. Recent data indicate that 48% of married women of reproductive age in Addis Ababa currently use a modern method of family planning (16). Of those using any contraceptive method, injectables are the most widely used, representing 34% of the method mix in the city (16). According to the 2016 Ethiopia Demographic and Health Survey, 32% of women in Addis Ababa had completed secondary education or higher and 88% were literate (17).
Study design
We used data from our recent study which aimed to generate information on the safety, acceptability, and feasibility of introduction and scale-up of DMPA-SC self-injection in select sub-cities in Addis Ababa, Ethiopia. This was achieved through a mixed-methods implementation research design, enrolling a prospective cohort of adult family planning users seeking injectable contraceptive services from public sector providers, who were followed-up over a three-month period.
Data collection
Four-hundred participants were recruited between August-December 2021. All family planning users at the pilot sites who were medically eligible and interested in DMPA-SC self-injection were invited by public health family planning providers to participate in the study, until the target sample size was reached. The target sample size of 400 was determined to be sufficient for assessing competency in DMPA-SC self-injection at three months, which was the primary outcome of the pilot study. Trained research assistants administered a tablet-based enrollment questionnaire to document participants’ sociodemographic characteristics and perceptions of self-injection prior to training. Providers then trained participants to self-administer DMPA-SC, who were invited to practice self-injection on a prosthetic, and then inject themselves. Providers assessed participants’ self-injection performance using a standardized observation form.(6, 9, 18-20) Those who demonstrated that they could competently self-inject were given one additional unit of DMPA-SC to take away with them to reinject three months later. These participants were eligible for follow-up. During the follow-up visit, the research assistants administered a tablet-based questionnaire to capture participants’ experience with self-injection training, 3-month self-injection, storage, disposal, and side effects, as well as their perceptions on acceptability of the method. Participants then demonstrated self-injection technique on a prosthetic while the research assistants assessed their performance using the standardized observation tool.
Reproductive empowerment was measured during both the enrollment and follow-up surveys using the validated Reproductive Empowerment Scale.(21, 22) This scale was selected because it meets the three levels of agency outlined by the Reproductive Empowerment Framework.(23) Specifically, this study utilized the reproductive health decision-making sub-scale as this component is most closely linked to individual behavior that may be related to DMPA-SC self-injection. The sub-scale is comprised of four questions scored from 1-4 for a total maximum score of 16, with higher scores representing higher levels of reproductive empowerment. Two of the questions employ Likert scales (from strongly disagree to strongly agree) focusing on one’s ability to use contraception or refuse sex when one’s partner is in opposition. The other two questions ask about one’s current and preferred contraceptive decision-makers such that those responding with “my partner and myself jointly” receive the highest score, then those saying “myself”, then “my partner” and finally all other options.(24) The mean response score was imputed for participants with two or fewer missing responses. Participants with three or more missing responses needed for the subscale were excluded from analysis.
Acceptability of DMPA-SC self-injection was assessed through the following questions (1) uptake of self-injection (measured through participation in the study), (2) re-injection at 3-months (dichotomous, measured at 3-month follow-up visit), and (3) desire to continue using DMPA-SC self-injection in the future (dichotomous, measured at 3-month follow-up visit). Related concepts of comfort and confidence with self-injection were also measured over time. Comfort with self-injection was measured using a 4-point Likert scale (very nervous, somewhat nervous, at ease, very at ease) and confidence was measured with a 3-point Likert scale (very confident, somewhat confident, not very confident). Participants were asked to consider their comfort with self-injection prior to self-injection training (measured during the enrollment survey prior to self-injection training), after training, and after reinjection approximately 3-months after training (both assessed in the follow-up survey). Confidence about self-injection was with respect to self-injection at enrollment, reinjection 3-months later, and regarding future self-injection, all of which were asked about during follow-up.
Data analysis
Descriptive analysis was performed to describe the characteristics of participants who completed a 3-month follow-up visit. We first assessed if there were changes in reproductive empowerment by performing a paired t-test to compare reproductive empowerment scores from enrollment to the 3-month follow-up visit. We performed this same paired t-test assessing changes in reproductive empowerment scores again among a restricted sample that only included participants who had already reinjected at the time of follow-up to explore if there was a difference for this subgroup. We then built a multivariable model to adjust for confounders and assess the factors related to the change in reproductive empowerment scores from baseline to endline. We used a data-driven approach to build the model, starting with separate regression analyses for each of the variables under consideration, with inclusion of baseline empowerment as the only confounding factor. Variables were considered for analysis based on theoretical plausibility that they may affect reproductive empowerment. Any variable with p<0.1 in this analysis was considered for inclusion in the final model. To avoid introducing endogeneity problems, variables measured at the time of follow-up were not eligible for inclusion in the model, with one exception. We considered self-injection competency at follow-up because this indicator was measured through independent observation of injection technique and may not be affected by the same factors as our measure of reproductive empowerment at follow-up. The model included the baseline reproductive empowerment score as a covariate to control for incoming levels of reproductive empowerment. We assessed the models using the variable inflation factor to identify multicollinearity problems, and AIC and R-squared, dropping additional variables based on this model assessment.
We selected one measure of acceptability of DMPA-SC self-injection to assess the relationship with empowerment. For this analysis, we considered desire to continuing using DMPA-SC self-injection as our main outcome of interest as it more closely reflects future behavior intentions, and we hypothesized that higher empowerment would lead to better acceptability. The modeling approach described below could be done for other acceptability measures.
To explore the relationship between empowerment and the desire to continue using DMPA-SC self-injection, we built an instrumental variable probit model. As described above, empowerment and measures of acceptability may be affected by similar unmeasured factors or have a reversed causality relationship. For example, is self-injecting empowering and/or are those with higher levels of empowerment more likely to continue self-injecting? This potential endogeneity could result in biased estimates unless an appropriate methodology is followed. To address the potential endogeneity, we identified instrumental variables of endline reproductive empowerment and then applied a two-stage regression approach to predict desire for continuation with an instrument for empowerment, controlling for other variables. For this exploratory analysis, we used a data-driven approach to identify suitable instrumental variables by first conducting a bivariate regression model for any variable that could potentially be related to desire to continue using self-injection, among those collected, and then selecting those with a p-value <0.1 for inclusion in the multivariable model. Variables were further selected based on measures of good fit including AIC and BIC. We compared multivariable models for our main outcome of desire to continue self-injecting and that of our potentially endogenous variable of endline reproductive empowerment. The model for endline reproductive empowerment was used rather than that of change in reproductive empowerment as we theorized that one’s level of reproductive empowerment at follow-up was more likely to be related to future use than the change in empowerment experienced over the last 3 months. However, the final models for both endline and change in reproductive empowerment resulted in the inclusion of the same factors (Supplemental material). Variables found to be associated with endline reproductive empowerment but not with desire to continue with DMPA-SC self-injection were selected as instruments.
To help explain the model results, we also explored changes in comfort and confidence to self-inject over time (not in relation to reproductive empowerment). We compared individual change in response to the question regarding how comfortable participants felt with self-injecting over time, among those who re-injected at 3-months. We also compared individual change in response to the question regarding how confident participants felt about self-injection over time, among those who re-injected at 3-months. We calculated the proportion of participants who shifted responses between time points and tested whether significant shifts in response categories occurred between time points using the Stuart-Maxwell marginal homogeneity test for symmetry. A visual depiction of these shifts is illustrated through Sankey diagrams. All analyses were conducted using Stata version 16 and final tests of associations with p<0.05 were considered significant for two-sided comparisons.(25)