Study sampling and participants
We used the 2019-20 Rwanda Demographic Survey (RDHS) for this analysis which employed a two-stage sample design, with the first stage involving cluster selection consisting of enumeration areas (EAs) [12]. The second stage involved systematic sampling of households in all the selected EAs leading to a total of 13,005 households [12]. In particular, the data used in this analysis were from the household and the woman’s questionnaires.
During this survey, the data collection period was from November 2019 to July 2020, taking longer than expected due to the COVID-19 pandemic restrictions [12]. Women aged 15-49 years who were either permanent residents of the selected households or visitors who stayed in the household the night before the survey were eligible to be interviewed. Out of the total 13,005 households that were selected for the survey, 12,951 were occupied and 12,949 were successfully interviewed leading to a 100% response rate [12]. This analysis included all women interviewed during the survey, and of the selected households,14,675 women aged 15-49 were eligible to be interviewed but 14,634 women were successfully interviewed leading to a 99.7% response rate [12].
Variables
Dependent variables
The study outcome variable was the usage of health insurance, and this was a binary variable directly coded yes or no.
Explanatory variables
Measures of women’s empowerment
Four indices were created to measure the empowerment of women: exposure to media, decision making, economic empowerment, and sexual empowerment. Women’s empowerment indices were measured as composite scores [4,28].
Exposure to media was considered as the women’s ability to have the opportunity to read a newspaper or a magazine, listen to the radio and watch TV. Responses were re-coded (1 if the woman was exposed to newspapers, radio or TV and 0 if the woman was not). We then created an index, by adding all the scores for each woman, with the total score ranging from 0 to 3, after which we finally categorized the scores into four groups [4]. A total score of 0 meant no access to any of the three media, while scores of 1(low), 2(medium) and 3(high) implied exposure to one, two, and three media channels respectively [4,28].
Decision making included women’s ability to be involved in making decisions regarding; their own health; large household purchases; visits to their family and control over family earnings [4]. We re-coded the responses to have two categories (1 = woman involved in decision making alone or with a partner, 0 = woman not involved in decision making). We then added all the scores to form an index score ranging from 0 to 4, and we finally categorized the score into four groups. The highest score was four which meant that the woman was involved in the decision making for the four used indicators. Medium decision-making ability meant that women were involved in 2 or 3 indicators, low decision making meant that the woman was involved in only one indicator and no decision making implied that the woman was not involved in any decision making [4,24,28].
Economic empowerment entailed women’s owning of a house, land and the type of earning from her work [4,28]. We re-coded the three indicators as 1-if the women owned a house or land (either alone or jointly with a partner) or received cash payment for their work and 0-if didn’t own a house, land or cash payment for work. An index was then created by summing the scores for each woman, with a total score ranging from 0 to 3, after which we categorized the score into four groups. The highest score of 3 implied that the woman owned a house, land, and earned cash for her work, while scores of 2, 1 and 0 meant medium, low and no economic empowerment, respectively.
Sexual empowerment referred to the women’s ability to refuse sex and ask a partner to use condoms [4,29]. Responses were coded (1 if the woman could refuse sex or ask for a condom and 0 if the woman could not) and sexually empowered women were those who were able to refuse sex or ask their partners to use condoms. We then created an index by adding the scores for each woman with a total score ranging from 0 to 2, after which we categorized the score into three groups. The highest score of 2 implied high sexual empowerment, while scores of 1 and 0 respectively meant low and no sexual empowerment.
Decision making and sexual empowerment had about 7,233 missing responses, while economic empowerment had about 3908 missing values, and this was because some of these questions were asked during the domestic violence survey sessions, yet not all women in the RDHS were included in the domestic violence module of the survey. These missing observations were assumed to be zero [4], thus we risked overestimating low subcategories of these composite indices/ variables. To ensure that this doesn’t affect our findings, we conducted a sensitivity analysis by considering only women sampled in the domestic violence model and excluded those with missing responses. However, this showed no significant difference from the original analysis and more details are included in the sensitivity analysis section of the results. Moreover, for background characteristics, we provided frequencies of these variables considering only women with valid responses.
Other explanatory variables or potential confounders
We included possible determinants of health insurance utilisation based on available literature and data [1,8,19-23]. Ten (10) variables were considered and of these, two were community-level factors that included; place of residence (categorized into rural and urban), and region of residence (categorized into Kigali, South, West, East and North). Three household-level factors included; household size which was classified into “less than six” and “six and above”, sex of household head (classified as female and male) and wealth index (categorized into five quintiles that ranged from the poorest to the richest quintile). Wealth index was calculated by RDHS from information on household asset ownership using Principal Component Analysis [12]. Five individual-level factors were also considered in the analysis, including; age (categorized as 15-24, 25-34, 35-44, 45-49), educational level (grouped as no education, primary, secondary and tertiary), working status (classified as working and not working), marital status (classified as married and unmarried), and religion (classified as Catholic, Protestant, Adventist, Moslem, and others).
Statistical analysis
We applied the DHS sample weights to account for the unequal probability sampling in different strata and ensure the representativeness of the study results [30,31]. We used SPSS (version 25.0) statistical software complex samples package incorporating the following variables in the analysis plan to account for the multistage sample design inherent in the RDHS dataset: individual sample weight, sample strata for sampling errors/design, and cluster number [12,30]. Frequency distributions were used to describe the background characteristics of the women, and cross-tabulations were used to examine the associations between health insurance utilisation and women empowerment indicators and various socio-demographic factors. Pearson’s chi-squared tests were then used to assess the significance of the cross-tabulated results, with the level of statistical significance set at a p-value < 0.25.
Bivariable logistic regression was also conducted and we presented crude odds ratio (COR), 95% confidence interval (CI) and p-values. Independent variables found significant at a p-value less than 0.25 were then included in the multivariable models. Moreover, independent variables that were non-significant at the bivariable analysis level but were associated with health insurance usage in previous studies were also included in the multivariable logistic regression models. We constructed two models in the multivariable analysis; one with only women empowerment variables and the final model that included the women empowerment indicators and other socio-demographic variables. Adjusted odds ratios (AOR), 95% confidence intervals (CI) and p-values were calculated and presented, with a statistical significance level set at p-value < 0.05.
Since questions of decision making and sexual empowerment were asked to only women selected for the domestic violence module, we conducted a sensitivity analysis where we considered only women with domestic violence module responses, excluding those with no (missing) such responses. All socio-demographic variables in the model were assessed for collinearity, which was considered present if the variables had a variance inflation factor (VIF) greater than 10 [32]. However, none of the variables had a VIF above 3.