1.1. Background of the Study

Globally, an estimated 36.7 million people were living with HIV by the end of 2016, many of whom were women (51.6%).1,2 According to the Joint United Nations Program on HIV/AIDS, one woman becomes infected with HIV every minute.3 Differences in the number of new HIV infections between men and women are most pronounced in younger age groups; new infections were 44% higher in women aged 15-24 compared to men in the same age group.2,4 HIV is the leading cause of death among women aged 30-49 and the third leading cause of death for women aged 15-29.5

Sub-Saharan Africa (SSA) accounts for 12% of the global population, yet 71% of the global burden of HIV infection is found within the region.6 More specifically, 76% of the total people infected with HIV, 76% of the total new HIV infections, and 75% of the total HIV/AIDS related deaths recorded in 2015 occurred in SSA.1,2 Young women bear a disproportionate burden of HIV infection. Of the 6.1 million people living with HIV in western and central Africa, 56% are women. In eastern and southern Africa, 19.4 million people are living with HIV and of those, 59% are women.1,2,6 The number of new infections of HIV in SSA have decreased by 33% from an estimated 2.2 million in 2005 to an estimated 1.5 million in 2013.7 Despite this declining trend, HIV infections continue to pose a problem in the region.

The disproportionate impact of HIV infection on women can be attributable to biological, social, cultural, economic, and structural factors.8 Within the Sub-Saharan region, women tend to play no part is sexual decision making as their needs and desires are not considered significant.9 Women in the region are expected to respect the choices of their husband, even if this means accepting of polygamous relationships and cheating on the part of the husband.10 Women who refuse sexual advances from their partners who reject the use of protection during sex may be subject to violence.9 Those who are unable to negotiate for safer sex practices are more likely to be infected by their partner.11

Late diagnosis of HIV and being unaware of one’s HIV positive status has numerous negative impacts. Not knowing that one has HIV increases the likelihood that the virus will be transmitted to others.12 Late diagnosis can also reduce an infected individual’s lifespan through an increase in viral load and reduction of the body’s CD4+ T-cell count.13 Early diagnosis of HIV is essential in order for those infected with HIV to receive proper treatment and care, as well as limit the transmission of the infection to others.14 Despite the host of negative outcomes that can be reduced with HIV testing, 80% of people from low- and middle-income countries have not been tested.15

For many women in Sub-Saharan Africa, fear of their partner’s reaction is a major barrier to their uptake of HIV testing as some partners discourage testing and even require that women need their permission in order to get tested.16 Literature has suggested that HIV prevention policies in SSA must consider women’s lack of negotiating power and gender inequality as one of the root causes contributing to the spread of HIV in the region.17 Several studies have identified the importance of women’s empowerment to increase HIV testing and prevention.18,19 The Sonagachi Project was a sustainable community intervention project in India that aimed to increase female sex workers’ social power through increase economic opportunities, creation of a female sex worker’s association and increased psychological empowerment and leadership.20 This 12-year project saw an associated decrease in HIV/AIDS risk among female sex workers.20

1.2. Objective

There is a growing body of evidence suggesting that women’s empowerment can help achieve better health outcomes. Achieving gender equality and empowering all women and girls are now recognized as powerful tools to enable accessibility to reproductive and sexual health care services for improved mother-and-child outcomes.21 So far, few studies have attempted to show the influence of women’s empowerment on HIV testing in Sub-Saharan Africa.18,22,23 However, our study significantly extended and improved the existing body of evidence on the women’s empowerment-HIV testing relationship. First, the two indicators of women’s empowerment in this paper are created in accordance with the DHS guideline, making our findings largely comparable to other similar papers whose methodology is aligned well with the DHS statistics guideline. Second, the use of methodologically rigorous and statistically sound analysis approaches are crucial to produce findings that are of policy relevance. In this regard, the use of IPD meta-analysis allowed us to statistically estimate the nature of the relationship women’s empowerment had with HIV testing in the context of SSA after first establishing the effect of women’s empowerment on HIV testing in each country. Unlike most prior studies on this area, our selection of confounding variables has been sufficiently informed by epidemiological principles; that is, all variables in the model other than the two indicators of women empowerment are confounders that have a demonstrated association with both women’s empowerment and HIV testing. This is important to get unbiased findings. Finally, given that there are only few studies on this increasingly important topic already, our findings are important to understand the role that women empowerment does play on the uptake of HIV testing for SSA women. To our knowledge, there is no meta-analysis study on this topic in the SSA region, yet findings that come from large sample sizes are important to accurately shed light on how women’s empowerment affects HIV testing in the region.

2.1. Data Sources and Study Design

In the SSA, we were able to access 39 DHS datasets. For our study, the DHS datasets from 33 SSA countries were used: Angola (2015), Benin (2018), Burkina Faso (2010), Burundi (2017), Chad (2015), Comoros (2012), Congo (2012), DR Congo (2014), Cameroon (2010), Côte d’Ivoire (2012), Ethiopia (2016), Gabon (2012), Gambia (2013), Ghana (2014), Guinea (2018), Kenya (2014), Liberia (2013), Lesotho (2014), Madagascar (2008), Mali (2018), Mozambique (2011), Namibia (2013), Niger (2012), Rwanda (2015), Sao Tome et Principe (2009), Senegal (2017), Sierra Leone (2013), South Africa (2016), Swaziland (2005), Togo (2013), Uganda (2016), Zambia (2016), Zimbabwe (2015). The most recent datasets were selected for each country. Excluded datasets included Nigeria, Sudan, Malawi, Tanzania, and Central African Republic because these datasets did not contain either the response variable or predictor variables or both.

DHS is a nationally representative household survey conducted at regular time intervals to provide countries with updated information on different health topics such as maternal and child health, reproductive health, fertility, nutrition, mortality, and HIV/AIDS.24 The United States Agency for International development (USAID) and each country’s statistical agency conduct the survey with technical support from Inner City Fund (ICF) International. Methodology and methods for DHS are described in detail in the final DHS report of each country. Concisely, it follows a two-stage stratified cluster sampling design. Big geographic areas, the enumeration area (EA), are selected in the first stage through the Probability Proportional to Size (PPS) approach,24 where a relatively larger EA is more likely to be selected than a smaller EA. In the second stage, a pre-calculated (28 to 30) number of households are selected from each EA. In the selected household, eligible participants (women aged 15 to 49 years and men 15 to 59 years) were interviewed about a wide range of topics using questionnaires that are comparable across countries. In Namibia DHS, data was collected for women aged between 15 to 64 years, and analyses was restricted to women age 15 to 49 years so that our analysis became similar in all DHS data included in this paper. For all 33 countries, the Individual Recode file (or women’s recode file) was used to complete our analysis. To capture some useful information on health topics that would be of relevance in a particular country only, country specific questions were also included in the model questionnaire.

2.2. Study Variables

The independent variable for the study is women empowerment. Following the DHS data analysis guidelines,25 women empowerment was measured using two indicators: decision making on three specific topics and disagreement with five reasons justifying wife-beating. For the first indicator, surveyed women were asked to report on participation in decision making regarding own health care (person who usually makes decisions on health care for yourself?), large household purchases (person who usually makes decisions on making major household purchases?), and visits to family or relatives (person who usually makes decisions on visits to your family or relatives?). Each of these decisions was changed into a binary variable with a value of 1 and 0, where 1 indicated the woman participated in decision making on that particular decision alone, or jointly with her husband. Finally, a three-category variable was formed out of the three binary variables with values 0 (no empowerment), 1-2 (moderate empowerment) and 3 (high empowerment). Prior evidence followed the same method of creating this variable.26

Women were asked if they agree that a husband is justified in beating his wife for the following reasons: a) burning food b) arguing with him c) going out without telling him d) neglecting the children and e) refusing to have sexual intercourse with him. If the woman disagreed to all of these reasons, they were assumed to be empowered. Therefore, an overall binary variable was created with a value of 1 and 0, where 1 indicated disagreement to all of the reasons, and 0 indicated she disagreed with only 4 or fewer reasons. These two indirect indicators of women’s empowerment were treated as separate exposure variables of HIV testing, and were associated with HIV testing separately.

HIV testing in the past 12 months was our response or outcome variable. In our analysis, potential confounding factors in the model were accounted for to determine the relationship between women empowerment and HIV testing. The following variables were included as confounders: age in years (15-19, 20-29, 30-39, 40-49); education (none, primary, secondary, higher); wealth (poorest, poorer, middle, richer and richest); employment for cash (employment for cash vs. not); regular media exposure (exposed to one or more of the following media at least once a week: reading newspaper or magazine, listening radio, and watching TV vs. not); and residence (rural vs. urban). Our analysis was restricted to women who were currently married at the time of survey administration.

2.3. Statistical Analysis

First, confounder adjusted logistic regression analysis was completed separately for each of the 33 DHS datasets to produce an adjusted odds ratio (OR) for the women empowerment-HIV testing association. The regression analysis strictly accounted for the three design elements (weight, cluster and strata) to produce an estimate that was representative of the respective countries. Second, meta-analysis to statistically summarize the effect of women empowerment on HIV testing was carried out through the Individual Participant Data (IPD) approach.27 The IPD meta-analysis has increasingly become an important alternative to the traditional Aggregate Data (AD) meta-analysis. This method of analysis offers researchers an important amount of control over the quality of the data to be analysed thereby boosting quality of evidence produced from the study. Meta-analysis using the IPD approach is unlikely to be affected by publication bias which is one of the most common limitations of AD meta-analysis.27

The country specific OR was pooled using a random effect model to enable generalization of findings to a setting the countries in the analysis represent. The random effect model was chosen over the fixed effect alternative because of dissimilarities between the 33 countries where many factors could affect the findings (i.e. culture). Furthermore, the random effect model allows researchers to assume that the true effect sizes (ORs in this study) vary randomly cross studies in the analysis. The inverse-variance approach was used to estimate weights for each study that were later used to produce the summary effect size. Unlike in the fixed effect model, the random effect model takes the between–study variability into account in the estimation of weights for each individual study. The between–study variability was estimated through the non-iterative, non-parametric methods of moments (MoM) estimator of DerSimonian and Laird.28 The effect of the country-level factors on the pooled effect estimates using sub-group analysis were investigated. The following country-level factors were considered: survey year, sample size, region and country’s income category. Analysis was completed in STATA v 14. The metan stata module was used for the meta-analysis. The report of findings was organized based on the Preferred Reporting of Items for Systematic Review and Meta-analysis for Individual Participant Data (PRISMA-IPD) guidelines and PRISMA flow chart.29

2.4. Ethical Approval

Ethical permissions were not required for this study since DHS datasets are publicly available. Ethical procedures were the responsibility of the institutions that commissioned, funded, or managed the surveys. All DHS surveys are approved by ICF international as well as an Institutional Review Board (IRB) in the respective country to ensure that protocols are in compliance with the U.S. Department of Health and Human Services regulations for the protection of human subjects. The dataset were extracted from MEASURE DHS website at http://dhsprogram.com/data/available-datasets.cfm.

3.1. Women’s Empowerment and HIV Testing Distribution by Countries

Table 1 presents percentage of women who are empowered in the SSA countries according to the two measures of women’s empowerment, and the proportion of women who tested for HIV. There was large variation in the number of women who were empowered across the countries studied. Among the studied countries, South Africa, Madagascar, Liberia, Lesotho, Angola, Namibia, Rwanda, Zimbabwe and Ethiopia had relatively higher percentage of women empowered in terms of participating in the three decision topics. The percentage of women who disagreed to all of the five wife-beating questions varied considerably from 0% in Chad to more than 94% in South Africa. We also showed huge between country disparity in terms of HIV testing, where the highest prevalence was in South Africa with nearly 76% and the lowest was in Senegal with nearly 32%. See table 1 for details.

3.2. Association of Women Empowerment and HIV Testing Uptake

The association between the two indicators of women’s empowerment and HIV testing uptake in each country, together with the pooled effect of women’s empowerment on HIV testing in SSA is presented in Figures 1-3 and Table 2. The association of women’s participation in decision making and their disagreement to wife-beating with HIV testing uptake was observed in 14 of the studied countries. Women empowerment was associated with higher odds of HIV testing in only five countries: Lesotho, South Africa, Niger, Namibia, and Sao Tome et Principe. In the remaining nine countries, there was a negative relationship between empowerment and HIV testing, where empowerment was associated with lower odds of testing for HIV. Most of the statistically significant associations were between decision making and HIV testing; attitude towards wife beating was associated with HIV testing in only six countries (See Figures 1-3).

Close

Figure 1

Forest plot showing the effect of having participated in one or two decisions related to women’s own health, visiting family or relatives and large household purchases on HIV testing

Export to PPT

Close

Figure 2

Forest plot showig the effect of having participated in three decisions related to women’s own health, visiting family or relatives and large household purchases on HIV testing

Export to PPT

Close

Figure 3

Forest plot showing the effect of wife beating attitude on HIV testing

Export to PPT

In the meta-analysis, compared to women who responded no to all of the three questions pertaining to the three different decisions, women who stated that they participated in making one or two decisions had marginally lower odds of HIV testing (0.89; 95%CI: 0.83, 0.97), though the prediction interval crossed the null (Figure 1). However, there was no statistically significant association between HIV testing and participation in all of the three decisions (0.92; 95%CI:0.84, 1.00; Figure 2) and attitude towards wife beating (0.99; 95%CI:0.94, 1.05; Figure 3).

Figure 1 shows the random-effects meta-analysis of the effect of having participated in one or two decisions related to women’s own health, visiting family or relatives and large household purchases on HIV testing. Country specific estimates are presented on the left side, and the overall OR is presented at the bottom of the graph. The prediction interval for a future study is displayed for overall.

Figure 2 displays the effect of having participated in three decisions related to women’s own health, visiting family or relatives and large household purchases on HIV testing. Country specific estimates are presented on the left side, and the overall OR is presented at the bottom of the graph. The prediction interval for a future study is displayed for overall.

Figure 3 highlights the effect of wife beating attitude on HIV testing. Country specific estimates are presented on the left side and the overall OR are presented at the bottom of the graph. The prediction interval for a future study is displayed for overall.

We further carried out sub-group meta-analysis to see whether the overall pooled estimate changes by different groups (Table 2). We completed the sub-group analysis for survey year, sample size, geographical region and country’s income. While geographical region and country’s income data are standard and known classifications, the survey year is the time of administration of the respective DHS and sample size was arbitrarily classified. A similar approach has been used in prior work.30 In the series of subgroup analyses, we found no evidence of differentials in the pooled association group.

4.1. Limitations

The study has a few strengths. The adoption of the IPD meta-analysis approach in this study elevated the quality of evidence as it allowed similar analysis strategies in each of the individual countries, which substantially reduced between-study variations in the effect size measured. The approach also allowed sufficient control over the data, something that the traditional meta-analysis technique does not.