Study population and design

Our primary analysis uses data from the 2010 RDHS [11] to assess the relationship between HIV and fertility among women in Rwanda. The 2010 RDHS is a nationally representative cross-sectional household survey of 13,671 women (15 to 49 years) and 6329 men (15 to 59 years) from 492 villages in Rwanda. The survey used a two-stage sampling design to produce separate estimates of key indicators for each of the 30 districts of Rwanda. All women aged 15 to 49 who were either permanent residents of the household or visitors in the household on the night before the survey were eligible to be interviewed.

In the 2010 RDHS, HIV testing was offered to 6952 eligible women, 99% of whom consented. Questions about fertility, however, were only asked to women who self-reported ever having sex and not being infertile. Fertility in women under age 20 (41 births per 1000 women) is much lower and influenced by legal age of marriage (21 years in Rwanda), secondary school attendance, and other factors not applicable to woman age 20 to 24 (189 births per 1000 women) and older [11], thus this analysis excluded women less than 20 years of age. The final analysis included the 4527 women who consented to HIV testing, responded to the fertility questions, and were aged 20 or older (Fig. 2).

For a secondary analysis, the 2005 RDHS [12] was used to evaluate if the relationship between HIV and fertility changed between 2005 and 2010 which would provide evidence of unmeasured factors, such as ART scale up, changing between these two points in time. The 2005 RDHS was collected using the same methods as the 2010 RDHS. The 2005 RDHS included 11,539 women (15 to 49 years), and HIV testing was offered to 5663 women with 98.8% consent rate, resulting in 3532 women age 20 or older in the secondary analysis.

Variables and analysis

Two outcomes were explored. The first outcome of interest was fertility in the last 5 years defined as having given birth in the last 5 years or currently being pregnant. The second outcome of interest was a desire to have more children, defined as a woman reports that she would like to have a (another) child. A literature review and development of a conceptual framework informed key variables to include in this analysis (Fig. 3). Socio-demographic variables included in the analysis were wealth index, urban/rural residence, family size, polygamous marriage, woman’s educational level, and woman’s employment status. Clinical and behavioral data for the women included HIV status, knowledge of MTCT, number of lifetime sex partners, polygamous marriage, current use of contraception, and fertility more than 5 years ago. Variables identified in the conceptual framework that were not available in the dataset included type and timing of past sexually transmitted infection (STI), knowledge of HIV status, and use of antiretroviral therapy. Only those variables which changed the coefficient of the primary predictor, HIV status, in one or more age group by 10% or more were considered for the final model as a potential confounder.

HIV prevalence, fertility in the last 5 years, and desire to have more children were estimated with 95% confidence intervals for all covariates and chi-square tests were used to describe differences between groups. Multiple logistic regression models were used to assess the association between HIV and each of the two outcomes. Since fertility patterns and drivers change over the life course, the effect of HIV was modeled in three separate age groups: 20–29, 30–39 and 40–49 years by including an interaction term of HIV status and age group. All potential confounders that were not collinear (r<=|0.7|) were included in the final multivariable model, and records with any missing data were dropped. We assessed each covariate for confounding the primary effects, and included it in the final model if it changed the effect of HIV within any age group by 10% or more. Models were built for each outcome separately.

For the secondary analysis, we produced multivariable regression models for 2005 as a comparison and presented adjusted odds ratios (AOR). In these 2005 models, we controlled for the same confounders identified in the 2010 analysis to ensure comparability. All analyses were completed in Stata v12 using the svy commands to account for sample weights, clustering of households, and stratification by district (2010) or province (2005).

Ethics

For both the 2005 and 2010 RDHS, informed consent was obtained from every study participant before the start of the interview, and a second consent was obtained before blood draw to test for HIV. The RDHS received ethical approval from the Rwandan National Ethics Committee and the US Centers for Disease Control and Prevention (CDC). We received permission from the MeasureDHS project to download and use these data for this analysis.

The relationship between HIV and fertility

The literature about the relationship between HIV infection and fertility is conflicting. The notion that HIV would lead to a stark decrease in fertility has been challenged by evidence that HIV infected women actually have more children than their HIV seronegative peers [13]. A study conducted in Kenya in 2010 showed an increase in fertility in HIV infected women due to an increased mortality in children which led HIV infected women to have more children [14]. However, the increase in fertility in HIV seropositive women may only be true for those women who are still in the early stage of their infection. There is increasing evidence suggesting that HIV infected women, particularly women with advanced HIV infections or AIDS, are less fertile than HIV uninfected women [15].

One possible explanation for no significant association between HIV and fertility in 2010 is the scale-up of ART. In Rwanda, it was estimated that 85% of HIV infected adults in need were on ART in 2010 (compared to 44% in 2005), and the median CD4 cell count at ART initiation among adult patients increased from 153 cells/mm³ in 2005 to 277 cells/mm³ in 2010 [Unpublished dissertation by Dr. Jean Pierre Nyemazi (2012) Outcomes of adult patients on Antiretroviral Therapy from 2004 to 2010 in National University of Rwanda, School of Public Health]. Thus expanded access to ART in Rwanda may explain the lack of significant differences in fertility that we observed in this analysis. Recent studies have reported an independent effect of ART on increasing pregnancy incidence over time in women on treatment compared with ART-naıve women [5, 6, 16]. In several settings where antiretroviral therapy has been scaled-up, pregnancy and births rates among HIV seropositive women have significantly increased [17]. The improved individual health through the availability of ART may result in improved sexual drive, influencing the behavior of HIV seropositive women [18]. Women on treatment may feel healthier, more optimistic about the PMTCT interventions, more positive about their own and children’s futures, and therefore engage in unprotected sexual activity and be more inclined to become pregnant [19].

Another possible explanation for the lack of a significant difference in fertility between HIV seropositive and HIV seronegative women in 2010 is the emphasis that Rwanda has put on promoting access and use of family planning services. Over the last 10 years, the country experienced a remarkable increase in modern contraceptive use. The contraceptive prevalence rate increased from 17 to 52% between 2005 and 2010 [11, 12]. Unmet need for family planning declined from 38 to 19%. The infant mortality rate has decreased from 105 deaths to 50 deaths per 1000 live births, and the total fertility rate dropped from 6.1 to 4.6 births [12, 20]. Thus in addition to increased fertility of HIV seropositive women, there may also be a greater drop in fertility among HIV seronegative women due to use of modern contraceptive methods.

A third potential explanation is that there is no relationship between HIV and fertility in Rwanda, which is supported by the results of the 2005 analysis. However, given the large improvements in survival with HIV, the lack of relationship in 2005 is likely influenced by survivor bias; women who were so sick they could not or chose not be become pregnant did not survive to be measured.

The relationship between HIV and the desire for more children

HIV infected women between 20 and 39 years of age were significantly less likely to want more children in the future than those not infected with HIV in 2010. This was similarly shown in some studies [21, 22], but not all [23]. There are several hypotheses for this finding. In the context of high service coverage, women infected with HIV have monthly contact with the health system through pre-ART or ART care to obtain medication. Therefore, with integrated HIV care and treatment services with family planning services, it is highly conceivable that through regular contact with the health system, infected women have increased awareness about the benefits of spacing and small family sizes thus decreasing their desire for more children. More research is needed to explore the plausibility of this hypothesis.

A competing hypothesis is that the lower desire for children among HIV seropositive women may be born from stigma or fear. In other studies, reasons for HIV infected women not wanting another child included the need to provide for their living children; fear that their poor health may prevent them from taking care of an additional child; the fear of leaving the child orphaned; the fear that they may not have any family support to raise the child; the fact that they were not yet fully convinced about the efficacy of antiretroviral triple therapy in mitigating MTCT; and fear that the physical demands of pregnancy and childbirth could negatively impact their own health [18, 22, 24].

We did not observe a similar effect between HIV and fertility desire in the 2005 secondary analysis. However, this can reflect both survivor bias, since very few people were on HIV treatment at the time, and the very low prevalence of women who knew their HIV status at that time (11.6%) [12].

Limitations

This study used data available through two recent RDHSs. With the cross-sectional nature of the data collection, it is impossible to ascertain the sequence of outcomes and exposures, specifically the length of HIV infection, whether or not a woman knew her infection status (either at the time of the survey or the time of pregnancy), and whether or not she was accessing HIV care. Our hypothesis that access to ART contributed to the non-significant difference in fertility rates between HIV infected and HIV uninfected women relies on the assumption that HIV infected women knew their status and accessed HIV care and treatment services. This is plausible in 2010 when an estimated 80% of eligible individuals in need were on ART, but tenuous in the full 5 years under study as these map to years of scale-up of the ART program. The amount that ART is contributing to non-differential fertility rates can be further explored by repeating this analysis with the 2015 RDHS underway at the time of this analysis, and the 2013 Rwanda AIDS indicator survey. The hypothesis that knowledge of HIV status decreases desire for more children (either through increased family planning knowledge or because of stigma/fear) hinges on the assumption that a woman knew her status at the time of the RDHS data collection. This is a plausible assumption given that in 2010, 77.2% of respondents had ever had an HIV test and 38.6% had a test in the last 12 months [11]. Another limitation resulting from use of secondary data was that not all variables contributing to the relationship between HIV and fertility were available. Specifically, data on past STIs and antiretroviral therapy use are important confounders that were unavailable for this analysis. We recommend that future iterations of the RDHS collect symptoms of common STIs to explore these hypotheses further, though current DHS policy prevents directly asking respondents to disclose HIV status by naming use of ART.