Violence against women has been identified as a major risk factor for HIV infection among women.[1], [2], [3], [4], [5] Studies have shown that a woman's exposure to violence, mainly intimate partner violence (IPV), is associated observationally with an increased risk for HIV infection in India,[6] and southern and eastern Africa.[7], [8], [9], [10], [11], [12], [13], [14] In addition, two cohort studies in South Africa have found reduced HIV risk behaviors[13] and reduced HIV incidence[15] following interventions to empower women. There are many mechanisms through which increased IPV could be related to increased risk for HIV infection among women. These include direct effects through higher levels of violent sexual intercourse and indirect effects through reduced ability to negotiate condom use and through an increased likelihood that women who have suffered past abuse are also likely to have more partners, more transactional sex, be less likely to test/disclose and may be less receptive to HIV awareness programs, reflecting an underlying power imbalance.[1], [4], [5], [16], [17] This is in addition to the likelihood that men who enact IPV will have more high-risk sexual behaviors,[18], [19], [20], [21], [22] and may be more likely to be HIV infected in the first place.[23]

Given the plausibility of a positive association between IPV and HIV among women, it is important to evaluate whether such an association is present across multiple country contexts using large, population-based samples. We test the hypothesis that women's self report of IPV is associated with their risk of being infected with HIV in the 10 lower income countries across three continents for which nationwide data on HIV and IPV are currently available.[24]

The overall prevalence of HIV among women in this study was 4.3% (Table 1; detailed results for each country in Table S3); the rate was highest in Zimbabwe (24.1%) and lowest in India (0.5%). Almost one-third (32.1%) of women surveyed reported having experienced some form of IPV in their most recent sexual relationship. Almost one-third of women reported experiencing physical IPV (30.0%) with their most recent partner, one in eleven reported experiencing sexual IPV (8.6%), and 6.5% of women reported experiencing both. The country-specific adjusted rates of physical and sexual IPV were strongly correlated with one another (ρ = 0.649; p = 0.04). The distribution of the covariates across countries overlapped substantially and there was evidence for demographic and socioeconomic patterning of IPV (pooled results shown in Table 2; detailed results for each country in Table S4).

In the pooled unadjusted regression analysis, a significant positive association between different categories of IPV experienced and HIV infection was observed, except for ‘sexual without physical violence’ (Table 3). The unadjusted odds ratios for HIV infection was strongest when comparing those who had experienced both physical and sexual IPV to those who had experienced no sexual IPV (OR 1.20; 95% CI 1.04, 1.39) and weakest when comparing those who had experienced sexual but not physical IPV to those who had experienced no sexual IPV (OR 1.06; 95% CI 0.97, 1.17). However, adjustment for covariates attenuated the point estimates for each of the IPV domains (Figure 1; detailed results of all covariate coefficient estimates in Table S5). In country-specific unadjusted analyses, Indian women who experienced any combination of IPV appeared to be at increased risk for HIV infection compared to those who did not, while Haitian women who reported experiencing either physical or sexual IPV were at lower risk of HIV infection in the adjusted analyses. In no other of the ten countries was there a statistically significant association between any combination of reported IPV and HIV before or after covariate adjustment.

Rerunning the analyses weighted either by the DV sample weights, the HIV sample weights or a combination of the two did not significantly affect the overall pooled results, although the results for individual countries did change in some cases (Table S6). Similarly, rerunning the pooled analyses excluding India made little difference; both the unadjusted and adjusted ORs were generally slightly lower than for all ten countries together (Table S7). Comparing the results specifically for India, by different forms of IPV, adjustments and weightings, the associations were similar for comparisons of women who had experienced physical or sexual, or sexual but not physical IPV, but were consistently larger for sexual and physical violence (Table S7).

Tests of homogeneity for each of the six pooled models (adjusted and unadjusted for each of the three outcomes) when rerun to test for variation in the IPV-HIV relationship across countries were marginally able to reject homogeneity in one case (adjusted analysis for physical or sexual IPV vs. no IPV: χ² = 1.90, p-value = 0.048), but unable to reject the null in the other five cases (not shown). It therefore appears that heterogeneity of effects was limited in this study. Separate pooled analyses for currently and previously married women showed that previously married women had slightly lower odds ratios for the IPV and HIV association, but no significant changes in the results were found (Table S8).

Using all currently available DHS surveys, based on laboratory tests of HIV and self reports of IPV we find no robust or consistent association between reported physical and sexual IPV and HIV infection amongst women in ten national population samples. In the pooled data, while unadjusted relationships found IPV to be associated with a small, significant increase in HIV prevalence (OR between 1.06 and 1.20), these relationships attenuated into insignificance once adjustment was made for demographic and social factors (OR between 1.02 and 1.05). This suggests that those experiencing IPV have a higher prevalence of HIV in this dataset, but that this association can be explained by precedent, common risk factors.

Country-specific regressions show that effect sizes were small in almost all cases – adjusted associational ORs were generally between 0.85 and 1.15 for all IPV exposures. The exceptions to this were India (with ORs of between 1.34 and 1.35) and Haiti (ORs between 0.41 and 0.48). Additionally, when comparing those experiencing physical and sexual IPV to those reporting no sexual IPV, Mali mimicked India, and Kenya and Liberia were closer to Haiti. Nevertheless, only one of thirty adjusted country-specific analyses recorded a significant association.

The study's findings should be interpreted with the following caveats. First, the cross-sectional nature of the data is of generic concern when considering any causal connection between variables, severely limiting our ability to ascertain the temporality of any association shown. For instance, it is possible that IPV is a consequence of a woman being HIV-positive. The data used here are stronger than some other cross-sectional data in that most women who tested HIV-positive will be unaware of their status prior to the interview,[37] thus reducing the connection between seropositivity and partner awareness. Second, there remains a possibility of selection bias, if those who participated in HIV testing or the domestic violence module differed from those who did not. Information on those refusing the domestic violence module was not available, however comparisons of those accepting and refusing an HIV test – which found those refusing an HIV test to report either similar or lower IPV rates than those accepting a test (Table S9) – suggest that any selection bias introduced would have acted to inflate rather than deflate our results.

Third, the measures of IPV collected by DHS are imperfect. The questions asked relate only to each woman's last “husband/partner”, and do not ask about frequency of IPV except in the categories of ‘often’ or ‘sometimes’. This makes the meaning of any reported association harder to interpret. The meaning of the question may have been systematically different for women who do and do not have a current partner; however our sensitivity analysis stratifying by marital status suggests that the results found are qualitatively similar for married and previously married women. Fourth, in several of the countries the number of women testing positive was small (fewer than 100 in the Dominican Republic, Haiti, Liberia, Mali and Rwanda), which may have reduced the power available to test for a significant association between IPV and HIV, such that a true association might not have been seen. However, in these five countries no adjusted effect size was larger than 1.14. This concern is also somewhat allayed by the results from the pooled regressions, which follow the pattern of the individual country regressions in showing a small, significantly positive relationship in the unadjusted regressions, which is then attenuated in the adjusted analyses. Furthermore, given that country-specific results are equally spread above and below the null value, increased power would not change our overall finding of no clear effect direction.

Fifth, the data analyzed are only for women who are, or have previously been, married or cohabiting with partners (24.7% of all DHS respondent women had never married). Consequently, inferences cannot be generalized to all women. As noted elsewhere [38], there is no consistent patterning of IPV by marital status or age, so it is unclear how this limitation of the sample affects how results can be extrapolated to all women, or compared to the existing literature. Nevertheless, with high lifetime marriage rates in most countries, the population covered includes the majority of the world's female population. Finally, although this study includes countries with a wide range of HIV epidemics – from the hyperendemic to the highly concentrated – reported IPV rates and models of gender-based behavior, the countries with all necessary data were not randomly selected with respect to the world's population. Generalizing from this dataset to other countries will require careful consideration of the state of HIV and IPV in other countries, as well as other normative beliefs and behaviors.

The findings of this analysis provide new evidence in the ongoing effort to understand the relationship between IPV and HIV. The differences between our findings from those previously reported [6], [7], [8], [9], [10], [11], [12], [14], [15] may – with the exception of those for India [6] and South Africa [9], [13], [15] – reflect differences in sampling technique since many of the previous studies were based on samples that did not cover the whole female population, often sampling women attending health facilities [7], [8], [11], [12], [14].

We note that the Indian study, using the same data as this analysis, reported a positive association between HIV and IPV for women who had suffered both physical and sexual IPV. However, our analysis suggests that this finding is not robust to different weightings (neither the DV nor the HIV weights provided by DHS are directly relevant to the subsample of women who were invited both to take an HIV test and complete the DV module), or to the addition of previously-married women to the sample (an additional 1801 women, or 6.0%), or to the inclusion of additional covariates (we added marital status, occupation and urban residence). Each of these three factors partially attenuated the initial association (Table S10). While the largest association we found in the Indian sample was for the types of IPV most likely to relate to HIV (experience of both physical and sexual violence compared to experience of no sexual IPV), the association was not significant at the 95% confidence level.

The null findings in this study may be a function of the geographic settings considered. While some of the previous studies have been conducted in India [6], and East African countries covered here [8], [12], [14], the strongest evidence for a relationship between IPV and HIV to date has been found in South Africa [7], [9], [13], [15], which does not have the DHS data to allow its inclusion in this analysis. The likelihood that these differences are due to country coverage is lessened by the fact that in Rwanda and Kenya positive associations were found in STI, antenatal and paediatric clinic attendees [8], [12], [14]. This suggests that the difference in results is more likely to be due to the differences in study design than geography in these two countries, although we cannot parse out the contribution of each elsewhere.

Further research is needed to determine whether the relationship between IPV and HIV in specific countries differs depending on whether the study population is clinic-based or uses a national sample, or whether geographical effect-modification is occurring. If clinic populations prove to have a stronger relationship between IPV and HIV, this argues for a more targeted approach to IPV based in clinical, rather than general, populations. If variation is by national setting, it will be important to first determine which factors (HIV-related, IPV-related or some third factor) are present when IPV and HIV are associated, and then to focus on intervening in such settings.

Although evidence for effect-modification by setting is marginal in this study, the findings of a significantly positive unadjusted association in India, and a significantly negative adjusted relationship in Haiti, raise the possibility that the IPV-HIV relationship in these countries may differ from the other eight countries included. If these effects are in fact causal, this may provide useful indicators for future research. In India, Decker and colleagues have found those men who commit IPV also engage more generally in higher levels of gendered HIV risk behaviors, including sexual infidelity, coercive condom practices and transactional sex[18]. In a society where HIV remains highly concentrated amongst Most At-Risk Populations (MARP), including sex workers [39], this combination of risk behaviors may act to put the partners of IPV perpetrators at raised risk of HIV. Research to determine whether similar patterns of behavior are seen in settings where IPV and HIV have been consistently linked – such as South Africa – but perhaps not seen in other countries in this study sample, would be of interest. In contrast, Haiti is a society where HIV is more broadly experienced, although still with higher levels of infection amongst MARPs [40]. It is also a country where social dislocation and economic stagnation loom large. It may be that in this specific context, those factors which put women at risk of IPV actually reduce the risk of their partner being HIV positive, and thus reduce the woman's risk of becoming HIV positive themselves. It is important to note that changes in HIV risk behavior will only affect HIV risk if HIV is present in a partnership's sexual network. Determining risk factors for the presence of HIV in a social network may help us understand why in some contexts IPV and HIV are associated, but elsewhere they are not.

Existing frameworks linking IPV and HIV posit multiple individual-level risk factors, including increased risk of violent intercourse, lower decision-making power and partners' high-risk sexual behaviors leading them to be more likely to be HIV-positive [1]. This study's results, if proven to be causal, suggest that such frameworks linking IPV and HIV may at a minimum be contingent on context-specific factors, such that these risk factors are not differentially distributed across women experiencing and not experiencing IPV. This does not necessarily undermine the relationship between gender inequity and sexual risk: inequity can increase IPV rates, and thus increase risky behavior, but if inequity does not also change the sexual networks within a setting such that HIV is more likely to be present in either partner, then no overall effect will be seen. Alternatively, it may be that focusing on individual- or relationship-level effects is to miss the point [41]: a structural explanation in which gender inequity is a driver of both increased IPV throughout society and HIV in both IPV and non-IPV experiencing populations in consistent is also consistent with our findings. In this case IPV would not be the mechanism through which gender inequity led to HIV, even though settings with more inequity might be linked to higher HIV rates.

Given the associational nature of these findings, and the various other potential limitations highlighted, any policy recommendations based on these results should be tempered by the need to replicate these results in longitudinal data. However, if the weak association between IPV and HIV suggested by our results prove to be causal, it would argue for focusing HIV prevention funds elsewhere than adult IPV prevention. This would not rule out gender empowerment more broadly as an HIV prevention intervention, or addressing childhood IPV in order to reduce adult risk of HIV infection [42]. These results also say nothing regarding the intrinsic importance of reducing IPV.

In summary, this study presents evidence that there is no consistent association between physical and sexual IPV and HIV among ever-married women in ten countries in the Americas, Africa and Asia, once we adjust for commonly considered confounders of the relationship. Given evidence elsewhere of plausible causal mechanisms between the two phenomena, further investigation is needed to understand in which circumstances the two may or may not be associated, and what mechanisms may be at play in determining the strength of the relationship. The null findings reported here do not negate the importance of reducing IPV as an intrinsically important public health goal and a basic human right, but do suggest that IPV is not consistently associated with HIV prevalence worldwide.