South Africa: Intimate male partner violence increases rural women’s incidence of HIV infection Print E-mail

 Volume 376, Issue 9734, Pages 6 - 7, 3 July 2010

Key to prevent HIV in women: reduce gender-based violence

Jay G Silverman a

In The Lancet today, Rachel Jewkes and colleagues1 provide the first prospective demonstration of the role of violence from male partners in increasing women's risk of incident HIV infection (Scroll down to read the complete research article). In women who were HIV negative at baseline, those experiencing violence from a male partner were more likely to become infected with HIV than were those not experiencing such violence. Previous large investigations (including those involving nationally representative data) across multiple countries in Asia and Africa have suggested that women who experience violence from male partners are significantly more likely to be infected with HIV and other sexually transmitted infections.2­5 Today's study brings clarity to the probable relation of male partner violence to HIV infection, and shows for the first time the temporal ordering of these negative events.

Notably, this pattern is not only seen across epidemics geographically, but also across HIV epidemics with different drivers and different demographics of those infected. For example, the Indian epidemic is driven by commercial sex, and men are the majority of those infected. The South African epidemic, by contrast, is more generalised, with women comprising the majority of those infected.6 Thus the association of men's violence against female partners and HIV infection does not seem to be context-dependent, which further elevates the potential positive effect of efforts to prevent partner violence as a means of reducing risk of HIV infection in women.

A nurse takes blood from a woman for an HIV test at Nthabiseng Clinic in Soweto, South Africa (Bloomberg via Getty Images)

In support of this contention, about one in seven of new infections in women in Jewkes and colleagues' study were attributed to either male partner violence or women's lack of relational power. In view of the facts that the fastest growing sector of those infected with HIV across both Asian and African epidemics is women whose main risk factor is sex with a male partner,6 we lack effective HIV-prevention programmes for women who fall outside of “most at-risk populations”, and that expansion of treatment with antiretroviral drugs cannot keep pace with the ever-increasing number of people infected,7 the global HIV community must move to make targeting such male behaviours a central focus of prevention efforts.

To create effective programmes, we must consider how violence and abusive behaviour by men creates women's vulnerability to HIV infection. Modelling of nationally representative data from India suggests that, in such cases, women suffer what can be labelled double jeopardy, in that their abusive male partner is both more likely to become HIV-infected because of his greater participation in high-risk extra-relationship sexual behaviours (eg, sex with female sex workers) and, once infected, is more likely to transmit the infection to a female partner than is a non-abusive HIV-infected man.8

On the basis of this model, men's violence seems to facilitate HIV transmission, which suggests that solely reducing sexual-risk behaviours in abusive men is probably insufficient to reduce the risk of HIV-infected men transmitting the infection to their main female partners. Instead, altering the gender-based abusive behaviours of men has been called for.9 Although some might question the feasibility of efforts to change entrenched social norms related to men's control over sex and women's lives, evidence suggests that this goal may be accomplished. A randomised trial of an intensive programme that targeted young men in South Africa found reductions in both abusive behaviour and sexual risk.10 Unfortunately, because of the relatively small investment in this area to date, this study is the only published randomised trial of such a programme.

As we move forward, we should consider that nature and social contexts of epidemics vary and so must interventions. Thus we need to develop and evaluate such work in nations beyond South Africa, considering cultural variations in partner violence, variations in HIV prevalence, as well as perceptions of HIV risk based on that prevalence. In the quest to develop effective, scaleable, and sustainable programmes to stem partner violence, the potential to place such intervention within various structural contexts (eg, health care, employment, media) must also be explored. Other widespread forms of gender-based violence that also increase female vulnerability to HIV infection (eg, violence against female sex workers11 and men's sexual coercion and rape of adolescent girls12) must also be included in this initiative.

A recent positive development is the commitment by the US Office of the Global AIDS Coordinator of US$30 million to implement and test prevention programmes for gender-based violence in three African nations currently receiving funding from the US President's Emergency Plan for AIDS Relief (Tanzania, Mozambique, and the Democratic Republic of Congo). In view of the scale of US efforts over the past decade to combat HIV, and the great potential for HIV prevention associated with reduction of gender-based violence, we must hope that this initial allocation will be followed by far greater investment.
I declare that I have no conflicts of interest.

1 Jewkes RK, Dunkle K, Nduna M, Shai N. Intimate partner violence, relationship power inequity, and incidence of HIV infection in young women in South Africa: a cohort study. Lancet 201010.1016/S0140-6736(10)60548-X. published online June 16. PubMed
2 Dunkle KL, Jewkes RK, Brown HC, Gray GE, McIntyre JA, Harlow SD. Gender-based violence, relationship power and risk of prevalent HIV infection among women attending antenatal clinics in Soweto, South Africa. Lancet 2004; 363: 1415-1421. Summary | Full Text | PDF(117KB) | CrossRef | PubMed
3 van der Straten A, King R, Grinstead O, Vittinghoff E, Serufilira A, Allen S. Sexual coercion, physical violence, and HIV infection among women in steady relationships in Kigali, Rwanda. AIDS Behav 1998; 2: 61-73. CrossRef | PubMed
4 Maman S, Mbwambo JK, Hogan NM, et al. HIV-positive women report more lifetime partner violence: findings from a voluntary counselling and testing clinic in Dar es Salaam, Tanzania. Am J Public Health 2002; 92: 1331-1337. CrossRef | PubMed
5 Decker MR, Gupta J, Raj A, Miller E, Silverman JG. Intimate partner violence and sexually transmitted disease symptoms in a national sample of married Bangladeshi women. Int J Gynecol Obstet 2008; 100: 18-23. PubMed
6 UNAIDS. AIDS epidemic update 2009. . (accessed June 1, 2010).
7 WHO. Towards universal access: scaling up priority HIV/AIDS interventions in the health sector. (accessed June 1, 2010).
8 Decker MR, Seage GR, Hemenway D, et al. Intimate partner violence functions as both a risk marker and risk factor for women's HIV infection: findings from Indian husband-wife dyads. J Acquir Immune Defic Syndr 2009; 51: 593-600. CrossRef | PubMed
9 Dunkle KL, Jewkes R. Effective HIV prevention requires gender-transformative work with men. Sex Transm Infect 2007; 83: 173-174. CrossRef | PubMed
10 Jewkes R, Nduna M, Levin J, et al. Impact of Stepping Stones on HIV, HSV-2 and sexual behaviour in rural South Africa: cluster randomised controlled trial. BMJ 2008; 337: a506. PubMed
11 Decker MR, McCauley HL, Phuengsamran D, Janyam S, Seage GR, Silverman JG. Violence victimization, sexual risk and STI symptoms among a national sample of female sex workers in Thailand. Sex Transm Infect 2010; 86: 236-240. CrossRef | PubMed
12 Jewkes R, Abrams N. The epidemiology of rape and sexual coercion in South Africa: an overview. Soc Sci Med 2002; 55: 1231-1244. CrossRef | PubMed
a Department of Society, Human Development and Health, Harvard School of Public Health, Boston, MA 02115, USA

 Volume 376, Issue 9734, Pages 41 - 48, 3 July 2010

Intimate partner violence, relationship power inequity, and incidence of HIV infection in young women in South Africa: a cohort study

Prof Rachel K Jewkes MD a b , Kristin Dunkle PhD d, Mzikazi Nduna MA c, Nwabisa Shai MPH a

Cross-sectional studies have shown that intimate partner violence and gender inequity in relationships are associated with increased prevalence of HIV in women. Yet temporal sequence and causality have been questioned, and few HIV prevention programmes address these issues. We assessed whether intimate partner violence and relationship power inequity increase risk of incident HIV infection in South African women.

We did a longitudinal analysis of data from a previously published cluster-randomised controlled trial undertaken in the Eastern Cape province of South Africa in 2002­06. 1099 women aged 15­26 years who were HIV negative at baseline and had at least one additional HIV test over 2 years of follow-up were included in the analysis. Gender power equity and intimate partner violence were measured by a sexual relationship power scale and the WHO violence against women instrument, respectively. Incidence rate ratios (IRRs) of HIV acquisition at 2 years were derived from Poisson models, adjusted for study design and herpes simplex virus type 2 infection, and used to calculate population attributable fractions.

128 women acquired HIV during 2076 person-years of follow-up (incidence 6·2 per 100 person-years). 51 of 325 women with low relationship power equity at baseline acquired HIV (8·5 per 100 person-years) compared with 73 of 704 women with medium or high relationship power equity (5·5 per 100 person-years); adjusted multivariable Poisson model IRR 1·51, 95% CI 1·05­2·17, p=0·027. 45 of 253 women who reported more than one episode of intimate partner violence at baseline acquired HIV (9·6 per 100 person-years) compared with 83 of 846 who reported one or no episodes (5·2 per 100 person-years); adjusted multivariable Poisson model IRR 1·51, 1·04­2·21, p=0·032. The population attributable fractions were 13·9% (95% CI 2·0­22·2) for relationship power equity and 11·9% (1·4­19·3) for intimate partner violence.

Relationship power inequity and intimate partner violence increase risk of incident HIV infection in young South African women. Policy, interventions, and programmes for HIV prevention must address both of these risk factors and allocate appropriate resources.

National Institute of Mental Health and South African Medical Research Council.

For two decades, gender-based violence and gender inequity in relationships have been associated with increased risk of HIV in women.1­3 Cessation of violence against women and girls is one of nine priority areas in the UNAIDS Outcome Framework for 2009­11,4 and many national HIV strategic plans acknowledge the need to address gender issues, albeit with uneven implementation. Although most new HIV infections in high prevalence areas are in women, prevention agendas remain dominated by promotion of male condom use, HIV testing, treatment for sexually transmitted infections (STIs), and more recently male circumcision and antiretroviral treatment. By not focusing on gender issues, these interventions provide little help for vulnerable women.5­7 Additionally, research linking gender inequity and gender-based violence to HIV is limited. The absence of longitudinal research on the topic has enabled sceptics to resist prioritising gender in resource allocation. Yet cross-sectional studies, from both low and high prevalence settings, show associations between partner violence and male controlling practices and HIV serostatus in women.8­11 These connections arise through multiple pathways (figure), predominantly in settings or populations with a high prevalence of HIV.

Figure Pathways through which gender-based violence and gender and relationship power inequity might place women at risk of HIV infection (STIs=sexually transmitted infections)

Violence is a consequence of gender power inequities, at both a societal and relationship level, and also serves to reproduce power inequities.12 Qualitative research has shown that the links between HIV/AIDS, gender inequity, and gender-based violence lie in the patriarchal nature of society, and ideals of masculinity that are based on control of women and that celebrate male strength and toughness.12, 13 These ideals readily translate into risky sexual behaviours, predatory sexual practices, and other acts of violence against women.2 Additionally, they allow men to have multiple partners and control their sexual encounters. Emerging evidence from South Africa and India shows that men who perpetrate violence are more likely to be HIV infected.11, 14 Although individual women might resist male power, women largely acquiesce to these practices, especially in developing countries. Violence prevents women from influencing the circumstances of sex, resulting in more frequent sex, and less condom use.15­17

The figure shows rape as a potential source of HIV infection. Yet even in high prevalence settings with injury, a single sexual act has a low risk of HIV transmission;18 thus, rape results in few HIV cases in women. From a population perspective, partner violence and gender inequity have a greater effect on risk of HIV through longacting indirect pathways. These pathways pertain in chronically abusive relationships, with repeated exposure to one individual, and in women who have had previous, but not necessarily continuing, exposure to violence (in childhood or as adults) and controlling practices.

In developing and developed countries, exposure to gender-based violence, including controlling behaviour of a partner, is associated with high-risk sexual behaviour, including multiple and concurrent sexual partnerships, substance use, transactional sex and prostitution, and less frequent condom use.8,15,19­24 This association partly results from psychological effects, which often last years after the incidents of violence.25 Women might agree to riskier sex, and be less able to refuse it, when drunk, drugged, dissociating, desperately seeking affection, or otherwise manipulated by controlling partners.23­25 Thus, there is a vicious cycle, with abuse enhancing risks of HIV infection and further abuse.

Although research has improved our understanding of potential connections between intimate partner violence, gender inequity, and HIV infection, convincing assertions of causality require epidemiological evidence on temporal sequence, which has not been available.26 The dataset from the evaluation of the HIV prevention intervention Stepping Stones in rural South Africa27, 28 presents an opportunity to examine hypotheses that exposure to sexual and physical intimate partner violence and gender power inequity in relationships at baseline predict incident HIV infections over 2 years of follow-up. We did a longitudinal analysis of data from 1099 young South African women from this trial who were HIV negative at baseline and had subsequent HIV test results.


Between 2002 and 2003, male and female volunteers aged 15­26 years were enrolled in a cluster-randomised controlled trial to assess the effect of an HIV prevention programme on incidence of HIV, herpes simplex virus type 2 (HSV2), and sexual behaviour ( number NCT00332878 ).28 70 locations (clusters) in the Eastern Cape province of South Africa were grouped into seven geographically defined strata. Within each stratum, equal numbers of clusters were randomly allocated to intervention or control. In each cluster, 15­25 male and 15­25 female volunteers were recruited from schools. Volunteers were eligible for enrolment if they were aged 16­23 years, normally resident in the village where they were at school, and mature enough to understand the study and the consent process. All participants gave written informed consent.

Participants in intervention clusters were assigned to receive Stepping Stones, a 50-h participatory intervention on sexual and reproductive health and HIV, delivered over 6­8 weeks. Participants in control clusters received a 3-h intervention on safer sex and HIV, delivered on one occasion. Apart from these interventions, participants in the two groups were not treated differently. Assessments at baseline, 12 months, and 24 months consisted of a face-to-face questionnaire and blood tests for HIV and HSV2. The cohort was maintained by use of details obtained at enrolment, with follow-up undertaken nationwide to trace migrant youth. Further information about all assessments, study recruitment, access, and ethical issues, including support for participants testing HIV positive, is published elsewhere.27, 28 Ethics approval for the trial was given by the University of Pretoria, Pretoria, South Africa.

We used data obtained in this trial to assess the effects of intimate partner violence and power inequity in relationships on incidence of HIV infection at 2 years of follow-up. For this longitudinal analysis, we excluded women who had HIV infection at baseline, women with missing data, and those who were lost to follow-up at both 12 months and 24 months.

Laboratory methods

HIV serostatus at baseline was assessed by use of two rapid tests.29 The Determine (Abbott Diagnostics, Johannesburg, South Africa) test was used for screening and samples with positive results were retested with Uni-Gold (Trinity Biotech, Dublin, Ireland). Indeterminate results were clarified by use of an HIV-1 screen ELISA (Genscreen; Bio-Rad, Steenvoorde, France) followed by two confirmatory ELISAs if the sample was positive for HIV (Vironostika; BioMérieux, Marcy l'Etoile, France and Murex 1.2.0; Murex Biotech, Dartford, UK). Towards the end of the second round of interviews, collection of blood as dried spots was introduced for some participants to ease logistics and improve acceptability. In the third round of interviews, most blood was obtained as dried spots. The samples were tested with a screen and confirmatory ELISA. In this analysis, 745 (68%) of the final HIV outcomes were from dried blood spots, equally distributed among participants who remained HIV negative (n=658, 68%) and those who seroconverted (n=87, 68%).

A glycoprotein G-based HSV2 ELISA was used to test for herpes infection (Kalon; Kalon Biological, Aldershot, UK). A CAPTIA HSV IgG type-specific ELISA was used to resolve discrepant results.


We recorded age and completed years of schooling. Socioeconomic status was assessed by use of a scale that encompassed household goods ownership, food, and cash scarcity. We asked about numbers of boyfriends, concurrency, condom use, sex or pregnancy, and time since first sexual intercourse.
A sexual relationship power scale (ten items, Cronbach's alpha 0·73) was used to measure gender power equity.8, 30 A typical item was “When (NAME OF BOYFRIEND) wants me to sleep over he expects me to agree”. Each item was assessed on a 4-point Likert scale and the measure was scored and categorised into tertiles of the measure. For the analysis, the tertile with lowest equity was compared with the middle and higher ones.
We used the WHO violence against women instrument to measure physical partner violence (five items) and sexual partner violence (four items) either over the past year or during the participant's lifetime.31 By use of the method described by Dunkle and colleagues,8 we coded physical or sexual violence into more than one episode versus none or only one. Women who disclosed being gang raped, being “forced or persuaded against their will” to have sex by a non-partner, or forced by more than one man to have sex were coded as experiencing rape by a non-partner.
We derived variables for possible time-varying covariates during follow-up. In each case we considered behaviour reported at either the 12 month interview (ie, between baseline and 12 months) or the 24 month interview (if the participant had not seroconverted at 12 months). Concurrency was defined as any reported khwapheni. This is an indigenous partner category that is, by definition, concurrent.8 Condom use and correctness of condom use (ie, without breakage, slipping off, or late use) was assessed for the last sexual intercourse. Partner numbers were categorised by whether two or more were reported. Transactional sex with a casual partner was measured from questions asking about sex motivated by expectations of receiving one of a range of items (as described by Dunkle and colleagues8).

Data analysis
Analyses were done with Stata version 10.0. All procedures took into account the study design, and considered the dataset as a cohort with a stratified, two-stage structure with participants clustered within villages. For each participant, we calculated the person-years of exposure as the time from baseline to the last negative HIV result if the person remained negative, or as the total time between any negative tests as well as half the time between the last negative and first positive HIV test results.

Social, demographic, and relationship characteristics, prevalence of HSV2 infection, and violence exposures were summarised as percentages (or means) with 95% CIs, by use of standard methods for estimating CIs from complex multistage sample surveys (Taylor linearisation). Pearson's 2 test was used to test associations between categorical variables.

To account for clustering of women within villages, random effects (multilevel) models were fitted. Random effects Poisson models were built to test the hypothesis that baseline partner violence and relationship inequity predicted HIV incident infections. Each model included variables for age, study treatment group, stratum, and person-years of exposure. Partner violence, rape, and relationship power equity variables were initially modelled separately. We then fitted models with both a measure of relationship power inequity and partner violence. For the initial models, we modelled partner violence exposure in women who had had a boyfriend at baseline. We tested for, and found no interactions between variables in the model, including treatment group. We repeated the modelling with adjustment for variables that we determined a priori to be potential confounders. HSV2 seropositivity at baseline was deemed to be a potentially important biological cofactor. We also tested the models for the possibility of confounding by age, education, socioeconomic status, pregnancy, and duration of sexual activity, but found none. We tested the HSV2-adjusted model for the range of possible behavioural covariates during follow-up and found none to be confounders. We tested goodness of fit by use of the Poisson test. We confirmed the findings of associations for each outcome variable by modelling survival time under observation with a Weibull model, with the same set of other variables included. To investigate whether results were robust to missing data, we undertook a sensitivity analysis with inverse probability weighting.

By use of Greenland's method,32 the population attributable fractions (PAFs) for partner violence and low levels of relationship power equity were calculated with the incidence rates (IRs) from the adjusted model and the formula PAF=([IR 1]/IR)×Pe where Pe was the proportion of the cases that had the exposure. CIs were calculated by use of the same formula, but with the upper and lower confidence limits of the incidence rate ratio (IRR).

Role of the funding source
The National Institute of Mental Health had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to the data, analysed it, and had final responsibility for the decision to submit for publication. The South African Medical Research Council's contribution to the study consisted of discretionary funds and staff time that were entirely under the control of the first author. These made up the difference between available funds from the National Institute of Mental Health and project funding needs.

Of the 1415 women who were enrolled into the trial, 316 were excluded from this analysis (159 women had HIV infection at baseline, one had missing data, and 156 were lost to follow-up at both 12 and 24 months). The 1099 women included in this analysis represent 88% of 1256 HIV-negative women in the trial.
Table 1 shows characteristics of women who were followed up and lost to follow-up. HIV-negative women who were lost to follow-up (156 of 1256, 12%) were older, more likely to have had a boyfriend and sex at baseline, and more likely to have low relationship power equity than were those who were followed up.

Table 1 Sociodemographic and behavioural characteristics of HIV-negative women who were followed up and lost to follow-up

At 12 months and 24 months, 41 (4%) of 976 women and 18 (2%) of 962 women had not had sex, respectively. Most women were in school (1079 of 1099, 98%), with most being two or more years from school completion (table 2). Generally, the participants' households were poor or very poor, for example 385 (35%) of 1099 often or sometimes went hungry. All participants were unmarried, but 197 (18%) of 1099 were mothers.

Table 2 Sociodemographic and behavioural characteristics of women who did and did not acquire HIV over 2 years

128 women acquired HIV during 2076 person-years of follow-up (HIV incidence 6·2 per 100 person-years). Participants' characteristics by HIV serostatus at 24 months are shown in table 2. There were no significant differences in age, education, or socioeconomic status between women who did and did not acquire HIV. Compared with women who did not acquire HIV, women who seroconverted were more likely to have had sex and a pregnancy at baseline; they had been sexually active for longer, and had more concurrent relationships, both at baseline and during follow-up. A higher proportion of women who acquired HIV during the study had reported violence and high gender power inequity at baseline than had women who did not acquire the infection. Although 517 (54%) of 965 sexually active women had used condoms, women who seroconverted were more likely to have used condoms in the year before the baseline survey and to report having done so correctly at last sexual intercourse before their final HIV test than were women who did not seroconvert. Prevalence of HSV2 infection at baseline was higher in women who acquired HIV than in women who did not acquire the infection. 25 women who were still HIV negative at 12 months acquired HSV2 infection, but this did not increase risk of later HIV seroconversion.
At baseline, women in the lowest power equity category were significantly more likely to have experienced more than one episode of physical or sexual intimate partner violence than were women in the higher category (94 of 325, 29%, vs 154 of 704, 22%; p=0·014). Women who had reported violence at baseline were more likely to have HSV2 infection at baseline than were women who had not reported violence (78 of 253, 30·8%, vs 167 of 846, 19·7%; p=0·0002).
The incidence of HIV and IRR derived from the adjusted multivariable Poisson models are shown in table 3. Women in relationships with low gender equity at baseline had a much higher incidence of HIV than did women in the middle or highest relationship equity category. HIV incidence was higher in women who had reported more than one episode of physical or sexual intimate partner violence at baseline than in women who reported one or no episodes of violence. Women who had been raped by a non-partner had a similar incidence of HIV to those who had not. Table 4 shows two Poisson models with exposure to violence and relationship power equity as independent variables. Both variables had an independent association with HIV incidence, after adjustment for age, study design, and HSV2 infection at baseline. Behavioural variables were not confounders (table 5).

Table 3 Incidence and relative incidence of HIV infection, by exposure to forms of violence and inequity

Table 4 Relative HIV incidence with exposure to both partner violence and relationship inequity

Table 5 Effects of behavioural variables on relative incidence of HIV

The population attributable fractions (table 6) show that 13·9% of incident HIV infections could be avoided if gender equity in heterosexual relationships was enhanced so that no women were in relationships with low power. Similarly, for violence, 11·9% of new HIV infections could be prevented if women did not experience more than one episode of physical or sexual partner violence.

Table 6 Population attributable fractions
A sensitivity analysis to investigate whether results were robust to missing data suggested that the potential effect of missing data was negligible (data not shown).

This study shows that in rural South Africa, women who experienced intimate partner violence and had high gender inequity in relationships had increased incidence of HIV infection. Risk of incident HIV infection was not associated with rape by a non-partner. Our results substantiate previous findings from cross-sectional studies.8 The population attributable fractions reported here show the importance of effectively addressing the HIV epidemic through programmes and interventions that address violence and gender inequity in relationships.

This study provides strong temporal evidence to support a causal association between intimate partner violence or relationship inequity and new HIV infection. The relation between these variables is plausible and coherent, and research from several settings has shown consistency and supports the strength of association. There is no experimental evidence to suggest that reducing women's exposure to violence reduces HIV incidence. Replicating this association in the context of trials to assess effective interventions should be a priority. There are some promising interventions that seek to empower women and change ideals of masculinity.7, 28 In addition to Stepping Stones, potentially effective interventions from around the world include Sexto Sentido from Nicaragua,33 Program H from Brazil,34 the Intervention with Microfinance for AIDS and Gender Equity (IMAGE) programme from South Africa,35 and Better Life Options for Boys from India.36 Research to assess and refine these and other such programmes is urgently needed.37

Incidence of HIV in our cohort (6%) was similar to that estimated nationally in women aged 15­24 years (6·5%).38 The results of two-way associations between HIV seroconversion and baseline sexual risk behaviour and HSV2 infection, and between seroconversion and concurrency during follow-up were as expected. The exception here was condom use, both frequency of use at baseline and during follow-up. Although possibly explained by measurement error, the consistency across several condom use indicators suggests another explanation. Since condom use is predominantly determined by male partners, it is more likely that men initiate use after a diagnosis of HIV or STI, but then become inconsistent in condom use.

Elimination of confounding variables, without over-adjusting the models,26 is difficult to achieve in observational epidemiology, especially with few new infections. We tested a range of possible confounders and found that none substantially altered the effect sizes of interest. Having tested the models extensively for confounding and interactions, we are confident that our models were appropriately constructed and fit the data well.
This study has strengths and limitations. The cohort consisted of volunteers, and although retention was high, there was some loss to follow-up and bias in the characteristics of those who were lost. We tested for robustness to missing data and our results suggested that the potential effect was small. We did not adjust for changes in exposure category during follow-up, in view of the short period of observation and uncertainty about temporality of infection and violence. We cannot exclude effects of exposure change on the findings, but expect that any effects would bias findings towards the null hypothesis. Follow-up data were interval censored and the model included the standard method of dealing with uncertainty about the time of HIV seroconversion in such data. We cannot know when seroconversion occurred and so cannot accurately estimate the errors in such an assumption, although we do not think that these would be biased by exposure to violence or gender inequity. We tried to ensure that this recognised and unavoidable problem39 did not result in spurious findings by confirming the results through two different approaches to model building. New HIV infections might have been slightly under-ascertained because of the period of study. Finally, the Stepping Stones intervention might have introduced biases in this analysis, but we adjusted for study treatment group and checked for interactions. There was no effect of treatment group on incidence of HIV in women.28

Exposure to physical and sexual intimate partner violence and low relationship power equity increase incidence of HIV in young women in rural areas of South Africa, and account for a substantial proportion of HIV infections. The association is likely to be causal because it accords with cross-sectional evidence from countries around the world. Organisations driving HIV prevention agendas for women, particularly UNAIDS and WHO, need to ensure that policies, programmes, and interventions to build gender equity and prevent partner violence are developed and widely implemented. Donors and researchers must invest efforts and resources in developing and testing new interventions.

RKJ designed the Stepping Stones study and was the principal investigator for the trial. RKJ analysed the data for this study and wrote the report. KD contributed to the design of this study, assisted with data management, and assisted in the interpretation of the data and drafting of the report. MN contributed to the design of the study and project management. NS contributed to study design and implementation. All authors saw and approved the final version of the report.

Conflicts of interest
We declare that we have no conflicts of interest.

This study was funded by the National Institute of Mental Health (grant numbers MH 64882-01, MH 64882-04S1A1, and 1R03MH085599) and the South African Medical Research Council. We thank all the members of the Stepping Stones study team and the National Institute for Communicable Diseases for quality control, testing, and storage of specimens; the staff who worked on the project; and members of the community advisory board and data safety and monitoring board. Carl Lombard assisted with data analysis.

1 Maman S, Campbell J, Sweat MD, Gielen AC. The intersections of HIV and violence: directions for future research and interventions. Soc Sci Med 2000; 50: 459-478. CrossRef | PubMed
2 Garcia-Moreno C, Watts C. Violence against women: its importance for HIV/AIDS. AIDS 2000; 14 (suppl 3): S253-S265. PubMed
3 UNAIDS, WHO. AIDS epidemic update 2005. (accessed May 11, 2010).
4 UNAIDS Joint Action for Results. UNAIDS Outcome Framework 2009­2011. (accessed May 11, 2010).
5 Wawer MJ, Makumbi F, Kigozi G, et al. Circumcision in HIV-infected men and its effect on HIV transmission to female partners in Rakai, Uganda: a randomised controlled trial. Lancet 2009; 374: 229-237. Summary | Full Text | PDF(166KB) | CrossRef | PubMed
6 Christofides N, Jewkes R. Acceptability of universal screening for intimate partner violence in voluntary HIV testing and counseling services in South Africa and service implications. AIDS Care 2010; 22: 279-285. PubMed
7 Greig A, Peacock D, Jewkes R, Msimang S. Gender and AIDS: time to Act. AIDS 2008 (suppl 2): S35-S43. PubMed
8 Dunkle KL, Jewkes RK, Brown HC, Gray GE, McIntyre JA, Harlow SD. Gender-based violence, relationship power, and risk of HIV infection in women attending antenatal clinics in South Africa. Lancet 2004; 363: 1415-1421. Summary | Full Text | PDF(117KB) | CrossRef | PubMed
9 van der Straten A, King R, Grinstead O, Vittinghoff E, Serufilira A, Allen S. Sexual coercion, physical violence, and HIV infection among women in steady relationships in Kigali, Rwanda. AIDS Behav 1998; 2: 61-73. CrossRef | PubMed
10 Maman S, Mbwambo JK, Hogan NM, et al. HIV-positive women report more lifetime partner violence: findings from a voluntary counselling and testing clinic in Dar es Salaam, Tanzania. Am J Public Health 2002; 92: 1331-1337. CrossRef | PubMed
11 Decker MR, Seage GR, Hemenway D, et al. Intimate partner violence functions as both a risk marker and risk factor for women's HIV infection: findings from Indian husband-wife dyads. J Acquir Immune Defic Syndr 2009; 51: 593-600. CrossRef | PubMed
12 Jewkes R, Morrell R. Gender and sexuality: emerging perspectives from the heterosexual epidemic in South Africa and implications for HIV risk and prevention. J Int AIDS Soc 2010; 13: 6. PubMed
13 Coovadia H, Jewkes R, Barron P, Sanders D, McIntyre D. The health and health system of South Africa: historical roots of current public health challenges. Lancet 2009; 374: 817-834. Summary | Full Text | PDF(1584KB) | CrossRef | PubMed
14 Jewkes R, Sikweyiya Y, Morrell R, Dunkle K. Understanding men's health and use of violence: interface of rape and HIV in South Africa. Technical report. Pretoria: Medical Research Council, 2009. (accessed May 11, 2010).
15 Jewkes R, Dunkle K, Nduna M, et al. Factors associated with HIV sero-status in young rural South African women: connections between intimate partner violence and HIV. Int J Epidemiology 2006; 35: 1461-1468. PubMed
16 Wood K, Maforah F, Jewkes R. “He forced me to love him”: putting violence on the adolescent sexual health agenda. Soc Sci Med 1998; 47: 233-242. CrossRef | PubMed
17 Pettifor AE, Measham DM, Rees HV, Padian N. Sexual power and HIV risk, South Africa. Emerg Infect Dis 2004; 11: 1996-2004. PubMed
18 Gray RH, Wawer MJ, Brookmeyer R, et althe Rakai Project Team. Probability of HIV-1 transmission per coital act in monogamous, heterosexual, HIV-1-discordant couples in Rakai, Uganda. Lancet 2001; 357: 1149-1153. Summary | Full Text | PDF(90KB) | CrossRef | PubMed
19 El-Bassel N, Gilbert L, Krishnan S, et al. Partner violence and sexual HIV-risk behaviors among women in an inner-city emergency department. Violence Vict 1998; 13: 377-393. PubMed
20 Gilbert L, El-Bassel N, Schilling RF, Wada T, Bennet B. Partner violence and sexual HIV risk behaviors among women in methadone treatment. AIDS Behav 2000; 4: 261-269. CrossRef | PubMed
21 Jewkes RK, Levin JB, Penn-Kekana LA. Gender inequalities, intimate partner violence and HIV preventive practices: findings of a South African cross-sectional study. Soc Sci Med 2003; 56: 125-134. CrossRef | PubMed
22 Zablotska IB, Gray RH, Koenig MA, et al. Alcohol use, intimate partner violence, sexual coercion and HIV among women aged 15­24 in Rakai, Uganda. AIDS Behav 2009; 13: 225-233. CrossRef | PubMed
23 Johnson SD, Cunningham-Williams RM, Cottler LB. A tripartite of HIV-risk for African-American women: the intersection of drug use, violence and depression. Drug Alcohol Depend 2003; 70: 169-175. CrossRef | PubMed
24 Wingood GM, DiClemente RJ. Rape among African American women: sexual, psychological, and social correlates predisposing survivors to risk of STD/HIV. J Womens Health 1998; 7: 77-84. PubMed
25 Wang SH, Rowley W. Rape: how men, the community and the health sector respond. Geneva: Sexual Violence Research Initiative and World Health Organization, 2007. (accessed May 11, 2010).
26 Doll R. Proof of causality. Deduction from epidemiological observation. Perspect Biol Med 2002; 45: 499-515. PubMed
27 Jewkes R, Nduna M, Levin J, et al. A cluster randomised controlled trial to determine the effectiveness of Stepping Stones in preventing HIV infections and promoting safer sexual behaviour amongst youth in the rural Eastern Cape, South Africa: trial design, methods and baseline findings. Trop Med Int Health 2006; 11: 3-16. CrossRef | PubMed
28 Jewkes R, Nduna M, Levin J, et al. Impact of Stepping Stones on HIV and HSV-2 and sexual behaviour in rural South Africa: cluster randomised controlled trial. BMJ 2008; 337: a506. PubMed
29 WHO. Rapid HIV tests: guidelines for use in HIV testing and counselling services in resource-constrained settings. Geneva: World Health Organization, 2004. (accessed May 11, 2010).
30 Pulerwitz J, Gortmaker S, De Jong W. Measuring sexual relationship power in HIV/STD research. Sex Roles 2000; 42: 637-660. PubMed
31 WHO. WHO Multi-country study on women's health and domestic violence: core questionnaire and WHO instrument. Version 9. Geneva: World Health Organization, 2000.
32 Greenland S. Applications of stratified analysis methods. In: Rothman KJ, Greenland S, eds. Modern epidemiology. Philadelphia: Lippincott, Williams & Wilkins, 1998.
33 Solórzano I, Pulerwitz J, Pena R, Ellsberg M, Banks A. Positive impact of an HIV prevention communication for social change program on youth in Nicaragua. . (accessed Jan 20, 2010).
34 Pulerwitz J, Barker G. Promoting more gender-equitable norms and behaviors among young men as an HIV/AIDS prevention strategy. (accessed Jan 20, 2010).
35 Pronyk PM, Hargreaves JR, Kim JC, et al. Effect of a structural intervention for the prevention of intimate-partner violence and HIV in rural South Africa: a cluster randomised trial. Lancet 2006; 368: 1973-1983. Summary | Full Text | PDF(126KB) | CrossRef | PubMed
36 Synergy, Men and reproductive health programs: influencing gender norms. Washington, DC: Synergy Project, 2003.
37 Barker G, Ricardo C, Nascimento M. Engaging men and boys to transform gender-based health inequities: is there evidence of impact?. Geneva/Rio de Janiero: World Health Organization/Institute Promundo, 2007.
38 Shisana O, Rehle T, Simbayi LC, et al. South African national HIV prevalence, HIV incidence, behaviour and communication survey, 2005. Cape Town: Human Sciences Research Council, 2005.
39 Lindsey JC, Ryan LM. Tutorial in biostatistics methods for interval-censored data. Stat Med 1998; 17: 219-238. CrossRef | PubMed

a Gender and Health Research Unit, Medical Research Council, Pretoria, South Africa
b School of Public Health, University of the Witwatersrand, Johannesburg, South Africa
c Department of Psychology, University of the Witwatersrand, Johannesburg, South Africa
d Rollins School of Public Health/Center for AIDS Research, Emory University, Atlanta, GA, USA

Correspondence to: Prof Rachel K Jewkes, Gender and Health Research Unit, Medical Research Council, Private Bag X385, Pretoria 0001, South Africa