#### Introduction

Systematic mass rape during armed conflict ^{[1–3]} now occurs so frequently that it has been recognized, and legally defined, by the United Nations as a weapon of war ^{[4]}. The highest number of armed conflicts occur in sub-Saharan Africa ^{[5]}, where many conflict-afflicted countries (CACs) have a high prevalence of HIV ^{[6]}. Consequently, it has been hypothesized by several authors ^{[7–11]} that mass rape may be increasing HIV epidemics in some CACs. This hypothesis has recently been questioned by Spiegel *et al.* ^{[12]}, who have assessed the link between mass rape and HIV prevalence by conducting a detailed review of the literature. In their review, they included studies from seven CACs in sub-Saharan Africa: the Democratic Republic of Congo (DRC), southern Sudan, Rwanda, Uganda, Sierra Leone, Somalia and Burundi ^{[12]}. Despite the fact that data show these countries have a high prevalence of HIV ^{[6]} and that wide-scale rape has occurred, the authors found no data to show rape has increased HIV prevalence ^{[12]}. Their finding has recently been corroborated by Anema *et al.* ^{[13]}, who used a risk equation model to estimate the potential impact of mass rape on HIV prevalence in the same seven CACs investigated by Spiegel *et al*. ^{[12]}. Their risk modeling shows that even under the most extreme circumstances, the impact of mass rape on prevalence could be expected to be negligible (i.e. they determined that prevalence would only increase by 0.023%) ^{[13]}. However, Anema *et al*. did not assess the potential impact of mass rape in CACs on increasing HIV incidence. Here, we use risk equation modeling to investigate the potential impact of mass rape on increasing HIV incidence (as well as prevalence) in the DRC, southern Sudan, Rwanda, Uganda, Sierra Leone, Somalia and Burundi.

To inform our modeling, we obtained parameter estimates from the literature. Reliable statistics on mass rape are difficult to obtain, but surveys have revealed that the number of women who are raped during conflicts can be extremely high. Nine percent of women in Sierra Leone reported war-related sexual assaults ^{[14]}, whereas 15% of women in Liberia reported rape or sexual coercion ^{[15]}. It has been estimated hundreds of thousands of women were raped during the genocide in Rwanda in the early 1990s ^{[16,17]}. Widespread sexual violence has been reported during the current conflict in the DRC ^{[3,18–21]}; for example, the organization Malteser International registered more than 20 000 female rape survivors during the 3-year period 2005–2007 ^{[19]}. For the period November 2008 to March 2009, it has been estimated that over a thousand women were raped each month in the DRC ^{[22]}.

The probability that a woman raped during an armed conflict will become infected with HIV depends on whether the rapist is infected with HIV or not, and the probability that HIV is transmitted during the rape. The probability the rapist is infected can be expected to be related to the prevalence of HIV in armed combatants. Several studies have found that in CACs with severe HIV epidemics, the HIV prevalence among the military forces can be significantly greater than in the general population. For example, a recent study of 21 African countries found, on average, HIV prevalence within the military was six times as high as that of males aged 15–24 years in the general population ^{[23]}. In Uganda, the HIV prevalence among military forces was 10 times higher than among males aged 15–24 years in the general population ^{[23]}. The probability that HIV is transmitted during an act of rape is unknown. However, this transmission probability is likely to be higher than for consensual sex, as mass rape often involves violence and genital trauma including widely reported cases of vaginal fistulae ^{[21,24,25]}. For consensual sex between heterosexuals in Africa, the per-act probability of HIV transmission (given that the partner is infected) has been estimated to be 0.0009 (range 0.0006–0.0012) and to rise to 0.0559 (range 0.0044–0.1073) if sexually transmitted infections (STIs) are present ^{[26]}. Notably, the highest rate of STIs occurs in sub-Saharan Africa ^{[27]}, and armed combatants often have high STI rates ^{[28,29]}. Taken together, the available data suggest that the probability that a woman raped during conflict would become infected with HIV could be fairly high.

#### Methods

We used the static risk equation model [Eq. (1)] developed by Anema *et al*. ^{[13]} to estimate the number of new HIV infections that could occur in the DRC, southern Sudan, Rwanda, Uganda, Sierra Leone, Somalia and Burundi (over 1 year) due to mass rape. Following Anema *et al.* ^{[13]}, we assume that the population at risk of rape is females aged between 5 and 49 years. Equation (1) shows the number of uninfected women and girls (5–49 years) that could become infected with HIV due to mass rape during armed conflicts (

) is calculated as the product of the number of women and girls (5–49 years) who are currently uninfected (*X*_{5–49}), the proportion of the female population (5–49 years) who are raped (*r*), the prevalence of HIV among the assailants (*P*_{r}) and the average probability of transmission per act of rape (*β*_{ar}):

The HIV prevalence among assailants (*P*_{r}) is assumed to be higher than the HIV prevalence in the rest of the male population (*P*), we defined *P*_{r} = *α* × *P* or *α* = *P*_{r}/*P*, where *α* is called the prevalence ratio and is assumed to be greater than or equal to 1.

To make calculations using Eq. (1), we estimated the parameters from data. We used data from UNAIDS/WHO ^{[30]} to determine HIV prevalence (i.e. *P*) in the DRC, southern Sudan, Rwanda, Uganda, Sierra Leone, Somalia and Burundi. We then estimated the number of uninfected women and girls aged between 5 and 49 years (i.e. *X*_{5–49}) in each of the seven CACs by subtracting the number of prevalent infections (determined from UNAIDS/WHO data ^{[30]}) from the total number of females aged 5–49 years (determined from demographic data in the US Census Bureau's International Database ^{[31]}). For the remaining parameters in Eq. (1), we explored, for comparative purposes, the same ranges that Anema *et al.* ^{[13]} used in their analyses. Hence, we assumed 1–15% of the female population (5–49 years) could be raped (i.e. *r* varied from 0.01 to 0.15), the prevalence of HIV in assailants would be one to eight times greater than in the general male population (i.e. *α* varied from 1 to 8) and the average probability of HIV transmission per act of rape (*β*_{ar}) varied between 0.0028 and 0.032.

We used Eq. (2) to calculate country-specific annual HIV incidence due to consensual sex among uninfected women and girls aged 15–49 years (

). This incidence is calculated as the product of several factors: the number of uninfected women and girls aged 15–49 years (*X*_{15–49}), the HIV prevalence in the general male population (*P*), the average probability of HIV transmission per partnership (*β*_{p}) and the average number of new consensual sex partners per year (*c*):

where

*n* is the number of sex acts per consensual partnership per year and *β*_{ac} is the average probability of HIV transmission per consensual sex act. We estimated the value for *X*_{15–49}, as we had for *X*_{5–49}, using HIV prevalence data from UNAIDS/WHO ^{[30]} and demographic data from the US Census Bureau's International Database ^{[31]}. We assumed that the probability of transmitting HIV during one act of rape is equal to or greater than that during one act of consensual sex. Therefore, we set *β*_{ac} to be the lowest value of *β*_{ar} (i.e. *β*_{ac} was set to 0.0028). We calibrated our risk equation model by setting ranges for *c* and *n*; calibration occurred when the country-specific HIV incidence (in times of no conflict) estimated by the model matched the UNAIDS estimate of incidence. At calibration, *c* ranged from 0.5 to 1 and *n* ranged from 20 to 100.

We then divided Eq. (1) (which specifies the number of HIV infections due to mass rape of 5–49-year-olds during conflict) by Eq. (2) (which specifies the number of HIV infections due to consensual sex in 15–49-year-olds in peace time) to determine the percentage increase in the annual number of HIV infections in women and girls due to mass rape during conflict:

To obtain quantitative estimates using the above expression, we used uncertainty analysis and sampled ranges on the following parameters: the proportion of the population who are raped, the prevalence ratio, the average probability of HIV transmission per act of rape, the average number of consensual partnerships per year and the average number of sex acts per consensual partnership per year. We assume that – in any particular year – the average probability of transmission per act of rape can be regarded as approximately constant. We make this assumption because the average value of this probability depends on the level of infectivity averaged over all the perpetrators in the population. Over 1 year, the infectiousness of some of these individuals may change (e.g. some individuals will move from primary to chronic infection), but the infectiousness of the majority of these individuals is unlikely to change substantially over this period. Consequently, it is reasonable to assume that – over a year – the average probability of transmission per act of rape remains approximately constant. However, if we were modeling the impact of rape over a period of several years, it would be necessary to include variability in infectiousness over time (i.e. to include variation in *β*_{ar} over time).

We then sampled all of the parameter ranges using Latin Hypercube Sampling ^{[32]} and used this sample to run 10 000 Monte Carlo simulations. We, therefore, made 10 000 estimates for each of the seven CACs.

#### Results

UNAIDS/WHO data ^{[30]} show that HIV prevalence and incidence in the general populations of the seven CACs we are investigating is high, with the lowest prevalence occurring in Somalia and the highest in Uganda (Table 1). Notably, calibration ensured that the incidence estimated by our model is in close agreement with the UNAIDS/WHO incidence data; compare column 5 in Table 1 with column 2 in Table 2. In terms of incidence rates, we estimate that mass rape could cause a median of five new HIV infections per 100 000 women per year in the DRC, southern Sudan, Sierra Leone and Somalia, double the number in Burundi and Rwanda and quadruple the number in Uganda (Table 2).

Table 1 Image Tools |
Table 2 Image Tools |

Due to differences in population sizes, we estimate that the annual number of women and girls (aged 5–49 years) who could become infected with HIV due to mass rape varies substantially among the seven CACs (Fig. 1). For two of these countries, our estimates show that this number would be relatively low; 127 [median (interquartile range [IQR] 55–254)] new HIV infections for Somalia and 156 [median (IQR 69–305)] new HIV infections for Sierra Leone (Fig. 1). However, in the DRC and Uganda, our estimates show the numbers could be high; 1120 [median (IQR 527–2360)] new HIV infections for the DRC and 2172 [median (IQR 1031–4668)] new HIV infections for Uganda. For Burundi, Rwanda and Sudan, the numbers are intermediate (Fig. 1). In total, we estimate that 4948 [median (IQR 2043–10 329)] women and girls could be infected with HIV due to mass rape each year in the seven CACs we have investigated.

Figure 2 shows, for the DRC (a) and Uganda (b), the estimates from the uncertainty analysis of the annual number of new HIV infections in women and girls (aged 5–49 years) due to mass rape. The graphs in Fig. 2 are constructed by varying all the parameters in the risk equation and then stratifying the estimates on the basis of the prevalence of rape. This figure illustrates that the distribution of the model-based estimates is highly skewed. The distribution is skewed as parameter ranges are sampled during the uncertainty analysis in order to account for heterogeneities in transmission risk due to the presence of STIs or other cofactors ^{[33,34]}. The skewed distribution in estimates also reflects the sampled variability in the prevalence ratio; the prevalence of HIV in assailants relative to the prevalence in the general population will substantially affect the influence of mass rape on increasing incidence. Notably, Fig. 2 shows that under extreme conditions, 10 000 women and girls could be infected per year in the DRC and 20 000 women and girls could be infected per year in Uganda.

Our analysis shows that mass rape could have a significant effect on increasing incidence, but would have a negligible impact on prevalence (Table 2). Specifically, we calculate that the median increase in incidence would be 6–7%, but the median increase in prevalence would be less than 0.02%. Figure 3 shows the relationship between the increase in incidence and the prevalence of rape. Our calculations show that in CACs where 15% of girls and women are raped, the annual HIV incidence could increase by a median of 15% (Fig. 3). However, the increase in incidence under these conditions could be significantly greater; 25% of the Monte Carlo simulations in our uncertainty analysis show that incidence could be increased by more than 27% (Fig. 3).

#### Discussion

The impact of mass rape on HIV epidemics in CACs is currently unknown. It is difficult to capture incidence data in contexts of crisis, thus there is a paucity of data available for assessing the consequences of mass rape on HIV incidence. We have addressed this problem by using a quantitative risk assessment framework to assess the potential effect of mass rape on HIV incidence. We have also assessed the potential effect on HIV prevalence. Our risk modeling indicates that even under extreme conditions, mass rape is likely to have a negligible impact on HIV prevalence. Our results, with respect to the potential effect of mass rape on HIV prevalence, are in agreement with those previously published by Anema *et al.* ^{[13]}, who calculated that prevalence would increase by only 0.023% even if 15% of the female population was raped, prevalence of HIV in the group of assailants was eight times greater than in the general population, and the transmission probability of HIV per act of rape was four times greater than the transmission probability per act of consensual sex ^{[13]}. However, Anema *et al.* ^{[13]} focused only on HIV prevalence, whereas we have assessed the potential impact of mass rape on HIV incidence as well as prevalence. Notably, our results show that although mass rape in CACs may not significantly affect prevalence, it could significantly affect incidence. Although our analysis is unlikely to have produced precise estimates of the impact of mass rape on incidence, it demonstrates the potential magnitude of the effect. Consequently, we can conclude that thousands of women and girls could be infected with HIV each year due to mass rape in CACs in sub-Saharan Africa. Furthermore, it should be note that the true value for the probability of transmission per act of rape could be as high as 0.1073 if STIs are present ^{[26]}. If the probability of transmission is this high, then the number of women and girls who are infected due to mass rape each year could be three times higher than what we have estimated.

Our analysis has only examined the consequence of mass rape on HIV incidence due to direct infection. Mass rape also indirectly increases HIV incidence through at least two other mechanisms: women who become infected through rape can transmit HIV to their future male partners and some survivors of rape may be infected with an STI (other than HIV) that increases their susceptibility to HIV. Moreover, our analysis does not include the fact that mass rape leads to unwanted pregnancies and subsequent HIV infections due to mother-to-child transmission (MTCT). Clearly, more sophisticated mathematical models that include these additional effects are needed in order to more adequately assess the impact of mass rape in CACs on HIV epidemics. As we have not included these effects in our current analyses, our estimates are likely to be underestimates of the effect of mass rape on increasing incidence (and ultimately prevalence) of HIV epidemics.

We stress that the type of modeling exercise that we have conducted in this study is only designed to model epidemiological processes. Such exercises cannot evaluate the impact of physical damage due to violent assault nor the unimaginable long-term suffering and psychological scarring due to rape ^{[35]}. In addition, modeling cannot be used to quantify the fact that survivors of sexual violence are often stigmatized and rejected by their partners and family ^{[35]}. Stringent protocols for supporting and treating women and girls subjected to rape in conflict situations are urgently needed. These protocols should include immediate medical care for traumatic genital injury caused by forced sex and treatment for STIs. However, it may not always be possible to implement such protocols. Even raising the issue of rape in conflict situations can itself prove hazardous. For example, in 2006, a Norwegian humanitarian organization was expelled from South Darfur following allegations of false reporting of rape cases and ‘promoting a foreign agenda’ ^{[36]}. Although 250 000 people had been killed in the conflict and a further 300 000 displaced, a government spokesman for South Darfur stated that no rape cases had occurred. A year previously, two members of Médecins Sans Frontières (MSF) had been arrested following an MSF report that detailed the treatment they had provided to hundreds of rape survivors in local hospitals ^{[35]}. Until 2005, women in Darfur could be threatened with imprisonment if they reported rape and healthcare staff risked incarceration for treating them. Although this policy has now changed, it has left a legacy of mistrust of the state, with women reluctant to report rape and healthcare workers cautious about treating rape survivors ^{[36]}.

Clearly, medical interventions are needed, on the basis of human rights, to ameliorate the experiences of women caught in the trauma of armed conflict and exposed to mass violence and rape. In our analysis, we have extended this perspective. By taking a quantitative risk assessment approach, we have shown (for the first time) that mass rape during armed conflicts may significantly increase HIV incidence. Our results have extremely important health policy implications. They indicate that HIV prevention and treatment strategies targeted to rape survivors could potentially have an impact on reducing HIV incidence in CACs. We recommend that interventions for prevention to reduce MTCT, postexposure prophylaxis and termination of unwanted pregnancies should be made widely available to survivors of rape in regions of armed conflict. During conflicts, many women can be forced to migrate and seek shelter in camps, often separating them from social and safety networks and exposing them to sexual violence ^{[37,38]}. Focusing targeted interventions in refugee camps may be an effective means for reaching rape survivors. Further research is needed to improve understanding of the impact of mass rape on HIV epidemics and to assist in the design of evidence-based medical interventions for rape survivors in CACs. With implementation of targeted, effective policies and service interventions, it may be possible to reduce HIV incidence during armed conflicts. Consequently, medical interventions and support for rape survivors during armed conflicts should be provided on both the basis of human rights and public health.

#### Acknowledgements

V.S. designed the study, carried out the computations, interpreted the results and contributed to writing the manuscript. Y.H. interpreted the results and contributed to writing the manuscript. S.B. conceived and designed the study, interpreted the results, and contributed to writing the manuscript. All authors saw and approved the final version of the manuscript.

S.B. and V.S. are grateful for the financial support of the US National Institutes of Health, National Institute of Allergy and Infectious Diseases (grant RO1 AI041935). S.B. also thanks the John Simon Guggenheim Foundation for funding. Y.H. did not receive any funding for this research.

There are no conflicts of interest.