Share this article on:

Highly efficient HIV transmission to young women in South Africa

Pettifor, Audrey Eb,d; Hudgens, Michael Gc; Levandowski, Brooke Ab; Rees, Helen Vd; Cohen, Myron Sa,b

doi: 10.1097/QAD.0b013e3280f00fb3
Epidemiology and Social

Background: Young women in sub-Saharan Africa are at very high risk of HIV acquisition, and high prevalence levels have been observed among women reporting one lifetime partner and few sexual contacts. Such findings have led to hypotheses that the probability of HIV transmission from men to women must be far higher than previously appreciated.

Methods: We used the data from a cross-sectional national household survey of HIV among South African women aged 15–24 years to estimate the per-partnership transmission probability from men to women. Estimates were obtained using maximum likelihood methods and a transmission probability model allowing for random error in the self-reported number of lifetime partners. Sensitivity analyses were employed to assess the robustness of the per-partnership transmission probability estimates to the assumed HIV prevalence in male partners.

Results: HIV prevalence in women was 21.2% (95% confidence interval 17.9–24.5). The mean reported number of lifetime partners was 2.3. A significant increase in prevalence was observed with increasing lifetime partner numbers (P = 0.02). For a range of plausible values of the partner prevalence, the estimated per-partnership transmission probability varied from 0.74 to 1.00 with 95% confidence intervals ranging from 0.56 to 1.00.

Discussion: The per-partnership probability of HIV transmission from men to women in this population was very high. Before this, the majority of studies examining per-partnership transmission probabilities estimated values below 50%. Identifying the factors that may drive the efficient spread of HIV in sub-Saharan Africa is essential for the development of effective prevention interventions.

From the aDivision of Infectious Diseases, School of Medicine, USA

bDepartments of Epidemiology, USA

cBiostatistics, University of North Carolina, Chapel Hill, North Carolina, USA

dReproductive Health and HIV Research Unit, University of the Witwatersrand, Johannesburg, South Africa.

Received 23 October, 2006

Revised 19 January, 2007

Accepted 29 January, 2007

Correspondence to Audrey Pettifor, Department of Epidemiology, CB #7435, McGavran-Greenberg Building, University of North Carolina at Chapel Hill, Chapel Hill, NC 27599-7435, USA. Tel: +1 919 933 5378; fax: +1 919 966 2089; e-mail:

Back to Top | Article Outline


Young women in sub-Saharan Africa are at very high risk of HIV acquisition [1]. We recently reported that among the general population of young women in South Africa, the prevalence of HIV infection increased from 4% among 15–16 year olds to 30% among 21 year olds [2]. These remarkable and tragic levels of infection were observed in the absence of any exceptionally risky sexual behavior. Similarly high HIV prevalence levels have been observed elsewhere in sub-Saharan Africa among young women reporting one lifetime partner and few sexual contacts [3,4]. Such findings have led to hypotheses that the probability of HIV transmission from men to women must be far higher than hitherto appreciated. Unfortunately, previous studies of transmission probabilities were limited by a focus on serodiscordant couples or study designs that could not capture the risk properly [5]. In this paper we use data from a national household survey of HIV among South African women to estimate the per-partnership transmission probability from men to women with the aim of better understanding young women's heightened risk of HIV acquisition.

Back to Top | Article Outline


Data on HIV prevalence and sexual behavior were collected as part of a cross-sectional nationally representative household survey of 11 904 young people aged 15–24 years conducted in South Africa in 2003. Details of the study methodology and main findings are described elsewhere [2]. Briefly, young people were asked to take part in a behavioral interview and to provide an anonymous oral fluid specimen for the detection of HIV infection. Oral fluid samples were collected using the Orasure HIV-1 oral specimen collection device (Orasure Technologies Inc., Bethlehem, Pennsylvania, USA) and tested for HIV-1/2 antibodies using the Vironostika Uni-Form II HIV-1/2 plus O MicroELISA System (Biomerieux, Durham, North Carolina, USA). This analysis focuses on a subset of sexually experienced young women who provided detailed information on their sexual behaviors.

All sexually experienced participants were asked about the number of lifetime sexual partners. In addition, detailed information about their first sexual partner and information on the last three sexual partners in the past 12 months was also collected. For each of these possible four partnerships, information was collected on the duration of the relationship (time from first to last sexual encounter), age of the sexual partner, whether the relationship was currently ongoing or ended, and the frequency of sexual contacts in the past month. The overall number of sexual contacts for the entire relationship was estimated by multiplying the number of months of each relationship with the number of sexual contacts reported in the past month. If the relationship was reported as ongoing and the number of sexual contacts in the past month was reported as zero, a rate of 0.5 contacts per month was imputed [3].

The male-to-female per-partnership transmission probability, i.e. the probability that an HIV-negative woman becomes infected during a sexual partnership with an HIV-infected man, was estimated based on the following models. Let P denote the male-to-female per-partnership transmission probability. For a particular woman, let a denote the average age of her male partners during their relationships and let π(a) denote the HIV prevalence in the population of sexually active men a years of age. Let n denote the reported number of lifetime male partners and τ be the time from sexual debut until the survey, such that c = n/τ is the rate of partner acquisition (i.e. partners per year of sexual activity). The hazard rate of infection at time t was modeled [6] as λ(t; c, a, x) = cpπ(a) exp(βx) where x denotes additional covariates (race, age of the woman, province of residence, and absence/presence of vaginal discharge), which might affect the risk of infection and β has the usual log hazard ratio interpretation. Letting Λ(c, a, x) = npπ(a) exp(βx) denote the cumulative hazard of infection since coital debut, the probability the woman escapes infection equals exp{−Λ(c, a, x)} and the probability of infection equals 1− exp{−Λ(c, a, x)}. The likelihood contribution thus equals



where δ equals 1 (0) if the woman is HIV positive (negative) at the time of the survey. We refer to this model as the conditional model because it is conditional on the reported number of lifetime sexual partners.

To adjust for possible random error in reporting the number of lifetime partners, an errors-in-variables (EIV) approach [7] was also considered. For this model, we assumed λ(t; ct, a, x) = ct p π(a) exp(βx) where ct is the latent ‘true’ rate of sexual partner acquisition with log(ct) = u + e 1 and log(c) = log(ct) + e 2 where e 1 and e 2 are normally distributed error terms with mean zero and variance V 1 and V 2, respectively. In words, V 1 is the variance between women with regard to their true partner rates on the log scale and V 2 is the variance of the reporting error. As ct is not observed, an empirical Bayes approximation [7] is used in constructing the likelihood by replacing c in (1.1) with E(ct|c), the expected true contact rate given the reported contact rate c. The conditional model corresponds to the special case of the EIV model when V 2 = 0.

All analyses were conducted allowing for the sampling design. For the transmission probability models, individual contributions to the log-likelihood were weighted inversely proportional to the probability of being sampled. Maximum likelihood estimates and confidence intervals for P were obtained assuming that the prevalence of HIV in the population of male partners (π(a)) was known (see Table 1) [2,10]. Sensitivity analyses were employed to assess the robustness of the maximum likelihood estimates to this assumption. All computing was performed using SAS (version 9.1; SAS Institute Inc., Cary, North Carolina, USA) and R [11].

Table 1

Table 1

Back to Top | Article Outline


All analyses were limited to the 3993 women with complete data on the number of lifetime sexual partners and the corresponding ages of those partners. The (weighted) mean reported number of partners was 2.3 [95% confidence interval (CI) 2.1–2.4]. The estimated overall prevalence of HIV among sexually active women was 21.2% (95% CI 17.9–24.5). Estimates of HIV prevalence ranged from 15.2% among women reporting one lifetime partner to 23.1% for two partners, 28.7% for three partners, and 28.5% among those with more than three partners. This increase in prevalence with the reported number of lifetime partners was significant (P = 0.02, logistic regression Wald chi-square test).

Back to Top | Article Outline

Per-partnership transmission probability estimates

The estimated per-partnership transmission probability from the conditional model was = 1.00 (95% CI 0.98–1.00), controlling for the woman's age, race, province of residence, and presence/absence of self-reported vaginal discharge. Given that the per-partnership transmission probability estimate is dependent on the number of lifetime partners, this result depends on the veracity of the sexual histories reported by women. To assess the effect of misreporting of the number of lifetime sexual partners, the EIV model was fit to the data and also yielded a per-partnership transmission probability = 1.00 (95% CI 0.96–1.00); we refer to this as ‘EIV model 1’. A Wald test of no reporting error (i.e. H 0: V 2 = 0) was highly significant (P < 0.0001), indicating the presence of reporting error. Goodness-of-fit analysis (not shown) also indicated superior fit of EIV model 1 compared with the conditional model. Estimates of V 1 and V 2 were 0.15 and 0.39, respectively, suggesting that approximately 70% of the variability in reported contact rates between women is attributable to measurement error. This finding is similar to the results of Satten et al. [7], who reported that 60% of the variability in reported contact rates with prostitutes among male Thai conscripts was attributable to measurement error. For women who reported two partners for every 3 years of sexual activity (the median reported rate), the empirical Bayes estimate of the true partner rate was 2.1 partners for 3 years. For women reporting one and three partners for 3 years of sexual activity, the estimated true rate of partner acquisition was 1.7 and 2.4 for 3 years, respectively.

Another factor that could affect the per-partnership transmission probability estimate is the accuracy of the assumed HIV prevalence in male sexual partners. Moreover, underestimation of male prevalence could lead to upwardly biased transmission probability estimates. For EIV model 1, the assumed male prevalence was based on prevalence estimates for sexually active men from household surveys conducted in 2002 and 2003 (see Table 1, column 2). Sensitivity analyses were conducted by fitting the EIV model assuming three alternative estimates of age-specific male prevalence. First we considered the possibility that men with more sexual experience may be more likely to be selected as a partner. Under EIV model 2 the assumed male prevalence was based on men reporting at least two lifetime sexual partners (Table 1, third column). For this model the per-partnership transmission probability estimate = 0.95 (95% CI 0.77–1.00) was slightly lower than for EIV model 1. For EIV model 3, an HIV prevalence from men with at least three lifetime partners (Table 1, fourth column) was assumed, yielding = 0.89 (95% CI 0.68–1.00). Finally, under EIV model 4 the assumed male prevalence was based on the upper limit of the 95% confidence interval of prevalence estimates for all sexually active men, giving = 0.74 (95% CI 0.56–0.92). For all three sensitivity analysis models, tests of H 0: V 2 = 0 indicated significant reporting error. The conditional empirical Bayes estimates of the true partner rates were also similar to EIV model 1. Transmission probability estimates and approximate 95% confidence intervals for all four EIV models are depicted in Figure 1.

Fig. 1

Fig. 1

Back to Top | Article Outline


The male-to-female per-contact transmission probability (p), i.e. the probability a susceptible woman becomes infected from a single sexual contact with an infected man, is also an important parameter in understanding the sexual transmission of HIV. Given that information on the duration and sexual contact rate for all lifetime partners of women in the study was not available, accurately estimating p from this study is not possible. Crude estimates, however, can be obtained as follows. First note p ≈ 1 −(1 − p)m where m is the average number of contacts for a relationship, i.e. the per-partnership transmission probability p is approximately equal to one minus the probability of escaping infection from m contacts. Given estimates and , a crude estimate of the per-contact transmission probability is given by = 1 − (1 − )1/JOURNAL/aids/04.02/00002030-200704230-00009/ENTITY_OV0471/v/2017-07-25T100145Z/r/image-png. For example, for the 3993 women included in the models above, the estimated mean (median) number of reported contacts per relationship was approximately 82 (22). Conservatively assuming = 0.74 from EIV model 4, would be in the range 0.02 to 0.06 per contact. These estimates suggest that the per-contact probability of HIV transmission are in the range of 1/50 to ∼1/16 contacts.

Back to Top | Article Outline


Our data indicate an incredibly high risk of HIV transmission per-partnership in young women in South Africa. These findings are supported by data from another large population-based study of young people in South Africa, which reported a per-partner HIV transmission probability estimate close to 1.0 for young women [4]. Similarly, Glynn et al. [3] concluded that the risk of HIV transmission at the first episode of sexual intercourse must be extremely efficient based on the high prevalence of infection observed among young women reporting one lifetime partner and few sexual contacts in Kisumu, Kenya. Taken together, these studies suggest that the probability of HIV acquisition among young women exposed to a single infected partner is extremely high.

Alternative explanations to our findings include: (i) systematic underreporting of the number of lifetime sexual partners; or (ii) underestimation of the prevalence in male partners. Regarding (i), the potential for error in the reporting of sensitive behaviors is an issue that plagues all research dependent on self-reported sexual behavior [8,9]. Research conducted by Nnko et al. [12] in Tanzania found that the overall rate of underreporting of sexual partnerships by women was of the order of 16%. Similarly, a recent study of young people in Zimbabwe found that young (15–29 years) unmarried women reported two to three times more sexual partners using an interview method that afforded greater confidentiality compared with face-to-face interviews [9]. One possible reason for systematic underreporting of the number of lifetime partners in our study may be the result of forced sex; 10% of women reported ever having been physically forced to have sex and it is likely that women may not report the perpetrators as sexual partners. As in other studies of self-reported behavior in Africa [3,4], we also found that 4% of young women who reported never having had sex were infected with HIV, indicating the presence of reporting error in this study. Likewise, the results from fitting the EIV models indicated significant reporting error in the number of lifetime partners. Whereas the EIV models do not allow for systematic bias, the empirical Bayes estimates of the true contact rates tend to shrink the reported rates towards the mean. The EIV models thus effectively allow for underreporting in women who report few partners and overreporting in women who report many partners. Nevertheless, the significant dose–response association observed between HIV infection and the reported number of lifetime partners supports the veracity of the self-reported data. Another factor that lends credibility to the self-reported sexual behavior data in our survey is that the prevalence of condom use and partner numbers are similar to those found in other surveys of young people in South Africa [10,13].

Regarding (ii), we do not know the true prevalence of HIV infection among the male partners of the women in this study. It is possible that the men mixing with these young women had a much higher prevalence of infection than in the general population of sexually active men. Our sensitivity analyses provide a partial assessment of the effect of underestimation of the male partner prevalence on the transmission probability estimates. Even for the most extreme of these models (EIV model 4), the per-partnership transmission probability estimate was high (0.74).

Our estimates of the per-partnership transmission probability are higher than many earlier studies of HIV transmission risk [14]. A recent meta-analysis that examined the per-partnership transmission probability from men to women found that the per-partnership estimates from resource-poor settings ranged from 0.20 to 0.67 [15]. Early studies of serodiscordant couples estimated male-to-female transmission probabilities that were very low [14]. Those studies, however, probably underestimated the true risk of HIV transmission for a variety of reasons, the primary one being that individuals most likely to become infected with HIV after few sexual contacts would inherently be excluded from such studies [5,16].

The male-to-female transmission probability parameter reflects both the infectiousness of the infected male partner and the susceptibility of the uninfected woman [5,16]. The infectiousness of HIV appears to vary by the stage of infection and the presence of other co-factors, most importantly sexually transmitted infections (STI) [16–18]. For example, Cohen and Pilcher [16] estimated the risk of transmission per coital act to be 1/50–1/250 during acute infection compared with 1/1000–1/10 000 during asymptomatic infection. As most young women in our study reported mixing with men in their twenties (the age group with the highest HIV incidence and high STI prevalence [4]), a substantial proportion of male partners may have been experiencing acute HIV infection. Unfortunately, as neither the stage of infection nor the presence of STI was known for the male partners in this study, our results should be interpreted as estimating average transmission probabilities over a population of HIV-infected men with varying stages of HIV infection and characteristics, such as STI, which may increase the probability that they transmit the virus. Factors that may increase young women's susceptibility to HIV infection, and that deserve further study, are also important and may include abnormal vaginal flora [19], the use of hormonal contraception [20], and pregnancy [21].

In conclusion, this study demonstrates that HIV transmission to young women in South Africa occurs at a rate far greater than hitherto appreciated. The results demonstrate extreme risk that demands a rethinking of HIV risk and prevention interventions. Prevention programmes targeting young women and their partners that accurately and effectively communicate the extreme risk young women face and that provide improved strategies for HIV prevention cannot be delayed further.

Back to Top | Article Outline


The authors would like to thank Dr Richard White and Dr Stephen Shiboski for their valuable comments on earlier drafts of the manuscript.

Sponsorship: Funding for the original study that generated the data used here was provided by the Kaiser Family Foundation.

Conflicts of interest: The authors do not have commercial or other associations that might pose a conflict of interest.

Back to Top | Article Outline


1. UNAIDS. AIDS epidemic update: December 2004. Geneva: Joint United Nations Programme on HIV/AIDS, World Health Organization; 2004.
2. Pettifor A, Rees H, Kleinschmidt I, Steffenson A, MacPhail C, Hlongwa-Madikizela L, et al. Young people's sexual health in South Africa: HIV prevalence and sexual behaviours from a nationally representative household survey. AIDS 2005; 19:1525–1534.
3. Glynn J, Carael M, Auvert B, Kahindo M, Chege J, Musonda R, et al. Why do young women have a much higher prevalence of HIV than young men? A study in Kisumu, Kenya and Ndola, Zambia. AIDS 2001; 15(Suppl. 4):S51–S60.
4. Auvert B, Ballard R, Campbell C, Carael M, Carton M, Fehler G, et al. HIV infection among youth in a South African mining town is associated with herpes simplex virus-2 seropositivity and sexual behaviour. AIDS 2001; 15:885–898.
5. Vernazza PL, Eron JJ, Fiscus SA, Cohen MS. Sexual transmission of HIV: infectiousness and prevention. AIDS 1999; 13:155–166.
6. Hudgens M, Longini IM Jr, Halloran M, Choopanya K, Vanichseni S, Kitayaporn D, et al. Estimating transmission probability of human immunodeficiency virus in injecting drug users in Thailand. Journal of the Royal Statistical Society 2001; 50:1–14.
7. Satten GA, Mastro TD, Longini IM Jr. Modelling the female-to-male per-act HIV transmission probability in an emerging epidemic in Asia. Stat Med 1994; 13:2097–2106.
8. Hewett PC, Mensch BS, Erulkar AS. Consistency in the reporting of sexual behaviour by adolescent girls in Kenya: a comparison of interviewing methods. Sex Transm Infect 2004; 80(Suppl. 2):ii43–ii48.
9. Gregson S, Zhuwau T, Ndlovu J, Nyamukapa CA. Methods to reduce social desirability bias in sex surveys in low-development settings: experience in Zimbabwe. Sex Transm Dis 2002; 29:568–575.
10. Shisana O, Simbayi L. Nelson Mandela/HSRC Study of HIV/AIDS South African National HIV Prevalence, Behavioural Risks and Mass Media Household Survey 2002. Cape Town: Human Sciences Research Council; 2002.
11. R Development Core Team. R: a language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing; 2005.
12. Nnko S, Boerma JT, Urassa M, Mwaluko G, Zaba B. Secretive females or swaggering males? An assessment of the quality of sexual partnership reporting in rural Tanzania. Soc Sci Med 2004; 59:299–310.
13. Reddy S, Panday P, Swart D, Jinabhai C, Amosun S, James S, et al. Umthenthe Uhlaba Usamila – the South African Youth Risk Behaviour Survey 2002. Cape Town: South African Medical Research Council; 2003.
14. Royce R, Sena A, Cates W, Cohen M. Sexual transmission of HIV. N Engl J Med 1997; 336:1072–1078.
15. Baggaley R. The impact of antiretroviral use in resource-poor settings: insights from mathematical models. London: University of London; 2006.
16. Cohen MS, Pilcher CD. Amplified HIV transmission and new approaches to HIV prevention. J Infect Dis 2005; 191:1391–1393.
17. Wawer MJ, Gray RH, Sewankambo NK, Serwadda D, Li X, Laeyendecker O, et al. Rates of HIV-1 transmission per coital act, by stage of HIV-1 infection, in Rakai, Uganda. J Infect Dis 2005; 191:1403–1409.
18. Leynaert B, Downs AM, de Vincenzi I. Heterosexual transmission of human immunodeficiency virus: variability of infectivity throughout the course of infection. European Study Group on Heterosexual Transmission of HIV. Am J Epidemiol 1998; 148:88–96.
19. Taha TE, Hoover DR, Dallabetta GA, Kumwenda NI, Mtimavalye LA, Yang LP, et al. Bacterial vaginosis and disturbances of vaginal flora: association with increased acquisition of HIV. AIDS 1998; 12:1699–1706.
20. Lavreys L, Baeten JM, Martin HL Jr, Overbaugh J, Mandaliya K, Ndinya-Achola J, Kreiss JK. Hormonal contraception and risk of HIV-1 acquisition: results of a 10-year prospective study. AIDS 2004; 18:695–697.
21. Gray RH, Li X, Kigozi G, Serwadda D, Brahmbhatt H, Wabwire-Mangen F, et al. Increased risk of incident HIV during pregnancy in Rakai, Uganda: a prospective study. Lancet 2005; 366:1182–1188.

disease transmission; HIV; South Africa; women

© 2007 Lippincott Williams & Wilkins, Inc.