The role of sexually transmitted infections in the evolution of the South African HIV epidemic

Authors


Leigh Johnson, Centre for Infectious Disease Epidemiology and Research, Faculty of Health Sciences, University of Cape Town, Anzio Road, Observatory, 7925, South Africa. Tel.: +27 21 406 6981; Fax: +27 21 406 6764; E-mail: Leigh.Johnson@uct.ac.za

Summary

Objectives  To assess the extent to which sexually transmitted infections (STIs) have contributed to the spread of HIV in South Africa and to estimate the extent to which improvements in STI treatment have reduced HIV incidence.

Methods  A mathematical model was used to simulate interactions between HIV and six other STIs (genital herpes, syphilis, chancroid, gonorrhoea, chlamydial infection and trichomoniasis) as well as bacterial vaginosis and vaginal candidiasis. The effects of STIs on HIV transmission probabilities were assumed to be consistent with meta-analytic reviews of observational studies, and the model was fitted to South African HIV prevalence data.

Results  The proportion of new HIV infections in adults that were attributable to curable STIs reduced from 39% (uncertainty range: 24–50%) in 1990 to 14% (8–18%) in 2010, while the proportion of new infections attributable to genital herpes increased. Syndromic management programmes are estimated to have reduced adult HIV incidence by 6.6% (3.3–10.3%) between 1994 and 2004, by which time syndromic management coverage was 52%. Had syndromic management been introduced in 1986, with immediate achievement of 100% coverage and a doubling of the rate of health seeking, HIV incidence would have reduced by 64% (36–82%) over the next decade, but had the same intervention been delayed until 2004, HIV incidence would have reduced by only 5.5% (2.8–9.0%).

Conclusions  Sexually transmitted infections have contributed significantly to the spread of HIV in South Africa, but STI control efforts have had limited impact on HIV incidence because of their late introduction and suboptimal coverage.

Abstract

Objectifs:  Evaluer la mesure dans laquelle les infections sexuellement transmissibles (IST) ont contribuéà la propagation du VIH en Afrique du sud et estimer dans quelle mesure des améliorations dans le traitement des IST ont réduit l’incidence du VIH.

Méthodes:  Un modèle mathématique a été utilisé pour simuler les interactions entre le VIH et six autres IST (herpès génital, syphilis, chancre mou, blennorragie, infection à Chlamydia et à Trichomonas) ainsi que la vaginose bactérienne et la candidose vaginale. Les effets des IST sur les probabilités de transmission du VIH ont été supposés être compatibles avec les résultats de méta-analyse d’études observationnelles et le modèle a été ajusté aux données de prévalence du VIH en Afrique du sud.

Résultats:  La proportion de nouvelles infections VIH chez les adultes, attribuables à des IST curables a diminué, allant de 39% (marge d’incertitude: 24–50%) en 1990 à 14% (8–18%) en 2010, tandis que la proportion de nouvelles infections attribuables à l’herpès génital a augmenté. Les programmes de prise en charge syndromique sont supposés avoir réduit l’incidence du VIH chez les adultes de 6,6% (3,3 à 10,3%) entre 1994 et 2004, où la couverture de la prise en charge syndromique était de 52%. Si la prise en charge syndromique avait été introduite en 1986, avec la réalisation immédiate d’une couverture de 100% et un doublement du taux des recours à la santé, l’incidence du VIH aurait été réduite de 64% (36–82%) au cours de la décennie suivante. Mais si la même intervention avait été retardée jusqu’en 2004, l’incidence du VIH aurait diminué de seulement 5,5% (2,8 à 9,0%).

Conclusions:  Les IST ont contribué de façon importante à la propagation du VIH en Afrique du sud, mais les efforts de contrôle des IST ont eu un impact limité sur l’incidence du VIH en raison de leur introduction tardive et de la couverture sous-optimale.

Abstract

Objetivos:  Evaluar hasta donde han contribuido las infecciones de transmisión sexual (ITSs) a la dispersión del VIH en Sudáfrica, y calcular hasta que punto las mejoras en el tratamiento de las ITSs han reducido la incidencia del VIH.

Métodos:  Se utilizó un modelo matemático para simular interacciones entre el VIH y otras seis ITSs (herpes genital, sífilis, chancro, gonorrea, infección por clamidias y tricomonas) al igual que la vaginosis bacteriana y la candidiasis vaginal. Se asumió que los efectos de las ITSs sobre la probabilidad de transmisión del VIH eran consistentes con las revisiones meta-analíticas de estudios observacionales, y el modelo se ajustó a datos de prevalencia del VIH en Sudáfrica.

Resultados:  La proporción de nuevas infecciones de VIH en adultos atribuibles a ITSs curables se redujo del 39% (rango de incertidumbre: 24–50%) en 1990 al 14% (8–18%) en 2010, mientras que la proporción de nuevas infecciones atribuibles al herpes genital aumentó. Se estima que los programas de manejo sindrómico han reducido la incidencia del VIH en un 6.6% (3.3–10.3%) entre 1994 y 2004, momento en el cual la cobertura del manejo sindrómico era del 52%. Si el manejo sindrómico se hubiese introducido en 1986, con un alcance inmediato del 100% de cobertura, y la tasa de búsqueda de salud se hubiese doblado, la incidencia del VIH se habría reducido en un 64% (36–82%) en la década siguiente; pero si la misma intervención se hubiese retrasado hasta el 2004, la incidencia del VIH se habría reducido solo en un 5.5% (2.8–9.0%).

Conclusiones:  Las ITSs han contribuido significativamente a la dispersión del VIH en Sudáfrica, pero los esfuerzos por controlar las ITS han tenido un impacto limitado sobre la incidencia del VIH por su introducción tardía y una ucobertura subóptima.

Introduction

South Africa has one of the highest HIV prevalence levels in the world, with an estimated 5.6 million HIV-positive citizens (UNAIDS 2010). It is important to assess the extent to which this high HIV prevalence is explained by different biomedical and behavioural factors. Sexually transmitted infections (STIs) have been hypothesized to play a major role in the spread of HIV, with observational studies showing that HIV-negative individuals with STIs are more susceptible to HIV than STI-uninfected individuals (Røttingen et al. 2001; Sexton et al. 2005). Observational studies have also demonstrated that HIV-infected individuals who are coinfected with STIs have increased levels of HIV in their genital tracts, rendering them more likely to transmit HIV (Johnson & Lewis 2008). Based on this observational evidence, mathematical modelling studies have suggested that a high proportion of heterosexual HIV transmission in Africa is directly attributable to other STIs (Robinson et al. 1997; Orroth et al. 2006; Freeman et al. 2007; Abu-Raddad et al. 2008).

However, trials of improvements in STI treatment have produced conflicting findings regarding the impact of STI treatment on HIV incidence in Africa. The first randomized controlled trial in Mwanza (Tanzania) found that in communities that were randomized to receive enhanced STI treatment according to syndromic management guidelines, HIV incidence was reduced by roughly 40% (Grosskurth et al. 1995; Hayes et al. 1995). A later trial in Masaka (Uganda) found that the introduction of syndromic management guidelines for STI treatment had no significant effect on HIV incidence (Kamali et al. 2003). Other trials of mass STI treatment (Wawer et al. 1999), periodic presumptive STI treatment (Kaul et al. 2004) and suppressive therapy for genital herpes (Celum et al. 2008, 2010; Watson-Jones et al. 2008) also failed to find a significant effect on HIV incidence. This has led to growing scepticism regarding the value of STI treatment as an HIV prevention strategy (Gray & Wawer 2008) and has called into question the hypothesis that STIs are major drivers of the HIV epidemic.

In an attempt to reconcile the divergent evidence from observational studies and randomized controlled trials, this study examines whether it is possible, using a mathematical model parameterized by observational data, to produce results consistent with the findings of the randomized controlled trials conducted in Mwanza and Masaka. The model is applied to South Africa, a country in which syndromic management guidelines were introduced in the mid-1990s. We aim to estimate the proportion of heterosexual HIV transmission attributable to different STIs at different stages in the South African HIV epidemic and to assess the impact of syndromic management interventions under various scenarios.

Methods

The model divides the South African population by age and sex, and projects the growth of the population over time, starting in 1985. Sexually active adults are divided into ‘high-risk’ individuals (who are assumed to have a propensity for concurrent partners) and ‘low-risk’ individuals, and are also stratified according to whether they are currently in a long-term cohabiting relationship and according to their number of current sexual partners. Individuals are assumed to move between these relationship strata over time as they form new partnerships and end partnerships. Three types of relationship are modelled: once-off interactions between sex workers and their clients, short-term (non-cohabiting) relationships and long-term relationships. Frequencies of sex and levels of condom usage are assumed to depend on age, sex and relationship type, and condom usage is assumed to increase over time in response to HIV communication programmes and condom distribution (Myer 2010). Sexual behaviour assumptions have been set based on South African data sources, and parameters have been fixed at the means of the distributions estimated in a previous analysis of sexual behaviour patterns in South Africa (Johnson et al. 2009).

The model is used to simulate the transmission of six STIs: genital herpes, syphilis, chancroid, gonorrhoea, chlamydial infection and trichomoniasis. For each STI, assumptions are made regarding probabilities of transmission per act of unprotected sex, proportions of cases that become symptomatic and the average duration of infection in the absence of treatment. The incidence and resolution of bacterial vaginosis and vaginal candidiasis in women are also modelled. Adults with symptoms of genital ulcers or discharges are assumed to visit healthcare providers at rates of 0.23 and 0.57 visits per week of symptoms, in women and men respectively. Treatment effectiveness is assumed to depend on proportions of healthcare providers who are using syndromic management protocols and on proportions of providers who have access to appropriate drugs. Assumptions regarding STI health-seeking behaviour and treatment practices are set based on South African surveys, with the proportion of STI patients treated according to syndromic management guidelines increasing from 0% in mid-1994 to 52% by mid-2004 (Johnson et al. 2011). The model has been fitted to South African STI prevalence data using a Bayesian approach, and the STI parameters in the present analysis are fixed at the medians of the posterior distributions estimated in the Bayesian analysis (Johnson et al. 2010).

The model is also used to simulate heterosexual HIV transmission. Adults infected with HIV are assumed to progress through four stages of infection (acute HIV, asymptomatic HIV, pre-AIDS symptoms and AIDS) before either dying or starting antiretroviral treatment. HIV transmission probabilities per act of unprotected sex with an HIV-positive partner are assumed to depend on the sex of the HIV-susceptible partner and the type of relationship. The transmission probabilities are also adjusted to allow the HIV transmission probability to depend on the stage of disease in the HIV-infected partner. Assumptions about relative levels of HIV infectiousness in different stages and average durations of different HIV stages are described elsewhere (Johnson et al. 2009).

If one partner is infected with an STI, the HIV transmission probability that would be expected in the absence of STIs is multiplied by an STI cofactor. This cofactor is assumed to depend on (i) the type of STI symptom present, (ii) whether the male or female partner has the STI, and (iii) whether the HIV-negative or HIV-positive partner has the STI. If one partner has multiple STIs, STI cofactors are assumed to ‘saturate’, meaning that only the highest of the STI cofactors is applied to the HIV transmission probability (Korenromp et al. 2001). If both partners in a relationship are infected with STIs, the HIV transmission probability that would be expected in the absence of STIs is multiplied by one cofactor for each partner. A mathematical description of the model of HIV transmission is provided in Appendix S1.

In the ‘Base-cofactor scenario’, STI cofactors are assumed to be consistent with odds ratios estimated in meta-analytic reviews (Røttingen et al. 2001; Freeman et al. 2006; Johnson & Lewis 2008). (Detailed justification of the choice of cofactor assumptions is provided in Appendix S1.) As HIV and other STIs share a common mode of transmission, the STI cofactor is difficult to measure accurately in observational studies (Korenromp et al. 2001). Because of this uncertainty regarding the biases in the observational data and because the 95% confidence intervals around the pooled odds ratios are wide, we consider two additional scenarios for the purpose of sensitivity analysis. In the ‘high-cofactor scenario’, the cofactor multiples are set equal to the cofactors in the ‘base-cofactor scenario’, raised to the power of 1.5. In the ‘low-cofactor scenario’, cofactors are assumed to be equal to those in the ‘base-cofactor scenario’, raised to the power of 0.5. The assumed cofactors in the three scenarios are shown in Table 1.

Table 1.   Assumed STI cofactors
ScenarioSymptomsSusceptibility cofactorInfectiousness cofactor
MenWomenMenWomen
  1. STI, sexually transmitted infection.

Base-cofactor scenarioUlcerative symptoms4.42.83.53.5
Discharge/dysuria symptoms2.61.63.11.7
Asymptomatic infection2.01.51.21.2
High-cofactor scenarioUlcerative symptoms9.24.76.56.5
Discharge/dysuria symptoms4.22.05.52.2
Asymptomatic infection2.81.81.31.3
Low-cofactor scenarioUlcerative symptoms2.11.71.91.9
Discharge/dysuria symptoms1.61.31.81.3
Asymptomatic infection1.41.21.11.1

In each scenario, the parameters determining the HIV transmission probabilities in the absence of STIs are fitted by maximum likelihood. The likelihood represents the closeness of the correspondence between model estimates of HIV prevalence and estimates of HIV prevalence in antenatal clinic surveys, household surveys in 2005 and 2008 and surveys of commercial sex workers. (A detailed explanation of the definition of the likelihood is provided in Appendix S1.) The likelihood is maximized using the downhill simplex method (Press et al. 1986).

For each STI, scenario and year, we calculate the proportion of new HIV infections in the relevant year that are attributable to the STI of interest. This proportion is calculated by (i) running the model up to the start of the year of interest, (ii) setting the prevalence of the STI at the start of the year to zero, (iii) calculating the number of new heterosexually acquired HIV cases over the year, (iv) obtaining the number of new heterosexually acquired HIV cases that would have been expected if the STI prevalence had not been set to zero and (v) expressing the difference between the numbers in steps (iii) and (iv) as a proportion of the number of new heterosexually acquired HIV infections that would have been expected if all STIs had been present. To ensure that proportions of new HIV infections attributable to different STIs are additive, partial attributable risks are calculated according to the approach described by Land et al. (2001) (further detail is provided in Appendix S1).

Results

Estimates of trends in STI prevalence in the base-cofactor scenario are summarized in Table 2. Gonorrhoea, chancroid and syphilis are estimated to have declined substantially in prevalence, because of the impact of syndromic management programmes and increased condom use. However, there has been little or no decline in the prevalence of other STIs. After fitting the HIV transmission probabilities that would be expected in the absence of STIs, by maximum likelihood, the model estimates of HIV prevalence are reasonably consistent with survey estimates (Figure 1), although survey estimates of HIV prevalence in commercial sex workers are more dispersed because of their limited sample sizes and geographical heterogeneity.

Table 2.   STI prevalence estimates in the base-cofactor scenario
 Women aged 15–49Men aged 15–49
1990 (%)1995 (%)2000 (%)2005 (%)1990 (%)1995 (%)2000 (%)2005 (%)
  1. HSV-2, herpes simplex virus type 2; STI, sexually transmitted infection.

HIV0.54.013.219.10.43.110.214.5
HSV-252.553.054.654.430.731.232.231.7
Chancroid1.21.00.10.01.00.80.10.0
Syphilis7.77.64.61.97.47.14.52.2
Gonorrhoea7.67.56.34.44.74.63.52.2
Chlamydial infection11.511.511.110.18.78.68.17.1
Trichomoniasis30.930.929.524.25.95.85.34.0
Bacterial vaginosis35.435.233.933.6    
Vaginal candidiasis41.040.840.941.2    
Figure 1.

 Correspondence between model estimates and survey estimates of HIV prevalence. Diamonds in panel (a) represent HIV prevalence data that were not used in the likelihood definition. Error bars represent 95% confidence intervals around survey estimates.

Proportions of new HIV infections attributable to STIs are shown in Figure 2. Gonorrhoea, chancroid and syphilis are estimated to have contributed significantly to HIV transmission in the early stages of the epidemic, accounting for 12%, 8% and 5% of heterosexual transmission in 1990, respectively, in the base-cofactor scenario. However, these STIs have fallen in significance as their prevalence has declined. Syndromic management has had less impact on the prevalence of trichomoniasis and chlamydial infection, and declines in proportions of HIV incidence attributable to these STIs have therefore been more moderate. In contrast, the contribution of genital herpes to HIV transmission has increased over time. Genital herpes is the most significant STI driving HIV transmission, accounting for 23% of heterosexual HIV transmission in 1990 and 29% of transmission in 2010, in the base-cofactor scenario.

Figure 2.

 Proportions of new adult HIV infections attributable to different STIs.

Altogether, curable STIs are estimated to have accounted for 39% of heterosexual HIV transmission in 1990 and 14% in 2010, in the base-cofactor scenario (corresponding proportions are 50% and 18% in the high-cofactor scenario and 24% and 8% in the low-cofactor scenario). The contribution of curable STIs is greater in youth (Figure 2d) than in older adults. Bacterial vaginosis and vaginal candidiasis, although not sexually transmitted, have also contributed significantly to HIV transmission, together accounting for 9% of heterosexual transmission, with this proportion remaining stable over time.

Compared with what would have been expected in the absence of syndromic management, the model estimates that the number of new HIV infections in South African adults, over the period from mid-1994 to mid-2004, has been reduced by 6.6% as a result of the adoption of syndromic management protocols (reductions are 10.3% and 3.3% in the high- and low-cofactor scenarios, respectively). Figure 3a shows the hypothetical 2-year reductions in HIV incidence that would have been expected if syndromic management protocols had instead been adopted immediately by 100% of health workers, and drug shortages immediately eliminated, at different stages in the HIV epidemic. The 10-year impact of the same changes, coupled with a concomitant doubling in the rate of health seeking (to bring average treatment delays in line with those in industrialized countries), is shown in Figure 3b. If these changes had occurred in 1986, new adult HIV infections over the next 10 years would have been 64% lower than would have been expected without change. However, if the same change had been delayed until 2004, the number of new HIV infections over the next decade would have been reduced by only 5.5%.

Figure 3.

 Effect of syndromic management interventions introduced at different stages in the South African HIV epidemic. Grey bars represent results from base-cofactor scenario, with each bar corresponding to a different scenario (scenarios are defined in terms of the year in which the intervention is introduced). Upper and lower error bars represent results from high- and low-cofactor scenarios, respectively.

Discussion

Consistent with model-based analyses of STI–HIV interactions in other African countries (Robinson et al. 1997; Orroth et al. 2006), we estimate that more than half of heterosexually transmitted HIV in the early stages of the South African epidemic was attributable to other STIs. In agreement with these analyses, we also estimate that the proportion of incident HIV attributable to other STIs has declined over time. This is partly because of the impact of increases in condom usage and the introduction of syndromic management protocols, starting in the mid-1990s, which substantially reduced the prevalence of curable STIs (Johnson et al. 2011). The decline may also be explained by increasingly high proportions of new HIV infections occurring in low-risk groups (in which STI prevalence is lower) in the later stages of the epidemic, or by saturation effects when HIV prevalence levels are high. Our model confirms other model-based analyses, which have suggested that 10–30% of new HIV infections in early-stage African epidemics are attributable to genital herpes (Freeman et al. 2007; Abu-Raddad et al. 2008) and that this proportion may increase over time. The significance of genital herpes is related to the high prevalence of herpes simplex virus type 2 (HSV-2), which occurs in over 50% of women and 30% of men in the 15–49 age group (Table 2). As genital ulcers are assumed to occur more frequently in individuals infected with HSV-2 who are HIV-immunocompromised, the prevalence of HSV-2 ulcers increases as the HIV epidemic evolves, thus accounting for an increasingly high proportion of HIV transmission.

Previous modelling studies have suggested that the impact of syndromic management interventions on the incidence of HIV is significantly greater in early-stage HIV epidemics than in more mature epidemics and that syndromic management has significantly less impact when preceded by substantial changes in sexual behaviour (White et al. 2004; Korenromp et al. 2005). The Mwanza trial of syndromic management, which was conducted in the early stages of the Tanzanian epidemic, prior to substantial behaviour change, achieved a 38% (95% CI: 15–55%) reduction in HIV incidence over 2 years (Hayes et al. 1995), which is consistent with our base- and high-cofactor scenario estimates of what the reduction in HIV incidence would have been if syndromic management had been introduced in the early stages of the South African HIV epidemic (Figure 3a). The Masaka trial of syndromic management, which was introduced at a more advanced stage in the Ugandan epidemic, when substantial behaviour change had already occurred, had no significant impact on HIV incidence (Kamali et al. 2003), and this too is consistent with our model estimates of what the impact of syndromic management would have been if introduced at a late stage in the South African epidemic, after significant increases in condom usage. The difference in the impact of syndromic management at different stages in the epidemic can be explained in terms of differences in both the direct impact of the intervention (because of the reduced HIV transmission probabilities) and the indirect impact (because of reductions in secondary HIV transmission from individuals who would have been infected with HIV because of other STIs). The former is likely to be greater when syndromic management precedes behaviour change, as reductions in risk behaviour can be expected to reduce STI prevalence and hence the proportion of new HIV infections attributable to other STIs. The indirect effect is likely to be more substantial in the early stages of the epidemic, because newly infected individuals are likely to have more HIV-susceptible partners in the early stages of an HIV epidemic (when HIV prevalence is low and most transmission is occurring in high-risk groups) than in a late-stage HIV epidemic.

This analysis suggests that the explosive spread of HIV in South Africa in the early 1990s could have been slowed significantly if appropriate improvements to STI services and education to promote health seeking had been introduced in the late 1980s. The extremely high HIV prevalence levels seen in South Africa today therefore need to be understood both in terms of behavioural factors and in terms of health service factors, in addition to biological factors such as the low prevalence of male circumcision. The actual impact of syndromic management is estimated to be a modest 6.6% reduction in the number of new HIV infections over the 1994–2004 period, relative to what would have been expected in the absence of any change in STI treatment. The effectiveness of this intervention has been limited by the low proportion of private practitioners adopting syndromic management protocols (Schneider et al. 2001) and the relatively late introduction of the intervention, at a time when significant increases in condom distribution were occurring (Myer 2010).

A limitation of this analysis is that STI cofactors have been assumed to depend on the type of STI symptoms, but not on the STI causing the symptoms. This may be unrealistic if certain STIs interact with HIV in unique ways that are unrelated to the presence of symptoms. For example, HSV-2 may inhibit the production of HIV-neutralizing IgA antibodies, which appear to protect against HIV acquisition (Hirbod et al. 2008). However, given the wide confidence intervals around the odds ratios estimated in published meta-analyses, and the sources of bias inherent in STI cofactor estimation (Korenromp et al. 2001), it is difficult to argue that there is significant heterogeneity between STIs in terms of their effect on per-contact HIV transmission probabilities, after controlling for symptoms. We have adopted a parsimonious approach in setting STI cofactor assumptions for each STI syndrome, as it would be difficult to set STI cofactor assumptions separately for each STI (both symptomatic and asymptomatic) on the basis of the limited data available.

In this analysis, it has been assumed that STI cofactors saturate at the individual level. This assumption is debatable, as there are two main biological mechanisms by which STIs increase the HIV transmission probability (increased immune activation in the genital tract and disruptions of epithelial barriers to HIV transmission), and it is not clear whether two STIs affecting the same biological mechanism have any more effect on the HIV transmission probability than one STI alone or whether there is overlap in the effect of different biological mechanisms (e.g. because of genital ulcers causing immune activation). In a sensitivity analysis to explore the effect of assuming that the multiplicative effect of each STI is independent of other STIs present, the model estimated a substantially higher proportion of new HIV infections attributable to other STIs but was found to provide a significantly poorer fit to South African HIV prevalence data (see Appendix S1). Results have therefore been presented only for the model that assumes saturating STI cofactors.

This analysis demonstrates that STI treatment interventions may be expected to have relatively little impact on HIV incidence in mature HIV epidemics, notwithstanding the significant role played by STIs in the early evolution of African HIV epidemics. Most randomized controlled trials are not adequately powered to detect these modest reductions in HIV incidence. The failure of recent STI treatment trials to produce statistically significant results should therefore not be interpreted as evidence against STI treatment as an HIV prevention strategy, or as evidence against the STI cofactor hypothesis. STI treatment may still be a cost-effective strategy for HIV prevention (White et al. 2008), even if it is insufficient by itself to achieve a substantial reduction in HIV incidence.

Ancillary