Skip Navigation

Determinants of the Impact of Sexually Transmitted Infection Treatment on Prevention of HIV Infection: A Synthesis of Evidence from the Mwanza, Rakai, and Masaka Intervention Trials

  1. E. L. Korenromp1,a,
  2. R. G. White2,
  3. K. K. Orroth2,
  4. R. Bakker1,
  5. A. Kamali3,
  6. D. Serwadda4,
  7. R. H. Gray5,
  8. H. Grosskurth2,
  9. J. D. F. Habbema1 and
  10. R. J. Hayes2
  1. 1Department of Public Health, Erasmus Medical Center, University Medical Center Rotterdam, Rotterdam, The Netherlands;
  2. 2London School of Hygiene and Tropical Medicine, London, United Kingdom;
  3. 3Medical Research Council Programme on AIDS in Uganda, Uganda Virus Research Institute, Entebbe, and
  4. 4Institute of Public Health, Faculty of Medicine, Makerere University, Kampala, Uganda;
  5. 5Johns Hopkins University School of Public Health, Department of Population and Family Health Sciences, Baltimore, Maryland
  1. Reprints or correspondence: Prof. R. Hayes, Dept. of Infectious and Tropical Diseases, London School of Hygiene and Tropical Medicine, Keppel St., London WC1E 7HT, UK (richard.hayes{at}lshtm.ac.uk)
 
Next Section

Abstract

Community-randomized trials in Mwanza, Tanzania, and Rakai and Masaka, Uganda, suggested that population characteristics were an important determinant of the impact of sexually transmitted infection (STI) treatment interventions on incidence of human immunodeficiency virus (HIV) infection. We performed simulation modeling of HIV and STI transmission, which confirmed that the low trial impact in Rakai and Masaka could be explained by low prevalences of curable STI resulting from lower-risk sexual behavior in Uganda. The mature HIV epidemics in Uganda, with most HIV transmission occurring outside core groups with high STI rates, also contributed to the low impact on HIV incidence. Simulated impact on HIV was much greater in Mwanza, although the observed impact was larger than predicted from STI reductions, suggesting that random error also may have played some role. Of proposed alternative explanations, increasing herpetic ulceration due to HIV-related immunosuppression contributed little to the diminishing impact of antibiotic treatment during the Ugandan epidemics. The strategy of STI treatment also was unimportant, since syndromic treatment and annual mass treatment showed similar effectiveness in simulations of each trial population. In conclusion, lower-risk behavior and the mature HIV epidemic explain the limited impact of STI treatment on HIV incidence in Uganda in the 1990s. In populations with high-risk sexual behavior and high STI rates, STIs treatment interventions may contribute substantially to prevention of HIV infection

On the basis of observed associations between sexually transmitted infections (STIs) and subsequent HIV acquisition and shedding, STIs are considered to be cofactors in HIV transmission, and STI treatment forms part of the strategy for prevention of HIV infection [1, 2]. The magnitude of the effect of improved STI treatment on HIV transmission at a population level was tested during the 1990s in 3 community-randomized trials in rural East Africa, where rates of HIV infection were high and STI control was suboptimal. Improved clinical management of symptomatic STIs was associated with a reduction in incidence of HIV infection of 38% (95% confidence interval [CI], 15%–55%) over the course of 2 years in the Mwanza region, Tanzania [3, 4]. Over a similar period and with seemingly comparable proportional reductions in treated STDs, periodic mass treatment in Rakai, Uganda, did not significantly reduce the incidence of HIV infection (3% reduction; 95% CI, −16% to 19%) [5]

Several hypotheses have been put forward to explain these contrasting findings. One of these hypotheses is that syndromic treatment is a more effective strategy than periodic mass treatment [2]. However, the results of a third trial, in Masaka district, Uganda, which neighbors Rakai, refuted this hypothesis, since syndromic treatment had no effect on incidence of HIV infection in Masaka (0% reduction, for syndromic treatment plus health education vs. the control arm; 95% CI, −58% to 37%) [6]

Attention has since been diverted to characteristics of the study populations, which appeared to differ between Mwanza, where syndromic treatment was effective, and the Ugandan sites, where neither syndromic nor mass treatment was effective. A clear difference between these populations was the more advanced and severe HIV epidemic in Uganda. Several mechanisms were proposed to explain why the impact of STI treatment would be lower in an advanced HIV epidemic than in an early-stage HIV epidemic [7]. First, cofactor STDs would become less important once HIV spreads outside the most sexually active core groups with the highest rates of STI. Second, HIV-related mortality among high-risk groups would lower the prevalence of cofactor STIs, owing to HIV-related deaths [8], which would be consistent with lower prevalences of curable STIs observed in the Ugandan trials, compared with Mwanza [9]. Recently, lower rates of cofactor STIs in Rakai and Masaka were recognized to have possibly resulted from a reduction in sexual risk behaviors that was observed in Uganda during the late 1980s and early 1990s [10–13]. Other hypotheses focus on random error; greater population mobility and, hence, dilution of the impact of interventions in Uganda; and artefacts relating to different trial designs and analyses [14–17]

The existence of some of these factors was supported by secondary analyses and comparison of data between the trials [9, 18]. We assessed the comparative role of each of these factors and the extent to which their combined effect could explain the full set of trial outcomes, by simulating scenarios reflecting all plausible hypotheses, using the microsimulation model STDSIM [19]. On the basis of this synthesis of the available evidence, we review the interpretation of the 3 trials and discuss the implications for HIV-prevention research and policy

Previous SectionNext Section

Methods

Hypotheses concerning the contrasting outcomes of the trials were identified from scientific publications by means of a PubMed search (http://www.ncbi.nlm.nih.gov/entrez/query.fcgi). Relevant conference proceedings also were included (table 1). The main contrast to be explained was the higher impact on incidence of HIV infection in Mwanza, relative to that in the 2 Ugandan sites, whereas no consistent difference in proportional reductions in treated STIs was found. Prior to simulations, hypotheses therefore were selected on the basis of the criterion that they must primarily relate to the reductions in incidence of HIV infection in Mwanza, as opposed to Uganda, given a certain reduction in STIs (table 1 [“Type of impact” column]). Second, we reviewed whether empirical evidence was consistent with each hypothesis (table 1 [“Empirical evidence” column]). The role of each factor that seemed relevant and plausible, on the basis of these 2 criteria, was then estimated by modeling, for which existing simulations of the Mwanza, Rakai, and Masaka trials [23] were extended to include counterfeit scenarios that explored each plausible factor in turn by its selective omission or addition

Figure 1
View larger version:
Figure 1

Prevalences of HIV infection (A–C) and gonorrhea (D–F) over time in the trial populations in Mwanza, Tanzania, and Masaka and Rakai, Uganda, in comparison arms. Consistent with available data, prevalences of HIV infection refer to men and women aged 15–54 years, and prevalences of gonorrhea refer to men and women aged 15–44 years in Rakai and Masaka and to men aged 15–54 years in Mwanza. Solid lines simulation model; dashed lines counterfeit scenarios for Masaka and Rakai in which behavior change in Uganda in the late 1980s was omitted; triangles empirical data; arrows onset of the respective trials

Figure 2
View larger version:
Figure 2

Ratios (between intervention and comparison arms) of incidence of HIV infection (top) and prevalence of gonorrhea (bottom) in trials and counterfeit scenarios for Mwanza, Tanzania (gray bars) and Masaka (black bars) and Rakai (white bars) Uganda. Consistent with available data, prevalence ratios for gonorrhea refer to 2 years after intervention in Rakai and Mwanza and to 4 years after intervention in Masaka; incidence ratios for HIV infection were calculated over 2 years of intervention in Rakai and Mwanza and over 4 years of intervention in Masaka. Reductions in incidence of HIV infection refer to men and women aged 15–54 years in Rakai and Mwanza and to men and women aged ⩾15 years in Masaka. Reductions in prevalence of gonorrhea refer to men and women aged 15–29 years in Rakai, men and women aged 15–44 years in Masaka, and men aged 15–54 years in Mwanza. Crosses with error bars indicate empirical data points with 95% confidence intervals (where available) or P value, for observed differences between intervention and control arms. “Data” for the impact of syndromic treatment (ST) in the Masaka trial (which was not evaluated separately from IEC [information, education, and communication]) are the average of the relative risks (RRs) for IEC alone (0.94) and IEC plus ST (RR, 1.0) for HIV infection, compared with the control arm, and the ratio of the RRs for IEC plus ST (0.25) and IEC alone (RR=0.64) for gonorrhea. “Behav. change,” reduction in risk behavior, as observed in Uganda; MT, mass treatment

Table 1
View larger version:
Table 1

Summary of proposed explanations for the larger observed impact of syndromic treatment (ST) of sexually transmitted infection (STI) on incidence of HIV infection in Mwanza, Tanzania, compared with observed impacts of ST and mass treatment (MT) in Masaka and Rakai, Uganda

STDSIM modelSTDSIM simulates the natural history and transmission of STIs and HIV infection in a dynamic population of interacting individuals [19]. We used this individual-based, stochastic model to simulate in parallel the spread of multiple interacting STIs, including herpes simplex virus (HSV) type 2 infection, and to properly represent sexual network effects that may have influenced STI interactions in the trial populations

The transmission of STIs is represented in STDSIM at the level of single sexual contacts between individuals in sexual relationships or during one-off (commercial) contacts. Gonorrhea, chlamydia, trichomoniasis, syphilis, chancroid, HSV-2 infection, and HIV infection all have separate quantifications for natural history and transmission probabilities [23]. Simulated episodes of STI are represented, when applicable, in multiple subsequent stages that differ in symptomatology, cofactor effect on HIV transmission, status at diagnostic testing, and/or infectivity to sex partners. This “real-time” representation ensures that biomedical and epidemiological interactions between HIV infection and STI are properly represented and allows for comparisons of the magnitude and timing of reductions in HIV and other STIs, between simulations and empirical (trial) data. For cofactor effects on HIV transmission, we assumed that chancroid and primary HSV-2 ulcers are associated with a 25-fold increase in the per-act probability of HIV transmission, recurrent HSV-2 ulcers with a 10-fold increase, primary syphilis with a 7.5-fold increase (averaged over the early phase of infection, including periods without ulcers), gonorrhea and chlamydia with a 3-fold increase, and trichomoniasis with a 2-fold increase. On the basis of recent reestimations [24], these values were somewhat lower than those in previous simulations [19]

Bacterial vaginosis (BV) was not modeled in this study. First, the literature review suggested that BV, which involves neither inflammation nor ulceration, is not a strong, independent cofactor for HIV transmission: observed associations between BV and enhanced HIV transmission are more likely to reflect associations between BV and cofactor STIs, such as chlamydia, trichomoniasis, and gonorrhea, or the reverse effect of HIV infection predisposing to BV [25–27]. Second, although BV was included among the treated STIs, it was not effectively controlled at any of the 3 intervention sites; for example, BV recurred rapidly after (mass) treatment in Rakai [5]

Simulation of the trialsThe detailed model quantifications of the trials are described elsewhere [23]. In brief, HIV was introduced in 1978 in the Masaka and Rakai populations [28] and in 1983 in Mwanza [29]. To explain the much higher prevalence of HIV in Rakai (15%–25% [30]) and Masaka (10%–15%) [6], compared with Mwanza (2.5% [19]) by 1990, we assumed that rates of formation of new sexual partnerships and of one-off sexual contacts in Uganda increased to levels higher than those in Mwanza, during the Ugandan civil war (1978–1986) [10–13]. A slightly higher prevalence of risk behaviors was specified, in turn, for Rakai, compared with Masaka. We furthermore assumed that subsequent socioeconomic stabilization and the adoption of a national AIDS-control program at the sites in Uganda resulted in reduction of risk behaviors after 1986 [10–13]. The assumed behavior change consisted of sustained reductions in rates of partner change and in one-off sexual contacts and, in 1990, the onset of condom use in 10% of casual and one-off sexual contacts. For Mwanza, no changes in sexual behavior over time were modeled, and no condom use was assumed. Quantifications of rates of partner change were based on numbers of partners reported by male participants in trial surveys, but adjustments were allowed so that the model could fit observed STI prevalences. Proportions for those married and for age and patterns of sexual risk behaviors adequately fitted the trial data [23]

Coverage and clinical efficacy of STI treatment interventions were estimated on the basis of data from the trial surveys. The Rakai intervention consisted of 2 rounds of mass treatment, in 1994 and 1995, at which 70% of individuals aged 15–59 years were treated and which resulted in an ∼95% cure rate for gonorrhea, chlamydia, chancroid, and syphilis and an ∼85% cure rate for trichomoniasis [5]

Coverage of syndromic treatment in Mwanza varied between STIs, depending on the proportion of symptomatic STIs, treatment-seeking behavior, and the clinical efficacy of drugs used. The intervention involved a sustained increase, starting in 1992, in the proportion of symptomatic STIs for which treatment was sought, from 15%–20% (the level of the comparison arm) to 40%–50% [31]. In addition, the clinical efficacy of treatment increased from 30% to 60% for gonorrhea and chlamydia, from 30% to 40% for trichomoniasis, from 50% to 60% for syphilis, and from 0% to 60% for chancroid [31]. Overall, the intervention increased the mean proportion of symptomatic STIs that were cured, from 5% to 25%

For Masaka, syndromic treatment was implemented in 1995, and we used the same quantifications as those used for Mwanza. The Masaka syndromic treatment intervention was implemented in the presence of an educational intervention to reduce sexual risk behaviors (arm A: education; arm B: education plus syndromic treatment of STIs; arm C: comparison) [6]. For the purpose of this analysis, however, we present results of simulations of STI treatment alone, without education, compared with an arm without any intervention

Design of simulations and counterfeit scenariosFor each simulation of the intervention or comparison arm of a trial, results were averaged over 150 runs, to eliminate the influence of random fluctuations associated with stochastic simulations. The simulated populations consisted of ∼20,000 individuals of all ages, but presented outcomes refer to the populations aged 15–54 years, 15–44 years, or ⩾15 years, according to the age group used for empirical rates and impact (see legends for figures 1 and 2)

Simulated impacts of STI treatment are expressed as proportional reductions in the prevalence of STI or the incidence of HIV infection in the intervention arm, compared with the comparison arm, over the course of a 2-year period after the onset of interventions in Mwanza and Rakai and over the course of a 4-year period after the onset of intervention in Masaka, which is analogous to the impacts reported from the trials. The possible influences of hypothesized determinants of impacts in the trials were assessed by comparing the intervention impacts in counterfeit scenarios against those in the trial base-case scenarios

Previous SectionNext Section

Results

Differences between Populations

Prevalence of risk behaviors and STIsComparisons of trial data that were standardized by age and adjusted for differences in diagnostic testing between the sites [32] revealed that the Mwanza trial population had prevalences of high-titer serological syphilis, gonorrhea, chlamydia, and trichomoniasis that were higher than those of the Ugandan trial populations; this difference was not observed only for chlamydia among male individuals [9]. Furthermore, sexual risk behaviors, such as the reported number of recent sex partners, were more prevalent in the Mwanza population than in the Ugandan populations. In contrast, data comparisons suggested that, in the 1980s, the prevalence of risk behaviors must have been higher in Uganda. For example, the excess in numbers of partners in Mwanza was larger for recent partners than for lifetime partners, and the higher prevalence of syphilis in Mwanza was limited to high-titer, recently acquired infection, whereas all-titer syphilis (including latent infection acquired many years before) was more prevalent in Uganda, among older participants [9]

Figure 1 shows how these population differences were simulated for the 3 trials. The high-risk behavior in Uganda in the 1980s, relative to that in Mwanza, explains the higher prevalences of HIV infection in Rakai and Masaka (figure 1A–C). The assumed reduction in risk behaviors in Uganda after 1986, to a level below that in Mwanza, resulted in lower rates of curable STIs in the Ugandan trials, relative to those in Mwanza, as shown for gonorrhea in figure 1D–F

Simulated reductions in incidence of HIV infection during the trials in the STI-intervention groups were 28% in Mwanza, 3% in Masaka, and 10% in Rakai (figure 2, top, A). Compared with the observed impacts (38% in Mwanza, 0% in Masaka, and 3% in Rakai), the simulated differences in risk behaviors and the epidemiology of HIV infection and STI, between Mwanza and Uganda, thus explain, to a large extent, the observed differences in impact on HIV infection

These population differences were not associated with corresponding differences in reductions in STIs across the sites, as illustrated for gonorrhea in figure 2(bottom, A). Depending on the intervention, the simulated impact on STIs was larger in Rakai or in Mwanza and did not depend on whether behavior changed, in the Ugandan populations, or on the timing of the intervention. For gonorrhea, the impact was systematically higher for mass treatment than for syndromic treatment, owing to the low probability of symptoms for gonorrhea [33, 21]. However, this was not the case for ulcerative STIs, particularly chancroid (data not shown), and, overall, the impact on the prevalence of cofactor STIs was relatively similar across populations and interventions

Stage of HIV epidemicThese base-case scenarios demonstrated that the population differences between Uganda and Mwanza could explain, to a large extent, the contrast in impact on HIV infection, between the trials, but do not establish whether behavior change in Uganda or the stage of the HIV epidemic per se was the main determinant. To determine this, we simulated a counterfeit scenario in which no behavior change in Uganda took place after 1986 (as indicated by the dashed lines in figure 1). Impacts on incidence of HIV infection in the Masaka and Rakai trials would then be much larger than those in the trial scenarios—that is, 21% and 19%, respectively (figure 2, top, C)—suggesting that behavior change was the more important determinant

Additional counterfeit scenarios in which the trial interventions were moved forward or backward in time, in relation to the HIV epidemic, confirmed that the timing of the intervention during the HIV epidemic had a similar but smaller effect (figure 2, top, D). For example, with implementation of the Masaka and Rakai trials in 1987, the 9th year of the HIV epidemic (which corresponds to the timing of the Mwanza trial, relative to the introduction of HIV), impact would increase to 17% and 12%, respectively. Conversely, had the Mwanza trial been postponed to 2000, the 17th year of the epidemic (which corresponds to the timing of the Masaka trial), impact would be reduced to 6%, according to simulations. Furthermore, previous work based on modeling of the epidemic in Rakai has shown that, in the absence of behavior change, only a gradual reduction in impact on HIV infection would have occurred as the epidemic matured, whereas impact decreased steeply after behavior change [8]

Increased incidence of HSV-2 ulceration in mature HIV epidemicsAn aspect of the more advanced HIV epidemic in Uganda that was not included in the above simulations was the increased rate of ulcers due to HSV-2 in patients with HIV infection/AIDS, owing to immunosuppression. To explore the influence of the dynamics of HSV-2 infection on the impact of STI treatment during the HIV epidemic, we simulated the effects of HIV infection on ulceration due to HSV-2, as described elsewhere [34]. As with the observed 3–4-fold increases in the number of days of HSV-2–positive culture results and in clinical incidence of ulcers [35–37], patients with AIDS and patients with HIV infection–related symptoms during the 2 years preceding onset of AIDS were assumed to experience twice the frequency of recurrent ulcers due to HSV-2 (4 episodes/year, instead of 2) and twice the duration of ulcers (2 weeks, instead of 1). In this scenario, the incidence of symptomatic herpetic ulcers among individuals aged 15–54 years increased, compared with that in the base-case scenario, from 12.7 episodes/100 person-years (py) to 17.7 episodes/100 py in Rakai and from 12.4 episodes/100 py to 16.2 episodes/100 py in Masaka, in 1995. In the less severe HIV epidemic in Mwanza, the effect on incidence of herpes would be small. However, no trial data were available to verify these predictions

Consistent with trial data [9], simulations showed that the effect of HIV infection on herpes did not considerably increase HSV-2 seroprevalence in Uganda. For example, simulated seroprevalences for individuals aged 15–54 years in Rakai, the site with the highest HIV prevalence, were 53% in the scenario allowing for an effect of HIV infection on herpes and 49% in the base-case scenario. This limited influence is explained by the high prevalence of HSV-2, compared with that of HIV, so that most HSV-2–infected subjects are HIV negative, and by saturation of HSV-2 transmission, so that enhanced ulceration in patients with HIV infection does not lead to increased HSV-2 transmission in the population [34]. For Rakai, the simulated impact of mass treatment on HIV infection decreased somewhat in a scenario allowing for an effect of HIV infection on HSV-2 infection (from 10% to 6%), compared with the base-case scenario, thus improving the fit to empirical data. For Masaka and Mwanza, impact on incidence of HIV infection did not change appreciably (data not shown)

Population mobilityA final proposed population determinant was greater population mobility in the Ugandan trials, which included larger villages on trading routes, relative to the Mwanza trial, in which most communities were more isolated and some were located on islands in Lake Victoria. Population mobility would enhance reintroduction of STIs into the study area after treatment, by allowing for more sexual contacts between participants and external nonparticipants. Especially in small-scale trials, the resulting artificial dilution of trial impact might be significant. However, comparison of trial data did not confirm a difference in migration rates between the 3 trials, although data were limited and did not pertain to short-term mobility [9]. Furthermore, the hypothesis lacks plausibility because mobility would dilute reductions in both HIV infection and STI. Moreover, the coverages of STI treatment achieved in all 3 trials were limited (see Methods), and reinfection from nonparticipants within the study area may have been more important than reinfection associated with migration and travel. This hypothesis, therefore, was discarded without performing counterfeit simulations

Differences between Interventions

Clinical syndromic management would be more effective than periodic mass treatment in the prevention of HIV infection if it achieved a higher coverage of STI episodes, owing to its continuous availability, despite treatment being limited to patients with symptoms. However, the results from the Masaka trial refuted this hypothesis, and the trial simulations shown in figure 2(top and bottom: A, C and D) illustrate that population characteristics can explain to a large extent the lack of impact of both mass treatment and syndromic treatment in the Ugandan trials. The hypothesis also lacked plausibility because it implied a larger impact on STIs in Mwanza and Masaka, compared with that in Rakai, which was not consistently observed. Simulations of mass treatment in Mwanza and Masaka and of syndromic treatment in Rakai, at the coverage levels estimated in the original trials, confirmed these conclusions (figure 2, top, B). In simulations, Mwanza always showed a larger impact on HIV infection, compared with that in Masaka and Rakai, irrespective of the STI treatment strategy, whereas the magnitude of reductions in STI was similar across sites (figure 2, bottom, B)

A more plausible mechanism for a greater impact of syndromic treatment on HIV infection, compared with mass treatment, might be that, at a given coverage of STIs comparable to that under mass treatment, syndromic treatment covers proportionately more symptomatic STIs, which are stronger cofactors for HIV infection than are asymptomatic STIs. Correlation between symptoms and cofactor effect was incorporated into the model to some extent, in that those infections and infection phases associated with more-severe inflammation or ulceration were attributed the highest cofactor effects. Thus, gonorrhea and chlamydia were assumed to have higher cofactor effects than was trichomoniasis, which is consistent with the&ranking of these 3 infections in terms of probability of symptoms. Similarly, the cofactor effect assumed for syphilis was lower than that for chancroid, which is more highly symptomatic [23]. No relationship between symptoms and cofactor effect was modeled for individual episodes of gonorrhea, chlamydia, or trichomoniasis, although HIV shedding (a proxy for infectivity) has been shown to correlate with signs of inflammation [38], which, in turn, may influence the probability of symptoms. However, this model simplification for the nonulcerative STIs is unlikely to have materially affected our predictions, since, in the Mwanza simulations, the impact on HIV infection was shown to depend almost entirely on the reduction in chancroid, the STI with both the highest cofactor effect and the highest probability of symptoms

In addition to diagnosis and treatment, syndromic management of STIs includes health education to encourage sexual abstinence or condom use until treatment is completed, to prevent infection of partners or reinfection of the patient [39]. A behavioral response by patients to the counseling may have contributed to the impact of syndromic treatment in Mwanza. The possible effects of such a behavioral response might be considerable at the population level, since patients with STI are at high risk of recent HIV acquisition (e.g., from the same high-risk sexual contact) and since HIV infectivity is highest in the weeks immediately after infection [40]. Available data from the trial do not provide clear evidence for or against this hypothesis. The intervention did not increase reported condom use, and uptake of condoms at clinic visits was very low (0.9%). Only 57% of patients with STI reported that they had received health education, and 30% reported that they had been offered condoms, suggesting that this aspect of syndromic management was not pursued very vigorously by health workers. However, at follow-up, participants in the intervention arm tended to report fewer recent (casual) partners, compared with participants in the comparison arm [3]. In addition, among men with STI symptoms, a lower proportion (32%) in the intervention arm reported having had sex when symptoms were present, compared with those in the comparison arm (50%) (P = .039). We cannot exclude the hypothesis that these reported differences were due to reporting bias among those who had received counseling. Conversely, more-subtle behavioral effects may have been undetected in the small-scale behavioral surveys [41, 42]. As a result of these limitations, we cannot draw firm conclusions regarding the contribution of behavior changes to the impact of syndromic treatment in Mwanza

Differences in Study Design

Although the study design was broadly similar across the 3 trials, subtle differences were of potential relevance to their comparative impacts. Of the 3 populations surveyed, the Rakai cohort was most vulnerable to rapid STI reinfection, since intervention areas were immediately bordered by nonintervention areas. In Mwanza and Masaka, by contrast, the survey cohorts were sampled from the inner parts of larger intervention areas. Had coverage of the interventions been higher, this design would have diluted impact in Rakai to below the level expected from large-scale implementation. However, as discussed above, given the relatively low intervention coverage in all 3 trials, populations were prone to rapid reinfection from within the study area, so that this difference in study design is unlikely to have appreciably influenced trial outcomes

In Mwanza, impact on incidence of HIV infection was measured in a closed cohort, whereas the Rakai and Masaka cohorts were open. The closed-cohort design may have influenced impact on STIs and HIV infection in Mwanza, by undersampling of more-mobile individuals who contribute most to the incidence of STIs and HIV infection. However, the direction of this influence was unclear. On one hand, mobile individuals may have used the improved health services the least; on the other hand, because of their high rates of HIV and other STIs, they might have benefited proportionately more. In any case, the representation of the trial populations in STDSIM, in which only low levels of migration were modeled [23], avoided this potential bias

Random Error

Because they were based on relatively small numbers of incident or prevalent cases, measured reductions in HIV and other STIs in the trials were subject to random error, as reflected in the large CIs around the point estimates. Within these CIs, simulations could adequately replicate all observed reductions in HIV and other STIs in all trials simultaneously (figure 2, top and bottom, A [23]). However, simulated reductions in measured STIs in Mwanza tended to be higher than those observed in the trial (except those for high-titer syphilis), but the simulations [23] were unable to fully replicate the observed reduction in HIV infection at this site (figure 2, top, A). Random error in the measured impact on HIV infection may have been partly responsible for this discrepancy

One factor of interest in Mwanza was the distribution of risk factors for HIV infection, between trial communities at baseline. By chance, baseline HIV prevalence in the intervention cohort was 3.8%, compared with 4.4% in the comparison communities [3]. This imbalance was paralleled by differences in the same direction in the prevalences of gonorrhea, chlamydia, and self-reported ulcers, although most covariates showed good balance between the study arms [3]. Although data limitations (surveys covering only small samples of the study area and differences that were mostly nonsignificant) imply that these findings cannot be regarded as conclusive, they may indicate that the incidence of HIV infection would have been lower in the intervention arm in the absence of an intervention effect

In simulations presented in the current study, these differences between arms at baseline have been ignored for simplicity, but we previously investigated the importance of imbalances in HIV prevalence at baseline in the Mwanza trial, in a model scenario that fitted HIV prevalence separately for both trial arms, by specifying lower-risk sexual behavior or a later introduction of HIV infection in the intervention arm relative to the comparison arm [15]. These simulations suggested that the resultant bias may be considerable, by inflating, for example, the reduction in HIV infection from a true impact of 17%–22% to an observed impact of 28%. By contrast, the more conventional approach of statistical adjustment, in which baseline HIV prevalence and other risk factors are incorporated in the logistic regression analysis, resulted in a smaller reduction in estimated impact, from 43% to 38% [4]. Both these approaches are subject to limitations. Statistical adjustment may have been incomplete because of inadequacies in measures of baseline risk factors and because not all factors were included in the adjustment. Conversely, the model-based adjustment may be too large because it assumed a homogeneous study population with low rates of migration, in which incidence of HIV infection might depend more strongly on baseline HIV prevalence than it did in reality

Previous SectionNext Section

Discussion

Of the numerous factors proposed to explain the contrasting outcomes of the 3 community-based trials of STI treatment for HIV prevention in Uganda and Tanzania, simulation modeling of STI and HIV transmission during and preceding the trials allowed us to describe the distinguishing characteristics most plausibly linked to the differing outcomes. The low impact of STI treatment on incidence of HIV infection in Rakai and Masaka in the 1990s is explained by low rates of curable cofactor STIs, which were due, in turn, to reduction in risk behaviors prior to the trials and, secondarily, to the advanced stage of the HIV epidemic in Uganda

Data from the trials demonstrate that risk behavior and rates of curable cofactor STIs were substantially higher in Mwanza than in Rakai or Masaka, and simulations confirmed that STI treatment interventions would be expected to have a much larger effect in Mwanza. The observed point estimate for the reduction in incidence of HIV infection in Mwanza (38%) was larger than predicted by the model (28%), on the basis of the intervention’s estimated effect on STI-cofactor burden. This discrepancy could be explained by random error in the impact measure from the trial. Unmeasured behavioral effects of syndromic management also may have contributed, but there is little empirical evidence to support this. Finally, the simulation model may not have included relevant mechanisms that possibly were present in reality, such as a positive association between individual risk behavior and symptom recognition or treatment seeking for STIs

Published estimates of population-attributable fractions (PAFs) for incidence of HIV infection due to the symptomatic STIs targeted in the Mwanza trial may provide further support for the contribution of random error [43]. The PAF estimate for the comparison arm was 26% (40% for men and 12% for women), so that even a full elimination of cofactor STIs would not have explained a 38% reduction in HIV transmission. However, PAF estimates based on observed associations between HIV and other STIs are an unsatisfactory measure of the population-level effect of STIs on the spread of HIV. These estimates cannot fully control for confounding by shared risk factors for HIV and other STIs (leading to overestimation) and capture only STI-cofactor effects on HIV acquisition but not on HIV infectivity (leading to underestimation [44]). At the coverage levels achieved in the trials, the strategy of STI treatment did not influence the overall magnitude of reductions in STI or HIV infection (figure 2, top and bottom, B), as had been shown before for Mwanza and Rakai specifically [8, 19]

The original purpose of this modeling study was to assess the relative effectiveness of STI treatment in the 3 populations, compared with other strategies for HIV infection prevention, and a major limitation is that the results remain fraught with uncertainty. Given the low PAF for incidence of HIV infection due to curable STIs in Uganda, the impact on incidence of HIV infection would have been low irrespective of the magnitude of STI-cofactor effects or the efficacy of the STI interventions. Our understanding of the efficacy of STI treatment in populations with high STI prevalence thus continues to depend mainly on the observations from Mwanza. Uncertainties in the interpretation of the Mwanza trial include not only the role of random error but also the intermediate impact achieved on the targeted STI. Our simulations have suggested that the impact on HIV infection in Mwanza depended almost entirely on a reduction in chancroid, the STI with the largest cofactor effect and for which the simulated reduction in prevalence was far larger (at a prevalence ratio of 0.19, between the intervention and comparison arms, after 2 years of intervention) than that for any of the other STIs. Because of its low basic reproductive number [45], chancroid is particularly susceptible to intervention, and anecdotal evidence from several populations in Africa [46, 47] suggests that even modest control efforts have resulted in the virtual elimination of chancroid as a public health problem. Nonlinear effects of syndromic treatment on the prevalence of this highly symptomatic STI are therefore plausible, but, unfortunately, there are no empirical data from the Mwanza trial to either refute or confirm our predictions

Reductions in STIs in all 3 trials were less than expected at the outset of the interventions [48]. In part, this reflects the moderate coverage achieved by the interventions. In Rakai, coverage was somewhat reduced by the frequent absence of residents during rounds of mass treatment. For syndromic treatment services in Mwanza and Masaka, true coverage was more difficult to ascertain, but coverage was certainly limited by poor recognition of symptoms in these poorly educated rural populations (despite some health education as part of the Mwanza intervention) [21]. Improving, through community-based health education, the proportion of patients with STIs who recognize symptoms and promptly seek treatment would greatly enhance the impact of improved treatment services. Improved clinical treatment services combined with periodic mass treatment to cover the remaining asymptomatic infections may be worthwhile [19]

The findings have several implications for the design and interpretation of future trials of HIV infection prevention. Because of the small number of randomization units (communities) and of incident cases of HIV infection in all trials, the possible role of random error must be recognized. To avoid important imbalances between intervention and comparison arms, increasing the number of randomization units or closer matching of baseline characteristics than was done in the trials considered here may be desirable. The interpretation of the trials was hampered by suboptimal data on reductions in STIs, particularly in Mwanza, where the targeted STIs were monitored only in subsamples of the trial population and by use of imperfect diagnostic tests. More-precise estimates of these intermediate outcomes would have helped to better distinguish between true and random effects. Better diagnostic tests are now available and should be used in future studies. Finally, in these trials study design probably had little influence on outcomes, but, with interventions of higher coverage, the possibility of underestimation of efficacy, owing to reinfection from outside small-scale intervention areas, must be considered

In conclusion, in the many populations in Africa and southern Asia with high-risk sexual behaviors, high rates of STI, and poor STI treatment services, syndromic and mass treatment of STIs may make an important contribution to HIV infection prevention. Improvement of treatment services should include health education on treatment seeking and should be implemented along with primary prevention aiming at reducing sexual risk behaviors. For advanced HIV epidemics in which behavior change has already occurred, interventions to prevent HIV transmission in stable relationships, such as microbicides, control of HSV-2 infection, voluntary counseling and testing, and vaccines, deserve urgent attention

Previous SectionNext Section

Acknowledgments

We thank the directors of the National Institute for Medical Research, Tanzania, the Uganda Virus Research Institute, and the African Medical and Research Foundation, Tanzania, for their support and assistance in carrying out this study. We are grateful to the study communities in Mwanza, Tanzania, and Rakai and Masaka, Uganda, for their participation and support. We also would like to thank all the members of the 3 trial teams in Uganda and Tanzania for their helpful contributions

Previous SectionNext Section

Footnotes

  • ↵Presented in part: 14th International AIDS Conference, Barcelona, 7–12 July 2002 (abstracts MoOrD1085, MoOrD1086, and MoOrD1087)

    Financial support: UK Department for International Development (grant RD499); Commission of the European Communities (for development of the STDSIM model [contracts B7.6211/96/010 and B7.6211/97/017])

  • ↵Present affiliation: HIV, TB, and Malaria Cluster, Roll Back Malaria Department, World Health Organization, Geneva, Switzerland

Previous Section
 

References

  1. ↵
    1. Fleming DT,
    2. Wasserheit JN
    . From epidemiological synergy to public health policy and practice: the contribution of other sexually transmitted diseases to sexual transmission of HIV infection. Sex Transm Infect 1999;75:3-17.
    Abstract
  2. ↵
    1. Grosskurth H,
    2. Gray R,
    3. Hayes R,
    4. Mabey D,
    5. Wawer M
    . Control of STDs for HIV prevention: understanding the implications of the Mwanza and Rakai trials. Lancet 2000;355:1981-7.
    CrossRefMedlineWeb of Science
  3. ↵
    1. Grosskurth H,
    2. Mosha F,
    3. Todd J,
    4. et al
    . Impact of improved treatment of sexually transmitted diseases on HIV infection in rural Tanzania: randomised controlled trial. Lancet 1995;346:530-6.
    CrossRefMedlineWeb of Science
  4. ↵
    1. Hayes R,
    2. Grosskurth H,
    3. Ka-Gina G
    . Impact of improved treatment of sexually transmitted disease on HIV infection [reply]. Lancet 1995;346:1159-60.
    Web of Science
  5. ↵
    1. Wawer MJ,
    2. Sewankambo NK,
    3. Serwadda D,
    4. et al
    . Control of sexually transmitted diseases for AIDS prevention in Uganda: a randomised community trial. Lancet 1999;353:525-35.
    CrossRefMedlineWeb of Science
  6. ↵
    1. Kamali A,
    2. Quigley M,
    3. Nakiyingi J,
    4. et al
    . A community randomised trial of sexual behaviour and syndromic STI management interventions on HIV-1 transmission in rural Uganda. Lancet 2003;361:645-52.
    CrossRefMedlineWeb of Science
  7. ↵
    1. Hitchcock P,
    2. Fransen L
    . Preventing HIV infection: lessons from Mwanza and Rakai. Lancet 1999;353:513-5.
    CrossRefMedlineWeb of Science
  8. ↵
    1. Korenromp EL,
    2. Bakker R,
    3. De Vlas SJ,
    4. et al
    . HIV dynamics and behaviour change as determinants of the impact of STD treatment on HIV transmission in the context of the Rakai trial. AIDS 2002;16:2209-18.
    CrossRefMedlineWeb of Science
  9. ↵
    1. Orroth KK,
    2. Korenromp EL,
    3. White RG,
    4. et al
    . Higher risk behaviour and rates of sexually transmitted diseases in Mwanza compared to Uganda may help explain HIV prevention trial outcomes. AIDS 2003;17:2653-60.
    CrossRefMedlineWeb of Science
  10. ↵
    1. Asiimwe-Okiror G,
    2. Opio AA,
    3. Musinguzi J,
    4. Madraa E,
    5. Tembo G
    . Change in sexual behaviour and decline in HIV infection among young pregnant women in urban Uganda. AIDS 1997;11:1757-63.
    CrossRefMedlineWeb of Science
  11. ↵
    1. Kilian AH,
    2. Gregson S,
    3. Ndyanabangi B,
    4. et al
    . Reductions in risk behaviour provide the most consistent explanation for declining HIV-1 prevalence in Uganda. AIDS 1999;13:391-8.
    CrossRefMedlineWeb of Science
  12. ↵
    1. Kamali A,
    2. Carpenter LM,
    3. Whitworth JA,
    4. Pool R,
    5. Ruberantwari A
    . Seven-year trends in HIV-1 infection rates, and changes in sexual behaviour, among adults in rural Uganda. AIDS 2000;14:427-34.
    CrossRefMedlineWeb of Science
  13. ↵
    1. Serwadda D,
    2. Wawer MJ,
    3. Musgrave SD,
    4. Sewankambo NK,
    5. Kaplan JE
    . HIV risk factors in three geographic strata of rural Rakai District, Uganda. AIDS 1992;6:983-9.
    MedlineWeb of Science
  14. ↵
    1. Boily MC,
    2. Lowndes CM,
    3. Alary M
    . Complementary hypothesis concerning the community sexually transmitted disease mass treatment puzzle in Rakai, Uganda. AIDS 2000;14:2583-92.
    CrossRefMedlineWeb of Science
  15. ↵
    1. Korenromp EL,
    2. De Vlas SJ,
    3. Habbema JDF
    . Program and abstracts of the XIII International AIDS conference. Stockholm: International AIDS Society; 2000. The effect of imbalances in risk factors between arms at baseline in community-randomized HIV prevention trials [abstract WePpC1313]; p. 108.
  16. ↵
    1. Hudson CP
    . Community-based trials of sexually transmitted disease treatment: repercussions for epidemiology and HIV prevention. Bull World Health Organ 2001;79:48-58.
    MedlineWeb of Science
  17. ↵
    1. O’Farrell N
    . Cost-effectiveness of services for sexually transmitted diseases in the prevention of HIV-1 [letter]. Lancet 1998;351:681-2.
    Medline
  18. ↵
    1. Korenromp EL
    . Cofactor magnitudes, syndromic management and a reappraisal of the Mwanza and Rakai trials [PhD thesis]. Rotterdam, The Netherlands: Erasmus University Rotterdam; 2001. Treatment of sexually transmitted diseases as an HIV prevention strategy; p. 264.
  19. ↵
    1. Korenromp EL,
    2. Van Vliet C,
    3. Grosskurth H,
    4. et al
    . Model-based evaluation of single-round mass STD treatment for HIV control in a rural African population. AIDS 2000;14:573-93.
    CrossRefMedlineWeb of Science
    1. O’Farrell N
    . Interpretation of the Mwanza and Rakai trials [letter]. Bull World Health Organ 2001;79:482.
    MedlineWeb of Science
  21. ↵
    1. Korenromp EL,
    2. Sudaryo MK,
    3. De Vlas SJ,
    4. et al
    . What proportion of episodes of gonorrhoea and chlamydia becomes symptomatic. Int J STD AIDS 2002;13:91-101.
    Abstract/FREE Full Text
    1. Dik J,
    2. Habbema JDF,
    3. de Vlas SJ
    . Impact of improved treatment of sexually transmitted disease on HIV infection [letter]. Lancet 1995;346:1157-8. (discussion: 1159–60).
    CrossRefMedline
  23. ↵
    1. White RG,
    2. Orroth KK,
    3. Korenromp EL,
    4. et al
    . Can the contrasting results of the Mwanza, Rakai and Masaka HIV/STD intervention trials be explained by population differences? A modelling study. J AIDS. (in press).
  24. ↵
    1. Korenromp EL,
    2. De Vlas NJ,
    3. Nagelkerke NJD,
    4. Habbema JDF
    . Estimating the magnitude of STD cofactor effects on HIV transmission: how well can it be done. Sex Transm Dis 2001;28:613-21.
    MedlineWeb of Science
  25. ↵
    1. Sewankambo N,
    2. Gray RH,
    3. Wawer MJ,
    4. et al
    . HIV-1 infection associated with abnormal vaginal flora morphology and bacterial vaginosis. Lancet 1997;350:546-50.
    CrossRefMedlineWeb of Science
  26. ↵
    1. Martin HL,
    2. Richardson BA,
    3. Nyange PM,
    4. et al
    . Vaginal lactobacilli, microbial flora, and risk of human immunodeficiency virus type 1 and sexually transmitted disease acquisition. J Infect Dis 1999;180:1863-8.
    Abstract/FREE Full Text
  27. ↵
    1. Zenilman JM,
    2. Fresia A,
    3. Berger B,
    4. McCormack WM
    . Bacterial vaginosis is not associated with circumcision status of the current male partner. Sex Transm Infect 1999;75:347-8.
    Abstract
  28. ↵
    1. Serwadda D,
    2. Sewankambo NK,
    3. Carswell JW,
    4. et al
    . Slim disease: a new disease in Uganda and its association with HTLV-III infection. Lancet 1985;2:849-52.
    MedlineWeb of Science
  29. ↵
    1. Killewo J,
    2. Nyamuryekunge K,
    3. Sandstrom A,
    4. et al
    . Prevalence of HIV-1 infection in the Kagera region of Tanzania: a population-based study. AIDS 1990;4:1081-5.
    MedlineWeb of Science
  30. ↵
    1. Wawer MJ,
    2. Serwadda D,
    3. Gray RH,
    4. et al
    . Trends in HIV-1 prevalence may not reflect trends in incidence in mature epidemics: data from the Rakai population-based cohort, Uganda. AIDS 1997;11:1023-30.
    CrossRefMedlineWeb of Science
  31. ↵
    1. Buvé A,
    2. Gavyole A,
    3. Mugeye K,
    4. et al
    . Program and abstracts of the 12th International AIDS Conference. Stockholm: International AIDS Society; 1998. A closer look at the intervention of the Mwanza trial: how many patients with an STD are cured by health services? [abstract 33234]; p. 639.
  32. ↵
    1. Orroth KK,
    2. Korenromp EL,
    3. White RG,
    4. et al
    . Comparison of STD prevalences in the Mwanza, Rakai, and Masaka trial populations: the role of selection bias and diagnostic errors. Sex Transm Infect 2003;79:98-105.
    Abstract/FREE Full Text
  33. ↵
    1. Paxton LA,
    2. Sewankambo N,
    3. Gray R,
    4. Serwadda D,
    5. McNairn D
    . Asymptomatic non-ulcerative genital tract infections in a rural Ugandan population. Sex Transm Infect 1998;74:421-5.
    Abstract
  34. ↵
    1. Korenromp EL,
    2. Bakker R,
    3. De Vlas SJ,
    4. Robinson NJ,
    5. Hayes R
    . Can behaviour change explain increases in the proportion of genital ulcers attributable to herpes in sub-Saharan Africa. A simulation modelling study. Sex Transm Dis 2002;29:228-38.
    MedlineWeb of Science
  35. ↵
    1. Schacker T,
    2. Zeh J,
    3. Hu HL,
    4. Hill E,
    5. Corey L
    . Frequency of symptomatic and asymptomatic herpes simplex virus type 2 reactivations among human immunodeficiency virus–infected men. J Infect Dis 1998;178:1616-22.
    Abstract/FREE Full Text
  36. ↵
    1. Mbopi-Keou F-X,
    2. Grésenguet G,
    3. Mayaud P,
    4. et al
    . Genital HSV-2 shedding is increased in HIV-infected women in Africa. AIDS 1999;13:536-7.
    CrossRefMedlineWeb of Science
  37. ↵
    1. Kaul R,
    2. Kimani J,
    3. Nagelkerke NJD,
    4. et al
    . Risk factors for genital ulcerations in Kenyan sex workers: the role of HIV-1 infection. Sex Transm Dis 1997;24:387-92.
    MedlineWeb of Science
  38. ↵
    1. Wang CC,
    2. McClelland RS,
    3. Reilly M,
    4. et al
    . The effect of treatment of vaginal infections on shedding of human immunodeficiency virus type 1. J Infect Dis 2001;183:1017-22.
    Abstract/FREE Full Text
  39. ↵
    1. World Health Organization (WHO)
    . Management of sexually transmitted diseases [WHO/GPA/TEM/94.1] Geneva: WHO; 1994.
  40. ↵
    1. Jacquez JA,
    2. Koopman JS,
    3. Simon CP,
    4. Longini I Jr.
    . Role of the primary infection in epidemics of HIV infection in gay cohorts. J Acquir Immune Defic Syndr 1994;7:1169-84.
    MedlineWeb of Science
  41. ↵
    1. Morris M
    . Telling tails explain the discrepancy in sexual partner reports. Nature 1993;365:437-40.
    CrossRefMedline
  42. ↵
    1. Wadsworth J,
    2. Johnson AM,
    3. Wellings K,
    4. Field J
    . What’s in a mean. An examination of the inconsistency between men and women in reporting sexual partnerships. J R Stat Soc [Ser A] 1996;159:111-23.
  43. ↵
    1. Orroth KK,
    2. Gavyole A,
    3. Todd J,
    4. et al
    . Syndromic treatment of sexually transmitted diseases reduces the proportion of incident HIV infections attributable to these diseases in rural Tanzania. AIDS 2000;14:1429-37.
    CrossRefMedlineWeb of Science
  44. ↵
    1. Garnett GP,
    2. Rottingen JA
    . Understanding the dynamic consequences of infectious disease interactions and fractions attributable to STDs. Int J STD AIDS 2001;12 Suppl 2:20-1.
  45. ↵
    1. Brunham RC,
    2. Plummer FA
    . A general model of sexually transmitted disease epidemiology and its implications for control. Med Clin North Am 1990;74:1339-52.
    MedlineWeb of Science
  46. ↵
    1. Steen R,
    2. DeCoito T,
    3. Ralepeli S,
    4. et al
    . Program and abstracts of the 13th Meeting of the International Society for Sexually Transmitted Diseases Research. Amsterdam: ISSTDR; 1999. Rapid control of chancroid in a South African mining company [abstract 337]; p. 191.
  47. ↵
    1. Steen R
    . Eradicating chancroid. Bull World Health Organ 2001;79:818-26.
    MedlineWeb of Science
  48. ↵
    1. Hayes R,
    2. Mosha F,
    3. Nicoll A,
    4. et al
    . A community trial of the impact of improved sexually transmitted disease treatment on the HIV epidemic in rural Tanzania. I. Design. AIDS 1995;9:919-26.
    MedlineWeb of Science
| Table of Contents

This Article

  1. J Infect Dis. (2005) 191 (Supplement 1): S168-S178. doi: 10.1086/425274
  1. AbstractFree
  2. » Full Text (HTML)Free
  3. Full Text (PDF)Free

Classifications

Share

  1. Email this article

Navigate This Article

Current Issue

  1. March 1, 2012 205 (5)
  1. Journal of Infectious Diseases
  1. Alert me to new issues

Most