Using Modeling to Explore the Degree to Which a Microbicide's Sexually Transmitted Infection Efficacy May Contribute to the HIV Effectiveness Measured in Phase 3 Microbicide Trials

Abstract

Background: 

Several microbicide candidates show activity against pathogens that cause sexually transmitted infections (STIs). This may increase a microbicide's impact on HIV in phase 3 trials. Modeling is used to estimate the degree to which a microbicide's STI efficacy contributes to the HIV effectiveness of a phase 3 microbicide trial.

Methods: 

An expression is derived and coupled with an STI model to estimate how much a microbicide's STI efficacy contributes to a trial's HIV effectiveness. The STI model estimates the decrease in STI prevalence that may occur in the trial's active gel arm for microbicides of different STI efficacy. Projections are produced for different STI cofactors and epidemiological settings.

Results: 

The model projects that if a microbicide is active against curable STIs with a combined prevalence of ≥10% among trial participants and the reduction in HIV incidence is <50%, then the STI activity could have substantially contributed to the trial's HIV effectiveness (>50% in some cases) if the per exposure multiplicative STI cofactor is 2.5 or greater. However, if the STI prevalence is <10% or the STI cofactor is <2.5 or if the reduction in HIV incidence is >50%, then the trial's HIV effectiveness will be mainly due to its direct HIV efficacy.

Conclusions: 

In high STI settings, phase 3 trials documenting a moderate impact on HIV incidence may partially result from a gel's activity against curable STI. Care should be taken generalizing these trial results to other settings. This is less important for trials documenting large reductions in HIV incidence.

BACKGROUND

Currently, there are many microbicide products under development, with 4 in or entering phase 2b/3 effectiveness trials.¹ Potential products range from vaginal defense buffers and absorption inhibitors to antiretroviral-based replication inhibitors. In vitro and animal models suggest that most non-antiretroviral microbicide candidates have activity against many of the pathogens that cause sexually transmitted infections (STIs).^2,3 Therefore, because STIs are cofactors for STI transmission,⁴ a microbicide's human immunodeficiency virus (HIV) impact is likely to result both from its direct activity on HIV and also from its indirect effect on other STIs.^4,5

If a microbicide is shown to be effective in reducing HIV transmission in a phase 3 trial, it is important to know the degree to which this was due to the microbicide's direct activity on HIV and its indirect effect on other STIs. If no difference in STI prevalence is observed in the active and placebo gel arms of the trial, then the trial's effectiveness can be assumed to be due to the microbicide's direct effect on HIV. However, if the STI prevalence is significantly lower in the active gel arm, and as the same STI treatment will have been provided in both trial arms, this reduction can be attributable to the active gel. It will then be important to estimate the extent to which a gel's STI activity contributed to reducing HIV transmission, as this will influence the degree to which the trial results can be generalized to settings with different STI prevalences.

A recent modeling study by Desai et al⁶ explored this issue for recently completed circumcision trials and concluded that the protective effect circumcision may provide for STI transmission is unlikely to have contributed much to the documented impact on HIV incidence. However, although they did consider a higher STI prevalence setting in a sensitivity analysis, the main focus of their analysis was a setting where the prevalence of curable STIs was low, ~8%. To explore how these projections may vary for microbicide trials among women who can have much higher prevalences of curable STIs,^7,8 this paper uses a static trial model coupled with a simple STI model to explore the degree to which a product's direct effect on STI transmission may contribute to a trial's HIV effectiveness through reductions in STI prevalence. Projections are produced for a range of potential products that provide protection for both women and their partners or just the women using them, for different STI cofactors and for hypothetical trials with different prevalences of curable STIs and levels of STI treatment. The implications for interpreting future microbicide trials are discussed.

METHODS

Trial HIV Effectiveness Model Derivation

As with other randomized controlled trials,^9-11 the primary outputs of a phase 3 microbicide trial are the rates of incident HIV infections per 100 person years in the placebo (λ_p) and active gel (λ) arms of the trial. At the simplest level, these outputs are used to estimate the microbicide's effectiveness in terms of the relative rate ratio (RR):

If the women in the active and placebo gel arms have the same sexual activity (m partners per unit time and n sex acts with each) and their male partners have the same HIV prevalence (y'), then the rate of incident HIV infections can be approximated by:

where b is the amount by which microbicide use directly decreases the per sex act HIV transmission probability (defined as the HIV efficacy of the microbicide) and M is the average consistency of gel use by women in the active gel arm of the trial. The parameters B and B_p are the average per sex act male to female HIV transmission probabilities in the active gel and placebo trial arms, including the cofactor effect of any STIs and any protection given by condoms. Substituting these expressions into equation 1 gives the following:

If the prevalence of different STIs and condom use is the same in both trial arms, then equation 3 simplifies to RR = Mb, implying the trial effectiveness (RR) equates to the product of the microbicide's HIV efficacy and the average consistency that it was used in the active gel arm. This product Mb is the degree to which the microbicide directly decreased the per sex act HIV transmission probability, which we define as the “HIV use effectiveness” of the microbicide.

However, if the prevalence of different STIs in each trial arm is significantly different, then B and B_p are likely to be different. If the prevalence of a specific STI in the active and placebo gel arms is S_j and S_pj among the women and S′_j and S′_pj among their male partners (different STIs having different subscripts j), then B and B_p can be estimated by the following expression if multiple STI cofactors are assumed to act in an additive fashion:

where α_j is the per sex act increase in HIV susceptibility associated with either partner being infected with STI j (α_j = 1 if STI j does not increase HIV transmission), β is the per sex act probability of male to female HIV transmission, f and f_p are the consistencies of condom use in each trial arm, and e is the HIV efficacy of condoms. Equation 4 is derived from the standard formulation for one STI-

Substituting these into equation 3 and assuming that f and f_p are equal gives the following:

where

and

are the average probabilities that at least one individual in each partnership has STI j in the active or placebo gel trial arms

STI Model Derivation

To explore the degree to which an STI efficacious microbicide may reduce the prevalence of a curable STI over the duration of follow-up of a trial, a simple deterministic STI transmission model is developed. The model simulates the transmission of an STI from women in the trial to their male sexual partners, and vice versa, and assumes that the male partners can also become infected from other partners not involved in the trial. Once a woman or man is infected, he/she remains infected until he/she is treated through the trial (enrolled women only), treated by another treatment provider, or self-cures, all of which are assumed to occur at constant rates. Once an individual is treated or self-cures, he/she is assumed to become susceptible to infection again. In the baseline simulations, the microbicide is assumed to reduce the transmission of the STI from males to females and vice versa. However, the implications of a microbicide only reducing STI transmission to the females are also explored. To simplify matters, the model simulates the cohort of women who complete the full duration of follow-up. In addition, because of the relatively short follow-up time used in microbicide trials, it is assumed that very few of their male partners die due to acquired immunodeficiency syndrome or other causes. The equations for the model are as follows:

where S and S′ are the prevalence of the STI in the female trial population and their male partners, respectively, and the other parameters are defined in Table 1. For each simulation, the model is first run to equilibrium without any trial STI treatment (ψ = 0) or microbicide use included (M = 0) and then it is run twice, first to simulate the effect of the STI treatment provided to participants in both trial arms and second to simulate the additional effect of microbicide use in the intervention arm. To estimate how any changes in STI prevalence may effect the overall transmission of HIV in both trial arms, the average STI prevalence over the duration of the trial is numerically estimated.

TABLE 1: Parameter Values for Uncertainty Analysis of STI Model

Analyses Undertaken

To obtain simple and conservative projections on how much a microbicide's STI efficacy may alter a trial's HIV effectiveness, this analysis considers the overall prevalence of curable STIs in each trial arm and attributes an average STI cofactor to them. This is likely to underestimate the impact of a microbicide on the overall STI prevalence among trial participants because, for example, it is normally easier to control 4 STIs at 5% prevalence compared to 1 STI with their combined prevalence. The analysis does not consider the effect of the microbicide on herpes simplex virus-2 transmission because any positive effect achieved through reducing herpes simplex virus-2 transmission is likely to be small over a trial's short time frame, both because it is incurable and also due to its high prevalence in many settings.¹²

Effect of Microbicide on STI Prevalence

Because there is uncertainty over the likely prevalence of STIs in different trial settings, especially among their male partners, and to make the results more generalizable to trials with different levels of STI treatment, the STI model was run over a wide range of parameter values to mimic the STI levels in different settings. The uncertainty analyses assumed that a microbicide either protects against STI transmission in both directions (bidirectional) or protects only susceptible women (unidirectional) and explored how microbicides of different STI efficacies may decrease the prevalence of curable STIs over the duration of a trial-assumed to be 12 months in this analysis. The parameter ranges used by the model are shown in Table 1 and were sampled randomly to produce 5000 different paired simulations of the STI model for each arm. For microbicides with different STI efficacies, and for different initial STI prevalences and levels of trial STI treatment, the output from each pair of simulations was used to estimate the relative decrease in the mean STI prevalence among trial participants using active gel compared with women using placebo gel. The results were used to produce best, upper and lower bound quadratic fits for the relationship between the microbicide's STI efficacy and the relative decrease in the mean STI prevalence achieved in the trial women and their male partners. The upper and lower bound quadratic fits were determined by ensuring that 97.5% of the model projections lay below the upper bound fit and 97.5% lay above the lower bound fit.

Effect of Microbicide on HIV Incidence

The quadratic fits obtained from each uncertainty analysis were then used with equation 5 to estimate the reduction in HIV incidence that should be achieved in the intervention arm compared with the placebo gel arm for microbicides of differing HIV and STI efficacy. These projections were used to explore the degree to which the STI efficacy of a microbicide can contribute to any decrease in HIV incidence observed in a trial. In addition, different projections were produced that assumed the mean STI prevalence in the placebo arm over the duration of the trial was 5%-20%. These scenarios translate to different combinations of initial STI prevalence and levels of STI treatment provided by the trial. For most of the analysis, a constant STI cofactor was assumed, α = 3.5,⁴ but a sensitivity analysis was also undertaken to explore the implications of assuming an STI cofactor of between 1 and 6.

RESULTS

Effect of a Microbicide's STI Efficacy on a Trial's Observed Decrease in STI Prevalence

The 5000 simulations of the STI transmission model produced a wide range of different STI epidemics, with the predicted STI prevalence before the trial ranging from 0.04% to 96%. For these simulations, the model projected that the STI treatment provided by a microbicide trial is likely to result in a noticeable reduction in STI prevalence but is unlikely to reduce the STI prevalence to very low levels because their male partners are not effectively treated. For example, if 50% of STIs are treated each month by the trial and the initial STI prevalence in the trial sites is 20%-30%, then the model predicts that the average prevalence in the placebo gel arm over the duration of the trial is likely to be 10%-20%. More general projections are shown in Appendix 1.

If only the simulations that predict an average STI prevalence in the placebo arm of less than 40% are considered, then Figure 1 shows the strong relationship between the microbicide's STI use effectiveness, assuming either unidirectional or bidirectional protection for STI transmission, and the predicted relative decrease in average STI prevalence in the microbicide arm of the trial compared with the placebo gel arm. For example, if the product is bidirectional, then the model projects that a 40% STI use effective microbicide will reduce the average STI prevalence in the microbicide arm of the trial by 48% (31%-65%) compared with the placebo gel arm, whereas an unidirectional microbicide will result in a 40% (27%-51%) reduction. Figure 1 also shows the best, upper and lower bound quadratic fits to the model output.

FIGURE 1: Projected decrease in STI prevalence among the women in the microbicide arm of the trial compared with the women in the placebo gel arm for different microbicide STI efficacies. A, Assuming microbicide reduces STI transmission to women in trial and their male partners. B, Assuming microbicide only reduces STI transmission to women in trial.

Effect of a Microbicide's HIV and STI Efficacy on a Trial's Observed Reduction in HIV Incidence

For a hypothetical trial where the trial participant's partners are assumed to have the same initial STI prevalence, Figure 2 shows how the effectiveness of a microbicide trial (in decreasing HIV incidence) is likely to vary for microbicides with different HIV and STI use effectiveness and whether the microbicide is uni- or bidirectional.

FIGURE 2: Best-fit projections for the decrease in HIV incidence observed in a trial for microbicides with different HIV and STI use efficacies, where the microbicide's protection for STI transmission is either unidirectional or bidirectional. The simulations assume that the average STI prevalence among the women in placebo gel arm and their male partners is 15% and STI cofactor of 3.5. A, Microbicide's protection for STI transmission is bidirectional. B, Microbicide's protection for STI transmission is unidirectional.

In general, Figure 2 shows that the degree to which the STI efficacy of a microbicide contributes to the decrease in HIV incidence observed by a trial can be large for microbicides with an HIV use effectiveness less than 50% but diminishes for microbicides with increased HIV efficacy, becoming negligible for microbicides with an HIV use effectiveness greater than 70%. Additionally, the Figure shows that the same reduction in HIV incidence can be achieved by microbicides with different HIV and STI use effectiveness profiles.

Figure 3 shows that, for other scenarios, where the average STI prevalence in the placebo arm is not 15%, but the STI cofactor is still assumed to be 3.5, the model projects that a microbicide's STI efficacy has little importance when the STI prevalence is 5% or less, but for STI prevalences of 10% or above, it plays an important role unless the HIV use effectiveness of the microbicide is 60% or greater. For instance, a bidirectional microbicide with 40% HIV and 80% STI use effectiveness results in a 48% (43%-51%) or 58% (46%-66%) reduction in HIV incidence over the trial if the average STI prevalence in the placebo arm is 5% or 20%, respectively, whereas if the microbicide was 60% HIV and 80% STI use effective then it would result in a 65% (62%-67%) or 72% (64%-78%) reduction in HIV incidence for the same STI prevalences.

FIGURE 3: Projected decrease in HIV incidence observed in a trial for uni- or bidirectional microbicides with different average STI prevalences in the placebo trial arm. The simulations assume an STI cofactor of 3.5 and use the best-fit relationship between the microbicide's STI efficacy and the relative decrease in the STI prevalence in the trial women and their male partners. A, Bidirectional microbicide. B, Unidirectional microbicide.

Lastly, Figure 4 shows that the STI cofactor plays a large role in determining the importance of a microbicide's STI efficacy. For example, the STI efficacy of a bidirectional 40% HIV and 80% STI use effective microbicide results in a 50% increase in the trial effectiveness at an STI cofactor of 6 and a 33% increase at an STI cofactor of 3 (average STI prevalence in placebo arm is 15%). However, if the STI cofactor is 2.5 or lower, then the STI efficacy of the microbicide will not play much of a role unless the STI prevalence is high (greater than 15%). For instance, if the STI cofactor is 2.5, then a bidirectional microbicide with 30% HIV and 70% STI use effectiveness would result in a 35% (32%-38%) or 44% (35%-52%) reduction in HIV incidence over the trial if the average STI prevalence in the placebo arm is 5% or 20%, respectively.

FIGURE 4: Projected decrease in HIV incidence observed in a trial for uni- or bidirectional microbicides while assuming different STI cofactors. The simulations assume that the STI prevalence in the placebo trial arm is 15% and use the best-fit relationship between the microbicide's STI efficacy and the relative decrease in the STI prevalence in the trial women and their male partners. A, Bidirectional microbicide. B, Unidirectional microbicide.

DISCUSSION

In this analysis, mathematical modeling is used to explore the potential contribution of a microbicide's STI efficacy on the reduction in HIV incidence observed in a phase 3 trial. The study has several limitations and made some simplifying assumptions. The model did not explicitly model each specific STI but considered a generic STI, and so the projections are approximate and could be conservative. Additionally, uncertainty in the likely value of the STI cofactor resulted in further uncertainty in our projections. A microbicide's potential role in averting new herpes infections was not incorporated because its effect on HIV transmission was thought to be small over a trial's duration. Because of a lack of data, simplifying assumptions had to be made for the STI prevalence among male partners of trial participants. Lastly, the model did not incorporate any heterogeneity in the sexual behavior or consistency of microbicide use among trial participants. This simplification was made because it was not the focus of this analysis.

Nevertheless, despite these limitations, the findings provide useful information for interpreting phase 3 trial results. First, the analysis illustrates that if a microbicide trial observes a <50% decrease in HIV incidence then, if the overall prevalence of curable STIs in the placebo gel arm was not low (overall prevalence >10%) and the STI cofactor is greater than 2.5, any decrease in STI prevalence due to the microbicide's STI efficacy may have contributed substantially to the observed reduction in HIV incidence. The magnitude of this contribution increases with the average STI prevalence in the placebo gel arm and the relative decrease in STI prevalence in the active gel arm. In these trial scenarios, a range of different HIV and STI efficacy combinations could have resulted in the observed trial effectiveness, and so care should be taken in generalizing the impact projections to settings with lower STI prevalences.

In contrast, if the trial observed a 50%-70% decrease in HIV incidence, then the influence of the microbicide's STI efficacy will only be important if it is relatively large (>50%) and resulted in a large relative decrease in STI prevalence in the active gel arm (~50%), and/or the STI prevalence in the placebo arm was high (>15%), or the STI cofactor is large (>3.5). Lastly, if the trial achieved a large decrease in HIV incidence (>70%), then the trial's effectiveness will be largely due to the direct HIV efficacy of the microbicide, irrespective of any decrease in STI prevalence.

These results highlight that when a microbicide has high HIV efficacy its STI efficacy is less important and contributes less toward impact. This is to be expected because STIs can only increase the probability of HIV transmission during an unprotected sex act and has implications for the next-generation antiretroviral-based microbicides, because if they have high HIV efficacy, it may be less important that they are not STI efficacious.

In general, the results of this analysis are qualitatively similar to the results of Desai et al⁶ for circumcision trials but suggest that the STI efficacy of a microbicide may play a greater role in contributing to the effectiveness of microbicide phase 3 trials than for circumcision trials. This is mainly due to the higher prevalence of curable STIs observed in some microbicide trial settings,⁷ the possibility that the STI protection afforded by microbicides may be bidirectional and the possibility that microbicide trials may observe a smaller decrease in HIV incidence. Compared to Desai et al,⁶ the projections of this analysis are comparable but slightly more conservative with respect to the degree to which a product's/circumcision's STI efficacy contributes to the decrease in HIV incidence observed in a trial. However, considering the different models and methods used, the differences between the projections are small and give a degree of validation to the model projections.

Lastly, obtaining accurate estimates of the contribution of STIs to the impact of microbicide trials requires much better estimates for the cofactor effects of different STIs and needs to be informed by STI data from trial participants and their partners. However, although all current trials routinely document and provide treatment for STIs among trial participants, with some having herpes as a secondary endpoint, none collect STI data from their partners. Our analysis highlights the importance of routinely collecting detailed STI data to help interpret and generalize trial results.

ACKNOWLEDGMENTS

The views expressed are those of the authors and cannot be taken to reflect the official opinion of the London School of Hygiene and Tropical Medicine, UK Department for International Development, or Microbicide Development Programme.