Disentangling Contributions of Reproductive Tract Infections to HIV Acquisition in African Women

Abstract

Objective: 

To estimate the effects of reproductive tract infections (RTIs) on HIV acquisition among Zimbabwean and Ugandan women.

Methods: 

A multicenter prospective observational cohort study enrolled 4439 HIV-uninfected women aged 18 to 35 attending family planning clinics in Zimbabwe and Uganda. Participants were interviewed, and tested for HIV and RTIs every 3 months for 15 to 24 months. They received HIV risk reduction counseling, male condoms, and treatment for curable RTIs.

Results: 

Despite HIV risk reduction counseling and regular screening and treatment for RTIs, the HIV incidence did not decline during the study. Positive HSV-2 serostatus at baseline (hazard ratio [HR] = 3.69, 95% confidence interval = 2.45–5.55), incident HSV-2 (HR = 5.35, 3.06–9.36), incident Neisseria gonorrhoeae (HR = 5.46, 3.41–8.75), and altered vaginal flora during the study (bacterial vaginosis [BV]: HR = 2.12, 1.50–3.01; and intermediate flora: HR = 2.02, 1.39–2.95) were independently associated with HIV acquisition after controlling for demographic and behavioral covariates and other RTIs (Treponema pallidum, Chlamydia trachomatis, Trichomonas vaginalis, and vaginal yeasts). For N. gonorrhoeae, C. trachomatis, T. vaginalis, and vaginal yeasts, the risk of HIV acquisition increased when the infection was identified at the visit before the HIV-detection visit or with the duration of infection. Population attributable risk percent (PAR%) calculations show that HSV-2 contributes most to acquisition of new HIV infections (50.4% for baseline HSV-2 and 7.9% for incident HSV-2), followed by altered vaginal flora (17.2% for bacterial vaginosis and 11.8% for intermediate flora).

Conclusions: 

A substantial proportion of new HIV infections in Zimbabwean and Ugandan women are attributable to RTIs, particularly HSV-2 and altered vaginal flora.

In Brief

Secondary analyses of the “Hormonal Contraception and Risk of HIV Acquisition” study showed that a substantial proportion of new HIV infections in Zimbabwean and Ugandan women are still attributable to reproductive tract infections, most importantly herpes simplex virus 2 followed by altered vaginal flora.

Observational studies have consistently shown that sexually transmitted infections (STIs) facilitate HIV transmission.^1–5 The evidence for endogenous vaginal infections (such as bacterial vaginosis [BV] and vaginal yeasts) as cofactors in HIV transmission is less clear but is mounting.^6–10 Several mechanisms have been proposed by which sexually transmitted and vaginal infections, also referred to as reproductive tract infections (RTIs), could increase HIV transmission. RTIs may enhance infectivity of HIV-positive patients by increasing HIV viral load and shedding of HIV virus in the genital tract or in semen.¹¹ They may enhance susceptibility to HIV by changing epithelial integrity and/or permeability and by causing inflammation and/or immune activation in the genital tract.^12,13

At a population level, most agree that STI control contributes to reduced HIV incidence, but perhaps with different effects in different stages of the HIV and STI epidemics and in different populations depending on the types of sexual networks.^14,15 Evidence of STI control reducing HIV incidence mostly comes from quasi-experimental, cohort, and ecological data; only 1¹⁶ of 7 randomized controlled trials of STI control (including herpes simplex virus type 2 [HSV-2] suppression) for HIV prevention demonstrated a statistically significant reduction in HIV incidence.^16–22 However, most of these trials were underpowered to measure modest effect sizes, STI control procedures may have been insufficient (treatment of some but not all relevant STIs, use of syndromic management in women, low frequency of screening and/or treatment, and lack of partner notification and treatment), and the HSV-2 suppression trials may have suffered from insufficient dosing and/or poor treatment adherence. Recent studies have shown that HSV-2 and BV are not only independently associated with HIV acquisition, but also with each other in a bidirectional manner.^23,24 To date, no randomized controlled trials have evaluated the effect of simultaneously controlling for viral and nonviral STIs and vaginal flora on the acquisition of HIV infection.

Between 1999 and 2004, when the data presented in this article were collected, both Zimbabwe and Uganda experienced generalized HIV epidemics.²⁵ In Zimbabwe, the HIV prevalence in pregnant women attending urban and peri-urban antenatal clinics declined from about 30% in 1999 to 20% in 2004.²⁶ In Uganda, HIV prevalence rates started declining in 1992 and stabilized in the early 2000s. Between 1999 and 2004, HIV prevalence rates were stable but lower than in Zimbabwe (10% among urban residents).²⁵

As part of a large prospective cohort study in Zimbabwe and Uganda, women were routinely tested for a variety of RTIs using state-of-the-art laboratory techniques. While STI control was not the main aim of the study, all study participants received voluntary counseling and testing for HIV and other RTIs, male condoms, and treatment for curable infections and symptomatic herpes outbreaks approximately every 3 months. In this article, we report on the prevalence and incidence of RTIs and HIV infection over time, relationships between each RTI and subsequent HIV acquisition, and the population attributable risk percentages (PAR%) for HIV due to each of the RTIs. While a previous article using data from the same cohorts examined relationships between prevalent and incident HSV-2 infection and HIV acquisition,⁴ this article also includes nonviral STIs and vaginal flora (BV, intermediate flora, and the presence of yeasts) in the same statistical models so that their independent effects on HIV acquisition could be evaluated.

MATERIALS AND METHODS

The effect of RTIs on HIV acquisition was examined through secondary analysis of the Hormonal Contraception and the Risk of HIV Acquisition (HC-HIV) study. Detailed methods of this study have been published previously.²⁷ The study was conducted between November 1999 and January 2004 and was approved by ethical review committees of collaborating institutions in the United States, Uganda, and Zimbabwe. All women provided written informed consent before participating in the study.

Study Population

All Zimbabwean and most Ugandan women were recruited from urban and peri-urban public family planning and mother-child health clinics.²⁷ HIV incidence in Uganda was initially lower than expected and recruitment was expanded to include higher-risk populations, such as patients at sexually transmitted disease clinics, sex workers, and military wives. At enrollment, all participants were aged 18 to 35 years, HIV-uninfected, sexually active, not pregnant, had not had a spontaneous or induced abortion in the previous month, had not injected drugs or had a blood transfusion in the previous 3 months, and reported using either combined oral contraceptives (COCs; low-dose pills containing 30 μg ethinylestradiol and 150 μg levonorgestrel), depot-medroxyprogesterone acetate (150 mg administered every 12 weeks), or a nonhormonal method of contraception (condoms, withdrawal, traditional methods, sterilization) continuously for at least 3 months and intending to continue use for an additional year. Each site enrolled roughly equal numbers of women into using COCs, depot-medroxyprogesterone acetate, and nonhormonal methods of contraception. Women were not eligible to participate in this study if they had a hysterectomy or an intrauterine device.

Data Collection Procedures

At screening, consenting women were counseled and tested for HIV and T. pallidum. Women returned within 15 days for their test results and possible enrollment. At enrollment, eligible consenting women were interviewed in their local language about demographics, sexual and contraceptive behavior, and reproductive health using structured questionnaires. They were counseled on HIV risk reduction and contraception, and received condoms and contraceptives free of charge. A standardized physical and pelvic examination was conducted, and specimens were collected to test for various RTIs (see following paragraphs). Follow-up visits were conducted every 12 weeks for 15 to 24 months. Follow-up procedures were similar to those at enrollment and included structured questionnaires, physical exams, and testing for HIV and other RTIs; women were tested for T. pallidum biannually. Stored sera from the enrollment and exit (if negative at enrollment) visits were tested for HSV-2 retrospectively. For women who were HSV-2 antibody negative at enrollment and positive at study exit, stored sera from follow-up visits were tested to determine the timing of HSV-2 seroconversion.

Laboratory Methods

At enrollment, all participants were determined to be HIV negative by ELISA. At each quarterly follow-up visit, HIV testing was done using ELISA, with positive results confirmed using rapid testing, followed by Western blot or polymerase chain reaction (PCR) testing; HIV DNA PCR results were the final arbiter of infection status. Incident HIV infections were confirmed by a subsequent blood draw using an ELISA or rapid test. For confirmed HIV infections, HIV DNA PCR was performed serially on previous visit specimens. HIV incidence was defined as the date of the first positive PCR result.

HSV-2 serostatus was determined by a type-specific serological IgG antibody enzyme-linked immunosorbent assay (ELISA, Focus Technologies). To optimize the specificity of the ELISA in Uganda, an index cutoff value of 3.4 was used to determine HSV-2 seropositivity.²⁸ All of the HSV-2 seroconversions were confirmed by ELISA at an external laboratory. HSV-2-positive serostatus was defined as the date of the first positive HSV-2 ELISA result. Women who acquired HSV-2 infection during the study follow-up period were classified as being HSV-2 negative at each visit before HSV-2 seroconversion and classified as having incident HSV-2 at the seroconversion visit and at each subsequent visit.

T. pallidum was diagnosed by Rapid Plasma test (RPR) with Treponema pallidum hemagglutionation test/Treponema pallidum particle agglutionation test (TPHA/TPPA) confirmation (Serodia, Japan). Incident T. pallidum was defined as testing positive for RPR and TPHA/TPPA after concordant negative or discordant test results at the previous visit. N. gonorrhoeae and C. trachomatis were identified by PCR (COBAS Amplicor). Incident N. gonorrhoeae and C. trachomatis were defined as a positive PCR test after a negative test at the previous visit. Wet mount microscopy of vaginal fluid was conducted on-site to diagnose T. vaginalis, BV by Amsel criteria (used for treatment purposes only), and vaginal yeasts. Incident T. vaginalis and vaginal yeasts were defined as the presence on wet mount after absence on wet mount at the previous visit. Vaginal smears were air-dried and stored for Gram staining and subsequent Nugent scoring, with BV defined as a Nugent score of 7 to 10 and intermediate flora as Nugent score of 4 to 6.²⁹ Incident BV was defined as a Nugent score of 7 to 10 after a previous Nugent score of 0 to 6. External quality control of Nugent scoring was conducted at the University of California at San Francisco.

Treatment for Curable Infections

Vaginal infections diagnosed by wet mount and genital ulcers diagnosed syndromically were treated on-site before the woman left the clinic. Women who were subsequently diagnosed with chlamydia, gonorrhea, active syphilis, or active herpes by laboratory testing were recalled to the study clinic for treatment if they had not already been optimally treated. Treatments were as follows: syphilis with doxycycline (100 mg twice daily for 14 days) or erythromycin (500 mg 4 times daily for 15 days) in Zimbabwe and benzathine penicillin (2.4 mega IU) in Uganda; herpetic lesions with acyclovir; other genital ulcer diseases with erythromycin (500 mg 4 times daily for 15 days); chlamydia with doxycycline (100 mg twice daily for 7 days) or erythromycin (500 mg 4 times daily for 7 days) for pregnant women; gonorrhea with intramuscular kanamycin (2 g) or norfloxacin (800 mg) in Zimbabwe, and with ciprofloxacin (500 mg) in Uganda; trichomoniasis and BV with metronidazole in both countries (different dosages were used in both countries: 2 g once, or 200 or 400 mg 2 or 3 times daily); and vaginal yeast with 1 application of clotrimazole or gynodaktarin cream per day for 3 to 7 days in Zimbabwe and with clotrimazole pessaries (100 mg for 3–7 days) or oral nystatin (100,000 U once or twice daily for 14 days) in Uganda. In both Zimbabwe and Uganda, most women preferred to take an extra dose of treatment home to give to their male partner(s). Only a few men attended the study clinic for treatment.

Statistical Analyses

Incidence density rates were calculated based on Poisson distributions with person time of follow-up as the denominator. Multivariable Cox proportional hazards models for repeated observations were used to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) for the effect of each RTI on time to HIV infection. Person-time was calculated as time from enrollment until time of first positive HIV test or last visit date for women remaining HIV-uninfected. In most models, predictors (including RTIs) were measured at the same visit as the HIV infection (current visit). However, we hypothesized that risk of HIV acquisition may depend on the timing and duration of RTIs (which in turn determine the timing and duration of immune activation and/or epithelial disruption in the genital tract) in relation to HIV exposure. Some models therefore considered the presence of RTIs at the visit before the visit at which the HIV infection was diagnosed (previous visit), or a combination of visits (previous or current visit, and previous and current visit). Statistical analyses were performed using SAS (version 9.3.1, SAS Institute, Cary, NC) for the 2 countries combined as well as for each country separately.

HSV-2 infection was categorized as a 3-part time-varying variable (HSV-2 seropositive at baseline, HSV-2 seroincident during the study, or HSV-2 negative) because women who have recently acquired HSV-2 infection may have more HSV-2 viral shedding and local immune activation than women with established HSV-2 infection.⁴ BV was also categorized as a 3-part time-varying variable (BV, intermediate flora, or normal flora) because a recent study has shown that intermediate flora is also associated with HIV acquisition.¹⁰T. pallidum, N. gonorrhoeae, C. Trachomatis, T. vaginalis, and vaginal yeast were categorized as time-varying dichotomous variables (positive or negative).

Covariates that were considered a priori to be potential confounders (such as “country” and “high-risk recruitment group in Uganda”), and covariates that were associated with HIV (P ≤0.05) in bivariable models, were assessed for confounding by adding them to the multivariable model one-by-one: if any of the RTI-HIV effect measures changed by 15% or more, the covariate was retained in the model³⁰. RTIs may cluster in individuals due to similar underlying risk factors. Spearman correlation coefficients for each pair of RTIs were therefore calculated but none was statistically significant. The effect of each RTI on other RTI-HIV relationships was also assessed, and the only RTIs that showed some evidence of clustering were N. gonorrhoeae and C. trachomatis. However, the change in effect measure of each RTI with HIV was 11% and all RTIs were therefore included in each multivariable model. Circumcision of the primary partner was not assessed for confounding because a previous analysis of the same dataset showed that it was not associated with HIV acquisition in women.³¹ To evaluate heterogeneity of RTI-HIV relationships among population subgroups (women recruited from family planning clinics vs. higher-risk settings in Uganda), interaction terms were examined. These were not statistically significant and therefore not included in the models. Adjusted hazard ratios from the final multivariable Cox models were used to calculate PAR% using the formula: PAR% = pd (HR − 1/HR) × 100%, where pd is the proportion of cases that was exposed during the study.³²

RESULTS

A total of 8346 women were screened to enroll 2240 women in Zimbabwe and 2199 women in Uganda. The 24-month retention rate was 92% in Zimbabwe and 97% in Uganda. Mean follow-up was 23.3 months; median time between study visits was 3 months. A total of 7776 person-years of follow-up was accrued over 31,197 follow-up visits.

The mean age of study participants at enrollment was 25 years, the median education was 9 years, and the median number of live births was 2 (Table 1). Most participants lived with their primary sexual partner (83%) and few women (4%) reported more than 1 sex partner in the 3 months before enrollment. All women were sexually active but the majority reported no (55% in Zimbabwe and 61% in Uganda) or inconsistent (21% and 17%) condom use at baseline. The Ugandan cohort included a higher proportion of women who engaged in risk behaviors, had partners who engaged in risk behaviors, reported a history of STIs, or reported RTI symptoms in the 12 months before enrollment than the Zimbabwe cohort due to the different recruitment strategies in each country. About one-fifth (20%) of women in both countries had been pregnant in the year before enrollment. Circumcision of the primary partner was reported by 36% of Ugandan women and 9% of Zimbabwean women.

TABLE 1: Participant Characteristics at Enrollment by Country

HIV prevalence at screening was 39% in Zimbabwe and 17% in Uganda. HSV-2 prevalence at enrollment was 53% and 49% in Zimbabwe and Uganda, respectively (Table 1). BV, intermediate flora, and vaginal yeasts were also common at enrollment with prevalence estimates of 29%, 23%, and 15% in Zimbabwe and 21%, 11%, and 5% in Uganda. Nonviral STIs were less common at baseline (Table 1).

Table 2 shows incidence rates for HIV infection and each RTI for each country over different observation periods. In Zimbabwe, HIV incidence was 4.1 per 100 woman-years of follow-up. Most incidence rates (including HIV) remained stable over the entire observation period but the incidence of the 3 vaginal infections (BV, yeasts, and trichomoniasis) declined slightly. In Uganda, HIV incidence was 1.3 per 100 woman-years of follow-up in the general population and 2.6 per 100 woman-years of follow-up in the high-risk population. The overall HIV incidence was 1.5 per 100 woman-years of follow-up and remained stable over the entire observation period. The incidence rate of HSV-2 infection increased over time, while the incidence of trichomoniasis and yeasts on wet mount declined slightly.

TABLE 2: Incidence (Per 100 Woman-Years of Follow-Up) of HIV and RTIs by Country and by Observation Period

In the multivariable Cox model for Zimbabwe and Uganda combined, positive HSV-2 serostatus at baseline (HR = 3.69, 95% CI = 2.45–5.55), incident HSV-2 (HR = 5.35, 95% CI = 3.06–9.36), incident N. gonorrhoeae (HR = 5.46, 95% CI = 3.41–8.75), and altered vaginal flora during the study (BV: HR = 2.12, 95% CI = 1.50–3.01; and intermediate flora: HR = 2.02, 95% CI = 1.39–2.95) were independently associated with HIV acquisition after controlling for demographic and behavioral covariates and other RTIs (T. pallidum, C. trachomatis, T. vaginalis, and vaginal yeasts) (Table 3). The magnitude of the associations with HIV acquisition for HSV-2, N. gonorrhoeae, and intermediate flora were similar for each country, but C. trachomatis, BV, and yeasts were associated with HIV acquisition in Zimbabwe but not in Uganda. T. vaginalis and T. pallidum were the only 2 infections that were not associated with HIV acquisition in any of the models in Table 3.

TABLE 3: Multivariable Associations Between RTIs and Incident HIV Infection Among Women in Zimbabwe and Uganda

Positive HSV-2 serostatus at baseline carried the greatest PAR% (50.4%), followed by altered vaginal flora (17.2% for BV and 11.8% for intermediate flora), incident HSV-2 (7.9%), and incident N. gonorrhoeae (5.3%) (Table 4). The PAR% for all other infections were 3% or below. PAR% were similar for Zimbabwe and Uganda, except for BV, intermediate flora, vaginal yeasts, and C. trachomatis: the PAR% of these conditions were larger in Zimbabwe (21.3%, 13.2%, 6.4%, and 2.6%) than in Uganda (8.5%, 9.5%, −9.8%, and −5.3%).

TABLE 4: HIV Population Attributable Risks Percents (PAR%) for RTI in Women in Zimbabwe and Uganda

The association with HIV acquisition was greater if N. gonorrhoeae was present at the current visit (HR = 5.46, 95% CI = 3.41–8.75) than at the previous visit (HR = 3.03, 95% CI = 1.67–5.48), and the strongest when present at both visits (HR = 7.09, 95% CI = 3.13–16.08) (Table 5). C. trachomatis, T. vaginalis, and vaginal yeasts were only significantly associated with HIV acquisition if the previous visit was included in the analysis.

TABLE 5: Temporal Relationships Between RTIs and HIV Among Women in Zimbabwe and Uganda

DISCUSSION

The incidence of HIV infection and RTIs remained high throughout the study period despite regular risk reduction counseling, active screening and treatment for curable RTIs, and availability of partner treatment. All RTIs were associated with HIV acquisition in at least 1 statistical model except for T. pallidum (probably due to its low prevalence). For N. gonorrhoeae, C. trachomatis, T. vaginalis, and vaginal yeasts, the risk of HIV acquisition increased when the infection was identified at the visit before the HIV detection visit (the previous visit) or at both the previous and current visit. HSV-2 and N. gonorrhoeae were associated with an increased risk of HIV acquisition in all models.

For outcomes with multiple risk factors that overlap, PAR% can sum to more than 100% because some individuals with more than 1 risk factor can have the outcome prevented in more than 1 way; these preventable cases are consequently counted more than once.³³ However, PAR% can still be interpreted as estimating the relative importance of each predictor. In this study, HSV-2 accounted for the largest PAR% (50.4% for seroprevalent and 7.9% for seroincident infections), followed by altered vaginal flora (17.2% for BV and 11.8% for intermediate flora), and N. gonorrhoeae (5.3%). We compared our results with those from other studies that measured PAR% of HIV due to RTIs. In the Mwanza intervention study, the PAR% for chancroid was estimated to be 40% (using mathematical modeling) and an additional 25% for all other curable STIs (syphilis, gonorrhea, chlamydia, and trichomoniasis).³⁴ In the Rakai intervention study, the PAR% for genital ulcer disease (GUD) was 9%, but only 8% of GUD cases were caused by syphilis or chancroid.³⁴ The PAR% for HSV-2 was estimated to be 23% (using mathematical modeling) and 20% for all curable STIs.³⁵ PAR% were not reported for the other 5 intervention studies. None of these PAR% calculations took both HSV-2 and altered vaginal flora into account.

Our study has a number of strengths and limitations. An important strength is that we tested study participants regularly for all common RTIs using state-of-the-art diagnostic tests (with 1 exception, see following paragraphs), and that we included all RTIs in each statistical model to avoid overestimation of a single RTI cofactor. However, despite the longitudinal nature of our study, temporal relationships between RTIs and HIV acquisition remain uncertain because of infection detection delays, RTI persistence in the genital tract, and the fact that some RTI episodes may have gone unnoticed because the interval between study visits was approximately 3 months. We tried to minimize the uncertainty by using HIV DNA PCR to better time the HIV transmission event. Furthermore, women were encouraged to visit the study clinic in case of symptoms, and were given treatment free of charge at the study clinic. Finally, statistical analyses were done for various observation periods before HIV detection.

STI cofactor values that are estimated from longitudinal observational studies are often inflated because of clustering due to the common mode of acquisition through sexual contact with an infected partner.¹⁵ While we adjusted our analyses for several underlying risk behaviors of the woman herself as well as her main sex partner, residual confounding cannot be excluded because we did not have complete risk-taking and exposure information for entire sexual networks. We only estimated cofactor values and PAR% for HIV acquisition in HIV-uninfected women. We did not determine the impact on HIV infectiousness of HIV-infected individuals, nor the indirect effects resulting from onward transmission of HIV and STIs in sexual networks.

Our data show that a sizable proportion of women who were diagnosed with an RTI had persistent or recurrent infection at the next visit. While the majority of women who needed treatment were in fact treated, unfortunately, treatment delays were common and tests of cure were not conducted after treatment. Also, while partner notification was offered to women, few partners came to the study clinics for treatment, and we do not know whether partner treatments given to women were in fact taken as prescribed by their male partner(s). In the case of N. gonorrhoeae, C. trachomatis, T. vaginalis, and syphils, reinfection rates may have been high due to low or delayed response to partner notification and treatment. In the case of altered vaginal flora and vaginal yeasts, treatment failure is common.³⁶ While we were able to show that detection of a specific infection at 2 subsequent visits increased the risk for HIV acquisition for some infections, we could not differentiate between persistent and recurrent infections.

State-of-the-art diagnostic tests were used for most infections, but the diagnostic test used for T. vaginalis and vaginal yeasts (wet mount) was suboptimal. PCR testing on a subset of specimens from this cohort revealed that wet mount diagnosis of T. vaginalis was less sensitive than PCR diagnosis,⁵ and external quality control procedures revealed that wet mount readings were less reliable in Uganda (done by clinicians) than in Zimbabwe (done by laboratory technicians). The latter may account for the larger PAR% estimates for vaginal yeasts in Zimbabwe than in Uganda. Furthermore, misclassification of active syphilis infections could have occurred because TPHA titers were not always available. In our study (as opposed to some of the intervention studies), women were not tested for Haemophilus ducreyi.

Even though only 1 of 7 randomized controlled trials of STI control (including HSV-2 suppression) for HIV prevention demonstrated a statistically significant reduction in HIV incidence,^16–22 our data suggest that aggressive RTI control is likely to make an important contribution to HIV prevention, especially if HSV-2 infection and altered vaginal flora are taken into account. Work on therapeutic and prophylactic HSV-2 vaccines should continue,^37,38 and better treatments or interventions are needed to allow women to maintain a healthy vaginal flora over long periods of time.³⁹ HIV and RTIs cause major morbidity and social harm and we can therefore not afford to stop efforts to bring them under control.