Intravaginal Practices, Bacterial Vaginosis, and HIV Infection in Women: Individual Participant Data Meta-analysis

Summary

Background:
Identifying modifiable factors that increase women's vulnerability to HIV is a critical step in developing effective female-initiated prevention interventions. The primary objective of this study was to pool individual participant data from prospective longitudinal studies to investigate the association between intravaginal practices and acquisition of HIV infection among women in sub-Saharan Africa. Secondary objectives were to investigate associations between intravaginal practices and disrupted vaginal flora; and between disrupted vaginal flora and HIV acquisition.

Methods and Findings:
We conducted a meta-analysis of individual participant data from 13 prospective cohort studies involving 14,874 women, of whom 791 acquired HIV infection during 21,218 woman years of follow-up. Data were pooled using random-effects meta-analysis. The level of between-study heterogeneity was low in all analyses (I² values 0.0%–16.1%). Intravaginal use of cloth or paper (pooled adjusted hazard ratio [aHR] 1.47, 95% confidence interval [CI] 1.18–1.83), insertion of products to dry or tighten the vagina (aHR 1.31, 95% CI 1.00–1.71), and intravaginal cleaning with soap (aHR 1.24, 95% CI 1.01–1.53) remained associated with HIV acquisition after controlling for age, marital status, and number of sex partners in the past 3 months. Intravaginal cleaning with soap was also associated with the development of intermediate vaginal flora and bacterial vaginosis in women with normal vaginal flora at baseline (pooled adjusted odds ratio [OR] 1.24, 95% CI 1.04–1.47). Use of cloth or paper was not associated with the development of disrupted vaginal flora. Intermediate vaginal flora and bacterial vaginosis were each associated with HIV acquisition in multivariable models when measured at baseline (aHR 1.54 and 1.69, p<0.001) or at the visit before the estimated date of HIV infection (aHR 1.41 and 1.53, p<0.001), respectively.

Conclusions:
This study provides evidence to suggest that some intravaginal practices increase the risk of HIV acquisition but a direct causal pathway linking intravaginal cleaning with soap, disruption of vaginal flora, and HIV acquisition has not yet been demonstrated. More consistency in the definition and measurement of specific intravaginal practices is warranted so that the effects of specific intravaginal practices and products can be further elucidated.

: Please see later in the article for the Editors' Summary

Introduction

HIV-infected women outnumber men in many sub-Saharan African countries [1]. Novel female-initiated preventive methods have, so far, proved elusive [2],[3], so identifying new modifiable factors that affect women's vulnerability to HIV might help in the development of new preventive interventions [4]. Women use a wide range of products, applied in a variety of ways inside the vagina, to manage their sexual relationships, menstruation, and to improve wellbeing [5].

It has been hypothesised that some intravaginal practices could increase the risk of HIV infection by causing physical abrasions [4] or by disrupting the vaginal epithelium and increasing the occurrence of bacterial vaginosis (BV) [6]–[8]. An association between BV and HIV has been shown in cross-sectional and prospective studies [9],[10] and, more recently, intermediate grades of vaginal flora have also been associated with an increased risk of HIV acquisition [11]. However, the evidence linking intravaginal practices, BV, and HIV infection is currently inconclusive [8],[12],[13], as is evidence of associations between these practices and BV [12]. A recent systematic review found that few prospective studies recorded intravaginal practices consistently and that there was substantial heterogeneity between studies reporting associations between intravaginal practices and HIV [12]. Even large individual studies lack statistical power to examine the effects of specific intravaginal practices on HIV acquisition. Combining individual participant data from different studies might overcome some of these problems because data can be analysed consistently across studies and statistical power and precision can be increased [14]. Our overall aim was to determine whether specific vaginal practices increase the risk of a woman acquiring HIV infection by facilitating disturbances in vaginal flora or vaginal epithelial disruption. The primary objective of this study was to pool individual participant data from prospective longitudinal studies to investigate the association between intravaginal practices and acquisition of HIV infection among women in sub-Saharan Africa. Secondary objectives were to investigate associations between intravaginal practices and disrupted vaginal flora; and between disrupted vaginal flora and HIV acquisition.

Methods

The study protocol specified hypotheses, inclusion criteria, and methods of analysis, and is available at http://www.ispm.ch/uploads/media/VP_IPD_protocol__final_090205_01.pdf. The review was reported according to the guidelines of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) group (Text S1).

Study Selection

Potentially eligible studies were cohort studies and randomised controlled trials conducted in sub-Saharan Africa that had collected data prospectively about both intravaginal practices and incident HIV infection in females aged 10 y or older. We did not consider studies examining associations between disrupted vaginal flora and HIV unless data on intravaginal practices were also collected. We excluded studies in which participants were asked not to use intravaginal practices during the study or where the primary vaginal practice was female genital cutting or genital surgery. We also excluded studies involving vaginal microbicides or placebo products, tampons, or other devices to deliver medication. However, the control group in such studies was eligible for inclusion if they did not receive these interventions.

We identified studies using the results of a systematic review of published studies, which has been described elsewhere [12]. We obtained published reports of all studies included in the previous review and asked experts in the field to identify additional studies that had collected data about both intravaginal practices and incident HIV infections. We tried to contact the corresponding author or principal investigator of all potentially eligible studies by e-mail to determine whether relevant data had been collected and to invite them to contribute to the individual participant data meta-analysis. We included eligible studies for which we had received both a signed agreement to participate and a dataset by 16th March 2009. All included studies were approved by relevant country-specific and institutional ethical review boards and all participants within each study provided written informed consent for the original studies.

Data Collection

We used protocols, questionnaires, and publications from included studies, and information from investigators to determine whether requested variables had been collected or could be derived. The final variable list included: HIV infection, intravaginal practices, vaginal flora status, herpes simplex virus infection (HSV-2) status, age, education, religion, marital status, employment, age at first sex, numbers of sex partners, sex in exchange for money, pregnancy and contraception, and general condom use (consistent, inconsistent, or never). A named investigator for each included study provided access to the dataset, answered questions about the study procedures and coding of variables, and contributed to interpretation of results and revising the manuscript. For most studies we obtained raw data and recoded these in house (MFC, KS, or SCF). For two studies, a data manager provided a dataset, coded according to our predefined scheme.

Outcome and Exposure Measures

The primary outcome was HIV infection diagnosed within the first 2 y, using the diagnostic criteria defined by each individual study. We included women with a negative HIV test at the baseline visit and at least one follow-up HIV test and estimated the date of HIV infection as the midpoint between the last negative and the first positive test.

Intravaginal practices were the main exposures and were based on self-reported data collected in face-to-face interviews in all but one study, which instead used audio computer-assisted self-interviewing [15]. Definitions were based on a published classification system [6]. After examining the data available in individual studies, we defined five separate intravaginal practices (Box 1). We used baseline data about current practices (that is, those from the 1 to 3 mo preceding the start of the study), since there were too few studies reporting repeated measures of these variables to consider changes in practices over time. The reference group was women not at increased risk of acquiring HIV infection through intravaginal practices, and included women using no intravaginal practice or water.

Vaginal flora status was considered as both an exposure and an outcome, and assessed using scores from Gram-stained vaginal smears [16],[17], or Amsel clinical criteria [18]. We used the Nugent score if results from more than one method were available [16]. BV was defined as a Nugent score of 7–10, Ison-Hay grade III, or the presence of three or more Amsel criteria. Intermediate vaginal flora was defined as a Nugent score of 4–6 [11] or Ison-Hay grade II. Two studies used Amsel criteria only and could not be included in analyses that included intermediate vaginal flora as an exposure or outcome [19],[20].

Assessment of the Risk of Bias

We assessed the potential for bias in each cohort study, arising from prespecified methodological domains [21]: description of participation and evidence of bias; definitions of diagnostic criteria; blinding of outcome assessment; conduct of follow-up; missing data; and measurement of main confounders.

Statistical Analysis

Association between intravaginal practices and HIV infection

For the primary objective we included all women and used Cox proportional hazards models to examine associations in each study between each intravaginal practice at baseline and HIV acquisition, and expressed these as the hazard ratio (HR) and 95% confidence intervals (CIs). Follow-up time was measured until the first of: estimated date of HIV infection; the last follow-up visit; the end of the study; or after 2 y of follow-up. We pooled data if the level of between-study heterogeneity was mild or moderate, defined as a value from the I² statistic below 50% [22]. We used two methods [14]: for all study objectives we used random effects meta-analysis to combine effect estimates from individual strata (two-stage method); for analyses with HIV acquisition as the outcome we also used stratified, fixed effects Cox regression, with studies as the strata (one-stage method). Both methods gave very similar results. The proportional hazards assumption, tested on the basis of Schönfeld residuals, was not violated in any model. All statistical analyses were conducted using Stata (version 10, Stata Corporation).

We looked for statistical evidence of confounding of the association between intravaginal practices and HIV infection in each individual study and for each intravaginal practice by comparing the univariable HR with the HR from bivariable models including baseline measures of each of the following prespecified factors: marital status; numbers of sex partners; condom use; contraception; educational level; religion; employment; age at first intercourse; having received goods or money in exchange for sex; pregnancy; and HSV-2 status. If inclusion of the variable resulted in the HR changing by more than 10% in one or more individual studies we considered it for inclusion in multivariable analyses [23]. We controlled for age in all multivariable models. We present results from models that also controlled for marital status and number of sex partners during the previous 3 mo as recorded at cohort entry because these were the variables that fulfilled our criteria most often. The results were very similar to those from models that controlled for a larger common set of variables, or controlled for different variables in each study.

We then examined the role of BV. We aimed to control for time-varying confounding by BV [24], but there were too many differences between studies in the frequency and timing of vaginal specimen collection to use fully time-updated measures. We therefore conducted two exploratory analyses in multivariable models to see, qualitatively, whether there was attenuation of the HR; using either the result of the baseline test or the last available test for BV before the estimated date of HIV infection, or before the date of censoring.

Associations between intravaginal practices and changes in vaginal flora status

To examine associations between intravaginal practices and short-term changes in vaginal flora status we included women who had normal vaginal flora at baseline, assessed using Gram-stain criteria, and a follow-up assessment within the first year of enrolment. We categorised vaginal flora status at follow-up as normal, intermediate, or BV. We used ordered logistic regression with a proportional odds model [25]. The assumption of the model is that the odds ratio (OR) associated with an intravaginal practice for the odds of intermediate vaginal flora or BV compared with normal vaginal flora is equal to the OR for BV compared with normal or intermediate flora. The model therefore estimates a single OR from the data.

Association between disrupted vaginal flora and HIV acquisition

To examine the association between disrupted vaginal flora and HIV acquisition we included all women with vaginal flora assessed by Gram-stain at baseline. We considered vaginal flora status as a three-level ordered exposure and used Cox proportional hazards regression to estimate the HR (and 95% CI) for incident HIV infection in the first 2 y of follow-up.

Results

Description of Studies

We assessed 22 studies for eligibility; 13 prospective studies conducted in sub-Saharan Africa included in our previous systematic review [8],[19],[20],[26]–[35] and nine identified through expert meetings [15],[36]–[43]. We excluded five studies for which we could not determine eligibility because we did not establish contact with the authors [26]–[28],[31],[33], three studies that did not include data on relevant exposures or outcomes [30],[32],[34], and four studies for which the investigators declined to take part, or could not send their agreement to participate or dataset by 16th March 2009 (Figure 1; Table S1) [29],[38]–[40]. We included ten studies [8],[15],[19],[20],[35]–[37],[41]–[43]. We analysed the data as 13 separate studies, stratifying results from three multicentre studies according to enrolment site [15],[19],[35]. Table 1 shows selected characteristics of the studies, which were done in six sub-Saharan African countries and included data from 14,874 women followed for 21,218 woman years, with 791 incident HIV infections in the first 2 y. The individual studies were generally assessed as having a low risk of bias (data available on request). Most studies involved women enrolled from community settings or clinics providing reproductive health services (studies 1, 6, 7, 8, 9, 10). Two studies in Tanzania enrolled women working as food sellers or in bars and other recreational facilities, amongst whom high prevalences of sexually transmitted infections and HIV have previously been reported (studies 4, 5). Two studies in Kenya enrolled only women who were self-identified sex workers (studies 2, 3). The mean age at enrolment was 29.4 y (standard deviation [SD] 8.7, range 15.2–67.0 y). The incidence of HIV infection in individual studies ranged from 1.6 (95% CI 1.2–2.0, study 1 Uganda) to 15.1 (95% CI 10.5–21.8, study 9) per 100 woman years. The median time from enrolment to the estimated date of HIV infection was 250 d (interquartile range [IQR] 130–415 d).

Frequency of Intravaginal Practices

The percentage of women reporting any current intravaginal practice at baseline ranged from 18% (study 8) to 95% (study 3, Table 2). Studies done in South Africa tended to have a low overall prevalence of any current vaginal practice (18%–27%, studies 8–10) and studies in Zimbabwe tended to have a high prevalence (69%–92%, studies 1, 6, 7). Female sex workers in Kenya (76%–95%, studies 2, 3) and high risk women in Tanzania (67%–76%, studies 5, 4) also reported high levels of any vaginal practice. Table 2 shows the proportions of women reporting any use of specific practices, i.e., whether or not they reported other practices. Cleaning inside the vagina with soap was the most common practice involving a specified product, and was reported by more than one-third of women in six studies in the most northern countries in the region (studies 1–6 Malawi). Reported use of other more abrasive household cleaning products was uncommon, ranging from 0.7% (study 8) to 7% (study 2). Reported use of cloth or paper to wipe out the vagina, or apply products, ranged from 0.3% (study 8) to 70% (study 7 Zimbabwe). Inserting products to dry or tighten the vagina was uncommon; this was most commonly reported in four studies conducted in South Africa and Zimbabwe, where the prevalence was 13%–20% (studies 1 Zimbabwe, 7 Zimbabwe, and 8 South Africa). Cleaning with water, with or without other practices, was reported by more than 60% of women in all but four studies in South Africa (studies 7 South Africa, 8–10). Where measurements of intravaginal practices were available at follow-up visits, the majority of reported practices were consistent with baseline data; 60% reported the same practice at all study visits at which data were collected, 34% reported either the same practice or no practice, and 6% reported different practices at all visits.

Associations between Intravaginal Practices and HIV Infection

Intravaginal use of cloth or paper to wipe out the vagina or apply products was the practice most strongly associated with HIV acquisition in univariable analysis and after controlling for age, marital status, and number of sex partners in the past 3 mo (pooled adjusted HR 1.47, 95% CI 1.18–1.83) (Figure 2). There was also an increased risk of HIV acquisition in women reporting intravaginal cleaning with soap and water (pooled adjusted HR 1.24, 95% CI 1.01–1.53) (Figure 3) and insertion of products to dry or tighten the vagina (pooled adjusted HR 1.31, 95% CI 1.00–1.71) (Figure 4). One study (study 2) did not ask about insertion practices. The use of household cleaning products other than soap was much less common; in four studies the HR could not be estimated (studies 1 Uganda, 4, 5, 8; Figure 5). In the remaining studies the pooled analyses showed no evidence of an increased risk of HIV acquisition (pooled adjusted HR 1.11, 95% CI 0.73–1.68). There was little evidence of between-study heterogeneity in any of these analyses (I² values 0%–14%). Results from the fixed effects models were the same or very similar to those from random effects models (Table S2).

There was no statistical evidence that intravaginal use of water alone increased the risk of HIV acquisition when compared with no practice in univariable (Table S3, pooled unadjusted HR 1.02, 95% CI 0.78–1.35, I² 3.9%) or multivariable analyses (pooled adjusted HR 1.03, 95% CI 0.76–1.40, I² 16.1%). Women who used only water intravaginally, compared with those using no practice, were younger (p<0.001), more likely to be currently married (p<0.001), more likely to have one partner in the last 3 mo (p<0.001), more likely to be using contraception (p<0.001), more likely to be Catholic or Muslim than Protestant (p<0.001), and less likely to never use condoms (p = 0.005). Controlling for these variables did not alter the strength of the associations found in univariable analyses (unpublished results).

Table 3 shows the results of analyses controlling for BV in addition to demographic and behavioural characteristics. These analyses exclude most participants from study 7, in whom vaginal flora was only assessed in a subset of women from one study site. Intravaginal use of cloth or paper remained associated with HIV acquisition after controlling for BV. For cleaning with soap and insertion of products to dry or tighten the vagina, the associations with HIV acquisition were weakened slightly after adjustment for demographic and behavioural variables. Adjustment for the presence of BV at baseline or at the last visit before the estimated date of HIV infection did not have any additional effect. When vaginal flora status was considered with intermediate vaginal flora and BV as separate categories, the pattern of results was similar to that seen when BV was included as a binary variable, but confidence intervals were wider because studies that used only Amsel criteria for diagnosis were excluded (unpublished results).

Associations between Intravaginal Practices and Disrupted Vaginal Flora

Table 4 shows the associations between intravaginal practices and the development of intermediate vaginal flora or BV amongst women with normal vaginal flora at the baseline visit and vaginal specimens examined by Gram-stain criteria at follow-up. Amongst women who cleaned intravaginally with soap and water, the incidence of disrupted vaginal flora at the next visit was increased in univariable and multivariable analyses (pooled adjusted OR from ordered logistic regression 1.24, 95% CI 1.04–1.47). There was a similar but weaker trend for the insertion of products to dry or tighten the vagina, but confidence intervals for these estimates were wider (pooled adjusted OR 1.29, 95% CI 0.98–1.71, p = 0.072). There was no evidence of an association between intravaginal cleaning with household products or use of cloth or paper and development of disrupted vaginal flora. There was no evidence of between-study heterogeneity (I² values 0.0% for all analyses).

Associations between Disrupted Vaginal Flora and HIV Infection

Vaginal flora status was associated with HIV incidence in univariable and multivariable analyses (Table 5). The risk of HIV acquisition was higher in women with BV than with intermediate vaginal flora. Controlling for potential confounders did not substantially attenuate the effect estimates. The strength of association between vaginal flora status and HIV acquisition was slightly weaker when using the vaginal flora status measured at the visit preceding the estimated date of HIV infection (median 51 d, IQR 40–147 d between vaginal flora assessment and HIV infection) than when using the baseline value (median 253 d, IQR 116–426 d).

Discussion

This study combined individual participant data from ten prospective studies in six sub-Saharan African countries. Intravaginal use of cloth or paper remained associated with HIV acquisition after controlling for age, marital status, number of sex partners in the past 3 mo, and in models that controlled for BV. Insertion of products to dry or tighten the vagina and intravaginal cleaning with soap were associated with HIV acquisition in univariable and multivariable analyses controlling for demographic and behavioural variables, but not in models that controlled for BV. Intravaginal cleaning with soap was also associated with the development of intermediate vaginal flora and BV at follow-up in women with normal vaginal flora at baseline. Disrupted vaginal flora measured at baseline or at the visit before the estimated data of HIV infection was associated with HIV acquisition in both univariable and multivariable analyses.

Strengths and Weaknesses

The main strength of this study was the collaboration of investigators from ten different studies, which allowed the collation of individual participant data from nearly 15,000 women and analysis using consistent definitions of intravaginal practices across studies. By pooling data we increased the power and precision of our analyses and adjusted for confounding, which is difficult or impossible in an aggregate data meta-analysis [14]. There was a striking lack of between-study heterogeneity in results, despite differences in study populations, designs, and questionnaires, which increases the robustness of our findings. In addition, we obtained very similar results using different statistical methods to pool data. Our results might be biased because we did not identify or include all eligible studies. We did, however, conduct a wide-ranging search and reasons for exclusion were not related to study results.

Limitations of this study were mainly due to data collection differences that could not be remedied by recoding. Questions about intravaginal practices were asked in different ways because there are no agreed upon definitions [5], there is no validated measurement instrument, and the purposes of the studies differed. We therefore had to limit the number of specific intravaginal practices from those originally planned and differences in wording of questions about intravaginal practices between studies might have affected our results. By grouping exposure categories, we might have masked harms (or benefits) of particular practices or products. Grouping of categories to obtain uniformly defined variables for confounding factors, or imprecision in the measurement of other variables included in the analysis might also have resulted in residual confounding. In addition, we cannot exclude the possibility of residual confounding from unmeasured factors, such as sexually transmitted infections, from our reliance on baseline measures of intravaginal practices that changed over time, or the motivation for performing certain practices, which can vary according to perceptions of risk [5].

A further limitation was the difficulty in definitively establishing the temporal sequence of intravaginal practices and changes in vaginal flora status; intravaginal practices could promote disruption of vaginal flora but symptoms related to those changes, such as vaginal discharge or fishy odour, could prompt intravaginal washing or wiping [24]. We tried to overcome this problem when examining the association between intravaginal practices and disrupted vaginal flora by including only women with normal vaginal flora at baseline. We could not, however, consider changes in exposure status over time because of data collection differences and uncertainty about the effects of treatment for BV, which was documented in some studies but not in others.

Comparison with Other Studies

This study is likely to be the largest to have examined associations between intravaginal practices and HIV acquisition and it allowed us to reexamine some previously observed inconsistencies between published studies. For example, McClelland and colleagues found a strong association between intravaginal cleaning with soap and incident HIV infection in Kenya [8] but van de Wijgert and colleagues and Myer and colleagues found no associations in their studies in Uganda and Zimbabwe [44] and South Africa [20]. In this analysis, estimates from the same studies were close to the published data; when pooled with the other studies, using the same definitions and multivariable model, the results were consistent with the overall finding of a modest increase in the risk of HIV acquisition (Figures 1–4). Differences in definitions of intravaginal practices and methods of analysis in individual studies make it difficult to compare findings directly and to synthesise results across studies. These differences meant that in our systematic review of aggregated published data we were limited to examining associations with intravaginal cleaning and insertion of products and could not draw conclusions about any specific practice [12], illustrating the advantages of individual participant data meta-analysis.