Bacterial vaginosis (BV) is the most common vaginal infection in women of reproductive age. Its prevalence varies from 9% to 50% 1 2 and may reach 70% among female sex workers (FSW) 3 . Besides its high frequency, BV is associated with many adverse health outcomes 4 5 6 7 8 9 10 including pelvic inflammatory disease, unfavourable pregnancy outcomes and recently HIV. BV is not only associated with female acquisition of HIV 9 10 but also with female-to-male transmission of HIV, as reported in a recent study 7 . Though it is increasingly clear that BV results from the replacement of the lactobacillus dominated normal flora by a predominantly anaerobic flora, no single causal agent has yet been identified. As a result, current treatment strategy aims at restoring the balance of the vaginal flora without specifically targeting any single causal agent. This may explain the high rate of treatment failure and recurrence that constitutes a major challenge for the clinical management and control of BV. Some authors 11 12 13 14 have suggested periodic presumptive treatment (PPT) as response to this challenge; however, the high recurrence rate makes the cost-effectiveness of this strategy questionable 15 . Knowing predictors of BV recurrence may help identify subgroups in whom PPT may be more efficient and prevent BV recurrences and their subsequent adverse health outcomes.

The purpose of this study was to identify behavioural and medical predictors of BV recurrence among women who experienced a prior BV episode during their follow-up at two African and two Indian sites of a randomized controlled microbicide trial.

Among these 440 HIV seronegative FSW with a BV episode at baseline, we found that 253 (57.5%) have had at least one BV recurrence during their follow-up. The incidence rate was 20.8 recurrences/100 person-months (95% CI=18.1 – 23.4) or 2.53 recurrences/person-year (95% CI=2.26 – 2.77).

Most prior studies on RBV looked only at the first recurrence, in the context of treatment regimen evaluation 20 21 22 23 . Therefore, their primary outcome was the proportion of treated subjects experiencing at least one BV recurrence in a given post-treatment length of time. This makes it difficult to compare the RBV frequency in the present study to those previously reported.

The two studies which reported BV incidence as multiple events were sub-studies of the same trial that evaluated the effect of PPT on the incidence of vaginal infections 12 . The first study, which was a secondary analysis focusing on the placebo group only, found a BV incidence rate of 361/100 person-years over the trial period 24 . The second study evaluated the post-trial effect of the PPT (first 120 days post-trial) and found RBV incidence of 260/100 person-years (95% CI=199-340) and 358/100 person-years (95% CI=286-448) for the intervention and placebo groups, respectively 25 . The RBV incidence of 253/100 person-years in the present study is somewhat lower than the three others, though only significantly for the two results reported from the placebo group (whether in- or post-trial). This discrepancy may be due to the fact that, unlike in the current study where all microscopically diagnosed BV were treated, (regardless of the presence of symptoms), only symptomatic BV were treated in the PPT and post-PPT studies. These two studies may have thus reported several times some persistently asymptomatic and thus untreated BV. Consequently we may rather be comparing incidence of “visits with BV diagnosis” to that of RBV.

In the multivariate analysis controlling for study site, recent vaginal cleansing, as reported at study entry, was a risk factor for RBV, whereas CCU with the primary partner and vaginal candidiasis were negatively associated with it.

While several previous cross-sectional or prospective studies found that intra-vaginal cleansing increased the risk of single BV, very few dealt with recurrent BV 24 26 27 . Schwebke et al. reported that vaginal douching increased the risk of RBV 26 , and the first sub-study on the PPT trial data found that the risk of RBV increased with vaginal washing frequency (p-value for trend=0.04) 24 . These findings are consistent with our results, though we did not collect data on vaginal washing frequency. However, some studies found no association between vaginal douching and BV 21 27 . In fact, although vaginal washing is a common practice, the frequency, techniques, products used and circumstances vary from one individual to another, and from culture to culture. This might explain the inconsistencies in studies results.

Compared to women who reported inconsistent condom use with their primary partners in the previous seven days, those who reported CCU were significantly at lower RBV risk, and interestingly their aHR was similar to that of women with no primary partner or no sexual act with the latter. These results support the idea that consistent and proper use of condoms may be as effective as sexual abstinence in protecting against BV recurrence. The non-significance of the association between CCU with other sexual partners (clients) and RBV may be due to a lack of statistical power resulting from the high percentage of condom use (97.6%) reported for this subgroup of sexual partners. Several studies have shown that CCU can reduce BV recurrence 24 26 28 29 . However, Yotebieng et al. found that condom use was protective against BV occurrence but not BV recurrence 30 . Unlike with a single BV episode, we did not find any association between STI or reproductive tract infections (except candidiasis) and RBV, and neither did Klatt et al. and Myer et al. 27 31 . However, in some studies, trichomoniasis was associated with RBV 24 32 .

Vaginal candidiasis (VC) during follow-up (but not at baseline) was inversely associated with RBV, consistent with the results of McClelland et al. 24 . The common explanation is that the two conditions are incompatible as they are inversely related to vaginal pH. Another study found that VC often preceded a BV episode though they rarely coexisted 33 . The reason of such a chronology remains unknown.

Finally, the study site exhibited a consistently strong association with BV recurrence, independently of all other factors, suggesting that RBV may be determined by a more complex network of site-specific unmeasured and probably interrelated factors. A striking finding in the current study is the overwhelming effect of the study site on the associations between investigated risk factors and RBV. With the exception of recent vaginal cleansing as self-reported at baseline, CCU with the primary partner and vaginal candidiasis (as time-dependent variables), all the associations which were initially significant lost their significance when controlled for study site. This suggests strong confounding by site due to a highly variable distribution of socio-demographic and medical variables across sites. Nevertheless, the univariate association between anal or oral sex (receiving penis in the mouth) and RBV deserves some attention.

The role of oral or anal sex in BV occurrence is supported by several studies 32 34 35 36 37 . In a recent prospective study 35 , women with incident BV were more likely to have previous colonization of anus or oral cavity with BV-associated bacteria. In another study 34 , the risk of periodontal disease was increased among women with BV (adjusted risk ratio =1.23; 95% CI: 1.08-1.40). In the same study, the risk for periodontal disease was 1.28 times (95% CI: 0.97-1.69) greater for receptive oral sex (ROS) with an uncircumcised partner, compared with ROS with a circumcised partner 34 . In another study, having vaginal sex after receptive anal intercourse was linked with acquisition of BV 37 .

The primary limitation of the current study resides in assuming for some BV episodes (in the absence of test of cure) that BV treatment, as administered per local guidelines, became effective seven days from the onset of the treatment. Some persistent BV may have thus been taken as RBV and this would have resulted in an overestimation of the RBV incidence. However the setting of the cure timeline of 7 days was based on the cure (Nugent score <7) rate of 80% to 90% generally reported in the literature for the same duration regarding the treatment regimens selected for our analysis 38 . Because of the relatively long periodicity of BV testing (3 months) we may have missed some BV episodes. We were not able to assess the effect of vaginal cleansing as reported during follow-up, because the proportion of women reporting the practice drastically declined over follow-up. This decrease resulted from the consistent counselling provided to women against vaginal cleansing at monthly visits as part of the instructions for the use of the study product. However, since data on vaginal cleansing were self-reported, those collected during follow-up, subsequently to several counselling sessions might be more subject to the effect of social desirability than those collected at study entry. Therefore, the former may be less appropriate than the latter in assessing the effect of vaginal cleansing on RBV. Also, some established BV risk factors, such as HSV-2 infection, smoking and alcohol consumption could not be included in the analysis because no data were collected about them in the trial. Finally, the current analysis did not cover biological predictors of RBV such as the presence and/or concentration of some specific BV related micro-organisms.

Nonetheless, this study presents several strengths. The primary one is the use of Andersen-Gill proportional risk modeling which allows for repeated events. To our best knowledge, the present study is the third analysis (after the first two by McClelland et al.) and the largest one that has used this approach to investigate predictors of RBV. Also, the consideration of BV clinical features to work out the at-risk period, instead of using the whole follow-up period, prevents underestimation of the RBV incidence. Like with other infectious diseases, the subject is not at risk for another BV episode until the previous resolves and this should be considered when calculating the person-time at risk (which was not done in previous studies). Other strengths include the large sample size, the use of the current gold standard for BV diagnosis (Nugent score) and the high level of the quality assurance system conferred by the clinical trial setting.

In summary, from this longitudinal analysis of data from a randomized clinical trial, we report a relatively high rate of RBV, though lower than in some previous studies 24 25 , among FSWs HIV seronegative at baseline. Some common risk factors for a single BV episode were associated with RBV while others were not. Predictors of RBV were primarily behavioural, particularly unprotected sex with primary sex partner and recent intravaginal cleansing. It is thus important to counsel high-risk women with RBV about the adverse effects of vaginal cleansing and the protective effects of condom use with all types of partners.

The study site exhibited a consistently strong association with RBV suggesting that research on risk factors of this condition may need to be as well approached as a socio-cultural environmental issue. Finally more prospective studies, specifically designed to identify risk factors of RBV are still warranted.

None of the authors has any conflict of interest to declare.

All authors were involved in the parent multicenter microbicide clinical trial that generated the data. For the present manuscript, Fernand A. Guédou (FAG) and Michel Alary (MA) conducted the statistical analyses and interpreted results. FAG wrote the first draft of the manuscript and MA revised it before further revision by other co-authors. In addition, Jennifer Deese (JD) and Marissa Becker (MB) edited the text. All authors revised and approved the present version of the manuscript.