Prevalence of bacterial vaginosis in postmenopausal women: a systematic review and meta-analysis

Keywords: bacterial vaginosis (BV), menopause, postmenopausal, prevalence, sexually transmissible, symptoms, vaginal discharge, vaginosis, women.

Bacterial vaginosis (BV) is the most common cause of abnormal vaginal discharge in women of reproductive age. Characterised by a disturbance of vaginal microbiota, there is a marked depletion of Lactobacillus species and an increase in diversity of organisms including Gardnerella vaginalis, Atopobium vaginae, and genital mycoplasmas.^1–6 The symptoms can include a thin homogeneous vaginal discharge and fishy odour and are accompanied by an increase in vaginal pH. Symptoms can be very distressing for some women, and recurrent BV can have a significant effect on a woman’s self-esteem and intimate relations, substantially affecting quality of life.⁷ However, a significant number of cases are asymptomatic.⁸

BV has been associated with poor obstetric and gynaecological sequelae in premenopausal women including preterm birth,^9–15 spontaneous abortion,^10,15 postpartum endometritis,¹⁶ and pelvic inflammatory disease (PID).¹⁶ There is considerable observational evidence that BV is sexually transmissible;^8,17 therefore, sexually active postmenopausal women continue to be at risk of acquiring BV. Furthermore, women diagnosed with BV are at an increased risk of acquiring sexually transmissible infections (STIs) including HIV.^8,18 Recurrence is common after treatment.¹⁹

BV is commonly diagnosed by one of two gold standard diagnostic methods: the Nugent score,²⁰ or Amsel criteria.²¹ The Nugent score is based on counting bacterial morphotypes identified under a microscope. Amsel criteria are based on the presence of at least three of the four following signs and symptoms: a thin homogenous discharge, raised pH, detections of amines, and presence of Clue cells seen in a wet preparation viewed under a microscope. These two methods were developed in women of reproductive age and there is ongoing debate about whether these are appropriate for diagnosing BV in postmenopausal women where the absence of oestrogen effect the vaginal epithelium and microbiome.²²

The global burden of BV in women of reproductive age is high. General population prevalence estimates range from 23 to 29% globally,²³ with a higher burden in women who have sex with women,^24–27 women of African ethnicity^28–33 and women who engage in intravaginal practices (e.g. douching or insertion into the vagina of items or products for hygiene, enhancement of sex or treatment of symptoms).^8,34 In contrast, there is very little prevalence data and no real understanding of the effect of BV in postmenopausal women. Women continue to be sexually active after menopause and are at risk of acquiring BV, so it is important to understand the burden of BV among these women. We undertook this systematic review and meta-analysis to determine the prevalence of BV in postmenopausal women.

The PRISMA statement was used to guide the reporting of results in this review and meta-analysis.³⁵ Our review protocol is registered on the International Prospective Register of Systematic Reviews (PROSPERO), registration no.: CRD42019146598 (https://www.crd.york.ac.uk/PROSPERO/).

The electronic databases PubMed, EMBASE, Web of Science, and The Cochrane Library were searched for English-language studies that provided BV prevalence data, published before 20 July 2020. The search terms included (prevalence OR survey OR proportion) AND ‘bacterial vaginosis’ (Supplementary Table S1). Citation lists were examined for additional references.

Studies were eligible for inclusion if they included BV prevalence data for postmenopausal women and the study was published in English. Women were defined as postmenopausal if they had not had a menstrual period for ≥12 months or were aged >50 years if menopausal status was not specified. Studies had to report on the prevalence of BV using an established, diagnostic method such as the Nugent score,²⁰ Amsel criteria,²¹ or Hay/Ison criteria.³⁶ Studies reporting on BV exclusively in women of reproductive age were excluded. Letters to the editor, reviews, editorial and discussion articles, conference abstracts without subsequent publication, or studies where the target population (postmenopausal women) sample size was <10 were excluded. Studies were reviewed by two authors (LLS and JC). Differences in opinion were resolved by consensus with other authors (JSH, LAV, CSB).

Authors were contacted if it was unclear whether the study included BV prevalence data for postmenopausal women. Given the timeframe of the search and that email addresses of authors were not available for several older publications, an a priori decision was made to only contact authors of papers published since 2009, focusing on the most recent decade of publications.

In addition, if the paper reported prevalence data on a sample of women across a broad age range, authors were only contacted if the study included either >100 total participants across all ages or if the sample size of postmenopausal women was ≥10. Fourteen authors were contacted, six responded and three were able to provide additional data eligible for inclusion (see Supplementary Table S3).

The data extracted included: (1) study details (author and year of publication reported); (2) country of study; (3) study population and setting (e.g. clinic or hospital or general population); (4) ethnicity; (5) study design (cross-sectional, cohort, case–control, randomised controlled trial); (6) age range of participants; (7) total sample size across all ages; (8) number of postmenopausal women; (9) number of postmenopausal women who tested positive for BV; (10) diagnostic method used; (11) BV prevalence in postmenopausal women; and (12) reason for visit. For the two community-based cohort studies and the randomised controlled trial, baseline data were extracted. For the population-based cohort study, the most recent data were used. One author (LLS) extracted the data, and a second author (JC) checked the extracted data. Discrepancies were resolved by consultation with a third author (JSH).

The primary outcome was BV prevalence among postmenopausal women. This was measured as a proportion, with the numerator being the number of women diagnosed with BV and the denominator being the number of women tested for BV.

Meta-analysis was used to calculate summary estimates of BV prevalence; random effects models were used. Where studies did not report 95% confidence intervals (CIs), these were calculated using available data. The I² test was used to calculate the proportion of total variability in prevalence estimates that could be attributed to underlying study heterogeneity alone and not simply due to chance. An I² value was considered moderate if >50% and high if >75%.³⁷ Subgroup analyses were undertaken to investigate factors associated with heterogeneity. Subgroup analyses included assessing BV prevalence by diagnostic method (Nugent score or Amsel criteria) and population type (classified as gynaecology populations, general population-based or screening populations [e.g. cervical cancer or employment health screening] or HIV-positive population). To investigate the effect of individual studies on the overall summary estimate, a sensitivity analysis was undertaken in which the effect of removing each study on the summary estimate was explored. Data were analysed using Stata 16 (Version 16; StataCorp) applying the Metaprop syntax using the Freeman–Tukey double arcsine transformation.

We assessed within-study bias using a combination of the following: evaluation criteria suggested by Sanderson et al.³⁸ and AXIS, the critical appraisal tool for cross-sectional studies developed by Downes et al.³⁹ We assessed the following parameters: (1) selection bias: setting and source population, the reporting of inclusion/exclusion criteria, sample representation of the source population/the use of consecutive sampling, reporting of response fraction, missing data; (2) measurement bias: appropriate diagnostic method, experience with diagnostic method; and (3) statistical bias: sample size and reporting of sample size calculations. Publication bias, assessed by funnel plot analysis, was not performed because of the low number of eligible studies.

Overall, 5693 articles were identified, of which 2461 were unique studies. A total of 328 full-text articles were assessed for eligibility, with 13 studies eligible for inclusion in the review (see Fig. 1). Table 1 shows that of 13 eligible studies, eight were cross-sectional studies,^22,40–46 three were cohort studies,^47–49 one was a nested case–control study⁵⁰ and one was a randomised controlled trial.⁵¹ Three studies were from the USA,^47–49 three were from Italy,^22,45,46 and one study from each of the UK,⁴¹ Kenya,⁴⁰ South Africa,⁵⁰ Brazil,⁵¹ Greece,⁴⁴ China⁴³ and India.⁴² The populations included were ethnically diverse, including Afro-Caribbean,⁴¹ Asian⁴⁶, Caucasian,^22,41,51 Central West African⁴⁶, East European⁴⁶, Han Chinese,⁴³ Hellenic,⁴⁴ Hispanic or Latina,^47,49 Indian subcontinent,⁴¹ Italian,⁴⁶ Non-Hispanic African American,⁴⁹ Non-Hispanic Black,⁴⁷ Non-Hispanic White,^47,49 North African,⁴⁶ South American⁴⁶ and West European.⁴⁶ Four studies recruited participants from gynaecological clinics or inpatient gynaecological wards,^41,44,45,51 two were from community screening programs (cervical screening, employment screening),^22,43 three were from HIV-positive populations,^40,42,50 one was a population cohort,⁴⁸ two were community-based cohorts^47,49 and one used archived laboratory data.⁴⁶ Seven studies used the Nugent score as the diagnostic method,^{22,40,42,47,48,50,51} four used Amsel criteria,^41,43,46,49 one required both the Nugent score and Amsel criteria to be positive to make a diagnosis of BV,⁴⁴ one used modified Amsel criteria,⁴⁵ and none used Hay/Ison criteria. Sample sizes of postmenopausal women varied from 14 to 1732 with an overall total of 4092 women.

Fig. 1.  Flowchart of study selection.

Table 1.  Included studies, their characteristics and BV prevalence.

Prevalence estimates ranged from 2.0 to 57.1%. The overall summary prevalence estimate for BV among postmenopausal women was 16.93% (95% CI 8.45–27.4%; I² = 97.9%; P < 0.01) with marked heterogeneity (Fig. 2).

Fig. 2.  Forest plot of BV prevalence. ES, effect size (prevalence). Studies ordered by year of publication.

Subgroup analyses found that BV prevalence was higher in studies that used the Nugent score to diagnose BV (22.90%; 95% CI 8.64–41.18; I² = 98.15; P < 0.01) than in those that used Amsel criteria (7.92%; 95% CI 4.05–12.76; I² = 74.64; P < 0.01). BV prevalence was highest among HIV-positive women (25.06%; 95% CI 7.49–48.27; I² = 94.6; P < 0.01) (Table 2).

Table 2.  Subgroup analyses – BV prevalence.

The sensitivity analysis found that removing each study had little effect on the summary estimates, with the overall summary BV prevalence estimates for the remaining studies ranging from 14.50 to 18.64% with marked heterogeneity for each estimate (I² > 93%).

Risk of bias is summarised in Table 3 and Supplementary Table S2. Overall, the included studies were at very high risk of selection bias with all but one⁴⁸ set in specialist clinic settings or high-risk populations (e.g. HIV-positive patients). Even the one population-based study had oversampled women of African American or Hispanic ethnicity.^48,52 Only two (15.4%) studies reported consecutive recruitment of patients^22,41 and none reported the retention of all data. Inclusion and exclusion criteria were described in four (30.8%) studies^22,43,45,51 and a further four (30.8%) studies^42,48–50 reported inclusion criteria alone. Overall, measurement bias was considered low because all studies used gold standard diagnostic methods. However, the risk of measurement bias was considered moderate in one study, which required that the Nugent score and Amsel criteria were both positive.⁴⁴ Only three studies focused on postmenopausal women,^45,48,51 with all other studies including adult women of all ages, reducing the available sample size for our target population.^{22,40–44,46,47,49,50} None of the studies reported any sample size calculations.

Table 3.  Summary assessment of study quality and bias.

To the best of our knowledge, this is the first systematic review and meta-analysis of BV prevalence specifically among postmenopausal women. Our review included a total of 4092 women and found an overall summary BV prevalence of 16.93%. However, the included studies were at considerable risk of bias; only one population-based study was identified,⁴⁸ and our pooled estimate had marked heterogeneity, highlighting the scarcity of robust BV prevalence data for postmenopausal women.

There are several limitations present in this review. First, by including studies with women who were aged >50 years and not just those who were postmenopausal, our review will have included some women who are still premenopausal or perimenopausal. However, this highlights the lack of studies that focus specifically on postmenopausal women. Second, by limiting our search to studies published in English, some important studies may have been omitted. Third, there were several studies that we were unable to extract data specifically for postmenopausal women, even after contact with authors, so some important data may have been excluded from our analysis. However, this highlights the need for further studies with a focus on postmenopausal women. Furthermore, our search only identified one population-based study and only three studies specifically focusing on postmenopausal women.^45,48,51 Finally, we chose to only include studies that used either the Nugent score,²⁰ Amsel criteria²¹ or Hay/Ison criteria³⁶ to diagnose BV.

Our subgroup analysis found that BV prevalence was higher in studies using the Nugent score than in studies that used Amsel criteria as the diagnostic method. It has been previously shown that the Amsel criteria has a lower sensitivity for BV than the Nugent score, and this may have contributed to the observed difference.^53–55 Furthermore, BV prevalence was highest among HIV-positive women. However, there are concerns about how well these diagnostic methods work after menopause^22,48 when the decline in oestrogen alters the vaginal epithelium and leads to a significant reduction or absence of lactobacilli.^22,56 Currently, there is no standardised recommendation for the diagnosis of BV in postmenopausal women. According to a recent study that compared the relative abundances of vaginal microbiota with Nugent score,⁵⁷ classic BV-associated organisms such as Gardnerella, Atopobium, Sneathia, and Megasphaera were strongly associated with an elevated Nugent score in premenopausal women, but in postmenopausal women, these same organisms were not. The authors concluded that a high Nugent score should not be used to infer a diagnosis of BV in postmenopausal women.⁵⁷ This highlights the challenges with understanding BV in postmenopausal women and may contribute to the heterogeneity in prevalence estimates.

The use of hormone replacement therapy (HRT) is relatively common among postmenopausal women to alleviate the symptoms of menopause.⁵⁶ Physiological changes in response to HRT can include significant increases in the relative abundance of Lactobacillus spp., a decrease in vaginal pH, as well as increased maturation of the vaginal epithelium.⁵⁸ However, it remains unclear how the use of HRT could affect the pathogenesis, symptoms, or diagnosis of BV in postmenopausal women when there are already ongoing concerns about how well diagnostic methods work after menopause^22,48 and, currently, there is no standardised recommendation for the diagnosis of BV in postmenopausal women. Only one study in our review compared the prevalence of BV in postmenopausal women who use HRT with those who did not, with the authors observing that the difference was not statistically significant.²²

BV is associated with poor reproductive and obstetric outcomes, significant gynaecological sequelae,^{9–14,16,41,59,60} and a greater risk of STI acquisition, including HIV^18,61–63 in women of reproductive age, but there is sparse discussion about its impact in postmenopausal women. However, women continue to be sexually active after menopause, often lacking knowledge of STIs and safer sexual practices,⁶⁴ and are less likely to use condoms.⁶⁵ A recent study found that STIs are increasing at a faster rate among older women than among younger women.⁶⁶ Given the considerable evidence suggesting that BV is sexually transmissible,^8,17 and the knowledge that women continue to be sexually active after menopause, postmenopausal women continue to be at risk of acquiring BV, which can increase their risk of other STIs including HIV.^{8,18,32,67–70}

This review highlights that there are few quality studies on BV prevalence in postmenopausal women. Most studies were subject to considerable selection bias and lacked sample size calculations. Even the single population-based study had some selection bias with women of African American or Hispanic ethnicity being oversampled.^48,52 A recent systematic review and meta-analysis by Peebles et al.,²³ of global BV prevalence and costs among women of reproductive age in the general population, found similar marked heterogeneity. Nevertheless, our review suggests that BV may be highly prevalent among postmenopausal women.

BV is a condition we take seriously in younger women, but know very little about in postmenopausal women. Our review highlights how under-researched and poorly understood this important area of public health is. Quality research is needed to further our understanding of BV in postmenopausal women and understand its effect on their lives.

Supplementary material is available online.

Data sharing is not applicable as no new data were generated or analysed during this study.

Jane Hocking is a Joint Editor of Sexual Health, but was blinded from the peer-review process for this paper.

JSH is supported by a National Health and Medical Research Council (NHMRC) Senior Research Fellowship (1136117) and CSB is supported by a NHMRC Investigator Grant (1173361). No other funds were used for this work.

Designed the search strategy: LLS, JSH, CSB and LAV. Analysed the data: LLS, JC and JSH. Wrote the paper: LLS and JSH. Contributed to drafting/review of final manuscript: all authors.