Association between bacterial vaginosis and fecundability in Kenyan women planning pregnancies: a prospective preconception cohort study

Abstract

STUDY QUESTION

Is bacterial vaginosis (BV) associated with fecundability?

SUMMARY ANSWER

Women with BV may be at increased risk for sub-fecundity.

WHAT IS KNOWN ALREADY

While BV has been associated with poor IVF outcomes, the association between vaginal microbiota disruption and non-medically assisted conception has not been thoroughly explored.

STUDY DESIGN, SIZE, DURATION

Kenyan women with fertility intent were enrolled in prospective cohort that included monthly preconception visits with vaginal fluid specimen collection and pregnancy testing. Four hundred fifty-eight women attempting pregnancy for ≤3 menstrual cycles at enrollment were eligible for this fecundability analysis.

PARTICIPANTS/MATERIALS, SETTING, METHODS

At monthly preconception visits, participants reported the first day of last menstrual period and sexual behavior, underwent pregnancy testing and provided vaginal specimens. Discrete time proportional probabilities models were used to estimate fecundability ratios (FRs) and 95% CI in menstrual cycles with and without BV (Nugent score ≥ 7) at the visit prior to each pregnancy test. We also assessed the association between persistent BV (BV at two consecutive visits) and fecundability.

MAIN RESULTS AND THE ROLE OF CHANCE

Participants contributed 1376 menstrual cycles; 18.5% (n = 255) resulted in pregnancy. After adjusting for age, frequency of condomless sex and study site, BV at the visit prior to pregnancy testing was associated with a 17% lower fecundability (adjusted FR (aFR) 0.83, 95% CI 0.6–1.1). Persistent BV was associated with a 43% reduction in fecundability compared to cycles characterized by optimal vaginal health (aFR 0.57, 95% CI 0.4–0.8).

LIMITATIONS, REASONS FOR CAUTION

Detection of vaginal microbiota disruption using Gram stain and a point-of-care test for elevated sialidase identified a non-optimal vaginal environment, but these non-specific methods may miss important relationships that could be identified by characterizing individual vaginal bacteria and bacterial communities using molecular methods. In addition, results may be subject to residual confounding by condomless sex as this was reported for the prior month rather than for the fertile window during each cycle.

WIDER IMPLICATIONS OF THE FINDINGS

Given the high global prevalence of BV and infertility, an association between BV and reduced fecundability could have important implications for a large number of women who wish to conceive. Multi-omics approaches to studying the vaginal microbiota may provide key insights into this association and identify potential targets for intervention.

STUDY FUNDING/COMPETING INTEREST(S)

This work was supported by a National Institutes of Health grant (NICHD R01 HD087346-R.S.M.). R.S.M. received additional support for mentoring (NICHD K24 HD88229). E.M.L. was supported by pre- and post-doctoral fellowships (NIAID T32 AI07140, NICHD F32 HD100202). Data collection and management were made possible using REDCap electronic data capture tools hosted at the University of Washington’s Institute of Translational Health Science supported by grants from NCATS/NIH (UL1 TR002319). The content of this paper is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. R.S.M. receives research funding, paid to the University of Washington, from Hologic Corporation, and has received honoraria for consulting from Lupin Pharmaceuticals. L.E.M. receives research funding, paid to the University of Washington, from Hologic Corporation, and has received honoraria for service on scientific advisory boards from Hologic and Nabriva Therapeutics.

TRIAL REGISTRATION NUMBER

N/A.

Introduction

Bacterial vaginosis (BV) is characterized by replacement of optimal Lactobacillus-dominated vaginal microbiota with diverse anaerobic and facultative bacteria (Brotman, 2011). This condition has been associated with increased risk of sexually transmitted infections (STIs) and pelvic inflammatory disease (PID), which in turn, are associated with ectopic pregnancy and tubal factor infertility (Taylor et al., 2013; Tsevat et al., 2017). During pregnancy, BV and vaginal microbiota disruption are associated with spontaneous abortion and preterm birth (Leitich and Kiss, 2007; Fettweis et al., 2019; Al-Memar et al., 2020). The effects of BV on fecundity are not well understood outside the context of medically assisted reproduction.

Several studies have found that infertile women have a higher BV prevalence compared to fertile women (Mania-Pramanik et al., 2009; Dhont et al., 2010; Haahr et al., 2019b). A meta-analysis of pregnancy outcomes associated with BV among women undergoing IVF found that while BV was not associated with lower incidence of biochemical pregnancy, clinical pregnancy or live birth, it was associated with early spontaneous abortion (Haahr et al., 2019b). In studies using molecular methods to characterize the vaginal microbiota in women seeking assisted reproduction, higher vaginal bacterial diversity was associated with decreased clinical pregnancy and live birth, but not biochemical pregnancy (Hyman et al., 2012; Amato et al., 2019; Haahr et al., 2019a). There are conflicting results regarding the association between Lactobacillus-dominated endometrial microbiota and IVF outcomes (Franasiak et al., 2016; Moreno et al., 2016).

Few prospective studies have examined the association between vaginal microbiota disruption and female fertility in non-medically assisted reproduction (Wiesenfeld et al., 2012; Haggerty et al., 2016). None have assessed fecundability, the per-menstrual cycle probability of pregnancy. The objective of this study was to assess the associations between BV and elevated sialidase, two measures of vaginal microbiota disruption, and fecundability in Kenyan women trying to conceive.

Materials and methods

Study design and population

Participants for this prospective fecundability analysis were in the Microbiota and Preterm Birth (MPTB) Study, which enrolled non-pregnant women with immediate fertility intent in Nairobi and Mombasa, Kenya (Lokken et al., 2020). Inclusion criteria included: planning to conceive, HIV-negative, ≤45 years old and having a menstrual period in the prior 3 months or recent discontinuation of the implant, hormonal intrauterine device (IUD) or depo medroxyprogesterone acetate (DMPA). Exclusion criteria included: pregnant at enrollment, contraception at enrollment other than condoms for HIV/STI prevention, DMPA injection in prior 3 months, history of cervical/uterine surgery (excluding cesarean section), autoimmune disease, antibiotic use in prior 4 weeks or history of infertility care-seeking. The institutional review boards of Kenyatta National Hospital and the University of Washington approved the study. All participants provided written informed consent.

Participants who contributed ≥1 menstrual cycle between 18 April 2017 and 18 March 2020 were eligible for this analysis. Women reporting a history of hospitalization for PID, ectopic pregnancy, polycystic ovarian syndrome or endometriosis were excluded to minimize inclusion of women with potential sub-fertility requiring medical intervention. Participants reporting >3 menstrual cycles of pre-enrollment conception attempt time were excluded to reduce bias associated with unobserved cycle time (Weinberg et al., 1994; Schisterman et al., 2013).

Enrollment and monthly preconception visits

At enrollment, participants completed a structured interview on demographics, socio-behavioral characteristics, and medical history. Participants underwent a pelvic examination with vaginal fluid collection for Neisseria gonorrhoeae, Chlamydia trachomatis and Trichomonas vaginalis detection by nucleic acid amplification tests, vaginal Gram stain and elevated sialidase detection. If participants were menstruating, examination was deferred.

At monthly preconception visits, participants completed a urine pregnancy test and interview reporting condomless sex frequency and other behaviors for the prior month. Women self-collected vaginal swabs for Gram stain and sialidase detection. Most women who remained non-pregnant after 6 months exited the study; those discontinuing DMPA within 6 months of enrollment were eligible for 9 months of follow-up due to delayed return to fertility (Yland et al., 2020).

Women received counseling on healthy preconception behaviors and an estimate of their next fertile window using calendar methods. Participants with genital symptoms were treated using syndromic management (National AIDS & STI Control Programme of Kenya, 2018). Treatment for laboratory-detected STIs was provided at the first monthly preconception visit.

Enrollment samples were tested for N. gonorrhoeae, C. trachomatis and T. vaginalis (Aptima Combo-2 CT/NG Detection System, Aptima T. vaginalis assay; Hologic Corporation; San Diego, CA, USA). Vaginal Gram stained slides from each visit were evaluated for BV using the criteria of Nugent and Hillier (Nugent et al., 1991). Detection of elevated vaginal sialidase concentrations was performed using a point-of-care diagnostic test for BV (≥0.25 µg; Diagnosit BVBlue; Gryphus Diagnostics; Knoxville, TN, USA).

Defining time-at-risk and outcome

At each visit, women reported the first day of their last menstrual period (LMP). Women who missed preconception visits were asked to report interim menstrual cycles; 130 participants were unable to report interim menstrual cycles for visits missed before July 2018 when this question was added. For these cases, cycles were derived based on menstrual cycle duration reported at enrollment. For women reporting irregular cycles, a cycle length of 28 days was used. Three percent of cycles (41/1376; 19 for irregular cyclers, 22 for regular cyclers) were derived and 6.3% (87/1376) were reported after missed visits.

Discrete menstrual cycles were modeled utilizing the LMPs reported by participants and the derived cycles (Supplementary Fig. S1). Only complete menstrual cycles with known outcomes and first incident pregnancies were included. Participants were considered pregnant in a cycle if the urine pregnancy test was positive. They were considered not pregnant in a cycle if their pregnancy test was negative and they reported a new LMP at their next visit, demonstrating they did not become pregnant later in the cycle.

Statistical analysis

Cumulative incidence of pregnancy was estimated using Kaplan–Meier methods. Discrete time proportional probabilities models with robust standard errors were used to estimate fecundability ratios (FRs) and 95% CI (Weinberg et al., 1989, 1994). The FR is interpreted as the per-cycle probability of pregnancy comparing exposed to unexposed cycles (i.e. BV vs no BV). To account for left truncation, participants reporting conception attempt time prior to enrollment were delayed entry into the analysis by the number of reported menstrual cycles of prior trying time (Weinberg et al., 1994). Censoring criteria included initiation of biomedical infertility treatment, resumption of contraception, participant withdrawal, loss to follow-up and completing preconception follow-up without pregnancy.

Two primary time-varying measures of vaginal microbiota disruption were assessed: BV (Nugent score ≥ 7) and elevated sialidase by the point-of-care test. BV served as an overall indicator of vaginal microbiota disruption. Sialidase is a mucin-degrading enzyme produced by some BV-associated bacteria (Wiggins et al., 2001), which may be one mechanism through which disrupted vaginal microbiota impacts fecundability. In primary analyses, BV and elevated sialidase status were lagged from the visit prior to each pregnancy test to align the vaginal microbiota assessment with the preconception period for each cycle. For cycles reported following a missed visit and for derived cycles, missing exposure and time-varying confounder data were imputed using data from the last visit carried forward and then lagged.

Age (<25, 25–29, 30–34, 35–39, 40–45) and frequency of condomless sex in the prior month (time-varying; none, 1–4, 5–8, ≥9) were included a priori in adjusted models due to associations with both vaginal microbiota disruption and fecundity (Gallo et al., 2011; Wesselink et al., 2017). Study site was also included a priori. Additional potential confounders were selected based on potential associations with both vaginal microbiota and fecundability. These included body mass index (Lokken et al., 2019), vaginal washing in last month (Baird et al., 1996; Ness et al., 2004), any condom use in last month, maternal education and household income. Potential confounding factors were evaluated for inclusion in the multivariable model using a manual forward stepwise model-building approach. None of the potential confounding factors changed the FR estimate by >10%, so none were retained in the final adjusted models.

A sensitivity analysis was conducted after excluding women with potential underlying subfertility based on more restrictive exclusion criteria. These included N. gonorrhoeae, C. trachomatis, T. vaginalis or PID diagnosis at enrollment; history of any PID diagnosis; history of treatment for N. gonorrhoeae, C. trachomatis, T. vaginalis or syphilis; self-report of fibroids or unknown uterine abnormality; DMPA use within prior 6 months (Yland et al., 2020), and having an HIV-seropositive partner (Iyer et al., 2019). Two sensitivity analyses were conducted examining the effects of deriving menstrual cycles and missing exposure and confounder data on the effect estimates by (i) excluding these cycles, and (ii) excluding participants who enrolled prior to the addition of the interim missed LMP question in the parent study. An additional analysis was conducted to evaluate effect modification by recent contraceptive method (Yland et al., 2020).

Pre-defined secondary analyses were conducted to further assess the association between vaginal microbiota and fecundability. Models were run utilizing vaginal microbiota status measured at enrollment only, as fecundability studies more commonly use time-independent measures. In addition, to assess stable vs dynamic vaginal microbiota, a variable was created combining the measurement from the visit prior to and the measurement at each menstrual cycle’s ultimate pregnancy test. This method resulted in four categories: negative at both visits (optimal vaginal health); positive at prior visit but negative at current visit; negative at prior visit but positive at current visit; and positive at both visits (persistent disrupted). Women conceiving in the first menstrual cycle were eligible for this analysis as they contributed an enrollment sample and a sample at their first monthly preconception visit. To further assess persistence, a sensitivity analysis was conducted by generating a variable combining the vaginal microbiota status from the two visits prior to each pregnancy test (i.e. a double-lag).

Results

Baseline and follow-up characteristics

Of the 701 women enrolled in the MPTB Study, 458 were eligible for this fecundability analysis (Supplementary Fig. S1 and Table SI). Included participants were a median of 29 years old (interquartile range (IQR) 25–34, Table I). Most had been pregnant before (93.9%, n = 430). One participant reported smoking (0.2%) and a minority drank alcohol (14.2%, n = 65). The most common methods of recent contraception were implant (38.2%, n = 175), copper IUD (27.7%, n = 127) and none (22.5%, n = 103). Five percent (n = 22) of participants reported that their male partner was living with HIV. C. trachomatis was detected in 34 (7.5%) women, three had N. gonorrhoeae (0.7%) and four had T. vaginalis (0.9%). BV (35.8%, n = 164) and elevated sialidase (35.2%, n = 161) were detected in one third of participants. BV and elevated sialidase prevalence were similar between eligible and ineligible participants (Supplementary Table SI). Compared to participants from Mombasa, participants in Nairobi were older (median 30 vs 27), more likely to have been pregnant before (96.8% vs 82.0%), and recently used an IUD (32.8% vs 6.7%). Participants from Nairobi were less likely to report no recent contraception (19.0% vs 37.1%), prior conception attempt time (14.9% vs 50.5%) and vaginal washing (28.7% vs 62.9%) (Supplementary Table SII).

Table I.

Enrollment characteristics for 458 Kenyan women planning pregnancies.

Characteristic	N	All eligible
		Total (n = 458)
Demographic and partnership characteristics
Age	458
<25		87 (19.0)
25–29		147 (32.1)
30–34		121 (26.4)
35–39		84 (18.3)
40–45		19 (4.2)
Education level	458
<8 years		29 (6.3)
8–11 years		137 (29.9)
12–15 years		207 (45.2)
≥16 years		85 (18.6)
Monthly household income (KSh)	455
<2500		13 (2.9)
25 000–10 000		136 (29.9)
10 000–30 000		197 (43.3)
30 000–75 000		70 (15.4)
>75 000		39 (8.6)
Partner’s age	454
<25		14 (3.1)
25–29		85 (18.6)
30–34		134 (29.3)
35–39		114 (24.9)
40–44		83 (18.1)
≥45		28 (6.1)
Partner’s HIV-serostatus	456
HIV-seronegative		345 (75.7)
HIV-seropositive		22 (4.8)
Unknown		89 (19.5)
Substances
Smoke cigarettes	458	1 (0.2)
Frequency of alcohol use	458
None		393 (85.8)
Monthly or less		52 (11.4)
2–4 times per month		10 (2.2)
2–3 times per week		2 (0.4)
≥4 times per week		1 (0.2)
Contraception
Most recent contraceptive methodi	458
None		103 (22.5)
Condoms		24 (5.2)
OCP		9 (2.0)
Injectable (DMPA)		18 (3.9)
Copper IUD		127 (27.7)
Implant		175 (38.2)
Other		2 (0.4)
Any DMPA injection in last 6 months	458	20 (4.4)
Reproductive history
Ever pregnant	458	430 (93.9)
Parous	458	413 (90.2)
Number of menstrual cycles of prior conception attempt timeii	458
0		367 (80.1)
1		59 (13.0)
2		19 (4.2)
3		13 (2.8)
Regular menses	458	245 (53.5)
History of PID (not treated in a hospital)	458	1 (0.2)
Abnormal uterusiii
Fibroids	457	3 (0.7)
Other/unknown	457	3 (0.7)
Sexual risk behavior in last 4 weeks
Any vaginal washing	458	162 (35.4)
Frequency of condomless sex	457
No condomless sexiv		43 (9.4)
1–4		152 (33.3)
5–8		115 (25.2)
≥9		147 (32.2)
Any condom use	456	23 (5.0)
Clinical
BMI	455
Underweight		8 (1.8)
Normal		171 (37.6)
Overweight		164 (36.0)
Obese		112 (24.6)
History of STIv	458	6 (1.3)
Vaginal dischargeiii	458	46 (10.0)
BV (Nugent ≥7)	458	164 (35.8)
Elevated sialidase	458	161 (35.2)
N. gonorrhoeae	454	3 (0.7)
C. trachomatis	454	34 (7.5)
T. vaginalis	453	4 (0.9)
Metronidazole prescription	458	4 (0.9)

BMI, body mass index; BV, bacterial vaginosis; DMPA, depo medroxyprogesterone acetate; IUD, intrauterine device; KSh, Kenyan shillings; OCP, oral contraceptive pills; PID, pelvic inflammatory disease; STI, sexually transmitted infection.

ⁱReported dates of IUD and implant removal and last date of DMPA were reviewed. Women reporting device removal >2 months prior to enrollment were re-classified as non-contraceptors (none) for the purpose of this analysis. Women reporting a last DMPA injection >6 months prior were reclassified as non-contraceptors.

ⁱⁱParticipants reporting >3 cycles were excluded from this fecundability analysis.

ⁱⁱⁱSelf-reported.

^ivOf the 43 women reporting no unprotected sex in the prior 4 weeks, 22 (51.2%) reported no vaginal intercourse and 21 (48.8%) reported 100% condom use with vaginal intercourse in the prior 4 weeks. Nine (20.9%) of these women had HIV-positive male partners. The most recent method of contraception reported by these participants included no contraception (n = 13, 30.2%), condoms only (n = 11, 25.6%), DMPA (n = 2, 4.7%), copper IUD (n = 8, 18.6%) and implant (n = 9, 20.9%). Only 5 (11.6%) reported actively trying to conceive prior to study enrollment.

^vSelf-report of syphilis, chlamydia, gonorrhea and/or trichomoniasis.

Participants attended 1266 preconception visits with a median of 29 (IQR 28–35) days between visits. Participants contributed 1376 menstrual cycles (n = 128 missing cycles derived or reported by participants). Enrollment characteristics did not differ between participants with and without these cycles (Supplementary Table SIII). There were 255 pregnancies (18.5% of cycles). The median time-to-pregnancy was four menstrual cycles (IQR 2–7) and the cumulative six-cycle pregnancy rate was 70.4% (95% CI 65.1–75.4; Supplementary Fig. S2).

By the end of the study period, 55.7% (n = 255) of participants were pregnant, 7.6% (n = 35) remained in preconception follow-up, 15.9% (n = 73) did not conceive during preconception follow-up, 3.7% (n = 17) had withdrawn and 17.0% (n = 78) were lost to follow-up (Supplementary Fig. S1). Compared to those who remained in the study, women who withdrew or were lost to follow-up were less likely to have been pregnant before, less likely to be parous, and less likely to have recently used contraception (Supplementary Table SIV). Three participants reported initiating biomedical infertility treatment.

Bacterial vaginosis and fecundability

Thirty-one percent (527/1724) of study visits were BV-positive (Nugent ≥7), though abnormal vaginal discharge (7.8%, 41/526 BV-positive visits) and BV treatment were rare (2.9%, 15/527 BV-positive visits). In unadjusted analysis, BV at the visit prior was associated with a 16% lower fecundability (FR 0.84, 95% CI 0.65–1.07) (Table II). Adjusting for age, frequency of condomless sex and study site did not change the association (adjusted FR (aFR) 0.83, 95% CI 0.64–1.07). These results were attenuated in sensitivity analysis excluding 88 participants who did not meet the more restrictive inclusion criteria (aFR 0.93, 95% CI 0.70–1.24) but were robust to sensitivity analyses excluding derived and reported missed menstrual cycles and excluding participants enrolled prior to addition of the interim missed LMP question (Supplementary Table SV). There was no association between BV and fecundability in a secondary analysis assessing BV at enrollment only (aFR 1.06, 95% CI 0.83–1.34; Table II), and no effect modification by most recent contraceptive method (P-value for interaction: P = 0.5; Supplementary Table SV).

Table II.

Bacterial vaginosis and fecundability.

Periconceptual exposure		Menstrual cycles exposed (n %)	Pregnancies (n %)	Unadjusted FR (95% CI)	Adjusted FR (95% CI)i
Primary analysis		n = 1376	n = 255
BV at the visit prior to pregnancy testii		460 (33.4)	76 (29.8)	0.84 (0.65, 1.07)	0.83 (0.64, 1.07)
Secondary analyses
BV at enrollment		477 (34.7)	93 (36.5)	1.07 (0.86, 1.35)	1.06 (0.83, 1.34)
Persistent BV: visit before and at pregnancy test
Prior visit	Current visit	n = 1376	n = 255
No	No	848 (61.6)	167 (65.5)	Ref	Ref
Yes	No	110 (8.0)	36 (14.2)	1.63 (1.17, 2.26)	1.55 (1.12, 2.14)
No	Yes	68 (4.9)	12 (4.7)	0.91 (0.52, 1.57)	0.89 (0.52, 1.55)
Yes	Yes	350 (25.4)	40 (15.7)	0.58 (0.42, 0.80)	0.57 (0.41, 0.79)
Persistent BV: two visits prioriii
Two visits prior	One visit prior	n = 679	n = 174
No	No	393 (57.9)	104 (59.8)	Ref	Ref
Yes	No	67 (9.9)	24 (13.8)	1.18 (0.86.1.62)	1.14 (0.82, 1.59)
No	Yes	54 (8.0)	12 (6.9)	0.82 (0.50, 1.33)	0.78 (0.48, 1.25)
Yes	Yes	165 (24.3)	34 (19.5)	0.81 (0.58,1.12)	0.81 (0.59, 1.13)

BV, bacterial vaginosis; FR, fecundability ratio.

ⁱAdjusted for age, study site and frequency of condomless sex in last 4 weeks.

ⁱⁱLagged, time-varying exposure.

ⁱⁱⁱUnadjusted and adjusted fecundability models in the analyses assessing the association between BV at the two visits prior to each pregnancy test were run using the Poisson family due to convergence issues with the adjusted models and reduced sample size.

When BV status was re-evaluated utilizing measurements across two visits including the visit prior to and the measurement at each menstrual cycle’s pregnancy test, 61.6% of cycles were BV-negative at both visits (n = 848), 25.4% (n = 350) had persistent BV and the remaining 12.9% (n = 178) were BV-positive at one but not the other visit (Table II). Compared to menstrual cycles with optimal vaginal health, persistent BV was associated with a 43% lower fecundability (aFR 0.57, 95% CI 0.41–0.79) (Table II). In comparison, BV at the visit prior to and no BV at the visit with pregnancy testing was associated with a 1.6-fold higher fecundability (aFR 1.55, 95% CI 1.12–2.14). Results were similar in sensitivity analyses (Supplementary Table SVI). When categorizing participants based on BV status at two preceding visits (double-lagged), there were 679 menstrual cycles and 174 pregnancies for analysis (Table II). Compared to cycles preceded by optimal microbiota, both persistent BV (aFR 0.81, 95% CI 0.59–1.13) and BV at the most recent visit only (aFR 0.78, 95% CI 0.48–1.25) were associated with reduced fecundability, though confidence intervals are wider. Participants with BV two visits prior but not at the visit prior, had a slightly elevated fecundability (aFR 1.14, 95% CI 0.82–1.59) (Table II).

Elevated sialidase and fecundability

Concordance between BV by Gram stain and elevated sialidase test results was 85.9% (1469/1710, kappa = 0.67). Of 241 discordant samples, 48.5% (117/241) were BV-negative but sialidase positive and 51.5% (124/241) were BV-positive but sialidase negative.

There were no associations between elevated sialidase at the visit prior (aFR 0.99, 95% CI 0.77–1.27) or at enrollment (aFR 1.03, 95% CI 0.81–1.31) and fecundability (Table III). When compared to menstrual cycles that were sialidase test negative at the visit prior to and at the current pregnancy test, persistent sialidase test positivity was associated with a 33% reduction in fecundability (aFR 0.67, 95% CI 0.48–0.94). Cycles that were sialidase test positive at the visit prior and negative at the current visit had a 1.8-fold increased per-cycle probability of pregnancy (aFR 1.76, 95% CI 1.30–2.38) (Table III). In sensitivity analysis categorizing status using the measures from two prior visits, these associations were attenuated and close to null (Table III).

Table III.

Elevated sialidase and fecundability.

Periconceptual exposure		Menstrual cycles exposed (n %)	Pregnancies (n %)	Unadjusted FR (95% CI)	Adjusted FR (95% CI)i
Primary analysis		n = 1376	n = 255
Elevated sialidase at the visit prior to pregnancy test ii		448 (32.6)	82 (32.2)	0.99 (0.77, 1.26)	0.99 (0.77, 1.27)
Secondary analyses
Elevated sialidase at enrollment		463 (33.7)	87 (34.1)	1.02 (0.81, 1.29)	1.03 (0.81, 1.31)
Persistent elevated sialidase: visit before and at pregnancy test
Prior visit	Current visit	n = 1373	n = 252
No	No	831 (60.5)	151 (59.9)	Ref	Ref
Yes	No	130 (9.5)	43 (17.1)	1.82 (1.34, 2.46)	1.76 (1.30, 2.38)
No	Yes	95 (6.9)	20 (7.9)	1.15 (0.75, 1.76)	1.13 (0.74, 1.72)
Yes	Yes	317 (23.1)	38 (15.1)	0.66 (0.47, 0.92)	0.67 (0.48, 0.94)
Persistent elevated sialidase: two visits prior iii
Two visits prior	One visit prior	n = 679	n = 174
No	No	385 (56.7)	98 (56.3)	Ref	Ref
Yes	No	77 (11.3)	25 (14.4)	1.09 (0.78, 1.51)	1.09 (0.78, 1.51)
No	Yes	70 (10.3)	18 (10.3)	1.00 (0.66, 1.51)	0.97 (0.65, 1.45)
Yes	Yes	147 (21.7)	33 (19.0)	0.90 (0.64, 1.28)	0.93 (0.66, 1.31)

FR, fecundability ratio.

ⁱAdjusted for age, study site and frequency of condomless sex in last 4 weeks.

ⁱⁱLagged, time-varying exposure.

ⁱⁱⁱUnadjusted and adjusted fecundability models in the analyses assessing the association between elevated sialidase at the two visits prior to each pregnancy test were run using the Poisson family due to convergence issues with the adjusted models and reduced sample size.

Discussion

This study is the first to assess whether vaginal microbiota disruption is associated with fecundability among women trying to conceive without medical intervention. BV at the visit prior to each pregnancy test was associated with a 17% reduction in fecundability. Persistent BV was associated with a 19–43% lower per-cycle probability of pregnancy compared to cycles with optimal vaginal health depending on how persistent BV was defined. There was no association between elevated sialidase at the visit prior to each pregnancy test and fecundability, but persistent detection of elevated sialidase was associated with reduced fecundability.

When interpreting these results, it is important to consider that frequency of condomless sex is associated with both BV and with becoming pregnant (Gallo et al., 2011). Without accurate ascertainment and adjustment for frequency of condomless sex, this confounder would introduce a bias toward finding an association between BV and increased fecundability. This is the opposite of the hypothesis under study.

When defining persistent BV as BV at both the visit prior to and at the visit with pregnancy testing, there was a 43% lower fecundability in cycles with persistent BV. In contrast, there was a 55% higher fecundability in cycles characterized by a transition from BV-positive to BV-negative, which is likely a result of reverse causality due to collection of the second sample during early pregnancy (for those who became pregnant). During pregnancy the vaginal microbiota shifts toward a Lactobacillus-dominated community (Romero et al., 2014; DiGiulio et al., 2015; MacIntyre et al., 2015; Stout et al., 2017; Serrano et al., 2019). The strength of the analysis including BV at the visit before and the visit concurrent with each pregnancy test is that participants with optimal microbiota or persistent BV across two points likely had similar vaginal microbiota at the time of conception.

Persistent disruption of the vaginal microbiota could impact fecundability through multiple mechanisms. First, BV and BV-associated bacteria have been associated with PID and tubal factor infertility (Taylor et al., 2013; Haggerty et al., 2016, 2020). Second, ascension of BV-associated bacteria into the upper reproductive tract and persistent sub-clinical inflammation may upset the carefully modulated fetal-maternal immune interaction leading to implantation failure and pre-clinical pregnancy loss (D’Ippolito et al., 2018; Moreno and Simon, 2018). Lastly, sialidase and other mucin-degrading enzymes produced by bacteria associated with BV can degrade cervical mucus, disrupting its functions (Wiggins et al., 2001). Healthy cervical mucus supports conception by reducing bacterial ascension to the uterus and supporting selection, movement and capacitation of sperm (Wiggins et al., 2001; Chrétien, 2003).

In this Kenyan preconception cohort, the six-cycle pregnancy rate was 70%, which is within the 58–81% range reported for prospective fecundability cohorts in the USA and Europe (Bonde et al., 1998; Gnoth et al., 2003; Wildenschild et al., 2014; Wesselink et al., 2017). Retrospective studies in South Africa and Ethiopia have reported slightly lower 6-month pregnancy rates of 50% and 66%, respectively (Bello et al., 2010; Kassa and Kebede, 2018) In Nigeria, the median time-to-pregnancy was estimated at 5.1 months using Demographic and Health Survey data and the current duration approach to estimate fecundability (Polis et al., 2017).

This fecundability analysis has several strengths. The prospective design is the gold standard for assessing fecundability (Buck Louis and Platt, 2011). In addition, monthly specimen sampling allowed for time-varying evaluation of the association between vaginal microbiota and fecundability, which is important given the temporal variation in Nugent score (Brotman et al., 2010; Thoma et al., 2011). The participant population was an additional strength; the first cycle at-risk for pregnancy was observed for a majority of participants. Participants also reported low rates of smoking and alcohol use, so these results are unlikely to be confounded by these behaviors. Lastly, most cases of BV detected by Nugent’s criteria were asymptomatic, so few received syndromic treatment with metronidazole. Therefore, it is unlikely that observed associations between BV and fecundability were due to treatment.

There were also a number of limitations. First, participants could not report interim menstrual cycles if they missed a visit in the study’s first year, so missed menstrual cycles were derived based on self-reported menstrual cycle regularity during this period. A 28-day cycle was assumed for irregular cyclers, which may have overestimated cycles to pregnancy; however, only 1.4% of cycles were affected reducing concern for bias. In addition, results were similar in sensitivity analyses excluding these visits. Second, condomless sex was reported for the prior month rather than for the fertile window during each cycle, so these analyses are subject to residual confounding by condomless sex frequency. Third, excluding women without a menstrual cycle in the prior 3 months (in the absence of hormonal contraception) may have induced selection bias by excluding women with long or irregular menstrual cycles. However, only 3% (26/940) of women screened for the study met this criterion, reducing the likelihood of enrolling a more fecund cohort. Fourth, about 20% of participants were lost to follow-up, which may bias study results if vaginal microbiota disruption or fecundability were associated with remaining in the study. While most enrollment characteristics were similar between women lost to follow-up and those remaining in the study, women who did not complete follow-up may have been less fertile as they were less likely to ever have been pregnant and more likely to report no recent contraceptive method. Fifth, detection of vaginal microbiota disruption using Gram stain and a point-of-care sialidase test are non-specific methods of detection that may miss important relationships that could be identified using molecular methods to characterize individual bacteria and bacterial communities. Lastly, this study did not characterize the endometrial microbiota, which may be mechanistically important (Mitchell et al., 2015; Chen et al., 2017).

This prospective study of Kenyan women attempting to conceive found an association between BV and reduced fecundability. Studies using multi-omics approaches to interrogate the vaginal and endometrial microbiota may provide additional insights into this association and could identify potential targets for intervention. Given the high global prevalence of BV and infertility (Mascarenhas et al., 2012; Peebles et al., 2019), these findings could have important implications for a large number of women.

Data availability

This study was conducted with approval from the Kenyatta National Hospital—University of Nairobi Ethics and Research Committee (KNH-UON ERC), which requires that we release data from Kenyan studies (including de-identified data) only after they have provided their written approval for additional analyses. As such, data for this study will be available from the authors upon request, with written approval for the proposed analysis from the KNH/UON ERC. Their application forms and guidelines can be accessed at http://erc.uonbi.ac. To request these data, please contact KRTC Administrator at kenyares@uw.edu.