BACTERIOLOGY

crossm

Prognostic Indicators of Recurrence of Bacterial Vaginosis

Jack D. Sobel,a Navkiranjot Kaur,b Nicole A. Woznicki,a Dina Boikov,a Tina Aguin,c Gurveer Gill,b Robert A. Akinsb

a Division of Infectious DiseasesWayne State University School of Medicine, Detroit, Michigan, USA
b Department of Microbiology, Immunology, and Biochemistry, Wayne State University School of Medicine, Detroit, Michigan, USA
c
Department of Obstetrics and Gynecology, Wayne State University School of Medicine, Detroit, Michigan, USA

Downloaded from http://jcm.asm.org/ on July 29, 2020 by guest

ABSTRACT Following all forms of therapy for bacterial vaginosis (BV), recurrence
rates are extremely high. Many diagnostic tests are available that differentiate bacte-
rial vaginosis from other types of vaginal disorders, but none predict recurrence af-
ter treatment, nor are any vetted for monitoring ongoing responses to treatment.
Our goal was to determine which tests, and at what optimal times, have prognostic
value in predicting recurrence. This prospective cohort study monitored 74 highly
recurrent BV patients for up to 9 months. Symptomatic BV patients were treated
with oral metronidazole and were evaluated at cessation of treatment and monthly.
Index tests included Amsel, Nugent, BV Blue, and Affirm VPIII, as well as a quantita-
tive PCR (qPCR)-based test under initial evaluation here. The qPCR-based LbRC (Lac-
tobacillus Relative Composition) assay predicted BV recurrence when performed
shortly after oral metronidazole treatment, with both 90% positive predictive values
(PPV) and 74% negative predictive values (NPV); the Nugent scores had 93% PPV
but poor NPV (57%). No test, at any other visit, was prognostic. The LbRC assay and,
to a lesser extent, Nugent tests scored a week after oral metronidiazole predicted re-
currence, suggesting that the recurrence in this cohort was predominantly by re-
lapse due to incomplete restoration of eubiosis soon after therapy. This is the first
study in an under evaluated population of recurrent BV patients that emphasizes
the need for and a pathway to a possible prognostic modality. Given the high recur-
rence rates of BV, prognostic tests that could influence individualized treatment al-
ternatives are urgently needed.

KEYWORDS Affirm VPIII, Amsel criteria, BV Blue, Lactobacillus, Nugent score,

bacterial vaginosis, metronidazole

O ral metronidazole is the standard of care for treatment of bacterial vaginosis (BV)
(1) because it has broad spectrum activity against anaerobic microbes (2–5),
minimal impact on lactobacilli and facultative anaerobes (1, 6), and rarely induces
Citation Sobel JD, Kaur N, Woznicki NA, Boikov
D, Aguin T, Gill G, Akins RA. 2019. Prognostic
indicators of recurrence of bacterial vaginosis. J
acquired resistance (4). However, all metronidazole regimens result in high recurrence Clin Microbiol 57:e00227-19. https://doi.org/10
.1128/JCM.00227-19.
rates of BV, 69 to 80% within 12 months, including patients with relapse or reinfection
Editor Erik Munson, Marquette University
(7, 8). Recurrence of BV is complex and may result from a variety of mechanisms that
Copyright © 2019 American Society for
differ among individuals (1, 8–13) and has prompted numerous alternatives to metro- Microbiology. All Rights Reserved.
nidazole (14–18). Predicting recurrence in individual patients with acute BV, especially Address correspondence to Jack D. Sobel,
patients with a history of recurrence, has not been addressed and has important [email protected], or Robert A. Akins,
[email protected].
implications for customizing acute and long-term maintenance therapies to break the
Received 14 February 2019
cycle of ongoing recurrence (15). Returned for modification 2 March 2019
In the present study, we demonstrate that a novel quantitative PCR (qPCR)-based Accepted 5 March 2019

test, along with LbRC (Lactobacillus Relative Composition) and Nugent scores, obtained Accepted manuscript posted online 6
March 2019
immediately after oral metronidazole therapy predict whether patients in remission Published 26 April 2019
from recurrent BV will sustain that remission or recur later.

May 2019 Volume 57 Issue 5 e00227-19 Journal of Clinical Microbiology jcm.asm.org 1

Sobel et al. Journal of Clinical Microbiology

MATERIALS AND METHODS

Patient enrollment. This single crossover prospective trial was performed at a Vaginitis Clinic at
Wayne State University in Detroit, MI, from 30 September 2014 to 30 May 2017. Women were enrolled
as recurrent BV patients if their patient histories indicated three or more episodes of symptomatic BV in
the previous year and had acute symptomatic BV at enrollment. Patients were categorized as BV patients
if they were both positive for ⱖ3 Amsel criteria (vaginal pH ⱖ 4.5, positive amine “Whiff” test, ⬎20% clue
cells, and grayish-white adherent discharge) (19) and symptomatic with odor, discharge, discomfort, or
itching.
A patient was included if she was premenopausal and ⱖ18 years old; had no mixed infections of
vaginal Candida, Trichomonas, herpes simplex virus, or cervicitis; was willing to refrain from using any
other vaginal products during the study period; and was heterosexual and agreed either to use condoms
for the duration of the study or to report unprotected sex and to abstain from coitus within 48 h of any
study visit and to abstain from alcohol during therapy. She was retained in the study if at any point she
became culture positive for Candida; if she had a history of vaginal candidiasis she was treated according
to the standard of care as defined by Centers for Disease Control and Prevention guidelines. She was
excluded from the study if she was pregnant, nursing or at risk of pregnancy; if she was on anticoag-

Downloaded from http://jcm.asm.org/ on July 29, 2020 by guest

ulation, lithium, or disulfiram therapies; if she was allergic to metronidazole; if she had used any vaginal
antibiotics or antifungals in the previous 10 days; if she would not agree to abstain from vaginal
douching or probiotics during the enrolled period; if she required treatment for an abnormal Pap smear
or genital cancer; or if she had vaginal bleeding at time of enrollment. Other forms of contraceptives
were not exclusionary. Consent and HIPAA forms were explained to each patient, who agreed and signed
these forms as a condition for enrollment. The protocol and forms were approved by the Wayne State
University Institutional Review Board (IRB 040314M1F). BV patients at enrollment were given the
standard of care regimen of oral metronidazole 500 mg twice daily for 7 days.
Reference and index tests. Four vaginal swabs were taken to determine index tests, including
Amsel (19) and Nugent (20) scores, two commercial tests, and our qPCR-based test, the LbRC assay. BD
Affirm VPIII (Becton, Dickinson, and Co., NJ; performed by the Detroit Medical Center Clinical Laboratory)
generates a positive hybridization signal if there are ⱖ200,000 cells per swab of G. vaginalis (21). pH was
visually scored using pH paper with a range 3.0 to 5.5 (Hydrion pH test strips). BV Blue (22, 23) (Gryphus
Diagnostics, LLC, TN) is an enzymatic assay for sialidase derived from BV-associated species in the swab;
the colorimetric assay if positive is blue or green. All tests other than BD Affirm VPIII were performed by
clinicians and study personnel.
There is no gold standard for the diagnosis of BV, so as reference standards, we alternatively used
Amsel and Nugent scores, and a third more rigorous reference we designate symptomatic BV (sBV),
defined as Amsel-positive patients with symptoms that warranted treatment. Clinical information and
Amsel scores were not available to the laboratory staff or readers of index tests; conversely, clinicians
scoring Amsel criteria had no access to the other test scores, nor did performers or readers of individual
tests have access to results of other tests at scoring. Patients returned 7 to 14 days following initiation
of oral metronidazole therapy for a follow-up exam to assess outcomes by Amsel and index tests and
again once a month to assess long-term outcomes. BD Affirm VPIII and BV Blue tests were repeated only
on follow-up, second visits. Patients returned upon request if they experienced symptoms consistent
with recurrence. Patients were monitored monthly for 9 months unless they chose to drop out or
acquired other exclusion criteria or because of symptomatic recurrence of BV. Recurrent patients were
offered either a high-dose vaginal metronidazole treatment or a maintenance regimen from the
literature (15), not included in this analysis. All patients also performed daily vaginal self-swabs according
to the provided instructions and recorded whether they engaged in sex, with or without a condom, and
whether they were in menses. The full protocol is posted online (https://docs.google.com/document/d/
1uUZvo0soD2rBCDtMBxsT1TmAJp72CD2E_E6FoWM9Ato/edit?usp⫽sharing).
Primary end points among BV patients were defined as follows. Bacteriological recurrence was
defined as a Nugent score of ⱖ7. Clinical recurrence, sBV, is defined as an Amsel score of ⱖ3, including
elevated pH, and symptoms presented by the patient, i.e., the need to treat. Remission was defined as
the absence of recurrence and considered long-term remission if this persisted ⱖ3 months after therapy.
Refractory was defined as the absence of remission immediately following therapy.
DNA extraction. Vaginal swabs for DNA extraction were submerged in 4 ml of 2-isopropanol 1n
15-ml conical tubes immediately after swabbing. These tubes were stored at room temperature; we
determined that cells were not viable and that DNA was not degraded under these conditions for
⬎30 days. Tubes were centrifuged 1,580 ⫻ g for 30 min, supernatants were discarded, and pellets were
drained 15 min. Pellets were resuspended in 300 ␮l of alkaline lysis buffer (100 mM KOH, 2% Triton X-114,
2 mM EDTA) and heated at 65°C for 2 h. Tubes were centrifuged again (1,580 ⫻ g for 10 min). Then,
130 ␮l of supernatant was distributed to pairs of wells in duplicate 96-well sterile polypropylene plates
containing 65 ␮l of neutralizing buffer (200 mM Tris [pH 8.3], 4 mM EDTA, 200 mM HCl). Plates were
sealed and stored at –20°C until used for qPCR. The quality of DNA from these preparations was sufficient
for qPCR or targeted next-generation sequencing.
qPCR methods. The LbRC assay estimates the relative concentrations of Lactobacillus species to the
total bacterial titer. It does this by measuring the bacterial titer with a broad-spectrum primer mix in the
absence versus presence of blocking oligomers LBB3 and LBB4, which specifically complement Lacto-
bacillus species and partially overlap or juxtapose the 3= end of the broad-spectrum primers, thereby
specifically inhibiting amplification of all Lactobacillus species by the broad-spectrum primers (24, 25).
The assay used in this study differs from the published assay only in that here we used an expanded set
of broad-spectrum primers to be more inclusive (see Table S1 in the supplemental material). ΔCq

May 2019 Volume 57 Issue 5 e00227-19 jcm.asm.org 2

Prognostic Indicators of Recurrent Bacterial Vaginosis Journal of Clinical Microbiology

TABLE 1 Patient characteristics at enrollment

Characteristic Data
Total no. of patients 90a

Mean (SD)
No. of visitsb 5.6 (2.8)
Days in study, rangeb 124 (90), 17–381
pH 5.7 (0.6)
African-American 5.8 (0.3)
Caucasian 5.6 (0.5)
Nugent score 8.6 (1.6)
African-American 8.7 (1.3)
Caucasian 8.4 (1.4)
Amsel score 3.9 (0.4
African-American 3.9 (0.4
Caucasian 3.9 (0.4)

Downloaded from http://jcm.asm.org/ on July 29, 2020 by guest

Age (yr), range 34 (7), 21–48
African-American 34 (7)
Caucasian 34 (4)

Race, no. (%)

African-American 81 (90)
White 7 (8)
Otherc 1 (1)
Not reporting 1 (1)

Ethnicity, no. (%)

Non-Hispanic 89 (99)
Hispanic 0 (0)
Not reporting 1 (1)
aIncludes16 who were lost to follow-up after enrollment.
bExcludesthose lost to follow-up.
cMore than one race.

(quantification cycle [Cq] of Lactobacillus-blocked sample ⫺ Cq of unblocked sample) measures relative

Lactobacillus content, given by the formula % Lactobacillus ⫽ (100 · ΔCq2)/(ΔCq2 ⫹ 1), so that a ΔCq score
of 0 indicates that a sample has no Lactobacillus, 1 indicates that the population was 50% Lactobacillus,
2 indicates 80% Lactobacillus, etc. (Fig. S1). Here, we incorporated a penalty into the ΔCq score if a sample
displayed non-Lactobacillus melt profiles to generate the LbRC/5 score, i.e., ΔCq/5. The qPCR conditions
are detailed in Table S1 in the supplemental material. Representative qPCR data that generate the LbRC/5
scores from acute, recurring, and remission patient samples illustrate the parameters that influence the
score (Fig. S2). Duplicate runs of 235 samples, in which the second run was separated by days to a month
from the first, showed excellent agreement of ΔCq and dominant and subdominant melting temperature
(Tm) values (Table S2).
Statistical methods. Statistical significance and other analyses followed the STARD guidelines (26),
as cited in text, and were performed using GraphPad Prism version 6.00 for Windows (GraphPad
Software, La Jolla CA). Categorical data for diagnostic accuracy were analyzed using the Fisher exact test
and associated tools to generate 95% confidence intervals. Comparisons between assays in which one
assay result was missing were excluded. Significance of comparisons of continuous data were analyzed
by t tests if the distributions passed the D’Agostino-Pearson omnibus normality test or by the Mann-
Whitney test if not (P ⱕ 0.05 was considered significant).

RESULTS
Patient profiles. The study enrolled 90 recurrent BV patients for a mean of 5.6
monthly visits, ranging from 2 to 11 visits, and excluded 16 from outcome classification
because they did not return after their initial visit (Table 1). BV patient mean scores at
enrollment were pH 5.7, Nugent 8.6, and Amsel 3.9. The majority of BV patients were
African-American (90%) compared to 8% Caucasian, but these groups were not signif-
icantly different in mean age, pH, and Nugent or Amsel scores at enrollment (Wilcoxon
matched-pair signed-rank tests). Patients lost to follow-up were not a biased subset of
the total cohort.
Response to treatment. After standard-of care-oral metronidazole (SOC) therapy
among 74 patients, 46 (62%) achieved at least short-term remission, and 28 (38%) were
refractory at the first follow-up visit. Among the 46 patients in initial remission, 28 (60%)
recurred at a subsequent visit, i.e., 38% of the total tracked cohort. Thus, overall

May 2019 Volume 57 Issue 5 e00227-19 jcm.asm.org 3

Sobel et al. Journal of Clinical Microbiology

FIG 1 Flow diagram of enrollment and responses to therapy. Only patients with at least three visits in
remission after a treatment were scored as remission, excluding patients who dropped from the study
before this while in remission. Recurrent, at least one posttreatment visit in remission before recurrence;

Downloaded from http://jcm.asm.org/ on July 29, 2020 by guest

refractory, symptomatic BV at first posttreatment visit.

response to metronidazole was poor; 56 of the 74 patients (76%) who were not lost to
follow-up recurred or had a refractory response (Fig. 1). The percentage of patients that
achieved long-term remission after SOC therapy (24%) was unexpectedly high, given
this cohort’s history of recurrence. The numbers of remission versus recurrent versus
refractory outcomes were not significantly different (P ⫽ 0.269), i.e., an individual
patient was approximately equally likely to respond in any of the three ways.
Diagnostic performance of BV tests. Although the purpose of the study was to
evaluate the prognostic abilities of diagnostic tests for BV, we first evaluated their
ability to correctly diagnose BV. We alternatively used Amsel, Nugent, or sBV as
reference standards, as well as each of these plus BV Blue, Affirm, pH alone, and the
LbRC scores as index tests.
Accuracies of index tests of all pooled samples relative to three reference standards
(Amsel, sBV, and Nugent) were compared (Fig. 2). For this analysis, Nugent scores 7 to
10 were categorized as positive, and Nugent scores 0 to 6 were categorized as negative.
The three reference tests were approximately equivalent for most index tests, whose
accuracy varied by ⬍5%. Amsel, pH, and Nugent scores were most accurate, ranging
from 84 to 94% accurate to their reference tests. LbRC/5 at an optimal threshold score

FIG 2 Accuracies of the index tests (x axis) relative to reference tests Amsel, sBV (symptomatic, Amsel
positive), and Nugent score. The Nugent score breakpoint as a reference test was defined as not BV if 0
to 6 and as BV if 7 to 10. Several breakpoints of Nugent scores were considered as index tests (#/#); the
first number is the upper limit considered not BV, and the second number is the lower limit considered
BV. LbRC/5 scoring is described in the text; threshold numbers following these are the maximum scores
defined as BV. wMean is the mean of all index scores per reference test, weighted by the number of
samples in each index test. Error bars indicat 95% CI. *, Significantly different from the optimal index
score per reference test; those using Amsel as the reference are compared to the Nugent 6/7 test, and
those using sBV or Nugent 6/7 as the reference tests are compared to the Amsel index tests. Comparisons
were made using the Fisher exact test (P ⬍ 0.05). Data for all patients at all visits were pooled for this
analysis.

May 2019 Volume 57 Issue 5 e00227-19 jcm.asm.org 4

Prognostic Indicators of Recurrent Bacterial Vaginosis Journal of Clinical Microbiology

Downloaded from http://jcm.asm.org/ on July 29, 2020 by guest

FIG 3 Distribution of immediate posttreatment visit scores grouped by patient outcome defined as
follows: F, refractory, i.e., Amsel positive and symptomatic requiring treatment, at the posttreatment visit;
R, recurrent if the patient was in remission posttreatment but became Amsel positive and needed
treatment at some later visit; M, remission if the patient did not become Amsel positive or need
treatment at any subsequent visit. Markers denote individual patients; means and standard deviations
are shown as bars. Comparisons that showed significant differences are indicated by asterisks. Tests were
evaluated by Kruskal-Wallis analysis, followed by Dunn’s multiple-comparison test.

of 1 ranged from 78 to 82% accurate, significantly lower than Amsel, pH, or Nugent
scores. BV Blue and Affirm VPIII tests were less accurate than the other standard tests
(Fig. 2).
Prognostic value of BV diagnostic tests at the posttreatment visit. Because
recurrence rates among BV patients are so high, we sought to determine whether any
of the diagnostic tests could predict subsequent recurrence of patients who initially
achieved remission following SOC therapy. For patient management, we defined
recurrence as either Amsel positive or Amsel positive plus symptoms requiring treat-
ment (sBV). Distributions of scores of individual patients at this visit (Fig. 3) showed that
only LbRC/5 scores differed significantly between patients in remission who would later
recur (R patients) versus those who would remain in remission (M patients). At this visit,
the LbRC/5 scores of patients in remission but destined to recur later were not
significantly different from the scores of patients who were refractory.
We adapted diagnostic statistics to this prognostic application; in this context a true
positive test means that a positive score of a remission patient at the posttreatment
visit (LbRC/5 ⬍ 5 or Nugent ⱖ 4) corresponds to recurrence at a later visit and,
conversely, that a true negative means that a negative score of a remission patient at
the posttreatment visit (LbRC/5 ⱖ 5 or Nugent ⬍ 4) corresponds to long-term remission
(remission for at least three subsequent visits or to the end of enrollment, whichever is
longer). It follows in this context that strong positive predictive values (PPV) mean a
positive score of a patient in remission at posttreatment strongly predicts later recur-
rence and that strong negative predictive values (NPV) strongly predict long-term
remission for a patient with a negative posttreatment score. Only the LbRC/5 and
Nugent tests had significantly different distributions of scores for predicting continued
remission versus later recurrence (P ⬍ 0.05; Table 2). LbRC/5 using an optimal break-

May 2019 Volume 57 Issue 5 e00227-19 jcm.asm.org 5

Sobel et al. Journal of Clinical Microbiology

TABLE 2 Prognostic performances of LbRC/5, Nugent, and pH at the posttreatment visita

LbRC/5 (5) Nugent (4)
Parameter sBV Amsel sBV Amsel pH (4.5)
P 0.0002 0.007 0.002 0.017 0.079

Odds ratio 23.8 6.0 17.3 5.3 0.3

95% CI 3.99–142 1.6–22 1.98–151 1.41–20 0.09–0.98

Sensitivity 0.77 0.67 0.52 0.57 0.24

95% CI 0.55–0.92 0.45–0.84 0.31–0.72 0.37–0.76 0.09–0.45

Specificity 0.88 0.75 0.94 0.80 0.48

95% CI 0.62–0.98 0.51–0.91 0.71–1.0 0.56–0.94 0.28–0.69

PPV 0.90 0.76 0.93 0.80 0.32

Downloaded from http://jcm.asm.org/ on July 29, 2020 by guest

95% CI 0.67–0.99 0.53–0.92 0.66–1.0 0.56–0.94 0.13–0.57

NPV 0.74 0.65 0.57 0.57 0.39

95% CI 0.49–0.91 0.43–0.84 0.37–0.76 0.37–0.76 0.22–0.58

LR 6.2 2.7 8.8 2.9 0.5

aAsymptomatic patients at the posttreatment visit were grouped by whether they remained in remission (17
patients) or recurred at some later visit (25 patients). Recurrence was defined either as sBV (Amsel positive
requiring treatment) or as Amsel positive. Tests were scored as positive or negative if they fell above or
below the breakpoints indicated in parentheses. Amsel, BV Blue, and Affirm VPIII tests were also analyzed,
but the results are not shown because the scores were distributed randomly with respect to outcome
(P ⬎ 0.19). 95% CI, 95% confidence interval; LR, likelihood ratio.

point of 5 had strong PPV (0.90) and useful NPV (0.74). Nugent scoring using an optimal
Nugent breakpoint of 3 to 4 had excellent PPV (0.93) but poor NPV (0.57). Similar
patterns were seen when only Amsel criteria were used to define later recurrence, but
tests had lower PPV relative to the sBV criteria. Breakpoints were optimized based on
distributions of scores in remission versus recurrent groups and on receiver operating
characteristic (ROC) analyses (Fig. 3 and Fig. S3). Vaginal pH generated nearly signifi-
cantly different distributions of remission versus recurrent outcomes (P ⫽ 0.079) but
had no predictive value.
Other than at the immediate posttreatment visit, no test was prognostic of clinical
outcome at any other visit, including the initial, pretreatment, symptomatic visit. Since
it is reasonable to suspect that vaginal bacterial compositions begin to shift in advance
of symptomatic BV, we specifically sought to determine whether scores from visits just
preceding recurrence visits were predictive of those recurrences. For 21 patients for
which data were available, the test scores for each patient at the visit preceding the
recurrence visit were compared to the mean scores of the same patient for all other
remission visits. Surprisingly, the test scores at the prerecurrence visits were not
significantly different from the mean of other visits in remission (P ⱖ 0.099, Table 3).
Collectively, these data indicate that at no specific point in time, other than
immediately after treatment, does an individual’s status, as defined by any of the

TABLE 3 Diagnostic test scores at the remission visits of recurrent and remission patientsa
Test score (SD) or P value
Category pH Amsel Nugent LbRC/5
Prerecurrent only 4.5 (0.5) 0.6 (1.2) 4.0 (3.0) 5.2 (5.9)
Not prerecurrent* 4.7 (0.6) 0.9 (1.3) 4.4 (2.6) 3.1 (3.5)
Prerecurrent vs not prerecurrent (P) 0.276 0.102 0.277 0.099
Remission† 4.4 (0.3) 0.3 (0.4) 2.0 (1.4) 8.1 (3.3)
Not prerecurrent vs remission (P) 0.251 0.564 0.002‡ ⬍0.0001‡
aResultsare presented as means (with standard deviations in parentheses) except where data are indicated
as P values. Comparisons were averaged across the visits of 21 recurrent patients when in remission,
excluding the prerecurrent visit (*), or across all remission visits of 17 remission patients (†). P values were
calculated by the Wilcoxon matched-pair signed-rank test, except where distributions were normal as noted
(‡), where they were calculated using unpaired t tests.

May 2019 Volume 57 Issue 5 e00227-19 jcm.asm.org 6

Prognostic Indicators of Recurrent Bacterial Vaginosis Journal of Clinical Microbiology

diagnostic tests, predict later recurrence. Despite the lack of prognostic value of visits
other than posttreatment, patients who eventually recur had higher mean Nugent
scores and lower mean LbRC/5 scores while in remission than did patients who
remained in remission (Table 3). This reflects more instability in these scores over time
in the recurrent group.

DISCUSSION
The problems faced by the clinician in managing BV patients are refractory re-
sponses and recurrence at high rates following SOC therapy. Conventional tests for
diagnosing BV are accurate, not excellent, both historically and in this study, but neither
these nor molecular tests have been vetted for their ability to predict which patients
will be refractory and which will respond initially but later recur. Having this informa-
tion when the patient first presents, or at the posttreatment visit, would enable the

Downloaded from http://jcm.asm.org/ on July 29, 2020 by guest

clinician to tailor alternative treatments as needed to reduce the probability of these
undesirable outcomes.
The most significant finding of this study was that the microbial composition of
bacterial vaginosis patients within 7 days of finishing conventional oral metronidazole
therapy was pivotal in whether they would remain in remission versus recur. This status
was best quantified by the qPCR-based assay, LbRC/5. Based on the threshold of
prognostic LbRC/5, asymptomatic patients in remission who host any detectable
species other than Lactobacillus, i.e., BV-associated species, immediately after treatment
have a 90% chance of recurring (95% confidence interval [CI] ⫽ 67 to 99%). Conversely,
patients in remission immediately after treatment for which Lactobacillus is ⬎96%
dominant (LbRC/5 ⱖ 5) have a 74% chance of remaining in remission (95% CI ⫽ 67 to
99%). A Nugent score of ⱖ4 at this early visit also predicts recurrence but has poor
negative predictive value; a patient with a Nugent score of 0 to 3 has only a 57% chance
of staying in remission. If these associations hold in subsequent larger studies, at least
among recurrent, predominantly African-American cohorts, they will warrant a trial to
apply extended or alternative treatments, including probiotics, to patients at posttreat-
ment visits in remission, based on poor LbRC/5 or Nugent scores.
We saw a lack of association of any diagnostic test score with clinical outcome at the
initial pretreatment visit, the visit just preceding recurrence, or at any posttreatment
visit other than immediately after treatment. This suggests that the initial vaginal
composition or status of the pretreated or the prerecurrent patient, at least as mea-
sured by these tests, is not a key determinant of clinical outcome. If there are pivotal
bacterial determinants at these times, they will likely derive from species or strain level
differences that assays currently in clinical use cannot distinguish. Three studies have
addressed this issue using qPCR (27, 28) or next-generation sequencing (29) ap-
proaches and were limited by short-term follow-up ⱕ46 days posttreatment, which
probably clustered patients that would have recurred into the nonrecurrent groups. No
species was found to be associated with recurrence pre- or posttreatment in all three
studies, and the associations that were found in individual studies were insufficiently
strong to be prognostic. These mixed results and our data together suggest that it is
less important which of the BV-associated species is found, only that whatever they are,
conventional therapy reduces them to or below the limit of detection of a qPCR assay.
The overall response of our cohort of recurrent patients to SOC therapy was poor in
that 76% of tracked patients either recurred or were refractory. Despite a history of
frequent and recent recurrences, 18% of patients achieved long-term remission. Several
explanations may account for this response. Some of these women may have been
recurrent due to reinfection via coitus and had changed partners before or early in the
study. Some may have been misdiagnosed prior to the study and had other types of
vaginitis and thus only had a single episode of BV at enrollment. Some may have been
one treatment away from long-term remission, i.e., each successive treatment moved
them incrementally to a more stable vaginal microbiome. The tests applied to the
cohort could not distinguish this subgroup from the others at their initial symptomatic
visit, so by these criteria they did not have a less severe form of BV. Pending next-

May 2019 Volume 57 Issue 5 e00227-19 jcm.asm.org 7

Sobel et al. Journal of Clinical Microbiology

generation sequencing studies may show whether there were differences in composi-
tion of this subgroup that associates with their long-term remission.
The LbRC test is essentially a molecular Nugent test, in that it measures the relative
abundance of Lactobacillus and reports whether even traces of non-Lactobacillus
species are detected. While it is adequate at diagnosing BV, it suffers from false-positive
scores because it is too sensitive to dysbiosis. The empirically optimized threshold score
for diagnosing BV was ⱕ1, approximately 50% or less Lactobacillus species; the test is
not sufficiently precise to resolve scores under 1, where the likely true diagnostic
differentiation lies. This disadvantage turns into an advantage in prognosis because the
optimized prognostic threshold LbRC score was ⱖ5, or approximately 95% Lactobacil-
lus. This explains why Nugent scores 0 to 3 did not predict remission at the posttreat-
ment visit even though Nugent scores 4 to 10 predicted recurrence. The bimodal
distribution of Nugent scores among patients who will later recur indicates that nearly

Downloaded from http://jcm.asm.org/ on July 29, 2020 by guest

half of these patients have low Nugent scores 0 to 3 who nevertheless have LbRC/5
scores below 5. These low Nugent scores were not prognostic because they do not
distinguish patients whose levels of Lactobacillus were in the 80 to 95% range from
patients with levels of Lactobacillus above 95%, i.e., those less likely to recur because
they have very few non-Lactobacillus species.
Our observation that only the posttreatment visit has this prognostic information is
logical at least in hindsight and tells a story about the pathobiology of BV. Primarily, it
means that eradication of non-Lactobacillus, BV-associated species by SOC therapy to
nearly undetectable levels is crucial to guarantee long-term remission. Second, the lack
of prognostic signatures at other remission visits means that the battle for dominance
among vaginal species is an ongoing process which at no specific single time is
decisive, until recurrence. Getting off to a good start, posttherapy, is the only path to
success long term. Finally, the unique association of this visit with long-term outcome
suggests that eventual recurrence was not driven by events unrelated to therapy. If
coitus or reinfection, for example, were driving recurrence in this cohort, the association
of response to therapy and long-term outcome would have blurred, and prognostic
indicators would not have been found.
Interpretations of findings in this study are to be extrapolated cautiously because of
its limitations. These include that our cohort was predominantly African-American, we
had high drop rates after therapy, coitus was self-reported, and data on the numbers
of partners were not collected. Nevertheless, our prognostic findings clearly warrant
larger studies on more diverse groups. If validated, clinical extensions of these findings
may be to apply extended, higher-dose, or combination therapies that include probi-
otics, specifically to patients in remission at the post-SOC treatment visit if they are at
risk of recurrence by the LbRC or even the Nugent score. Preemptive therapy is not
standard of care for BV, but it has been applied successfully, for example, as twice-
weekly regimens of high-dose vaginal metronidazole/miconazole (15). Patients do well
during the treatment regimen but recur after therapy ends. We suggest that the
secondary treatment should occur at the posttreatment visit for only patients that score
as high risk for recurrence, before the non-Lactobacillus species that are still present
have the opportunity to retake the vaginal battleground.

SUPPLEMENTAL MATERIAL
Supplemental material for this article may be found at https://doi.org/10.1128/JCM
.00227-19.
SUPPLEMENTAL FILE 1, PDF file, 0.6 MB.

ACKNOWLEDGMENTS
We thank the many current and former Wayne State University undergraduates and
staff who contributed to patient and data management and sample scoring, including
Manell Aboutaleb, Naila Baydoun, Sarah Codreanu, Kay Dedicatoria, Manar Edriss,
Tamara Jebry, Michelle Hudson, Carly Malburg, Samantha Mathews, Roshan Nizamud-
din, Nasheen Nizamuddin, and Sharmi Purkayestha.

May 2019 Volume 57 Issue 5 e00227-19 jcm.asm.org 8

Prognostic Indicators of Recurrent Bacterial Vaginosis Journal of Clinical Microbiology

We report no conflicts of interest.

This study was funded in part by NIH/NIAID grant R21AI111103.

REFERENCES
1. Sobel R, Sobel JD. 2015. Metronidazole for the treatment of vaginal 16. Aguin TJ, Akins RA, Sobel JD. 2013. High-dose vaginal metronidazole for
infections. Expert Opin Pharmacother 16:1109 –1115. https://doi.org/10 recurrent bacterial vaginosis: a pilot study. J Low Genit Tract Dis https://
.1517/14656566.2015.1035255. doi.org/10.1097/LGT.0b013e31829a5558.
2. Brook I. 2016. Spectrum and treatment of anaerobic infections. J Infect 17. Peixoto F, Camargos A, Duarte G, Linhares I, Bahamondes L, Petracco A.
Chemother 22:1–13. https://doi.org/10.1016/j.jiac.2015.10.010. 2008. Efficacy and tolerance of metronidazole and miconazole nitrate in
3. Brook I, Wexler HM, Goldstein EJ. 2013. Antianaerobic antimicrobials: treatment of vaginitis. Int J Gynaecol Obstet 102:287–292. https://doi
spectrum and susceptibility testing. Clin Microbiol Rev 26:526 –546. .org/10.1016/j.ijgo.2008.04.014.
https://doi.org/10.1128/CMR.00086-12. 18. Reichman O, Akins R, Sobel JD. 2009. Boric acid addition to suppressive
4. Freeman CD, Klutman NE, Lamp KC. 1997. Metronidazole: a therapeutic antimicrobial therapy for recurrent bacterial vaginosis. Sex Transm Dis
review and update. Drugs 54:679–708. https://doi.org/10.2165/00003495 36:732–734. https://doi.org/10.1097/OLQ.0b013e3181b08456.
-199754050-00003.

Downloaded from http://jcm.asm.org/ on July 29, 2020 by guest

19. Amsel R, Totten PA, Spiegel CA, Chen KC, Eschenbach D, Holmes KK.
5. Lofmark S, Edlund C, Nord CE. 2010. Metronidazole is still the drug of 1983. Nonspecific vaginitis: diagnostic criteria and microbial and epide-
choice for treatment of anaerobic infections. Clin Infect Dis 50:S16 –S23. miologic associations. Am J Med 74:14 –22. https://doi.org/10.1016/0002
https://doi.org/10.1086/647939. -9343(83)91112-9.
6. Ling Z, Liu X, Chen W, Luo Y, Yuan L, Xia Y, Nelson KE, Huang S, Zhang 20. Nugent RP, Krohn MA, Hillier SL. 1991. Reliability of diagnosing bacterial
S, Wang Y, Yuan J, Li L, Xiang C. 2013. The restoration of the vaginal vaginosis is improved by a standardized method of gram stain interpre-
microbiota after treatment for bacterial vaginosis with metronidazole or tation. J Clin Microbiol 29:297–301.
probiotics. Microb Ecol 65:773–780. https://doi.org/10.1007/s00248-012 21. Cartwright CP, Lembke BD, Ramachandran K, Body BA, Nye MB, Rivers
-0154-3. CA, Schwebke JR. 2013. Comparison of nucleic acid amplification assays
7. Bradshaw CS, Morton AN, Hocking J, Garland SM, Morris MB, Moss LM, with BD affirm VPIII for diagnosis of vaginitis in symptomatic women. J
Horvath LB, Kuzevska I, Fairley CK. 2006. High recurrence rates of bac- Clin Microbiol 51:3694 –3699. https://doi.org/10.1128/JCM.01537-13.
terial vaginosis over the course of 12 months after oral metronidazole
22. Kampan NC, Suffian SS, Ithnin NS, Muhammad M, Zakaria SZ, Jamil MA.
therapy and factors associated with recurrence. J Infect Dis 193:
2011. Evaluation of BV® Blue test kit for the diagnosis of bacterial
1478 –1486. https://doi.org/10.1086/503780.
vaginosis. Sex Reprod Healthcare 2:1–5. https://doi.org/10.1016/j.srhc
8. Hay P. 2009. Recurrent bacterial vaginosis. Curr Opin Infect Dis 22:82– 86.
.2010.11.002.
https://doi.org/10.1097/QCO.0b013e32832180c6.
23. Myziuk L, Romanowski B, Johnson SC. 2003. BV Blue test for diagnosis of
9. Beigi RH, Austin MN, Meyn LA, Krohn MA, Hillier SL. 2004. Antimicro-
bacterial vaginosis. J Clin Microbiol 41:1925–1928. https://doi.org/10
bial resistance associated with the treatment of bacterial vaginosis.
.1128/JCM.41.5.1925-1928.2003.
Am J Obstet Gynecol 191:1124 –1129. https://doi.org/10.1016/j.ajog
24. Lambert JA, Kalra A, Dodge CT, John S, Sobel JD, Akins RA. 2013. Novel
.2004.05.033.
PCR-based methods enhance characterization of vaginal microbiota in a
10. Schuyler JA, Mordechai E, Adelson ME, Sobel JD, Gygax SE, Hilbert DW.
2016. Identification of intrinsically metronidazole-resistant clades of bacterial vaginosis patient before and after treatment. Appl Environ
Gardnerella vaginalis. Diagn Microbiol Infect Dis 84:1–3. https://doi.org/ Microbiol 79:4181– 4185. https://doi.org/10.1128/AEM.01160-13.
10.1016/j.diagmicrobio.2015.10.006. 25. Lambert JA, John S, Sobel JD, Akins RA. 2013. Longitudinal analysis of
11. De Backer E, Verhelst R, Verstraelen H, Claeys G, Verschraegen G, vaginal microbiome dynamics in women with recurrent bacterial
Temmerman M, Vaneechoutte M. 2006. Antibiotic susceptibility of vaginosis: recognition of the conversion process. PLoS One 8. https://
Atopobium vaginae. BMC Infect Dis 6:51. https://doi.org/10.1186/1471 doi.org/10.1371/journal.pone.0082599.
-2334-6-51. 26. Bossuyt PM, Cohen JF, Gatsonis CA, Korevaar DA, Group S. 2016. STARD
12. Mayer BT, Srinivasan S, Fiedler TL, Marrazzo JM, Fredricks DN, Schiffer JT. 2015: updated reporting guidelines for all diagnostic accuracy studies.
2015. Rapid and profound shifts in the vaginal microbiota following Ann Transl Med 4:85.
antibiotic treatment for bacterial vaginosis. J Infect Dis 212:793– 802. 27. Hilbert DW, Smith WL, Paulish-Miller TE, Chadwick SG, Toner G,
https://doi.org/10.1093/infdis/jiv079. Mordechai E, Adelson ME, Sobel JD, Gygax SE. 2016. Utilization of
13. Srinivasan S, Liu C, Mitchell CM, Fiedler TL, Thomas KK, Agnew KJ, molecular methods to identify prognostic markers for recurrent bac-
Marrazzo JM, Fredricks DN. 2010. Temporal variability of human vaginal terial vaginosis. Diagn Microbiol Infect Dis 86:231–242. https://doi
bacteria and relationship with bacterial vaginosis. PLoS One 5:e10197. .org/10.1016/j.diagmicrobio.2016.07.003.
https://doi.org/10.1371/journal.pone.0010197. 28. Marrazzo JM, Thomas KK, Fiedler TL, Ringwood K, Fredricks DN. 2008.
14. Sanchez S, Garcia PJ, Thomas KK, Catlin M, Holmes KK. 2004. Intravaginal Relationship of specific vaginal bacteria and bacterial vaginosis treat-
metronidazole gel versus metronidazole plus nystatin ovules for bacte- ment failure in women who have sex with women. Ann Intern Med
rial vaginosis: a randomized controlled trial. Am J Obstet Gynecol 191: 149:20 –28. https://doi.org/10.7326/0003-4819-149-1-200807010-00006.
1898 –1906. https://doi.org/10.1016/j.ajog.2004.06.089. 29. Wang B, Xiao BB, Shang CG, Wang K, Na RS, Nu XX, Liao Q. 2014. Molecular
15. Aguin T, Akins RA, Sobel JD. 2014. High-dose vaginal maintenance analysis of the relationship between specific vaginal bacteria and bacterial
metronidazole for recurrent bacterial vaginosis: a pilot study. Sex Transm vaginosis metronidazole therapy failure. Eur J Clin Microbiol Infect Dis
Dis 41:290 –291. https://doi.org/10.1097/OLQ.0000000000000123. 33:1749–1756. https://doi.org/10.1007/s10096-014-2128-5.

May 2019 Volume 57 Issue 5 e00227-19 jcm.asm.org 9