Bacterial vaginosis and preterm birth: a systematic review and meta-analysis

Bacterial vaginosis and preterm birth:

a systematic review and meta-analysis

Elisabeth-Ann Vandenwyngaerden

Supervisor(s): Prof. Dr. Hans Verstraelen, Prof. Dr. Piet Cools

A dissertation submitted to Ghent University in partial fulfilment of the requirements for the degree of Master of Medicine in Medicine

Deze pagina is niet beschikbaar omdat ze persoonsgegevens bevat.

Universiteitsbibliotheek Gent, 2021.

This page is not available because it contains personal information.

Ghent University, Library, 2021.

Acknowledgements

Completing this thesis has been an incredibly valuable experience as I have learned so much and grown as a student and a future doctor. Therefore, I would like to thank everybody who made this possible.

A sincere thank you goes to Prof. Piet Cools, who sparked a keen interest in me for an

underexposed topic in public health. He guided me closely through my first research project and it was a real pleasure to learn from his extensive knowledge in this field. His feedback, time and enthusiasm for this project contributed to this being a truly positive experience for me, for which I am very grateful.

Secondly, a special thanks goes out to Prof. Brecht Devleesschauwer, for conducting the statistical analyses of this thesis. His willingness and patience to teach me about the world of statistics, gave this thesis as a learning opportunity an extra dimension to me.

It was also an honor to work with Prof. Dr. Hans Verstraelen in his field of expertise. I would like to thank him for his time and his constructive feedback, that pushed me to be critical and explore more about the vaginal microbiota.

I also would like to give a word to my fellow student Judith, for starting this project with me and conducting the review of reviews together that gave grounds to this thesis. I want to thank her for the fruitful collaboration.

Finally, I would like to thank my incredible support system. A lot has happened in my personal life while I was studying, causing that I will unfortunately never look back at this period in my life as the most carefree, but rather as the most transformational. I would honestly not be where I am today, finishing my thesis, without the most loving people in my life encouraging me.

An immense thank you goes to my wonderful friends for always having my back, in the good and the bad. A special thanks goes to Roos, for being my mentor and attentive supporter. I am also forever grateful to my family in law, for making me feel a part of their family and making Ghent my new home.

I would like to dedicate this thesis to my mum, Marie-Angélique. She was so ecstatic that I passed the entrance exam and my first year of medicine, and it saddens me she can’t be here anymore to see what I have achieved since then. She would be so proud of me, flourishing in her city Ghent after so many setbacks on all fronts, but the truth is I am most proud of her. I admire her strength, kindness, determination and perseverance, and I would not be who I am without her example. She may not be here anymore, but her dedication to motherhood and my upbringing give me a lifetime of support. I am eternally grateful for this gift.

Finally, my love goes out to my boyfriend Sam who is my biggest supporter, my best friend, my home.

Elisabeth-Ann Vandenwyngaerden December 2019

Inhoudsopgave

1.7. CONTEXT OF THIS STUDY ... 9

1.7.1. Global Burden of Disease (GBD) ... 9

1.7.2. Disease model of bacterial vaginosis and aim of this study ... 10

1.8. PRETERM BIRTH ... 11 2. METHODS ... 13 2.1. SEARCH STRATEGY ... 13 2.2. SELECTION PROCESS ... 13 2.3. ELIGIBILITY CRITERIA ... 13 2.3.1. Inclusion criteria ... 13 2.3.2. Exclusion criteria ... 14 2.5. DATA COLLECTION ... 14 2.6. DATA MANAGEMENT ... 15 2.7. STATISTICAL ANALYSES ... 15 3. RESULTS ... 16 3.5. STUDY SELECTION ... 16 3.6. DESCRIPTION OF STUDIES ... 17

3.7. ASSOCIATION BETWEEN BV AND PTB ... 24

4. DISCUSSION ... 27

4.1.INTERPRETATION OF STUDY FINDINGS ... 27

4.2.BV AND PTB: POSSIBLE MECHANISMS ... 28

4.2.1. Intrauterine infection and inflammation ... 28

4.2.2. Vaginal microbiota composition... 29

4.3.IMPLICATIONS FOR CLINICAL PRACTICE, RESEARCH AND POLICY ... 30

4.3.1. Treatment in pregnancy aiming to reduce the risk of PTB ... 30

4.3.2. To screen or not to screen?... 32

4.3.3. Implications for research and policy ... 32

5. CONCLUSION... 35

REFERENCES ... 36

6. ADDENDUM ... 42

Abstract

Introduction: Bacterial vaginosis (BV) is defined as an imbalance of the vaginal microbiota and is the most common vaginal condition of women of reproductive age. BV is associated with a plethora of adverse health outcomes, among which an increased risk of preterm birth (PTB). A recent meta-analysis of the strength of association between BV and PTB is lacking, but an essential part of our current effort to estimate the total burden of BV. Therefore, we performed a systematic review and meta-analysis to assess the association between BV and PTB.

Methods: MEDLINE, Web of Science and Embase databases were searched for original studies documenting an association between BV and PTB. Selection criteria were: (i) the data appeared in prospective observational cohort studies, (ii) the studies reported on the estimates of the associations between BV and PTB or reported the data to calculate these, (iii) BV diagnosis was specified and limited to microscopy (such as the Nugent score), Amsel criteria, Spiegel criteria, BVBlue test, Pap smear and quantitative PCR. A summary univariate odds ratio (OR) and the corresponding 95% confidence interval (CI) was calculated using a random effects meta-analysis model.

Results: A total of 37 studies were included, reporting on 25,435 women. Women with BV were twice as likely to give birth preterm (summary OR, 1.98; 95% CI, 1.55-2.54; p < 0.001).

Conclusion: BV is an important risk factor for PTB. The presence of BV as a contributor to PTB can be explained by several mechanisms, however a consensus has yet to be reached regarding its pathophysiology. Further research is needed to understand the etiology of BV and its role in PTB better, to inform trials aiming to develop interventions to prevent PTB.

Samenvatting

Introductie: Bacteriële vaginose (BV) wordt gedefinieerd als een onevenwicht van de vaginale microbiota en is de meest voorkomende vaginale aandoening bij vrouwen van reproductieve leeftijd. BV is geassocieerd met een overvloed aan nadelige gezondheidsgevolgen, waaronder een verhoogd risico op vroeggeboorte. Een recente meta-analyse van de sterkte van associatie tussen BV en vroeggeboorte ontbreekt, maar is een essentieel deel van onze huidige inspanning om de totale ‘burden’ van BV te schatten. Bijgevolg voerden wij een systematische review en meta-analyse uit om de associatie tussen BV en vroeggeboorte te beoordelen.

Methode: MEDLINE, Web of Science en Embase databanken werden doorzocht voor originele studies die documenteren over een associatie tussen BV en vroeggeboorte. Selectiecriteria waren: (i) de data verschenen in prospectieve observationele cohort studies, (ii) de studies rapporteerden over de geschatte associatie tussen BV en vroeggeboorte of rapporteerden de data om deze te berekenen, (iii) BV diagnose was gespecificeerd en gelimiteerd tot microscopie (zoals de Nugent score), Amsel criteria, Spiegel criteria, BVBlue test, Pap smear en kwantitatieve PCR. Een ‘summary univariate odds ratio (OR)’ en het bijhorende 95% betrouwbaarheidsinterval (BI) werd berekend met behulp van een ‘random effects’ meta-analyse model.

Resultaten: Een totaal van 37 studies werd geïncludeerd, die rapporteerden over 25,435

vrouwen. Vrouwen met BV hadden twee keer zoveel kans om preterm te bevallen (summary OR, 1.98; 95% CI, 1.55-2.54; p < 0.001).

Conclusie: BV is een belangrijke risicofactor voor vroeggeboorte. De bijdrage van de

aanwezigheid van BV aan vroeggeboorte kan verklaard worden aan de hand van verscheidene mechanismen, maar een consensus over de pathofysiologie dient nog bekomen te worden. Verder onderzoek is vereist om de etiologie van BV en zijn rol in vroeggeboorte beter te begrijpen, om verdere proeven te instrueren met als doel interventies te ontwikkelen om vroeggeboorte te voorkomen.

Abbreviations

aOR Adjusted odds ratio

APO Adverse pregnancy outcome

BV Bacterial vaginosis

BVAB1, -3 Bacterial vaginosis associated bacteria 1, -3

BW Birth weight

CDC Centers for Disease Control and Prevention

CI Confidence interval

CST Community state type

DALY Disability-adjusted life years

GBD Global burden of disease

HIV Human immunodeficiency virus

HPV Human papillomavirus

HSV-2 Herpes simplex virus type 2 IVF In vitro fertilization

LBW Low birth weight

NGS Next-generation sequencing

NSS Nugent screening score

OR Odds ratio

PCR Polymerase chain reaction

PID Pelvic inflammatory disease

PPROM Preterm premature rupture of membranes

PRISMA Preferred Reporting Items for Systematic Reviews and Meta-Analysis

PROM Premature rupture of membranes

PTB Preterm birth

PTD Preterm delivery

PTL Preterm labor

qPCR Quantitative polymerase chain reaction

SE Standard error

SED Sexually enhanced disease

SES Socioeconomic status

sPTB Spontaneous preterm birth

sPTL Spontaneous preterm labor

STD Sexually transmitted disease STI Sexually transmitted infection TNF-α Tumor necrosis factor α

1. Introduction

Bacterial vaginosis (BV) is a condition defined as an imbalance of the vaginal microbiota. BV is the most common vaginal condition of women of reproductive age (1)with a general population prevalence varying across regions, from less than 10% in Finland to 58% in South-Africa (2). The prevalence among pregnant women varies from 12% to 49% depending on region and ethnicity (3). Yet, BV is incompletely understood and has been referred to as one of the most prevalent enigmatic conditions in the field of medicine (4).

1.1. Pathophysiology

The human vagina has a symbiotic relationship with its associated vaginal microbiota (1). The healthy vaginal microbiota in women of reproductive age is dominated by lactobacilli and has an important role in protecting the vagina from pathogenic microorganisms (5). This is largely achieved through the production of lactic acid by lactobacilli (6), which lowers the vaginal pH to 3,5-4,5 and has emerged to also have direct microbicidal, viricidal and immunomodulatory effects as such (7). Secondly, some lactobacilli also produce hydrogen peroxide (H2O2 ), an oxidizing agent which has been hypothesized to act as a growth inhibitor of catalase-negative bacteria such as most anaerobic microorganisms (5). However, conflicting results in vivo were found and it is a subject of debate (5). Finally, these lactobacilli also produce bacteriocins, antimicrobial

peptides with a variety of killing mechanisms (5). These three mechanisms protect the vaginal environment against pathogenic infection.

BV is characterized as a disturbance of this equilibrium (1) where the prevailing lactobacilli are replaced by an overgrowth of non-lactic-acid producing bacteria such as Gardnerella vaginalis and Atopobium vaginae (8). In BV, the protective mucus layer on the vaginal epithelium is broken down (9), the pH rises and these anaerobes attach to the vaginal epithelial cells, which are called ‘clue cells’ as such, when desquamated (Fig. 1).

Figure 1. The vaginal microbiota of women of reproductive age (10). Panel A. The spectrum of the vaginal microbiota goes from a healthy microbiota dominated by lactobacilli (left) to a bacterial vaginosis (BV) microbiota lacking lactobacilli and dominated by anaerobes such as Gardnerella vaginalis and Atopobium vaginae (right).In the classical view of the vaginal microbiota, an intermediate microbiota is thought of as the transition between a healthy vaginal microbiota and BV. Panel B. In the disturbed BV microbiota, the protective mucus layer on the vaginal epithelium is broken down by anaerobes. These anaerobes attach to

the vaginal epithelial cells in large numbers, which are named ‘clue cells’. Panel C.Microscopic view of Gram-stained vaginal smears (10x10x magnification) of a healthy vaginal microbiota (left), intermediate

microbiota (middle) and BV (right).

1.2. Clinical features

BV is often asymptomatic. However, symptoms do occur in up to 35% (3)of the affected women and have an impact on their well-being. Signs most commonly include an increased thin, gray-white vaginal discharge and a ‘fishy’ vaginal malodor (11). As a result of these manifestations, most symptomatic women complain of an impact on their emotional health and their sex life (12).

1.3. Risk factors

The epidemiology and cause of BV remain largely unresolved (2). Some risk factors have been identified, of which the strongest are vaginal douching, smoking and lack of condom use (1).

vaginal epithelium

mucus

layer

clue

cell

lactobacilli dominated

microbiota

disturbed microbiota

bacterial vaginosis

-intermediate

microbiota

lactobacilli

Nugent score 4-6

Gardnerella vaginalis

Atopobium vaginae

Nugent score 7-10

Nugent score 0-3

However, many women have BV without presenting these risk factors. In pregnant women, there is evidence that past induced abortions and more than one sexual partner are risk factors for BV (13).

Secondly, there are remarkable racial disparities in prevalence. Regardless of geographic location, BV is significantly more present in Black and Hispanic women compared to other ethnic groups (2,3).

Also in women who have sex with women, BV is estimated at 33,5% (3), and the condition is associated with an increased number of female partners and a partner diagnosed with BV (14). This addresses the controversial hypothesis that BV could be regarded as a sexually transmitted disease (STD). Further supporting evidence is given by a meta-analysis that reported a significant association of BV and sexual contact with new and multiple partners of both genders, and

furthermore, that decreasing the number of unprotected sexual encounters may reduce the incidence and recurrence of BV (15). However, evidence to prove that BV acts as an STD is incomplete, and also often contradictory of which the prevention paradox in BV epidemiology is the best example (16). Other researchers now suggest to consider an alternative infectious disease model, in which BV is regarded as a sexually enhanced disease (SED) (16) with frequency of intercourse as a critical factor (17).

1.4. Diagnosis

There are many different diagnostic criteria used to assess BV, the relevant diagnostic methods for this research are described here.

1.4.1. Amsel criteria

The criteria described by Amsel and colleagues (18) are used in clinical settings. BV is diagnosed when minimum three of the following four signs are present:

• Thin, white, yellow homogeneous discharge

• Clue cellson each of 10 separate low-power wet mount microscopic fields, described as vaginal epithelial cells heavily coated with bacilli

• A vaginal fluid pH > 4,5

• A fishy odor elicited when adding 10% potassium hydroxide (KOH) solution to the vaginal secretions. This is commonly known as the ‘whiff test’.

However, some literature suggests a modification that the presence of two signs would suffice to test positive for BV (19).

1.4.2. Spiegel criteria

Spiegel and coworkers (20) were the first to validate a Gram stained direct smear of vaginal fluid as a method to diagnose BV. In this technique, different microbial morphotypes are counted, i.e. Lactobacillus (large Gram-positive rods), Gardnerella (small Gram-variable rods) and others are categorized by morphology only, e.g. gram-positive bacilli, curved rods, fusiforms. All

morphotypes are quantitated and scored under oil immersion by this scheme: • 1+ : <1 per field

• 2+ : 1 to 5 per field • 3+ : 6 to 30 per field • 4+ : >30 per field

The smear is considered as normal when lactobacilli are predominant (3 to 4+) with or without G. vaginalis. On the other hand, when more mixed flora is found consisting of Gram-positive, gram-negative or Gram-variable bacteria and the Lactobacillus morphotype is absent or present in only low numbers (0 to 2+), the smear is considered as consistent with BV.

1.4.3. Nugent score

To provide an objective diagnostic tool that could be used in large multicenter studies, the Nugent scoring system was developed (21). Unlike the Spiegel criteria, it allows gradations in severity of dysbiosis. Today, it is considered the gold standard to diagnose BV in research settings. Gram-stained vaginal smears are examined microscopically under oil immersion for the presence of lactobacilli (Gram-positive bacilli), G. vaginalis/Bacteroides morphotypes (Gram-variable coccobacilli) and Mobiluncus morphotypes (small curved Gram-variable rods). Each of these three categories receive a score based on the number of these morphotypes counted. These scores are added for a total score ranging from 0-10. The scoring is the following (21):

• 0-3: normal microbiota • 4-6: intermediate microbiota • 7-10: bacterial vaginosis

Variations on this Nugent scoring system have been described, such as the smear score by Ison and Hay (22) and Schmidt and Hansen (23). This score also weighs morphotypes, but the smear is not Gram-stained, aiming to provide a more practical way of diagnosis.

However, it should be noted that some findings suggest that Bacteroides and Mobiluncus morphotypes on Gram-stained smears do not accurately reflect the actual presence of these bacteria in the human vagina. Bacteroides morphotypes are more likely to represent Prevotella or Porphyromonas spp. and Mobiluncus morphotypes are often BV-associated bacteria 1 (BVAB1), leaving some researchers to suggest a different listing of bacterial morphotypes in the Nugent score (24).

1.4.4. Molecular diagnosis

The Nugent score does not include identification of several bacteria that later have been associated with BV such as A. vaginae (25). This could lead to misidentification of bacteria that lack a cell wall (such as Mycoplasma species) or present with a variable morphology (such as A. vaginae) or to false negative BV diagnoses (26). With qualitative polymerase chain reaction (qPCR) technology, it is possible to develop tools that target several microorganisms, aiming to provide a more accurate diagnosis of BV (26). Molecular techniques based on qPCR and next-generation sequencing (NGS) are now increasingly used for research and to a lesser extent as a diagnostic tool in clinical settings (27).

1.5. Treatment

In most women, symptoms of BV resolve on their own without intervention (1). However, when treatment is necessary, with antibiotics such as metronidazole (oral or vaginal gel), tinidazole (oral) or clindamycin (vaginal cream) as recommended by the U.S. Centers for Disease Control and Prevention (CDC) (28). Treatment of male sexual partners is not recommended (29),

considering that BV is not regarded as a STD. Efficacy for cure four or more weeks after therapy ranges from 48% to 85% (30). However, when recurrence after one month is investigated, clinical trials show a high recurrence of BV in 58% by 12 months (31). The estimated annual global economic burden of treating symptomatic BV is $4.8 billion, with more than half of costs due to recurrent BV cases (3) thus this is an issue that has to be addressed. This recurrence could be partly explained by the biofilm that some G. vaginalis strains form (32).With standard treatment, the biofilm is not eradicated but only temporarily suppressed into a dormant state, which

reactivates when the medication is ceased (32). New therapeutic strategies should be developed, and there is evidence that suggests probiotics may have a place as an adjuvant to antibiotic treatment to eradicate the biofilm (32). Some clinical trials have obtained positive results with the use of lactobacilli to restore the vaginal microbial ecosystem in the therapy for BV. However,

these studies have been conducted with such heterogeneity that there is not sufficient evidence for or against recommending probiotics for the treatment of BV (33).

1.6. Sequelae

A considerable part of the large burden of BV can be attributed to its sequelae. BV is associated with a higher risk for the acquisition and/or transmission of sexually transmitted infections (STIs). It is consistently associated with an increased risk of human immunodeficiency virus (HIV) infection and transmission (34,35) and has a positive association with uterine cervical human papillomavirus (HPV) (36), chlamydia and gonorrhea (37). Further, women are at risk for acquiring Trichomonas vaginalis (38) and a link between BV and herpes simplex virus type 2 (HSV-2) also has been reported (39).

Secondly, BV has been associated with many adverse gynecological and obstetric outcomes. BV has been significantly associated with the increased risk of preterm birth (PTB) (40), as well that of late miscarriages, maternal infection (41) and low birth weight (42). In infertility patients, BV is a risk factor for preclinical pregnancy loss and there is strong circumstantial evidence that supports a causal link between BV and tubal infertility (43). Certain BV-associated organisms have been associated with pelvic inflammatory disease (PID), but a causal association needs to be

established in further studies (44).

Because of the high prevalence and a plethora of sequelae, there is a concomitant very high economic burden. The US economic burden of BV was nearly tripled when costs of BV-associated PTBs and HIV incidence due to BV were included (3).

1.7. Context of this study

This thesis is part of a larger study that aims to estimate the global burden of BV, by calculating the disability adjusted life years (DALY).

1.7.1. Global Burden of Disease (GBD)

The Global Burden of Disease study (GBD) is to date the most comprehensive effort to generate worldwide estimates of mortality and morbidity (45). It uses DALY to quantify health loss, allowing to compare diseases, health risks and injuries across time and populations and provides as such

a tool to quantify health loss (46). The GBD estimates give decision makers insight into the health challenges they are facing and are of great importance regarding policy and funding of research. Unfortunately, despite being the most prevalent gynecological condition and strong evidence of a plethora of sequelae of BV, BV was not included in this study, nor in any other burden of disease study. Therefore, this thesis is part of a larger study that aims to estimate the true burden of BV.

1.7.2. Disease model of bacterial vaginosis and aim of this study

In a preliminary study, we worked on a disease model where all known or possible sequelae were documented (Fig. 2). The total burden of BV can then be calculated when combining these data with the estimated global prevalence of BV (3) and the known burden of these sequelae. Second, a review of reviews was conducted to identify lack of systematic reviews and meta-analysis of certain sequelae. To guarantee an accurate estimate of the global burden of BV, it is essential that estimates of the strengths of the associations between BV and each of the sequelae are up to date. PTB was identified as one of the sequelae of which a recent estimate of the association was lacking. Such a review/meta-analysis has been completed three times in the past to our knowledge, i.e. in 1999 by Flynn and coworkers (47), in 2003 (40) and in 2007 (41) both by Leitich and coworkers. Therefore, this thesis aimed at performing a systematic review and meta-analysis on the association between BV and PTB.

Figure 2. Disease model of bacterial vaginosis (BV).

CT: chlamydia trachomatis, HIV: human immunodeficiency virus, HPV: human papillomavirus, HSV-2: herpex simplex virus type 2, NG: Neisseria gonorrhoeae.

1.8. Preterm birth

PTB is defined by the WHO as babies born alive before 37 weeks of pregnancy are completed (48). Subcategories can be made, based on the weeks of gestational age:

• <28 weeks: extremely preterm • 28 to <32 weeks: very preterm

• 32 to <37 weeks: moderate to late preterm • 35 to <37 weeks: mild preterm

Approximately 70% of PTBs happen spontaneously, but they can also be indicated for medical reasons and be induced or delivered by prelabor caesarian section (48,49). The pathways to spontaneous preterm birth (sPTB) typically include preterm labor (PTL) and preterm premature

rupture of membranes (PPROM) (49). The causes of PTB are however complex and the pathophysiology that triggers PTB is largely unknown.

An estimated 15 million babies are born too soon, of which 60% occur in Africa and South Asia, and this number increases each year (48). PTB is the leading cause of neonatal morbidity and mortality and 40% of deaths under the age of five are in newborns. Preterm born babies who survive, have an increased risk of disability, which also results in a burden on their families and health systems (48). It is clear that research is needed to understand the mechanisms of PTB better, and consequential action is urgent.

2. Methods

This systematic review and meta-analysis was carried out using an a priori defined study protocol (Addendum 1) and following the Preferred Reporting Items for Systematic Reviews and Meta-analysis (PRISMA) guidelines (50).

2.1. Search strategy

The study was carried out searching MEDLINE (through PubMed), Web of Science and Embase databases. The search strategy consisted out of the search terms ‘bacterial vaginosis’ and ‘preterm birth’.

2.2. Selection process

After removal of duplicates, studies were selected in a three-stage process. First, all studies were screened on relevance to the subject of BV based on the manuscript title alone. Subsequently, the abstracts of manuscripts withheld at first stage were considered to identify studies for full text evaluation. Finally, the full texts of these manuscripts were evaluated in order to include or exclude articles based on the predefined criteria. Reasons for excluding studies were recorded. Disagreements on eligibility, for example resulting from unclarities on certain diagnostic methods, were resolved by consensus with both promotors.

2.3. Eligibility criteria

Studies were selected according to predefined inclusion and exclusion criteria.

2.3.1. Inclusion criteria

Studies were included if they reported on the estimates of the associations (such as relative risk or odds ratio) between BV and PTB, or gave the raw data to calculate these. Only prospective observational cohort studies were included, where the study group is defined as BV-positive pregnant women and the control group as BV-negative pregnant women. Papers were

considered for inclusionwhen BV was diagnosed using at least one of the following diagnostic tools or variations thereof: the Nugent screening score (NSS), Spiegel criteria, Amsel criteria, BVBlue test, Pap smear or molecular diagnosis (qPCR). There were no restrictions in terms of

geography, ethnicity, setting or starting date. The end date limitation was 1st of April 2019. Only papers in English, French, Dutch, Spanish and German were considered.

2.3.2. Exclusion criteria

Studies were excluded when the sample size was not reported. Studies were also excluded if papers did not provide an estimate for the association of BV and PTB, or did not provide the data to calculate these. We excluded reviews, conference abstracts, comments, guidelines, case reports or case series, unpublished articles or multiple reports of the same data.

Retrospective studies and studies that did not report the diagnostic method for BV were also excluded, so were studies that diagnosed BV as abnormal flora or as certain community state types (CSTs) using molecular techniques. Some researchers suggest classifying the vaginal microbiota into five discrete CSTs depending on which lactobacilli dominates, with CST IV defined as diverse communities that are not dominated by Lactobacillus and sometimes used as a definition of BV (51). Even though Tabatei and coworkers found a 4.22 positive association with PTB using this definition (52), we argue that this approach entails a lack of distinction between some CSTs and propose using the quantitative frequencies of key taxa. For this reason, the definition of BV as CST IV was an exclusion criterium in our meta-analysis.

2.5. Data collection

In a predefined data extraction form, following information was abstracted from each individual study: title, first author, year of publication, country, study design, start and end date, study population, total number of participants, age of the participants, all criteria used to diagnose BV and the associated positive scoring cutoff and categorization of intermediate microbiota,

symptomatic and/or asymptomatic population, number of symptomatic cases, whether and which treatment was installed, gestational age at enrollment, moment of BV diagnosis in pregnancy, total number of live births, the definition of PTB (gestational age and whether delivered

spontaneous, induced or both) and correction for confounding. The measure of association was recorded dichotomously in 2 x 2 tables for all four study outcomes, univariate and multivariable odds ratio/relative risk/P-value, lower and upper confidence limits and confidence interval.

2.6. Data management

Endnote reference manager was used to remove duplicates and to create a database of the studies that are identified for review. The duplicates that Endnote missed, were identified in Microsoft Excel. Rayyan, a systematic review web-based application, was used to screen the articles for in- or exclusion and to keep track of the reasons to exclude. The data wereextracted in a form in Excel.

2.7. Statistical analyses

All eligible studies were included in the subsequent meta-analysis. For each individual study, the study-specific odds ratio (OR) and corresponding standard error (SE) were calculated. If the study reported the number of cases and controls, and the occurrence of PTB in both groups, these values were used to calculate the OR and SE post hoc. For the remaining studies that did not provide the number of cases and controls, the reported OR and 95% confidence interval (CI) were used to back-calculate the OR and SE. An optimization process was designed in which the log-transformed OR was fitted to a normal distribution and the sum of squared differences between the observed an the fitted 95% CI was minimized. This approach was applied to the provided univariable OR, or, when this was not available, to the multivariable OR.

A summary univariate OR and the corresponding 95% CI was calculated using a random effects meta-analysis model with restricted maximum likelihood used to weight studies. Heterogeneity was assessed using the I2 statistic and Cochran’s Q test. To evaluate publication bias statistically, Begg and Mazumdar’s rank correlation test (53) and Egger’s regression test (54) was used. This was confirmed by visual inspection of the possible asymmetry of a funnel plot created by plotting the logarithm of the ORs against the inverse of the standard error. Given the results of these tests, it was not deemed necessary to explore the impact of publication bias on the model estimates.

All statistical analyses were carried out in R using the ‘metafor’ package (55).

3. Results

3.5. Study selection

In total, 1419 unduplicated records were identified and screened (Fig. 3). A total of 1003 studies were irrelevant to the subject of BV and thus excluded. Hence 416 citations were assessed for eligibility, on the basis of their abstract and to the utmost extent also of their full text. Of these, 228 reported on a study following a design not meeting the eligibility criteria e.g. interventional studies, reviews, diagnostic methods not disclosed, categorizing BV as abnormal flora or

reporting on another pregnancy outcome (APO) than PTB. A total of 121 were excluded because of the wrong publication type such as conference abstracts, book chapters and letters to editors. Four studies reported on previously published data and one study was in Slovenian, and thus met the criteria for exclusion. Two manuscripts characterized the vaginal microbiota in community-state types using next-generation sequencing technology, making it difficult to assess cases of BV. Seven papers could not be found. Finally, 37 studies were included in the meta-analysis.

From: Moher D, Liberati A, Tetzlaff J, Altman DG, The PRISMA Group (2009). Preferred Reporting Items for Systematic Reviews and Meta-Analyses: The PRISMA Statement. PLoS Med 6(7): e1000097. doi:10.1371/journal.pmed1000097

For more information, visitwww.prisma-statement.org.

PRISM A 2009 Flow Diagram

Sc re en in g In cl u d ed El ig ib ili ty

Records after duplicates removed (n = 1419 ) Titles screened (n = 1419) Records excluded, irrelevant to BV (n = 1003)

Abstracts, and if relevant, full-text articles assessed

for eligibility (n = 416)

Records excluded (n = 379), with reasons:

• Study design not meeting the eligibility criteria (n = 244)

• Wrong publication type (n = 121)

• Multiple reports of the same data (n = 4) • Language restriction

(n = 1)

• Eligible, but unaccepted molecular BV diagnosis (n = 2)

• Full text not found (n = 7) Studies included in

quantitative synthesis (meta-analysis)

(n = 37)

Web Of Science database (n = 1113) PubMed database (n = 555 ) Embase database (n = 608) Id e n ti fi ca ti o n

3.6. Description of studies

The 37 included studies, representing a total of 25,435 women, are summarized in Table 1. Of these studies, twelve took place in North America, two in Africa, sixteen in Europe, six in Asia (including the one study performed in Turkey which is also part of Europe) and one in South America. The study populations varied, but the majority of pregnant women were unselected for a certain risk (low or high) and were healthy, meaning that they had no medical condition other than symptomatic or asymptomatic BV (18/37 studies or 49%). In three studies (8%), the population was unselected as well, but the health condition of the participants (other than BV) was not specified (42,56,57). In seven studies (19%), the population was selected to be low-risk for PTB (58–65), in contrast to two studies that specifically studied healthy women at risk for PTB (66,67).

Five studies (14%) examined women already hospitalized for PTL (68–72) One study was

conducted in a population of pregnant in vitro fertilization (IVF) patients (73). All pregnancies were singleton, however, in three studies exclusion criteria were not specified (73–75) and in two studies exclusion criteria were defined but multiple gestation was not one of them (76,77), thus leaving five studies in which multiple gestation might be possible.

The diagnostic methods to assess BV was the Nugent score in 18 (49%) studies, of which two used the combination of Nugent scoring and the clinical criterion of a pH > 4,5 (77,78). BV was diagnosed using the Amsel criteria in nine studies (24%) and one study used both Nugent and Amsel (70). The criteria defined by Spiegel were applied in five (14%) studies, in one of these, the criteria were combined with two other diagnostic criteria, i.e. the presence of clue cells on wet mount and a positive culture of BV-associated bacteria (69). Some diagnostic methods were only used in a single study: Schmidt and Hansen criteria (42),DNA concentrationsof A. vaginae (≥ 10^8 copies/mL) or G. vaginalis (≥ 10^9 copies/mL) with qPCR (73), microscopic wet mount (59) and clue cells on Pap smear (79). If multiple diagnostic criteria were used separately, the data for the Nugent score where used given it is considered as the gold standard. More than two third (68%, 25/37) of the studies corrected for different potential confounders such as maternal gestational age, smoking, parity, history of PTB and race.

The cut-off for PTB was defined at 37 weeks of gestation in almost all manuscripts, more precisely in 32 out of 37 manuscripts (86%). Of these, six studies specified the birth as

spontaneous and two also specified a birthweight (BW) of <2500g as an extra criterium on top of the birth < 37 weeks of gestational age. Five studies (14%) did not adhere to the standard definition of preterm as gestation of <37 weeks, but reported on a more preterm cut-off: two studies defined PTB at 35 weeks (68,77) and three at 36 weeks (60,80,81), of which one also stipulated a BW of ≤2500g as an extra criterium on top of the 36 weeks threshold (60).One study did not elaborate on their definition of PTB (75).

Author, year

(reference) Country N Study population BV criteria

Definition of

PTB Correction for confounding

Afolabi, 2016 (74) Nigeria 246 unselected healthy

pregnanta Nugentb <37 weeks no

Bothuyne-Queste,

2012 (76) France 1317

unselected healthy

pregnantc Nugentb,d <37 weeks

randomization & adjustment for maternal tobacco addiction and level of education Bretelle, 2015 (66) France 764 healthy pregnant, at risk

for PTB Nugentd <37 weeks

cox proportional hazard models, adjusting for gestational age

Çakiroğlu, 2015

(82) Turkey 250

unselected healthy

pregnant Nugent <37 weeks no

Crane, 1999 (83) Canada 140 unselected healthy

pregnant Nugent

<37 weeks, spontaneous

multiple logistic regression adjusting for race, maternal age and parity

Daskalakis, 2006

(58) Greece 1197 low-riskhealthy pregnant Nugent

<37 weeks, spontaneous

multivariate logistic analysis adjusting for age, ethnicity, height, weight, gravidity, history of miscarriage or pregnancy termination, smoking Donders, 2009

(59) Belgium 641 low-risk healthy pregnant Wet mounte <37 weeks no

Donders, 2000

(84) Belgium 196

unselected healthy

pregnant Amsel 20-37 weeks

multivariate regression adjusting for age, history of previous abortion, recent use of antibiotics, culture results, lactobacillary grades, signs of cervical inflammation and clinical diagnosis of bacterial vaginosis

Elliott, 1990 (60) Kenia 276 low-risk pregnant, almost

all low SES Spiegel

≤ 36 weeks & BW ≤ 2500 g

multivariate logistic regression for age, rupture of membrane, hypertension, gonococcal infection and marital status

French, 1997 (80) USA 511 unselected healthy

pregnant Amsel 22-36 weeks

stratification for first-trimester bleeding, BV, presence of genital-tract microorganisms and PTB

French, 1997 (81) USA 510 unselected healthy

pregnant Amsel 22-36 weeks

logistic regression adjusting for maternal ethnicity, age, smoking, nulliparity history of PTB, gestational age at enrolment and concurrent infection with T. vaginalis or C. trachomatis

Goffinet, 2003 (68) France 278 hospitalized for PTL,

intact membranes Nugent <35 weeks no

Gomez, 2010 (85) USA 743 unselected healthy

pregnant Nugent

<37 weeks, spontaneous

adjustment for race, periodontal disease status, current diagnosis of STD, history of spontaneous PTB

Guerra, 2006 (67) Italy 190 healthy pregnant, with

history of a previous PTB Nugent <37 weeks

multivariate logistic regression adjusting for each of the different patterns of vaginal flora together with all the other putative factors

Harper, 2012 (86) USA 1453 unselected healthy

pregnant Nugent <37 weeks

randomization & multivariable logistic regression adjusting for black race and history of

Hillier, 1995 (78) USA 577 unselected healthy

pregnant Nugent + pH > 4,5

<37 weeks & BW <2500g

multivariate logistic regression adjusting for smoking, race, previous delivery of LBW infant, loss of earlier pregnancy, gravidity, maternal age, marital status, use of antibiotics,

colonization with C. trachomatis, N. gonorrhoeae, T. vaginalis or group B streptococci Holst, 1994 (69) Sweden 49 hospitalized for PTL, without recognizable cause Spiegel + clue cells on wet mount and stained smear + culturef <37 weeks no Jacobsson, 2002 (79) Sweden 852 unselected healthy pregnant

Pap smear with clue cells

<37 weeks, spontaneous

logistic regression adjusting for primi/multiparity and use of antibiotics during pregnancy

Kurki, 1992 (61) Finland 733 low-risk nulliparas Spiegel <37 weeks no Lata, 2010 (75) India 159 unselected healthy

pregnanta Amsel not stated no

Laxmi, 2011 (70) India 60 hospitalized for sPTL Amsel + Nugent <37 weeks no Macones, 2004

(65) USA 375

healthy pregnant, without

history of previous sPTB Amsel

<37 weeks,

spontaneous stratification Mangot-Bertrand,

McGregor, 1990

(87) USA 135

unselected healthy

pregnant Spiegel <37 weeks

logistic regression adjusting for variables

frequently associated with PTB in this univariate analysis or other studies

Meis, 1995 (77) USA 2929 unselected healthy

pregnantc Nugent + pH > 4,5

<35 weeks, spontaneous

multivariable logistic regression adjusting for race, parity, smoking during pregnancy and BV at 28 weeks.

Nelson, 2014 (88) USA 386 unselected healthy

pregnant Nugentd

21-36 + 6 weeks, spontaneous

randomization & stratification for prior PTB

Oakeshott, 2004

(62) UK 848 low-risk healthy pregnant Nugentd

<37 weeks,

spontaneous adjustment for smoking

Pereira, 2016 (56) Brazil 272 unselected pregnant Nugent <37 weeks

adjustment for maternal age, economic class, hypertension during pregnancy, use of illicit drugs during pregnancy and previous PTB Povlsen, 2001

(89) Denmark 484

unselected healthy

pregnant Amsel <37 weeks logistic regression adjusting for BW Pratiksha, 2016 (90) India 96 unselected healthy pregnant, in third trimester Amselh <37 weeks no

Purwar, 2001 (63) India 938 low-risk healthy pregnant Nugent <37 weeks

multiple logistic regression adjusting for

socioeconomic stratum, history of previous PTB, candidiasis, vaginal microflora showing

Subtil, 2002 (72) France 102 hospitalized for PTL Amsel

<37 weeks, spontaneous or induced

no

Svare, 2006 (42) Denmark 3262 unselected pregnantc Schmidt and

Hansen <37 weeks

multivariate analysis adjusting for previous PTB, previous conization, smoking and gestational diabetes mellitus

Tellapragada,

2016 (91) India 710

unselected healthy

pregnant Nugent <37 weeks

multivariate regression adjusting for the maternal baseline, physical and infectious covariates.

Thorp, 2008 (57) USA 1352 unselected pregnant Nugentb

<37 weeks, spontaneous & BW <2500g

multinomial logistic regression adjusting for marital status, maternal age, maternal education, poverty index, pre-pregnancy BMI, parity,

number of lifetime sex partners, self-reported STI during pregnancy, self-reported yeast infection during pregnancy, smoking and pH

Thorsen, 2006

(64) Denmark 2151

very low-risk healthy

pregnant Amsel <37 weeks

stratification & logistic regression adjusting for last delivery LBW, public assistance recipient, unemployed, physically strenuous work, smoking >10 cigarettes daily and coagulase negative staphylococci

a: No exclusion criteria mentioned, and thus presumably there is no exclusion of multiple gestations. b: Intermediate microbiota categorized as normal microbiota c: Twin pregnancies are not excluded. d: Self-sampling e: BV was determined when predominant granular microflora with uncountable bacteria are present all over the

3.7. Association between BV and PTB

The logarithm of the ORs of the individual studies as well as the logarithm of the OR resulting from the meta-regression analysis are shown in Figure 4. PTB showed to be significantly

associated with BV: the meta-regression analysis (Fig. 4) showed that women with BV were twice as likely to give birth preterm (OR, 1.98; 95% CI, 1.55-2.54; p < 0.001). The heterogeneity among these studies was substantial (I2 = 74,44%) and this was confirmed by Cochrane’s Q test (p < 0.001), confirming that the use of a random effects model was required. Visual inspection of the funnel plot (Fig. 5) showed fairly strong symmetry, suggesting no publication bias was present in the included studies. This was supported by the statistical tests results, indicating no significant publication bias (Eggers regression test p = 0.7373; Begg and Mazumdar’s rank relation test p = 0.3163). It was not deemed necessary to explore the impact of publication bias on the model estimates.

Figure 4. Forest plots of estimates of association between bacterial vaginosis (BV) and preterm birth (PTB). Studies are plotted alphabetically according to the last name of the first author, followed by

publication year. Each study is represented by a square and a horizontal line, which respectively correspond to the OR and the 95% CI. Individual ORs are considered statistically significant if the 95% CI

does not include the null value. The magnitude of the square reflects the weight of the study (determined by random effects analysis) in the meta-analysis. The diamond shape below represents the summary

estimate with its width indicating the 95% CI. On the right, two columns document for each study the number of PTB cases on the total of BV-positive and of BV-negative women. a: only aOR reported, b: crude

OR was used for analysis.

BV+ BV -16/64 17/182 24/196 88/1121 6/19 214/745 8/53 18/197 1/31 8/109 16/95 88/1102 4/30 50/611 3/16 20/180 30/57 115/219 23/150 26/361 unknowna 6/24 50/254 32/306 36/437 33/95 17/95 70/792 54/661 77/118 291/168 9/12 13/37 unknownb 11/162 6/571 19/41 6/118 27/30 22/30 39/69 86/306 1/7 7/70 34/52 27/124 1/24 3/111 unknownb 42/348 4/38 5/110 38/738 15/40 75/232 10/70 74/414 5/48 7/48 32/115 40/823 6/14 43/88 37/533 134/2729 7/42 47/668 16/175 70/1177 8/293 62/1858 p-value < 0.001

Figure 5. Funnel plot to asses publication bias among the included studies assessing the association between bacterial vaginosis (BV) and preterm birth (PTB). The dots represent the estimates of association of the 37 included studies. On the horizontal axis, the log odds ratio is plotted against the inverse of the standard error (SE) of the odds ratio on the vertical axis. The vertical line in the middle of the plot indicates the random-effects

4. Discussion

4.1. Interpretation of study findings

In this study, we aimed to estimate the association of BV as a risk factor for PTB by means of a systematic review and meta-analysis. We found a total of 37 papers ranging from 1988 untill 2016 representing a total of 25,435 women. The results of current meta-analysis show that BV is significantly associated with PTB, and that pregnant women with BV have a two-fold higher risk of PTB.

Our results are in line with those from three previous meta-analyses that assessed the relation between BV and PTB. The meta-analysis of Flynn and coworkers (1999) reported a summary OR of 1.85 (95% CI; 1.62-2.11, p < 0.001) (47). This analysis represented a total of more than 17,000 women and included nineteen individual studies using similar eligibility criteria as our study. Flynn and coworkers also pooled the adjusted ORs and found a 60% increased risk of PTB in pregnant women with BV. They further performed several sub-analyses (e.g. stratified by study design, baseline population risk of prematurity, method and time of BV diagnosis in pregnancy) and BV remained significantly associated with PTB.

In 2003, Leitich and colleagues also performed a systematic review and meta-analysis on BV and PTB (40), in which eighteen studies were included, reporting on over 20,000 women. They found a summary estimate OR of 2.19 (95% CI, 1.54-3.12; p < 0.05). In their sub-analysis, they

stratified on the time of BV diagnosis in pregnancy. They reported significantly higher risks for BV screened at <16 weeks or <20 weeks of respectively OR 7.55 (95% CI, 1.80-31.65) and OR 4.20 (95% CI, 2.11-8.39) compared to BV screened at ≥20 weeks of gestation. A substantial

heterogeneity was found in their main analysis, which did not completely disappear in the sub-analyses although it was reduced strongly.

In 2007, Leitich and Kiss updated this meta-analysis themselves (41), in which they included fourteen new studies with results for more than 10,000 women, bringing the total study population to over 30,500 patients in 32 studies. They reported an OR of 2.16 (95% CI, 1.56-3.00; p < 0.05) for PTB in pregnant women with asymptomatic BV, and in a sub-analysis for patients with

symptoms of PTL they reported an OR of 2.38 (95% CI; 1.02-5.58, p < 0.05).

Additionally, some researchers calculated a higher risk of PTB when BV occurs early on in pregnancy (40),even though others contradicted this with their findings (92).

Although theoretically it is possible that confounders account for or contribute to the found association and a meta-analysis cannot prove causality, we do find that several of the Bradford Hillcriteria suggesting causality are met (93). These include the strength of the reported

association, the temporality that BV precedes the outcome of PTB (with the exception of a few studies that enrolled women at hospitalization for PTL) and biological plausibility.

Our study design was rather elaborate and strict, enabling that only well-defined studies reporting on BV diagnosed by predefined tools were included. In addition, studies reporting on PTL but not on subsequent PTB, or studies defining PTB as abnormal or intermediate vaginal microbiota were excluded. For example, in studies that also reported cases of PTL but delivery at term, these cases were assigned to the control group. In addition, abnormal vaginal flora or intermediate flora was not accepted as a definition of BV. This ensures that this analysis strictly examined the association between BV and PTB, which increases the power of our results.

In our results, no publication bias was observed to further strengthen our findings. However, substantial heterogeneity was found. This is not unexpected given that over the 37 studies, women of different gestational ages were included, from fifteen different countries, varying diagnostic methods and definitions of PTB were used.

4.2. BV and PTB: possible mechanisms

The etiology of PTB is multifactorial (49) and intra-uterine infections are thought to account for up to 40% of PTB (49).

4.2.1. Intrauterine infection and inflammation

Vaginal bacteria often identified in BV have been discussed, but the mechanistic association with PTB is most likely derived from the ascent of these bacteria into the uterus (94). Gardnerella, Mobiluncus and Bacteroides bacteria are commonly found in the uterus before rupture of the membranes, and have likely ascended before or early in pregnancy (49,95).

Many researchers have shown that BV is frequently associated with subsequent chorioamnionitis, in which similar organisms can often be found (95). Secondly, amniotic fluid infection is often present in women with PTL and intact membranes (95), inducing labor by either production of bacterial exotoxins or by the presence of endotoxins (94). A maternal and fetal tissue immune response is provoked, including an increased production of various cytokines and chemokines

(96). Prostaglandin synthesis is stimulated, leading to provoke uterine contractions and initiate metalloproteases that attack the chorioamniotic membranes to rupture (96). However, the mechanism linking BV to chorioamnionitis and PTB is unclear (94).

Macones and coworkers (65) studied the hypothesis whether a certain polymorphism of cytokine genes may be related to sPTB given the hypothesis that this variability may be associated with cytokine production and the severity of inflammatory responses. One of the identified cytokines is tumor necrosis factor α (TNF-α) (94). They identified a single nucleotide polymorphisms of TNF-α that has been related to different transcription. It was reported that maternal carriers of the rarer allele (TNF-2), were at a significantly increased risk of sPTB. This association was modified by the presence of BV, resulting in a 6-fold increased risk of PTB when women presented both the a ‘‘susceptible’’ genotype and BV.

Secondly, phospholipase 2 was identified as a pathogenic factor for the development of PTB in women with BV by French and coworkers (81). Phospholipase can be produced by bacteria associated with BV, which can result in the activation of the prostaglandin synthesis, which is as stated a step in the physiology of labor activation (96).

Finally, BV is associated with increased mucinase and sialidase concentrations in vagina and cervix (96). The cervical mucus layer consists out of mucus glycoproteins or mucins forming a matrix acting as a defensive barrier. This mucus may be degradated by mucin-degrading enzymes produced by certain Gardnerella species (97). It has been proven that sialidase, β-galactosidase and β-N-acetylhexosaminidase, such musinases, in vaginal fluid were associated with BV (98,99). Howe and coworkers suggested that these enzymes in BV-associated

microbiota may help to break down the host defensive mucus to allow bacteria to ascend to the uterus (98), further contributing to PTB.

4.2.2. Vaginal microbiota composition

The vaginal microbiota is highly dynamic (100). Studies have shown that in pregnancy, the composition of the vaginal microbiota shifts to lesser diversity and, to a lesser degree, richness (100). The microbiota in pregnancy are proportionally increasingly dominated by Lactobacilli such as L. iners, L. crispatus, L. jensenii and L. johnsonii, while many other usual vaginal microbiota members become less prevalent (100). The notable increased dominance of certain lactobacilli in

the vagina in pregnancy may be important to preserve the integrity of the microbiota to reduce risk of ascending infection (100).

Some studies researched the vaginal microbiota of BV-positive pregnant women with higher resolution. Researchers reported that most notably L. crispatus abundance was found to be decreased (101–103). Kindinger (104), Nelson (103) and coworkers corroborated a L. crispatus dominance to be protective against PTB. The study of Kindinger differed in its findings by reporting a positive association between L. iners dominated communities and PTB and no

significant association between BV-like communities and PTB (104). However, other researchers (101,102) state these discrepancies can be explained by methodological differences rather than biological differences and project the importance of these taxa in further research.

Secondly, abundance of BV associated bacteria such as Gardnerella (101,103,105–107) ,

Atopobium (101) and Mobiluncus (105) was confirmed to be associated with a higher risk of PTB. More specifically, it was discovered a certain subspecies clade of G. vaginalis explained the genus association with PTB (102). In addition, BVAB3 (108) was found to decrease the risk of PTB (105). This is an unexpected finding taking into account that it is viewed as sufficient to cause BV and thus should be further investigated.

Finally, higher total bacterial loads can be detected in women who delivered preterm compared to their at term equals (109) and estimate over a ten-fold increased risk when comparing the lowest to the highest tertile of log levels of G. vaginalis (103). This addresses the dose response

question and the possible importance of assessing the bacterial loads.

On the other hand, some studies find no difference when Gram stain score 7 to 8 and 9 to 10 are compared (92).

4.3. Implications for clinical practice, research and policy

4.3.1. Treatment in pregnancy aiming to reduce the risk of PTB

Currently, CDC guidelines advise to treat all symptomatic pregnant women to reduce the signs and symptoms of infection (28). The most commonly used drug is metronidazole, an antibiotic and antiprotozoal medication, which does not show an association with teratogenic or mutagenic effects in newborns when used in pregnancy (28). This medication, alongside with clindamycin was also used in some of the included studies when treatment was given. Clindamycin is an

antibiotic that also can be used safely in pregnant women (110). There are two ways to

administer antibiotics: either vaginal or oral. There can be found logic in both: delivering a high antibiotic load to the vagina where the abnormal overgrowth exists seems sensible, but so does treating systemically to ensure that microorganisms that already accessed the decidua can be eradicated (111).

In a 2013 Cochrane systematic review, antibiotic therapy was shown to be effective at eradicating BV in pregnancy and thus eliminating its symptoms. However, this effect was defined as an eradication of BV on examination, without giving a timeframe (112) and thus not taking into account the high recurrence rates. Asides from treating to improve maternal outcomes, it is of great interest whether antibiotics could reduce the risk of PTB in women with BV. The results of this systematic review and meta-analysis show that treatment of BV in pregnancy did not reduce the risk of PTB, however two included studies showed a 47% decrease in risk when screening and treating criteria were broadened to include women with intermediate microbiota (112). There was no difference found on the outcome of PTB regarding oral versus vaginal antibiotics, but oral therapy has a slight advantage on the prolongation of gestational age (112). Recently, a second multicenter, double-blind randomized trial that assessed the impact of clindamycin treatment of BV-positive women with low-risk pregnancies, confirmed the lack of risk reduction of spontaneous very PTB (28 to <32 weeks) (PREMEVA trial (113)).

Given that BV in early pregnancy is possibly associated with a higher risk of PTB compared to BV later in pregnancy, researchers have sought to investigate whether early treatment could reduce the risk of PTB. The findings on efficacy of treatment in early pregnancy to prevent PTB are conflicting: the Cochrane systematic review and PREMEVA trial reported that treatment before respectively 20 weeks’ gestation and 15 weeks’ gestation does not result in a decreased risk of PTB (112,113), however Lamont and coworkers did find a significant reduced risk of sPTB when administering clindamycin at < 22 weeks of gestation to women with abnormal flora (114). Treatment for women at high-risk for PTB (e.g. with a history of previous PTB) did not have a positive effect on PTB (112,113) in both the Cochrane and PREMEVA meta-analyses. We can thus conclude evidence is not sufficient to recommend early antibiotic treatment to these groups of pregnant women at higher risk for PTB. However, further research is both ethically justified and needed to make recommendations for women with a history of PTB (115).

Nevertheless, these studies also have their limitations (115): PREMEVA’s outcome of PTB occurred less than anticipated and compliance was 80% in the clindamycin group. This can be possibly attributed to the 3% reported adverse effects and the lack of cure; the effectiveness to resolve BV was not assessed (in contrast to the Cochrane review). Thus, the reasons for failure of antibiotic treatment in regard to the prevention of PTB should be further investigated

thoroughly.

One explanation for this inefficacy relies possibly in the high rates of antibiotic resistance among BV-associated bacteria (116). Consequently, antibiotics might not only kill the targeted bacteria, but also reduce the protective Lactobacillus population leading to a larger imbalance of vaginal microbiota (117). Secondly, conventional techniques are probably failing at finding all who is at risk. Genomics could help us refine our understanding (115). Furthermore, it is possible that the host response to infection is also crucial to comprehend the infection and if so, the treatment could be adjusted accordingly (115).

4.3.2. To screen or not to screen?

There is a lack of evidence of a reduction in PTB prevalence after treatment for BV and an increased risk of side-effects sufficiently to stop or change treatment exists (112). Evidence is thus insufficient to recommend routine screening of all pregnant women to treat asymptomatic BV with the aim to prevent PTB. This has been confirmed for the screening of low-risk pregnancies (112,113). There may be a benefit of screening women at high risk of PTB (118), but this is still under debate.

Other screening methods have been suggested such as self-testing of vaginal pH, but this failed to demonstrate the efficacy for the prevention of PTB as well (113).

4.3.3. Implications for research and policy

A deeper understanding of the vaginal microbiota and the pathophysiology of BV is essential. It is required to develop more specialized tools to diagnose BV such as taxonomic finetuning of bacteria (97) and to identify individual susceptibility of women for PTB. Further, the role of microbiota in the etiology of PTB remains enigmatic and requires additional research. More answers are needed to inform further trials on screening recommendations and therapies to aim to prevent PTB. Subsequently, recommendations can be made concerning screening for BV to prevent PTB, after also evaluating its cost-effectiveness.

4.4. Limitations

The first limitation can possibly be found in the lack of an elaborate search string: synonyms for PTB such as ‘preterm delivery’ or ‘prematurity’ have not been included in the search, which could lead to the oversight of some studies that could have been included. However, this is because our research initially conducted search strings for all APOs, among which PPROM and LBW, and was later narrowed down to the one search string of PTB because of the magnitude of results. On the other hand, it seems unlikely that qualitative prospective studies would not include the term PTB in their abstract, which reduces the impact of this restriction on our estimation.

Secondly, the studies that were included were not yet assessed for quality. This is necessary to critically appraise the results.

Regarding the used eligibility criteria, the issue of molecular diagnosis arose. We decided to include qPCR with a definition of BV when DNA concentrations were found of A. vaginae (≥ 10^8 copies/mL) or G. vaginalis (≥ 10^9 copies/mL). On the other hand, we excluded studies that diagnosed BV on CSTs. However, the molecular era of mapping the vaginal microbiota remains a field in continuous development. It is still full of questions at this time, but the molecular definition of BV should be constantly revisited and the newly acquired knowledge critically examined. Likewise, discussion arose regarding the inclusion of one study that reported diagnosis of BV on the basis of clue cells on Pap smear. The validation of Pap smear and molecular definitions to assess BV should be further discussed. Another eligibility criterium in this study that should be questioned, is the definition of PTB. Although we discussed the BV criteria in detail, we did not do the same for PTB. However, this is the outcome of our study and it should have been as carefully considered to not use any other definition than the only accepted one of the WHO, i.e. birth <37 weeks of gestational age. If the study reported preterm delivery results at multiple gestational cut-offs, we used only the 37-week cutoff, but five studies were included that used a definition other than 37 weeks i.e. more preterm. Their data should have been noted for sub-analyses and not have been included for the summary OR of this meta-analysis.

The major limitation of any meta-analysis is the suitability of combining results from different studies. One way to address the heterogeneity was the use of the random effects model, that accounts for variability between studies when estimating the risk.

A second possibility would be to research the sources of the predicted heterogeneity by the means of sub-analyses, but these have not been conducted yet. Further research, for example categorizing women according to their a priori risk or health status, trimester in which BV was

diagnosed, criteria of diagnosis or presence of symptoms, could help detect potential sources of heterogeneity and define whether the estimated association differs from the summary estimate in certain subgroups. Previous research (40) found evidence for higher risks when BV was

diagnosed at <16 weeks and at <20 weeks of gestation, confirming the relevance of these

subgroups. But even when stratification is conducted, generalization of these results still needs to be handled with caution. The studies mostly took place in North America and Europe, both very developed continents with mostly high-income countries. There is an underrepresentation of middle- and low-income countries viewing that only one study in Latin-America was included, along with two studies in Africa and that all five studies in Asia were in India. This could lead to bias taking into account that approximately 85% of PTB are concentrated in Africa and Asia (119).

Finally, the concern of confounding should also be addressed. Several studies corrected for confounding, but the crude data was used to assess a summary estimate. There was only one study in which this was not available and an aOR was used. Arguments for doing so is to eliminate that each study was corrected in different ways, and this could cause even more heterogeneity without being able to identify it. On the downside, the resulting OR is not corrected for any sort of confounding which raises the possibility of a misidentification of association.

Concluding on these shortcomings, we aim to rethink our search strings and the exclusion criteria concerning the diagnosis of BV and the definition of PTB. Further, a quality assessment of the included papers is necessary to determine with which confidence we can justify our conclusions. We plan to validate by using the Newcastle-Ottawa scale and adjusting this how we deem it to be relevant. Finally, substantial heterogeneity was found and further studies need to explore the sources of heterogeneity. We aim to conduct various further sub-analyses such as categorization for trimesters in which BV was diagnosed and the used method to diagnose BV.

These amendments will be conducted out of the scope of this thesis, to be able to present a more differentiated image of BV as a risk factor for prematurity.

5. Conclusion

We conducted a systematic review and meta-analysis to update the estimated association

between BV and PTB. Our results were in line with the three previous meta-analyses and showed a two-fold increased risk of PTB in pregnant women with BV. Our study had certain limitations that we will address further outside the scope of this thesis, aiming to present a more

differentiated image of BV as a risk factor for prematurity.

BV remains an enigmatic condition in many of its facets, and the pathophysiology of BV as a risk factor for PTB is no exception. It can be explained through several possible mechanisms,

however a consensus has not yet been reached.

Therefore, recommendations for clinical practice are incomplete. It is recommended to treat BV in symptomatic pregnant women with antibiotics to improve maternal outcomes. However, there is a lack of evidence for a reduction of the risk of PTB with current therapies. Subsequently,

insufficient evidence exists to screen and treat any group of pregnant women to prevent PTB. Finalizing this research, many questions remain, emphasizing the need of further research of BV and the role of vaginal microbiota in PTB. Further knowledge is needed to initiate trials aiming to develop therapies effective at preventing PTB. By estimating the burden of BV, the larger project that frames this thesis aims to inform clinicians, researchers and policy makers about its