752

Bacterial vaginosis (BV) is a common condition inwhich the normal vaginal lactobacilli are replaced by amixture of anaerobic bacteria and mycoplasmas, includ-ing Gardnerella vaginalis, Mobiluncus, Bacteroides spp, andMycoplasma hominis. Clinical diagnosis is based on its clas-sic appearance as a white thin homogenous dischargewith a characteristic “fishy” odor, the presence of cluecells, and a vaginal pH >4.5.1 BV may affect a substantialproportion of women during pregnancy, and many ofthese cases will never develop into symptomatic disease.2

The condition may also resolve spontaneously but ismore likely to be persistent throughout pregnancy.

A large body of literature indicates that intrauterine in-fection plays a major role in the pathogenesis of preterm

birth, which itself is responsible for 70% of perinataldeath and almost one half of long-term neurologic mor-bidity.3 Although the role of BV itself in the pathogenesisof preterm labor and delivery is not well understood, theassociation between BV and preterm birth has been con-firmed repeatedly and consistently.4,5

Because the screening and treatment of BV in earlypregnancy would be a simple and straightforward strat-egy to prevent preterm birth, several studies have alreadytested this approach. Oral treatment regimens that usemetronidazole alone or metronidazole and erythromycincombined have shown to be effective in lowering the rateof preterm deliveries in high-risk patients with a history ofat least one spontaneous preterm birth.6-8 In a systematicreview in the Cochrane Library,9 the conclusion was thatthere is some suggestion that the detection and treat-ment of BV in high-risk women may decrease the rate ofpreterm deliveries. It was also concluded that there is nobenefit in screening and treating all pregnant women forBV to prevent preterm birth.

In a recent, large study of the National Institute ofChild Health and Human Development,10 however, oralmetronidazole did not prevent preterm deliveries in a

From the Departments of Obstetrics and Gynecologya and Medical Com-puter Sciences,b Section of Clinical Biometrics, University of Vienna.Received for publication February 28, 2002; revised September 30,2002; accepted November 5, 2002.Reprint requests: Harald Leitich, MD, PhD, Department of Obstetricsand Gynecology, University of Vienna, Währinger Gürtel 18-20, 1090Vienna, Austria. E-mail: harald.leitich@univie.ac.at© 2003, Mosby, Inc. All rights reserved.0002-9378/2003 $30.00 + 0doi:10.1067/mob.2003.167

Antibiotic treatment of bacterial vaginosis in pregnancy:A meta-analysis

Harald Leitich, MD, PhD,a Mathias Brunbauer, MD,a Barbara Bodner-Adler, MD,a

Alexandra Kaider, MSc,b Christian Egarter, MD,a and Peter Husslein, MDa

Vienna, Austria

OBJECTIVE: The purpose of this study was to evaluate the effectiveness of antibiotic treatment of bacterialvaginosis in pregnancy to reduce preterm delivery.STUDY DESIGN: We performed a meta-analysis of published, English-language, randomized, placebo-con-trolled clinical trials of antibiotic treatment of bacterial vaginosis in pregnant women with intact amnioticmembranes at <37 weeks of gestation. Primary outcomes included preterm delivery, perinatal or neonataldeath, and neonatal morbidity.RESULTS: Ten studies with results for 3969 patients were included. In patients without preterm labor, antibi-otic treatment did not significantly decrease preterm delivery at <37 weeks of gestation, in all patients com-bined (odds ratio, 0.83; 95% CI, 0.57-1.21) nor in high-risk patients with a previous preterm delivery (oddsratio, 0.50; 95% CI, 0.22-1.12). In both groups, significant statistical heterogeneity was observed. A signifi-cant reduction in preterm delivery and no statistical heterogeneity were observed in 338 high-risk patientswho received oral regimens with treatment durations of �7 days (odds ratio, 0.42; 95% CI, 0.27-0.67). Non-significant effects and no statistical heterogeneity were observed in low-risk patients (odds ratio, 0.94; 95%CI, 0.71-1.25) and with vaginal regimens (odds ratio, 1.25; 95% CI: 0.86-1.81). In one study antibiotic treat-ment in patients with preterm labor led to a nonsignificant decrease in the rate of preterm deliveries (oddsratio, 0.31; 95% CI, 0.03-3.24).CONCLUSION: The screening of pregnant women who have bacterial vaginosis and who have had a previ-ous preterm delivery and treatment with an oral regimen of longer duration can be justified on the basis ofcurrent evidence. More studies are needed to confirm the effectiveness of this strategy, both in high-risk pa-tients without preterm labor and in patients with preterm labor. (Am J Obstet Gynecol 2003;188:752-8.)

Key words: Bacterial vaginosis, preterm delivery, antibiotic treatment, meta-analysis

Volume 188, Number 3 Leitich et al 753Am J Obstet Gynecol

mixed population of pregnant women, and, more impor-tant, it increased rather than decreased the rate of pre-term deliveries in the high-risk subgroup. After thepublication of this study, uncertainty about the role of BVin preterm birth and the usefulness of screen-and-treatstrategies increased. A recent review of screening for BVin pregnancy, which was carried out for the US PreventiveServices Task Force, could not explain the heterogeneousresults among the studies, including high-risk patients.11

The authors suggested that high-risk patients themselvesconstitute a heterogeneous population, which consists ofmore general high-risk patients for whom antibiotic treat-ment would not be effective and more selected high-riskpatients for whom treatment might still be effective.Based on the results of this review, the US Preventive Ser-vices Task Force concluded that the evidence is insuffi-cient to recommend for or against routine screening ofhigh-risk pregnant women for BV.12 The Task Force alsorecommended against routinely screening average-riskasymptomatic pregnant women for BV.

Similarly, in a recent clinical treatment guideline thatwas published by the American College of Obstetriciansand Gynecologists,13 the statement was made that thereare no current data to support the use of BV screeningstrategies to prevent preterm birth.

We carried out the present meta-analysis to re-evaluatethe role of antibiotic treatment of BV in pregnancy. Ourspecific goals were to include new evidence and to putspecial emphasis on the identification of differences inscreen-and-treat strategies among the existing studies toidentify strategies that will produce more consistent re-sults. This would allow us to draw more reliable conclu-sions about these strategies and to separate them fromother strategies that provide inconsistent results and thatare in need of continued research activity.

Material and methods

In June 2001, we searched MEDLINE from 1966, EM-BASE from 1988, and SCIENCE CITATION INDEX EX-PANDED from 1997 to identify all literature that wasincluded under (vaginitis or vaginosis or vagina* infec-tion or vagina* inflammation) and (pregnan* or gravid*or birth? or labor?r?) and (antibiotic* or antimicrob*).The following criteria were used to select studies for in-clusion: article (original published English-language re-port), study design (randomized placebo-controlledclinical trial), population (women entered <37 weeks ofgestation with intact amniotic membranes and with BV,diagnosed either by clinical criteria or by criteria basedon Gram stain findings [positive cultures for BV-associ-ated microflora alone were not sufficient]), intervention(antibiotic treatment), and ≥1 of the following out-comes: preterm delivery by any available definition, peri-natal or neonatal death, or any information regardingneonatal morbidity. We scanned all of the abstracts from

the computer printouts, the retrieved full-text reports,review articles, and the references from each retrievedreport to determine whether studies met our inclusion criteria. The literature search and the applica-tion of inclusion criteria were done by two independentinvestigators. Results were then compared, and a con-sensus was reached.

To determine the internal validity of the trials that areincluded, we extracted information regarding the processof randomization including concealment, the similarity ofbaseline characteristics between study groups, the num-bers of patients excluded after randomization and thereasons for exclusion, completeness of outcome assess-ment, and assessment of patient compliance.

To determine the clinical homogeneity of the studiesincluded, we extracted patient inclusion and exclusioncriteria, mean gestational ages at randomization, criteriafor a diagnosis of BV, treatment characteristics that in-cluded the administration of antibiotic treatment and thetreatment of patients with concomitant genitourinary in-fections, and the definitions of outcomes. If mean gesta-tional ages at randomization were missing in individualstudies, we estimated them from patient inclusion criteria.

For all outcomes in each individual study, we extractedtotal numbers of patients with results and the numbers oftrue-positive, true-negative, false-positive, and false-nega-tive results. Primary outcomes included preterm deliveryby any available definition, perinatal or neonatal death,and any information regarding neonatal morbidity. As asecondary outcome, we also extracted data regarding ma-ternal infectious morbidity. When the numbers of true-positive, true-negative, false-positive, and false-negativeresults were not available, we attempted to recalculatethem from the reported results.

For all outcomes in each individual study, we calculatedodds ratios with 95% CIs to determine the efficacy of theantibiotic treatment.14 We also calculated the odds ratiosand 95% CIs for all studies combined and for subgroupsof studies according to the following characteristics: pa-tient inclusion criteria (patients admitted with symptomsof preterm labor, patients without preterm labor at low orhigh risk for preterm delivery, and inclusion of patientswith multifetal gestation), gestational age at randomiza-tion, choice of antibiotic treatment, and route of admin-istration.15

A statistical test of homogeneity was carried out foreach analysis to address the statistical validity of the ag-gregation of the trials.15 This test assumes that differ-ences in the results of the individual trials are due tochance alone (ie, that all trial results are homogenous).

To calculate overall results for each outcome and allsubgroups,15 we generally used a fixed-effects model. Insuch models, the overall effect is obtained as a weightedaverage of the individual trial effects. However, if signifi-cant heterogeneity was observed (which we defined as a

754 Leitich et al March 2003Am J Obstet Gynecol

probability value of the homogeneity test of ≤.10 becauseof the low power of the test), we used a random-effectsmodel instead. Trials with zero cells in both the study andcontrol groups were not included in the calculation ofcombined results.

Results

Ten studies were included in this meta-analysis.6-8,10,16-21

Study populations of two studies overlapped, so resultswere available for 3969 individual patients.17,18 Becausethe study of Kurkinen-Räty et al18 was the report of onestudy center that was included in a larger multicenterstudy reported by Kekki et al,17 we generally extractedinformation from the larger multicenter study only. Anyinformation that was not available in the multicenter,but only in the single-center, study was extracted fromthe latter.

Four studies were excluded.22-25 In one study, resultsfor preterm delivery and other obstetric complicationswere reported in text, but not in numeric form.22 In theother three studies of antibiotic treatment of pregnantwomen23 or patients with preterm labor,24,25 subgroupsof patients with BV were identified, but the results werenot reported for these subgroups.

All studies that were included followed a randomized,double-blind, placebo-controlled design. Baseline char-acteristics between the treatment and control groupswere reported to be similar in all but one study, in whichthe percentage of teenage pregnancies in the placebogroup was larger than in the antibiotics group.7 In twostudies, patients were excluded after randomization.8,19

Reasons for exclusion were fetal distress before receivingstudy medication (one patient), patients who de-clined treatment (two patients), clinical chorioamnioni-tis (three patients), exclusions according to studyexclusion criteria (six patients), pharmacy error (one pa-tient), and twins (one patient) in one study19 and follow-up losses (five patients), a lack of compliance (sixpatients), or the need of additional antibiotic therapy forrenal or pulmonary disease (three patients) in the otherstudy.8 Results were available for 98.3%,10 98.7%,6

91.4%,16 100%,17 97.5%,7 88%,19 90.8%,20 85.1%,8 and100%21 of patients in the individual studies. Assessmentof compliance was described in all except one study.17 Infour studies, compliance in the treatment and controlgroups was reported to be similar.6,7,10,21

Patient inclusion criteria, mean gestational age at ran-domization, criteria for a diagnosis of BV, and treatmentregimens of the studies that were included are shown inTable I. In three studies, patients with a history of at leastone preterm delivery (high-risk patients) were included.In two studies, only patients without a previous pretermdelivery (low-risk patients) were included; in four studies,both low-risk and high-risk patients were included. Pa-tients with symptoms of preterm labor were included inonly one study. Multiple pregnancies were excluded fromall studies.

Patient exclusion criteria in individual studies mostlycomprised the use of antibiotic therapy before study initia-tion,6-8,10,16,19,20 severe fetal anomalies,8,10,16,19-21 significantmaternal conditions,6-8,10,16,19-21 concomitant infections,7,8,10

current or planned cervical cerclage,7,10,19,20 symptoms of

Table I. Patients and methods of studies included

Patients with Mean gestational age Study results (No.) Patient inclusion criteria at randomization (wk)

Carey et al, 200010 1919 Singleton pregnancy between 16-23 wk 19.7

Hauth et al, 19956 258 Singleton pregnancy between 22-24 wk and either 22.9previous preterm delivery or prepregnancy weight <50 kg

Joesoef et al, 199516 681 Singleton pregnancy between 14-26 wk 20.3

Kekki et al, 200117* 375 Singleton, low-risk pregnancy between 10-17 wk 13.5Kurkinen et al, 200018* 101 Singleton, low-risk pregnancy at 12 wk 12.4McDonald et al, 19977 480† Singleton pregnancy between 16-26 wk 24.1

McGregor et al, 199119 25‡ Singleton pregnancies at ≤34 wk with preterm labor 30.7

McGregor et al, 199420 129‡ Singleton pregnancy between 16-27 wk 21.5Morales et al, 19948 80 Singleton pregnancy between 13-20 wk and spontaneous 16.5

preterm delivery in the preceding pregnancyVermeulen and Bruinse, 199921 22‡ Singleton pregnancy <26 wk and preterm delivery in 20.0

the preceding pregnancy

*Study populations are overlapping.†Patients with abnormal Gram stain only.‡Patients with BV only.

Volume 188, Number 3 Leitich et al 755Am J Obstet Gynecol

threatened preterm delivery,6,8,10,16,20 and incompetentcervix or uterine anomalies.8,16,19-21

Mean gestational ages at randomization ranged from12.4 to 30.7 weeks. In all except one study,10 mean gesta-tional ages at randomization had to be estimated from pa-tient inclusion criteria, either because they were not statedfor the whole study8,16,17,20 or for the subgroup of patientswith BV.6,7,19,21 Criteria for a diagnosis of BV were clinicalin 1 study, based on Gram stain findings in five studies andbased on combinations of clinical and Gram stain findingsin four studies. Therapies used in individual studies in-cluded intravenous clindamycin followed by oral clin-damycin, vaginal clindamycin, oral metronidazole, or oralmetronidazole and erythromycin combined.

In three studies, patients with symptomatic vaginosis orvaginitis,7,10 a positive culture of Chlamydia trachomatis,10

Trichomonas vaginalis,8,10 Treponema pallidum,10 or Neisseriagonorrhoeae10 or patients with asymptomatic bacteriuria8

were not admitted to the studies. In six other studies, ac-tive screening and treatment of study patients for the fol-lowing microorganisms or infections were reported: Ctrachomatis,16,20 T vaginalis,6,16 group B streptoccocci,8,19

T pallidum,16 N gonorrhoeae,16,20 yeast,6,19 or urinary tractinfections.19,20

The results of the quantitative analyses of individualstudies are displayed in Table II. Results for the outcomeof preterm delivery were reported in all studies. The re-sults for maternal infectious morbidity were available inonly one study,17 and the results for neonatal death ormorbidity were available in none. One study reportedthat there was no significant difference in maternal in-

fections or neonatal outcome between both studygroups, but no numeric data were displayed.10 In an-other three studies, results regarding maternal infec-tions19 or neonatal outcome7,19,21 were not reported forthe BV subgroup.

A significant treatment effect was observed in high-riskpatients that were included in two studies only.6,8 In allother studies, antibiotic treatment did not lead to a lowerrate of preterm delivery or maternal infections.

In eight studies, the rates of BV after antibiotic orplacebo treatment were presented. For one study,21 theelimination rates were published separately in a sec-ondary analysis of the study.26 With intravenous or oraltreatment regimens, antibiotic treatment eliminated78%,10 70%,6 75%,7 or 89%8 of BV cases, and placebotreatment was associated with spontaneous resolution ofBV in 38%,10 18%,6 or 14%8 of cases, respectively. Withvaginal treatment regimens, the elimination rates in theantibiotics groups were 86%,16 66%,17 96%,20 or 67%,26

and the corresponding rates of spontaneous resolution inthe placebo groups were 56%,16 34%,17 45%,20 and43%.26

In Table III, the results for all studies are combined.One study that used the outcome of delivery at <35 weeksof gestation was pooled with other studies as the com-bined outcome of delivery at <34 weeks of gestation.10

In patients without preterm labor, no significant reduc-tion of the rate of preterm deliveries was observed if allstudies were pooled together. Large, yet insignificant,beneficial treatment effects were observed for high-riskpatients and with oral antibiotic treatment. However, the

Diagnosis of BV Administration of antibiotics

Abnormal Gram stain (Nugent score ≥7)32 and 2 Doses of oral metronidazole (2 g) or placebo 48 hours apart, vaginal pH >4.5 treatment or placebo repeated at follow-up visit at 24-29 wk in all

patientsClinical BV (Amsel criteria ≥3)1 and/or abnormal Oral metronidazole (250 mg), 3 times daily for 7 days and 333 mg

Gram stain (Spiegel definition)33 of oral erythromycin, 3 times daily for 14 days or placebo; treatment or placebo repeated after 2-4 wk in persistent BV

Abnormal Gram stain (Nugent score ≥7)32 and Vaginal clindamycin (5 g) or placebo, once daily for 7 daysvaginal pH > 4.5

Abnormal Gram stain (Spiegel definition)33 Vaginal clindamycin (5 g) or placebo, once daily for 7 daysAbnormal Gram stain (Spiegel definition)33 Vaginal clindamycin (5 g) or placebo, once daily for 7 daysAbnormal Gram stain (Spiegel definition)33 Oral metronidazole (400 mg) or placebo twice daily for 2 days,

treatment or placebo repeated after 4 wk in persistent BVAbnormal Gram stain (Nugent score ≥7)32 Intravenous clindamycin (900 mg) or placebo every 8 hours for

3 days, followed by oral clindamycin (300 mg) or placebo 4 times daily for 4 days

Clinical BV and abnormal Gram stain (Nugent score ≥7)32 Vaginal clindamycin (5 g) or placebo once daily for 7 daysClinical BV Oral metronidazole (250 mg) or placebo, 3 times daily for 7 days

Abnormal Gram stain (Nugent score ≥7)32 Vaginal clindamycin (5 g) or placebo once daily for 7 days at26 and 32 wk

756 Leitich et al March 2003Am J Obstet Gynecol

degrees of heterogeneity within those two subgroups ofstudies were also high. Thus, a random effects model wasused in these cases that resulted in wide 95% CIs of the ef-fects odds ratios.

Within the group of studies that used oral antibiotictreatment, heterogeneity was strongly reduced, if studieswere grouped according to treatment duration. A largeand significant effect was seen with longer treatment,and a nonsignificant effect was seen with shorter treat-ment.

Heterogeneity was not reduced if studies were groupedaccording to gestational age of treatment initialization.Both with earlier and later treatment, heterogeneity re-mained significant, and the treatment effects were non-significant. Heterogeneity was much smaller amongstudies that included low-risk patients or studies that usedvaginal antibiotic regimens, and the results in these sub-groups consistently showed no treatment effect. Hetero-geneity was also small if preterm delivery was definedmore restrictively as delivery at <34 or <32 weeks of gesta-tion. With either of these outcomes, no significant reduc-tion of preterm births could be observed.

Comment

The results of this meta-analysis show that strategies ofscreening and treating pregnant women for BV remaincontroversial. Such a strategy was suggested generally forhigh-risk patients with a previous preterm delivery. How-ever, there is now a large degree of heterogeneity amongthe studies that include high-risk patients and thus a largedegree of uncertainty about the true effect that the an-

tibiotic treatment of BV would have in this patient popu-lation.

The situation is much clearer for low-risk patients with-out a previous preterm delivery. The consistent result inthe respective subgroup analysis shows that antibiotictreatment of BV does not decrease the rate of preterm de-liveries in this population.

The subgroups of studies as defined in our meta-analy-sis are not mutually independent, but—to a smaller orlarger degree—overlap each other. High-risk patients, forexample, were treated almost exclusively with oral regi-mens. The subgroup of studies that used oral regimens oflonger duration (the only subgroup with both a signifi-cant treatment effect and absent heterogeneity) con-tained only high-risk patients. Nothing, practically, can besaid about the effect of vaginal clindamycin therapy inhigh-risk patients because the only study that used thisstrategy included 22 high-risk patients with BV.21

Low-risk patients, on the other hand, have beentreated both with oral and vaginal regimens, and no ap-parent effect was seen in either route of administration.In the respective subgroup, no statistical heterogeneitywas observed that allows us to draw this conclusion witha larger degree of certainty. Because vaginal clindamycintherapy was used almost exclusively in low-risk patients,the consistent results in the subgroup of studies thatused vaginal regimens are, in fact, the consistent resultsof studies that used vaginal clindamycin in low-risk pa-tients. For this patient population, it can be concludedreliably that vaginal clindamycin does not have a treat-ment effect.

Table II. Results of individual studies

Outcome Antibiotics Control Odds ratio Study Patients (wk) group group (95% CI)

Carey et al, 200010 All Delivery <37 116/953 121/966 0.97 (0.74-1.27)All Delivery <35 48/953 49/966 0.99 (0.66-1.49)All Delivery <32 22/953 26/966 0.85 (0.48-1.52)Low risk Delivery <37 86/852 95/857 0.90 (0.66-1.23)High risk Delivery <37 30/101 26/109 1.35 (0.73-2.49)

Hauth et al, 19956 High risk Delivery <37 54/172 42/86 0.48 (0.28-0.82)Joesoef et al, 199516 All Delivery <37 51/340 46/341 1.13 (0.74-1.74)

All Delivery <32 16/340 9/341 1.82 (0.79-4.18)Kekki et al, 200117* Low risk Delivery <37 9/187 7/188 1.31 (0.48-3.59)

Low risk Peripartum infections 21/187 33/188 0.59 (0.33-1.07)Kurkinen et al, 200018* Low risk Delivery <34 1/51 0/50 NC

Low risk Delivery <32 1/51 0/50 NCMcDonald et al, 19977 All Delivery <37 11/242 15/238 0.71 (0.32-1.57)

Low risk Delivery <37 10/225 9/221 1.10 (0.44-2.75)High risk Delivery <37 1/17 6/17 0.11 (0.01-1.09)

McGregor et al, 199119 Preterm labor Delivery <37 11/15 9/10 0.31 (0.03-3.24)McGregor et al, 199420 All Delivery <37 9/60 5/69 2.26 (0.71-7.16)Morales et al, 19948 High risk Delivery <37 8/44 16/36 0.28 (0.10-0.76)

High risk Delivery <34 2/44 4/36 0.38 (0.07-2.21)Vermeulen and Bruinse, 199921 High risk Delivery <34 1/11 1/11 1.00 (0.05-18.30)

NC, Not calculable.*Study populations are overlapping.

Volume 188, Number 3 Leitich et al 757Am J Obstet Gynecol

BV early in pregnancy has been identified to be astrong risk factor for preterm delivery.27,28 However,within the subgroups of studies with earlier or later treat-ment (defined as treatment before or after week 20 ofgestation), statistical heterogeneity was not diminished,and no significant treatment effect was observed in eithersubgroup. Thus, the results of these subgroup analyseswere not helpful to make a more reliable statement aboutwhen treatment should be commenced.

Rates of BV elimination after treatment are used oftento measure the effectiveness of antibiotic regimens in pa-tients with BV. We did not see obvious differences in therates of BV elimination between different antibiotic regi-mens and especially not between oral and vaginal regi-mens. Rates of BV elimination that were determined byclinical or Gram-based assessment of vaginal dischargemay only reflect changes in vaginal bacterial colonizationbut do not predict necessarily the elimination of bacteriathat have already ascended to the uterine cavity. Systemictreatment of longer duration may be necessary to treat in-trauterine infection, which has been described as achronic infection starting early in pregnancy.3 It may befor this reason that oral regimens of longer duration ap-peared to be effective in our meta-analysis, but regimensof shorter duration were not.

Regarding choice of antibiotic agents, oral treatmentaimed at the BV set of microbes, including genital my-coplasmas and anaerobes, with oral clindamycin or acombination of oral enteric-coated erythromycin andmetronidazole has been recommended specifically totreat BV in pregnancy.5 Unfortunately, the data pre-sented in this review do not allow us to make recommen-dations for specific (combinations of) antibiotic agents.

Because multiple pregnancies were excluded from allthe studies that were included in this review, there are no

data to support strategies of screening and treating BV inthis population. Also, BV was not found to be associatedsignificantly with preterm delivery in two studies,29,30 in-cluding patients with twin gestations.

Strategies to screen and treat BV in patients who wereadmitted for preterm labor are studied only poorly. Onlyone study with 25 patients with preterm labor and BV wasincluded in our meta-analysis; although the results forthose patients were promising, the effect, because ofsmall sample size, did not reach statistical significance.19

The paucity of evidence for this strategy is surprising be-cause it is used often as part of the infection workup ofpatients who were admitted for preterm labor. Surely,more clinical evidence is needed to confirm the effective-ness of this approach.

More controlled studies are also needed to make morereliable conclusions about the effectiveness of antibiotictreatment of BV in high-risk patients. A specific goalwould be to determine whether the rate of preterm deliv-eries (with the use of a more restrictive definition, such asdelivery at <34 or <32 weeks of gestation) would be dimin-ished in this population. With current knowledge, such astudy should treat high-risk patients early in pregnancywith an oral regimen of longer duration. Also, the promis-ing results of a subgroup analysis31 of the largest study in-cluded in this review,10 which showed a nonsignificantreduction in preterm birth in women with both BVand/or T vaginalis and a positive fetal fibronectin test,must be confirmed. No more resources, on the otherhand, should be allocated to study or implement screen-and-treat strategies in a general low-risk population.

In the meantime, before more evidence is available,screen-and-treat strategies for high-risk patients (as usedin those studies) that show clear treatment effects can bejustified, even on the basis of current knowledge.

Table III. Results of all studies combined in patients without and with preterm labor

Studies Patients Test of Test of included included heterogeneity Model Odds ratio treatment

Patients (No.) (No.) (P value) used* (95% CI) effect (P value)

Patients without preterm laborDelivery at <37 wk

All patients 7 3922 <.02 Random 0.83 (0.57-1.21) .33Low-risk patients 3 2530 .74 Fixed 0.94 (0.71-1.25) .69High-risk patients 4 582 <.01 Random 0.50 (0.22-1.12) .09Oral treatment 4 2737 <.02 Random 0.61 (0.36-1.03) .06Oral treatment at ≤2 days 2 2399 .47 Fixed 0.94 (0.72-1.21) .62Oral treatment ≥7 d 2 338 .38 Fixed 0.42 (0.27-0.67) <.0003Vaginal treatment 3 1185 .54 Fixed 1.25 (0.86-1.81) .25Treatment starting at <20 wk 3 2374 <.05 Random 0.75 (0.36-1.58) .46Treatment starting at >20 wk 4 1548 <.03 Random 0.88 (0.49-1.57) .66

Delivery at <34 wk: all patients 4 2122 .54 Fixed 0.96 (0.65-1.42) .85Delivery at <32 wk: all patients 3 2701 .22 Fixed 1.13 (0.71-1.79) .61Maternal infection: all patients 1 375 — — 0.59 (0.33-1.07) —

Patients with preterm laborDelivery at <37 wk: all patients 1 25 — — 0.31 (0.03-3.24) —

*Random denotes random effects model; fixed denotes fixed effects model.

758 Leitich et al March 2003Am J Obstet Gynecol

REFERENCES

1. Amsel R, Totten PA, Spiegel CA, Chen KSC, Eschenbach D,Holmes KK. Nonspecific vaginitis: diagnostic criteria and microbi-ologic and epidemiologic associations. Am J Med 1983;74:14-22.

2. Eschenbach DA. History and review of bacterial vaginosis. Am JObstet Gynecol 1993;169:441-5.

3. Goldenberg RL, Hauth JC, Andrews WW. Intrauterine infectionand preterm delivery. N Engl J Med 2000;342:1500-7.

4. Flynn CA, Helwig AL, Meurer LN. Bacterial vaginosis in preg-nancy and the risk of prematurity: a meta-analysis. J Fam Pract1999;48:885-92.

5. McGregor JA. Bacterial vaginosis in pregnancy. Obstet GynecolSurv 2000;55(Suppl):S1-19.

6. Hauth JC, Goldenberg RL, Andrews WW, DuBard MB, CopperRL. Reduced incidence of preterm delivery with metronidazoleand erythromycin in women with bacterial vaginosis. N Engl JMed 1995;333:1732-6.

7. McDonald HM, O’Loughlin JA, Vigneswaran R, Jolley PT, Har-vey JA, Bof A, et al. Impact of metronidazole therapy on pretermbirth in women with bacterial vaginosis flora (Gardnerella vagi-nalis): a randomised, placebo controlled trial. Br J Obstet Gy-naecol 1997;104:1391-7.

8. Morales WJ, Schorr S, Albritton J. Effect of metronidazole in pa-tients with preterm birth in preceding pregnancy and bacterialvaginosis: a placebo-controlled, double-blind study. Am J ObstetGynecol 1994;171:345-9.

9. Brocklehurst P, Hannah M, McDonald H. Interventions fortreating bacterial vaginosis in pregnancy (Cochrane Review).Cochrane Library, Issue 1, Oxford: Update Software; 2002.

10. Carey JC, Klebanoff MA, Hauth JC, Hillier SL, Thom EA, ErnestJM, et al. Metronidazole to prevent preterm delivery in pregnantwomen with asymptomatic bacterial vaginosis. N Engl J Med2000;342:534-40.

11. Guise JM, Mahon SM, Aickin M, Helfand M, Peipert JF, WesthoffC. Screening for bacterial vaginosis in pregnancy. Am J Prev Med2001;20:62-72.

12. The U.S. Preventive Services Task Force. Screening for bacterialvaginosis in pregnancy: recommendations and rationale. Am JPrev Med 2001;20:59-61.

13. American College of Obstetricians and Gynecologists. Assessment ofrisk factors for preterm birth. Washington (DC): American Collegeof Obstetricians and Gynecologists; 2001. Practice bulletin No.: 31.

14. Demets DL. Methods for combining randomized clinical trials:strengths and limitations. Stat Med 1987;6:341-8.

15. Whitehead A, Whitehead J. A general parametric approach tothe meta-analysis of randomized clinical trials. Stat Med1991;10:1665-77.

16. Joesoef MR, Hiller SL, Wiknjosastro G, Sumampouw H, LinnanM, Norojono W, et al. Intravaginal clindamycin treatment forbacterial vaginosis: effects on preterm delivery and low birthweight. Am J Obstet Gynecol 1995;173:1527-31.

17. Kekki M, Kurki T, Pelkonen J, Kurkinen-Räty M, Cacciatore B,Paavonen J. Vaginal clindamycin in preventing preterm birthand peripartal infections in asymptomatic women with bacterialvaginosis: a randomized, controlled trial. Obstet Gynecol2001;97:643-8.

18. Kurkinen-Räty M, Vuopala S, Koskela M, Kekki M, Kurki T,Paavonen J, et al. A randomized controlled trial of vaginal clin-

damycin for early pregnancy bacterial vaginosis. Br J Obstet Gy-naecol 2000;107:1427-32.

19. McGregor JA, French JI, Seo K. Adjunctive clindamycin therapyfor preterm labor: results of a double-blind, placebo-controlledtrial. Am J Obstet Gynecol 1991;165:867-75.

20. McGregor JA, French JI, Jones W, Milligan K, McKinney PJ, Pat-terson E, et al. Bacterial vaginosis is associated with prematurityand vaginal fluid mucinase and sialidase: results of a controlledtrial of topical clindamycin cream. Am J Obstet Gynecol 1994;170:1048-60.

21. Vermeulen GM, Bruinse HW. Prophylactic administration ofclindamycin 2% vaginal cream to reduce the incidence of spon-taneous preterm birth in women with an increased recurrencerisk: a randomized placebo-controlled double-blind trial. Br JObstet Gynaecol 1999;106:652-7.

22. Duff P, Lee ML, Hillier SL, Herd LM, Krohn MA, EschenbachDA. Amoxicillin treatment of bacterial vaginosis during preg-nancy. Obstet Gynecol 1991;77:431-5.

23. McGregor JA, French JI, Richter R, Vuchetich M, Bachus V, SeoK, et al. Cervicovaginal microflora and pregnancy outcome: re-sults of a double-blind, placebo-controlled trial of erythromycintreatment. Am J Obstet Gynecol 1990;163:1580-91.

24. Newton ER, Dinsmoor MJ, Gibbs RS. A randomized, blinded,placebo-controlled trial of antibiotics in idiopathic pretermlabor. Obstet Gynecol 1989;74:562-6.

25. Newton ER, Shields L, Ridgway LE III, Berkus MD, Elliott BD.Combination antibiotics and indomethacin in idiopathic pre-term labor: a randomized double-blind clinical trial. Am J ObstetGynecol 1991;165:1753-9.

26. Vermeulen GM, van Zwet AA, Bruinse HW. Changes in the vagi-nal flora after two percent clindamycin vaginal cream in womenat high risk of spontaneous preterm birth. Br J Obstet Gynaecol2001;108:697-700.

27. Hay PE, Lamont RF, Taylor-Robinson D, Morgan DJ, Ison C,Pearson J. Abnormal bacterial colonisation of the genital tractand subsequent preterm delivery and late miscarriage. BMJ1994;308:295-8.

28. Kurki T, Sivonen A, Renkonen O-V, Savia E, Ylikorkala O. Bacte-rial vaginosis in early pregnancy and pregnancy outcome. ObstetGynecol 1992;80:173-7.

29. Goldenberg RL, Iams JD, Miodovnik M, Van Dorsten JP, Thur-nau G, Bottoms S, et al. The preterm prediction study: risk fac-tors in twin gestations. Am J Obstet Gynecol 1996;175:1047-53.

30. Wennerholm U-B, Holm B, Mattsby-Baltzer I, Nielsen T, Platz-Christensen JJ, Sundell G, et al. Interleukin-1�, interleukin-6and interleukin-8 in cervico/vaginal secretion for screening ofpreterm birth in twin gestation. Acta Obstet Gynecol Scand1998;77:508-14.

31. Goldenberg RL, Klebanoff M, Carey JC, MacPherson C, Na-tional Institute of Child Health and Human Development Ma-ternal-Fetal Medicine Units Network. Metronidazole treatmentof women with a positive fetal fibronectin test result. Am J ObstetGynecol 2001;185:485-6.

32. Nugent RP, Krohn MA, Hillier SL. Reliability of diagnosing bac-terial vaginosis is improved by a standardized method of Gramstain interpretation. J Clin Microbiol 1991;29:297-301.

33. Spiegel CA, Amsel R, Holmes KK. Diagnosis of bacterial vagi-nosis by direct Gram stain of vaginal fluid. J Clin Microbiol1983;18:170-7.