Cerclage Journal Pubmed- 2007

Seediscussions,stats,andauthorprofilesforthispublicationat:http://www.researchgate.net/publication/227526576

Systematicreview:Cervicalstitch(cerclage)forpreventingpregnancyloss:individualpatientdatametaanalysisARTICLEinBJOGANINTERNATIONALJOURNALOFOBSTETRICS&GYNAECOLOGYNOVEMBER2007ImpactFactor:3.86DOI:10.1111/j.1471-0528.2007.01515.xSource:PubMed

CITATIONS30

DOWNLOADS77

VIEWS132

5AUTHORS,INCLUDING:

AndreaJorgensenUniversityofLiverpool41PUBLICATIONS933CITATIONS

SEEPROFILE

CatrinTudurSmithUniversityofLiverpool84PUBLICATIONS2,199CITATIONS

SEEPROFILE

Availablefrom:CatrinTudurSmithRetrievedon:24June2015

Cervical stitch (cerclage) for preventing pregnancyloss: individual patient data meta-analysisAL Jorgensen,a Z Alfirevic,b C Tudur Smith,a PR Williamsona; on behalf of the cerclage IPD

Meta-analysis Groupa Centre for Medical Statistics and Health Evaluation, University of Liverpool, Liverpool, UK b Department of Obstetrics and Gynaecology,

University of Liverpool, Liverpool Womens Hospital, Liverpool, UK

Correspondence: Mrs AL Jorgensen, Centre for Medical Statistics and Health Evaluation, University of Liverpool, Shelleys Cottage,

Brownlow Street, Liverpool L69 3GS, UK. Email [email protected]

Accepted 6 August 2007. Published OnlineEarly 28 September 2007.

Background Several observational studies have claimed high

success rates for cerclage in women with cervical insufficiency. A

recent Cochrane review found no conclusive evidence of benefit,

although significant heterogeneity was present for some of the

important clinical outcomes.

Objectives We undertook an individual patient data (IPD) meta-

analysis to examine effect of cerclage on neonatal and maternal

outcomes. In an attempt to explain the heterogeneity, we

investigated whether obstetric factors including multiple gestation

are associated with effectiveness.

Search strategy Search methods described in the original

Cochrane review were adopted and updated to December 2005.

Selection criteria This IPD systematic review and meta-analysis

was of randomised trials comparing cervical cerclage during

pregnancy with expectant management or no cerclage in women

with confirmed or suspected as having cervical insufficiency.

Analysis Multilevel logistic regression models stratified by trial

with random treatment effects were fitted to investigate the impact

of obstetric factors and multiple gestation on treatment effect.

Primary outcome measures were pregnancy loss or death before

discharge from hospital and absence of neonatal morbidity.

Main results The meta-analysis included seven trials and 2091

randomised women. In singleton pregnancies, the reduction in

pregnancy loss or death before discharge from hospital following

cerclage failed to reach statistical significance (OR 0.81; 95% CI

0.601.10). Cerclage was found to have a detrimental effect on the

outcome of pregnancy loss or death before discharge from hospital

in multiple gestations (OR 5.88; 95% CI 1.1430.19), although

only a small number of multiple pregnancies were included in the

analysis. Neither indication for cerclage nor obstetric history was

found to have a statistically significant impact on the effect

of cerclage.

Conclusions Cerclage may reduce the risk of pregnancy loss or

neonatal death before discharge from hospital in singleton

pregnancies thought to be at risk of preterm birth, but further

large trials are needed to elucidate the risk-benefit ratio precisely.

Cerclage in multiple pregnancies should be avoided. The efficacy of

cerclage was not influenced by either indication for cerclage or

mothers obstetric history.

Keywords Cervical cerclage, cervical stitch, individual patient

data meta-analysis, neonatal death, neonatal morbidity, neonatal

mortality, preterm delivery, pregnancy loss, randomised

clinical trials.

Please cite this paper as: Jorgensen A, Alfirevic Z, Tudur Smith C, Williamson P; on behalf of the cerclage IPD Meta-analysis Group. Cervical stitch (cerclage) for

preventing pregnancy loss: individual patient data meta-analysis. BJOG 2007;114:14601476.

Introduction

Cervical cerclage is a surgical procedure involving suturing

the neck of the womb (cervix) with a purse type stitch to keep

the cervix closed during pregnancy. This has been used widely

in the management of pregnancies considered at high risk of

preterm birth.

Several observational studies in the past 50 years have

claimed high rates of successful pregnancy outcome in

women with poor obstetric history attributed to cervical

insufficiency in whom cerclage was used. A recent Cochrane

review of randomised trials analysing outcomes includ-

ing miscarriage, perinatal loss, maternal infection, maternal

morbidity, antepartum haemorrhage and preterm birth

found no conclusive evidence of such benefit.1 However,

significant statistical heterogeneity was present for some of

the important clinical outcomes. This heterogeneity was

attributed to the inconsistency in clinical definitions

1460 2007 The Authors Journal compilation RCOG 2007 BJOG An International Journal of Obstetrics and Gynaecology

DOI: 10.1111/j.1471-0528.2007.01515.x

www.blackwellpublishing.com/bjogSystematic review

employed in the trials (e.g. varying cutoff points for defining

preterm birth) and in the different patient populations stud-

ied, however, neither meta-regression nor subgroup analyses

was undertaken.

Practically, methods of undertaking meta-analyses involve

collecting either aggregate data or data on each woman indi-

vidually. The advantages of the latter approach, known as an

individual patient data (IPD) meta-analysis and described as

the yardstick,2 include the potential to ensure a more con-

sistent definition of outcomes across trials, as well as a more

powerful analysis of whether treatment is more or less effec-

tive in particular subgroups.3 Although previous subgroup

analyses4 suggested that cerclage would be of benefit to the

subgroup of women with three or more second trimester

miscarriages or preterm births, the number of women con-

tributing to each obstetric history subgroup, and hence the

power to detect treatment effect within each, was small.

An IPD meta-analysis investigating the effects of cervical

cerclage has previously been published.5 This analysis included

only women found to have short cervix on ultrasound and did

not investigate the effect of cerclage on neonatal morbidity.

The data were analysed as although obtained from a single

large trial recruiting from the same population, and an

assumption of homogeneity of treatment effect between trials

was made. Our work includes methods of random-effects

meta-regression and multilevel logistic regression models to

detect and allow for heterogeneity of effect. Widening our

inclusion criteria to include trials that recruited based on

obstetric history and using the aforementioned analysis tech-

niques enabled us to not only examine the effect of cervical

cerclage in the general population of women at risk of pre-

term birth but also to investigate the impact that previous

obstetric history or cervical length may have on this effect.

Methods

We undertook an IPD meta-analysis to examine the effect

of cerclage on our prespecified neonatal and maternal

outcomes.6

SearchingThe search methods described in the original Cochrane

review1 were adopted and updated to December 2005.

Selection and study characteristicsThe types of studies considered for inclusion in the analysis

were randomised trials comparing cervical cerclage during

pregnancy (Shirodkar technique, McDonald technique, trans-

abdominal and transvaginal methods), with expectant

management or no cerclage in women with confirmed or

suspected as having cervical insufficiency. Quasi-randomised

studies in which allocation was transparent (e.g. use of alter-

native allocation or medical record numbers) were excluded.

Data abstraction and validity assessmentTwo reviewers independently assessed inclusion eligibility of

trials with any difference of opinion being resolved through

discussion. The methodological quality of each trial was

assessed in terms of method of generating randomisation

sequence, method of allocation concealment and potential

impact of losses to follow up. For each eligible trial, we requested

data on trial methods, treatment allocation, patient character-

istics and outcome data. The data provided were cross-checked

against any published report of the trial, and where possible, the

chronological randomisation sequence was reviewed, as was the

balance of prognostic factors at baseline. Any queries were fol-

lowed up with a nominated individual.

The primary outcomes were pregnancy loss or death before

discharge from hospital and absence of neonatal morbidity,

and secondary outcomes were preterm delivery (PTD) and

maternal morbidity. The impact of obstetric history and cer-

vical length on the effect of cervical cerclage was also assessed.

We asked trialists to provide all outcome data collected and

not just those reported in publications to avoid bias due to

within-study selective reporting.7

Standardising pregnancy loss or death beforedischarge from hospital outcome across trialsThe primary outcome was pregnancy loss or neonatal death

before discharge from hospital. This outcome includes all mis-

carriages, stillbirths and neonatal deaths before discharge, and

the IPD available for each trial were standardised as summar-

ised in Table 1. The use of a composite outcome appeared

justifiable here on the grounds that all events lead to the loss

of a baby, the prevention of which is the ultimate goal of using

cervical cerclage. The composite outcome was defined in accor-

dance with ICH E9 guidelines since analysing the outcomes

separately would not be addressing the primary question of

interest.14 It was not possible to analyse the outcome pregnancy

loss or neonatal death at any time since most trials4,8,9,11,13

followed up to hospital discharge only. For trials where length

of follow up was unclear,10 we assumed that follow up was to

hospital discharge only. Furthermore, it was confirmed that

although follow up continued after hospital discharge, all

deaths in the trial of Rust et al.12 occurred before discharge.

In the trials of To et al.,13 Berghella et al.,9 MRC,4 Rust

et al.,12 Rush et al.11 and Althuisius et al.8 only viable preg-

nancies were included. This could not be directly confirmed

for the trial of Ezechi et al.,10 and so an assumption had to be

made that this was indeed the case.

Standardising neonatal morbidity outcomeacross trialsDiffering neonatal morbidity outcomes were recorded in the

trials, and so an alternative outcome of baby healthy when

discharged from hospital was chosen, representing the absence

Cervical cerclage for preventing pregnancy loss

2007 The Authors Journal compilation RCOG 2007 BJOG An International Journal of Obstetrics and Gynaecology 1461

Table

1.Stan

dardisingoutcomemeasuresacross

trials

Althuisiusetal.8

Berghellaetal.9

Ezechietal.10

MRC4

Rush

etal.11

Rust

etal.12

Toetal.13

Neo

natal

mortality

Intrau

terinefetal

deaths(IU

Fs)

andneo

natal

deathswere

recorded

.No

inciden

cesofIUFs.

Neo

natalmortality

therefore

included

anyneo

nataldeaths

Neo

nataldeaths

wererecorded

.Th

is

included

miscarriages,

stillbirths

anddeathsafter

birth.Weclassified

as:miscarriages:

neo

nataldeathsat

,24weeks1

noNICU;

stillbirths:neo

natal

deathsat

24weeks

1noNICU;neo

natal

deaths:neo

natal

death1

NICU

Stillbirthsonly

wererecorded

.

Apparen

t

from

published

pap

erthat

totalp

erinatal

deathseq

ual

tototal

stillbirths.We

classified

as:miscarriages:

stillbirth

at,24

weeks;stillbirths:

stillbirth

at2

4

weeks;neo

natal

deaths:none

Liveborn,viab

le;

liveb

orn,dead

andstillbirth/

abortionwere

recorded

.We

classified

as:miscarriages:

stillbirth/abortion

at,24weeks;

stillbirths:

stillbirth/abortion

at2

4weeks;

neo

nataldeaths:

liveb

orn,dead

Miscarriages,stillbirths

andneo

nataldeaths

allw

ererecorded

specifically

withsame

classificationas

forthis

meta-an

alysis

Perinataldeaths

wererecorded

.

Notpossible

toclassify

into

subcategories

of

miscarriages,

stillbirthsan

d

neo

nataldeaths

butthiscomposite

outcomewas

sufficien

tfor

ourprimary

outcome

Stillbirthsan

dwhether

bab

ywas

alive

atfollow

uprecorded

.

Weclassified

as:

miscarriages:stillbirth

at,24weeks;

stillbirths:stillbirth

at

24weeks;neo

natal

death:notastillbirth

andnotaliveat

follow

up

Neo

natal

morbidity

Necrotising

enterocolitis,

RDS,

IVHan

d

neo

natalsepsis

wererecorded

.

Trialistsconfirm

ed

notnecessarily

case

that

bab

y

was

healthyat

dischargeifall

thesepathologies

reported

neg

ative.

Hen

ce,trialexcluded

from

analysisof

thisoutcome

IVH,RDS,

NEC

andsepsiswere

recorded

.Trialists

confirm

edthat

ifall

thesemarked

neg

ative,

can

assumebab

ywas

healthyat

discharge

Notavailable

Notrecorded

Anyserious

complicationsof

prematurity

wererecorded

,an

d

soifnone

was

recorded

,

canassume

bab

ywas

healthy

atdischarge

Perinatalmorbidity

was

recorded

in

term

sof

seriousnessof

complications.

Bab

ycountedas

healthy

atdischargeifit

did

notsuffer

from

anyserious

complicationsof

prematurity

IVH,positive

bloodcultures,

retinopathyof

prematurity

andBPD

wererecorded

.

Trialistsconfirm

ed

that

ifallthesemarked

neg

ative,

canassume

bab

ywas

healthy

atdischarge

Spontaneo

us

labour

Notrecorded

Dataonindication

fordeliverywere

collected

.Allwomen

markedspecifically

as

spontaneo

uslabour

werecountedas

having

spontaneo

uslabour

Unclearif

recorded

ornot

Spontaneo

us

labourstatus

was

recorded

directlyin

trial

Spontaneo

us

labourstatus

was

recorded

directlyin

trial

Notrecorded

Typeoflabour

was

recorded

directly.

Allwomen

marked

asspontaneo

us

countedas

having

spontaneo

uslabour

(continued

)

Jorgensen et al.


Table

1.(Continued

)

Althuisiusetal.8

Berghellaetal.9

Ezechietal.10

MRC4

Rush

etal.11

Rust

etal.12

Toetal.13

Pyrexia

Notrecorded

Notrecorded

Unclearif

recorded

ornot

Whether

woman

had

temperature

of.38recorded

directly

Temperature

of

mother

recorded

.

If.38,we

classified

aspyrexia.

Datadoes

notmatch

published

report

(onemore

ineach

treatm

entgroup).

Analysisisbased

onIPDdataheld

Notrecorded

Whether

mother

had

feverornotwas

recorded

directly

Chorioam

nionitis

Recorded

directly

Notrecorded

Unclearif

recorded

ornot

Recorded

only

incerclagegroup

asad

verseeven

t

from

interven

tion.

Asnotrecorded

incontrolg

roup

also

dataexcluded

from

analysis

Notrecorded

Directlyrecorded

Notrecorded

PPROM

Recorded

directly

Recorded

directly

Unclearif

recorded

ornot

Recorded

only

incerclagegroup

asad

verseeven

t

from

interven

tion.

Asnotrecorded

incontrolg

roup

also

dataexcluded

from

analysis

Recorded

directly.

Thedatadoes

notmatch

published

report(oneless

in

cerclagegroupin

datathan

inpublished

report).Analysisis

based

onIPDheld

Recorded

directly

Recorded

directly

Needfor

induction

and/orneed

forplanned

caesarean

Methodof

deliverywas

recorded

but

typeoflabour

notso

trial

excluded

from

analysisofthis

outcome

Indicationfordelivery

andmethodof

deliverywere

recorded

;however,

nodistinctionwas

mad

ebetween

elective

and

emergen

cy

caesareansection.

Therefore,

trialexcluded

from

analysisofthis

outcome

Unclearif

recorded

ornot

Spontaneo

uslabour

andmethodof

deliverywere

recorded

.Women

classedas

not

havingspontaneo

us

labourorhaving

spontaneo

uslabour

followed

by

emergen

cyor

elective

caesarean

classified

asyes

forthisan

alysis

Spontaneo

uslabour,

inducedlabour,

emergen

cy

caesareanan

d

elective

caesarean

wererecorded

.

Women

having

either

induced

labour,em

ergen

cy

caesareanorelective

caesareanclassified

as

yesforthisan

alysis

Methodofdelivery

was

recorded

but

typeoflabournotso

trialexcluded

from

analysisofthis

outcome

Indicationfor

deliveryan

dmethod

ofdeliverywere

recorded

;however,

nodistinctionwas

mad

ebetweenelective

andem

ergen

cy

caesareansection.

Therefore,trial

excluded

from

analysisofthis

outcome

BPD

,bronchopulm

anarydysplasia;

IVH,intraven

tricularhaemorrhag

e;NEC

,nectrotisingen

terocolitis;NICU,neo

natal

intensive

care

unit;RDS,

respiratory

distresssyndrome.



of any detectable neonatal morbidity at discharge. Table 1

summarises the data recorded on this outcome in each trial,

and how these were classified for the purposes of this analysis.

The lead author for the trials of To et al.,13 Berghella et al.,9

Rust et al.12 and Rush et al.11 confirmed that if a baby had

suffered from any morbidity at all then this would have been

recorded. For the trial of Althuisius et al.,8 only neonatal diag-

noses specifically requested were recorded if present. It does

not automatically follow that a baby with none of these specific

diagnoses was necessarily healthy at discharge; therefore, the

data from this trial were excluded. Neonatal morbidity data

were not collected in the trial of MRC,4 and confirmation

either way has not been obtained for the trial of Ezechi

et al.,10 hence these two trials have also been excluded.

As the outcome of interest was baby healthy when dis-

charged from hospital, for the trials where it was certain or

there was a possibility that follow up continued after

discharge (Rust et al.),12 we made the assumption that any

neonatal morbidity recorded first occurred prior to discharge

from hospital.

The analysis was two-fold: first, a composite outcome was

analysed, the event of interest being defined as not suffering

from any of the following: miscarriage, stillbirth, neonatal

death before discharge from hospital or some pathology

recorded. As well as making sense clinically, this approach

ensured our analysis included all women randomised, thus

the balance achieved from randomisation was preserved. Sec-

ond, an analysis was undertaken where all miscarriages, still-

births and neonatal deaths prior to discharge were omitted.

Hence, in this second analysis, only babies still alive at dis-

charge were included and so enabled conclusions to be drawn

relating to neonatal morbidity conditional on survival.

Standardisation of maternal morbidity andother outcomes across trialsThe maternal morbidity outcomes of pyrexia and chorio-

amnionitis were analysed. Preterm prelabour rupture of

membranes (PPROM), spontaneous labour and need for

induction or a planned caesarean were also examined. Table 1

summarises the data recorded on these outcomes.

Treatmentcovariate interactionsAs mentioned above, one of the aims of our study was to inves-

tigate whether a womans obstetric history influenced the effect

of cervical cerclage. For this purpose, women were categorised

into one of the following mutually exclusive categories:

1 No previous PTD or second-trimester loss (STL) and no

previous cervical surgery.

2 One previous PTD or STL and no previous cervical surgery.

3 Two previous PTDs or STLs and no previous cervical surgery.

4 Three or more previous PTDs or STLs and no previous

cervical surgery.

5 Previous cervical surgery.

These categories were chosen to reflect the subgroup analy-

ses undertaken in the MRC trial4 since this trial had found

a significant treatment effect (P < 0.05) on the outcome of

PTD before 33 weeks of gestation in a subgroup of women

with no previous cervical surgery but with three or more

previous PTDs or STLs.

It was possible to undertake this categorisation in five4,8,1113

out of the seven included trials. For the trial of Ezechi et al.,10

cervical surgery history was not recorded and the numbers of

previous PTDs or STLs were not recorded separately in the

database available from the trial of Berghella et al.9 Hence,

these two trials were excluded from the analyses of interaction

between obstetric history and cerclage.

We were also interested in investigating whether a womans

cervical length influenced the effect of cerclage, and this was

possible again for five8,9,1113 of the seven included trials.

Statistical analysisA study protocol6 and detailed statistical analysis plan (avail-

able on request from first author) were prepared in advance.

All analyses were conducted according to the analysis plan,

and the intention-to-treat principle was applied as far as

possible.

Clinical heterogeneity was assessed by reviewing differences

across trials in characteristics of randomised women. Statis-

tical heterogeneity was assessed using forest plots, the I2 sta-

tistic and chi-square test as set out in the analysis plan. The I2

statistic estimates the proportion of total variability in effect

estimates that can be explained by heterogeneity. Pooled odds

ratios were calculated using Petos method.15 Since the trials

could be partitioned into two distinct groups with respect to

what the main indication was for intervention of cerclage

(either short cervix on ultrasound or obstetric history),

meta-regression incorporating a trial-level covariate repre-

senting main indication was also undertaken to investigate

whether this accounted for any observed heterogeneity.

To examine the impact that a womans obstetric history

and cervical length may have on the effect of cerclage, in

addition to accounting for some of the observed heterogene-

ity, regression models were built stratified by trial. These were

two-level logistic regression models16 as explained in greater

detail in the analysis plan, and the models were fitted using

version 2.02 of the MLwiN software package for multilevel

modelling. In summary, the models included a fixed-effects

indicator variable for each trial to account for any trial-

specific characteristics. An indicator variable was also included

to represent treatment group; however, this was a random-

effects variable since it is assumed that treatment effect will be

similar, although not identical, across trials. Fixed-effect indi-

cator variables to represent both treatmentobstetric history

and treatmentcervical length interaction effects were also

introduced to the models to examine the impact of these

two obstetric factors on treatment effect. To assess the effect

Jorgensen et al.


of a variable on outcome, models both with and without that

variable were compared using the likelihood ratio test.

Where IPD were not available, the reason was assessed for

the potential of bias. Results using aggregate data from these

trials were then compared with results using aggregate data

from trials where IPD had been supplied. The analysis plan

was reviewed in light of the availability of IPD but prior to any

comparative analyses.

Multiple pregnanciesTwin or triplet pregnancies have been excluded from the main

analyses because: (a) the prognosis for PTD and associated

health problems is considered to be different among twins/

triplets and (b) outcomes for such babies are not deemed to

be independent of one another. However, to investigate the

impact of multiple gestation on treatment effect for neonatal

outcomes, data on all babies (from singleton, twin and triplet

pregnancies) were used to fit three-level logistic regression

models. These models included treatment effect assumed to

be random at both the trial and the mothers level, a binary

covariate representing multiple gestation and also a treat-

mentmultiple gestation interaction term. Similar models

but these times limited to two levels with treatment assumed

random only at the trial level were also fitted to assess the

impact of multiple gestation on maternal outcomes. For each

outcome, a Wald test to assess the statistical significance of

including the interaction term was undertaken to assess

whether the effect of cerclage on outcome is indeed different

in multiple pregnancies.

Data for multiple gestation were available for 66 mothers

and 138 babies (Berghella et al.9: 4 twin pregnancies, MRC4:

28 twin pregnancies, Rust et al.12: 28 twin pregnancies and

6 triplet pregnancies).

Women entered into the trials more than onceIt was apparent that women were entered more than once into

two trials (Rust et al.12: three women entered twice, MRC4:

exact number entered more than once unknown), and there

was a possibility that some women in the trial of Rush et al.11

were also entered more than once. No woman was entered

more than once into the trials of To et al.,13 Berghella et al.9

and Althuisius et al.,8 and we have not obtained confirmation

either way regarding the trial of Ezechi et al.10 Since it was not

always clear which women were entered more than once, we

have assumed that all pregnancies are independent regardless

of the fact that in some instances the same woman contrib-

uted with more than one pregnancy.

Results

Description of studiesThe search identified 17 potential trials, and overall agree-

ment between reviewers on eligibility was good. There was

some debate between reviewers regarding the eligibility of the

trial of Kassanos et al.17 However, since women randomised

to the no cerclage group in this trial were initially followed

up weekly with vaginal ultrasonograms with the possibility of

cerclage if a short cervix was found, it was decided that these

control women were not comparable with those in other

included trials, and the trial was excluded on this basis.

In total, nine trials were identified as being eligible for

inclusion, all published. Table 2 describes these trials and

summarises the results of assessing their methodological qual-

ity. For the trials where randomisation procedure was explic-

itly clarified,4,8,9,1113,19 the methods described were robust,

although for the majority of these trials we did not have

sufficient information to check that the methods had been

applied correctly. On inspecting key baseline characteristics

(Table 3), however, these appeared well balanced between the

two treatment groups for all trials. Due to the nature of the

intervention, it was not possible to blind patients or clinicians

to treatment for any of the trials.

Although the cerclage intervention varied with seven trials

using a McDonald type suture,812,18,19 one trial using a

Shirodkar type13 and one trial using a combination of more

than one type of suture4 undertaking a meta-analysis was

deemed appropriate. For two of the eligible trials, the authors

subsequently confirmed that IPD were no longer available,

and hence these trials have been excluded from our IPD

analyses (Lazar et al.,19 Dor et al.18).

Of the seven trials for which IPD were available, the main

indication for cerclage was the detection of short cervix on

ultrasound in four trials (Althuisius et al.,8 Berghella et al.,9

Rust et al.12 and To et al.13) and obstetric history in the

remaining three trials (Ezechi et al.,10 MRC4 and Rush

et al.11) For ease of interpretation, the forest plots have been

ordered such that the four trials where main indication was

ultrasound appear at the top, with the remaining three trials

appearing at the bottom.

Details of the eight excluded trials can be seen in the Quorum

diagram, Figure 1. A further three trials have been identi-

fied as continuing and therefore have not been included in

this analysis (Shennan A., pers. comm.; CIRCLE trial;26 Owen,

www.clinicaltrials.gov/;27 Silver, www.enh.org/).28

Baseline characteristicsA summary of baseline characteristics for the two treatment

groups in each trial can be seen in Table 3. Generally, most

characteristics were well balanced between the two interven-

tion groups within trials, and any small imbalances observed,

as well as those between trials, were given consideration when

accounting for any observed statistical heterogeneity.

Replicating published results from IPDThe IPD analysis replicated the published data by Althuisius

et al.,8 Berghella et al.,9 MRC4 and To et al.13 For the trial of



Table

2.Characteristicsofincluded

studies

Study

Randomisation

procedure/allocation

concealm

ent

Resultsofchecking

randomisation

procedure

Blinding?

Follow

upafter

hospital

discharge?

Location

ofstudy

Inclusioncriteria

Intervention

Primary

outcomes

Althuisiuset

al.8

Balan

cedblocks

stratified

fordifferent

inclusioncriteriaan

d

twoparticipating

hospitals.Allocation

bytelephone

Notpossibledate

ofrandomisation/

randomisation

number

not

provided

No

No

TheNetherlands

Singletonpregnan

cies;

considered

athigh

risk

ofPTDbecau

se

cervicallength

,25mm

before

27

weeks

ofgestation

McD

onaldtype

suture

with

bed

restvs

bed

restalone

PTD,34weeks

ofgestation;

neo

natalsurvival;

neo

natalmorbidity

Berghellaet

al.9

Computer-gen

erated

balan

cedblocks.

Allocationby

sequen

tially

numbered

opaq

ue,

sealed

envelopes

Firstblock

imbalan

ced.Authors

confirm

edan

overlookederror

No

No

USA

Either

athigh

risk

ofPTDbased

onpreviousobstetric

history

andiden

tified

duringultrasound

screen

ingbetween14

and23weeks

6days

ofgestationas

having

funnellingorashort

cervix;orat

low

risk

butfoundinciden

tally

tohaveshortcervix

McD

onaldtype

suture

with

bed

restvs

bed

restalone

Preterm

birth

,35weeks

Ezechietal.10

Notstated

Notpossibledate

ofrandomisation/

randomisation

number

not

provided

No

Notstated

Nigeria

One1

previousPTD

McD

onaldstype

suture

vsno

interven

tion

Gestational

ageat

delivery;PTD

MRC4

Balan

cedblocks

gen

erated

by

randomisation

service.

Allocation

bytelephoneorpost

Notpossible

tocheck

No

No

UK,Fran

ce,

Hungary,Norw

ay,

Italy,Belgium,

Zimbab

we,

South

Africa,

Icelan

d,

Irelan

d,Netherlands

andCan

ada

Obstetrician

uncertainwhether

ornotto

use

cervical

cerclagebecau

seof

previous:tw

oormore

second-trimester

miscarriages/PTD

s,

cervicalsurgery,

term

inationof

pregnan

cyorfirst-

trim

estermiscarriage;

orcurren

tcervical/

uterineab

norm

ality;

or

twin

pregnan

cies

Suture

vs

controlled

man

agem

ent.

More

than

one

typeofsuture

was

used

Length

of

pregnan

cy;vital

statusofbab

y

followingdelivery

(continued

)

Jorgensen et al.


Table

2.(Continued

)

Study

Randomisation

procedure/allocation

concealm

ent

Resultsofchecking

randomisation

procedure

Blinding?

Follow

upafter

hospital

discharge?

Location

ofstudy

Inclusioncriteria

Intervention

Primary

outcomes

Rush

etal.11

Computer-gen

erated

random

allocation.

Allocationin

opaq

ue,

sealed

envelopes

open

ed

byclinician

Notpossibledate

ofrandomisation/

randomisation

number

not

provided

No

Notstated

South

Africa

Two,three

orfourprevious

pregnan

cies

ended

spontaneo

usly

before

37weeks

aswellasoneormore

previous

pregnan

cyen

ded

spontaneo

usly

between14an

d36

weeks

ofgestation

McD

onaldtype

suture

vsno

suture

Gestationalag

e

atdelivery;

deliverybefore

37weeks

Rustet

al.12

Computer-gen

erated

random

allocation.

Allocationin

opaq

ueen

velopes

open

edat

patients

bed

side

Notpossibledate

ofrandomisation/

randomisation

number

not

provided

No

Yes

insome

cases

USA

Dem

onstrable

dilationofinternal

osan

deither

prolapse

of

mem

branes

of

.25%

total

cervicallength

or

distalcervicallength

of,2.5

cmbetween

16an

d24weeks

ofgestation

McD

onaldtype

suture

vsno

interven

tion

Gestational

ageat

delivery,

neo

natal

morbidity

Toet

al.13

Balan

cedblocks

stratified

bycentre.

Allocationby

telephone

Notpossible

randomisation

number

not

provided

No

No

UK,Brazil,

South

Africa,

Slovenia,Greece

andChile

Singletonpregnan

cies;

cervicallength

of

15mm

orless

atbetween

22weeks

and

24weeks

6days

Shirodkarsuture

vsexpectant

man

agem

ent

Deliverybefore

33weeks

Doret

al.18

Notstated

Notpossible

tocheck

No

Notspecified

Israel

Conceptionafter

inductionofovulation;

twin

pregnan

cies

McD

onaldtype

suture

vsno

suture

Durationof

pregnan

cy

Lazaret

al.19

Ran

domisation

procedure

not

stated

.Allocation

byway

ofsealed

envelopes

Notpossible

tocheck

No

Notspecified

Fran

ceScore

based

onobstetrichistory,

previouscervicalsurgery

history

andother

cervicalfactorsiswithin

aprespecified

range

McD

onaldtype

suture

vsno

suture

Obstetric

man

agem

ent;

durationof

pregnan

cy



Table

3.Comparingbaselinecharacteristicsacross

trials

Althuisiusetal.8

Berghellaetal.9

Ezechietal.10

MRC4

Rush

etal.11

Rust

etal.12

Toetal.13

Cerclage

Nocerclage

Cerclage

Nocerclage

Cerclage

Nocerclage

Cerclage

Nocerclage

Cerclage

Nocerclage

Cerclage

Nocerclage

Cerclage

Nocerclage

Treatm

ent

allocated

19(54%)

16(46%)

28(49%)

29(51%)

39(48%)

42(52%)

635(50%)

629(50%)

96(49%)

98(51%)

104(50%)

103(50%)

127(50%)

126(50%)

Compliantwith

treatm

ent

allocated:yes

19(100%)

14(88%)

27(87%)

28(93%)

Notstated

Notstated

586(92%)

581(92%)

95(99%)

97(99%)

104(100%)

103(100%)

122(96%)

124(98%)

Bed

rest:yes

19(100%)

16(100%)

28(100%)

29(100%)

Notstated

Notstated

227(36%)

(21missing)

168(27%)

(24missing)

9(9%)

3(3%)

104(100%)

103(100%)

0(0%)

0(0%)

Previous

cerclage:

yes

4(21%)

2(13%)

Notstated

Notstated

Notstated

Notstated

134(21%)

(1missing)

116(19%)

(2missing)

0(0%)

0(0%)

Notstated

Notstated

2(2%)

2(2%)

Previouscervical

surgery:yes

3(16%)

2(13%)

3(11%)

2(7%)

Notstated

Notstated

193(30%)

179(28%)

0(0%)

0(0%)

16(15%)

25(24%)

7(6%)

9(7%)

Funnelling:yes

10(53%)

11(69%)

Notstated

Notstated

Notstated

Notstated

Notstated

Notstated

Notstated

Notstated

Notstated

Notstated

121(95%)

117(93%)

Ethnicorigin:

Non-Cau

casian

:

yes

9(47%)

8(50%)

25(89%)

23(79%)

Notstated

Notstated

Notstated

Notstated

96(100%)

98(100%)

35(34%)

31(30%)

68(54%)

78(62%)

Smoker:yes

2(11%)

0(0%)

9(32%)

8(28%)

Notstated

Notstated

Notstated

Notstated

Notstated

Notstated

24(25%)

(7missing)

25(26%)

(7missing)

10(8%)

17(13%)

Fetalfi

bronectin:

yes

Notstated

Notstated

Notstated

Notstated

Notstated

Notstated

Notstated

Notstated

Notstated

Notstated

29(28%)

31(30%)

7(6%)

(1missing)

8(6%)

Bacterialvaginosis:

yes

6(32%)

4(25%)

Notstated

Notstated

Notstated

Notstated

Notstated

Notstated

Notstated

Notstated

19(18%)

20(19%)

13(10%)

(2missing)

12(10%)

(2missing)

Chlamydia:yes

Notstated

Notstated

Notstated

Notstated

Notstated

Notstated

Notstated

Notstated

Notstated

Notstated

1(1%)

1(1%)

Notstated

Notstated

Bishopscore

.4:yes

Notstated

Notstated

Notstated

Notstated

12(31%)

10(24%)

Notstated

Notstated

Notstated

Notstated

Notstated

Notstated

Notstated

Notstated

Meanag

eat

randomisation

(SD)

30.53(4.57)34.50(4.93)

27.81(6.4)

29.93(6.85)24.56(4.81)25.79(4.81)

27.69(5.07)

27.72(4.98)

Notstated

Notstated

28.03(6.10)

28.88(6.79)

29.85

(6.06)

29.30(5.90)

Meangestational

ageat

cerclage

procedure

(SD)

20.95(2.93)

N/A

Notstated

N/A

Notstated

N/A

Notstated

N/A

20.00.(1.41)

N/A

20.67(2.12)

N/A

23.85

(0.71)

N/A

Meancervical

length

(SD)

19.90(2.87)19.56(4.29)15.69(9.20)16.67(8.01)

Notstated

Notstated

Notstated

Notstated

16.47(3.71)18.42(2.92)

16.11(7.72)

17.59(6.24)

9.60(3.46)

9.33(3.57)

MeanBMI(SD

)Notstated

Notstated

Notstated

Notstated

Notstated

Notstated

Notstated

Notstated

Notstated

Notstated

Notstated

Notstated

26.45(5.49)

25.96(5.60)

Meangestational

ageat

entry(SD)

Notstated

Notstated

19.61(2.40)

19.03(2.2)

Notstated

Notstated

14.63(4.83)

14.89(5.10)

17.56(3.59)14.87(5.14)

20.67(2.12)

21.15(2.25)

23.52(0.69)

23.49(0.73)

Prim

igravida:

yes

10(53%)

9(56%)

7(23%)

12(40%)

0(0%)

0(0%)

Notstated

Notstated

0(0%)

0(0%)

14(13%)

13(13%)

32(25%)

33(26%)

(continued

)

Jorgensen et al.


Ezechi et al.,10 the numbers randomised to the cerclage and

no cerclage groups have been reported in the paper as 38 and

43, respectively, whereas in the IPD, the corresponding num-

bers in each group are 39 and 42 and the data have been

analysed as such. The IPD obtained for the trial of Rush

et al.11 were handwritten in pencil and, due to its age, some-

times difficult to read. It was, therefore, not possible to rep-

licate published results for some of the variables. Data on

Table

3.(Continued

)

Althuisiusetal.8

Berghellaetal.9

Ezechietal.10

MRC4

Rush

etal.11

Rust

etal.12

Toetal.13

Cerclage

Nocerclage

Cerclage

Nocerclage

Cerclage

Nocerclage

Cerclage

Nocerclage

Cerclage

Nocerclage

Cerclage

Nocerclage

Cerclage

Nocerclage

Previousdelivery

atgreater

than

37

weeks:yes

8(42%)

3(19%)

Notstated

Notstated

Notstated

Notstated

Notstated

Notstated

44(46%)

37(38%)

40(38%)

34(33%)

57(45%)

49(39%)

PreviousSTL:yes

Notstated

Notstated

15(54%)

13(45%)

Notstated

Notstated

285(45%)

(1missing)

260(41%)

(2missing)

78(81%)

79(81%)

33(32%)

19(18%)

Notstated

Notstated

Previousearly

spontaneo

us

loss:yes

4(21%)

5(31%)

20(71%)

20(69%)

Notstated

Notstated

201(32%)

(1missing)

193(31%)

(2missing)

47(49%)

61(62%)

40(38%)

42(41%)

54(43%)

61(48%)

(2missing)

PreviousPTD:yes

Notstated

Notstated

19(68%)

16(55%)

39(100%)

42(100%)

277(44%)

(1missing)

258(41%)

(2missing)

40(42%)

32(33%)

Notstated

Notstated

69(54%)

76(61%)

BMI,bodymassindex.

Potentially eligible studies identified bysearches (duplicates removed) n = 17

Studies excluded sincecomparison of cerclagetechnique only (n =1)

(Caspi et al.22)

Studies excluded sincewomen already included in

another of the included trials(n = 2) (Szeverenyi et al,24

Althuisius et al.20)

Studies retrieved (n =16)

Studies excluded sincecomparison of cerclage vs

pessary (n = 1)(Foster et al.23)

Studies retrieved (n = 15)


Studies excluded sincewomen not randomised

(n = 1) ) (Varma T.R., pers.comm.)


Studies excluded sincecontrol group

subsequently received elective cerclage (n = 2) )(Kassanos et al.17, Beigi

and Zarrinkoub25)


Studies excluded sincecomparison of inpatient vsoutpatient cerclage (n =1)

(Blair et al.21)

Studies included in the review (n = 9)

Figure 1. Quorum diagram.



such variables have been excluded from the analyses, with the

exception of maternal pyrexia and PPROM for which there

were very small discrepancies (Table 1). We did not attempt

to replicate the results for the trial of Rust et al.12 since the

most recent paper published included only a subset of the

women available for IPD.

Pregnancy loss or death before dischargefrom hospitalSingleton pregnancies onlyThe trial-specific odds ratios, together with the pooled odds

ratio and corresponding 95% CI are displayed in Figure 2.

These figures suggest little heterogeneity in treatment effect

across trials. The result of the meta-regression, introducing

a covariate representing main indication for cerclage (obstet-

ric history versus short cervical length) was not statistically

significant (P = 061).Introducing interaction terms between treatment and

obstetric history in a two-level logistic regression model did

not have a significant effect. The same applies for a treatment

cervical length interaction term (Table 4).

Multiple gestationsIncluding a treatmentmultiple gestation interaction effect in

a three-level logistic regression model including data on all

babies gave a significant result (Table 4), suggesting that cerc-

lage has a detrimental effect on the outcome for such babies.

Calculating risk scores (data not shown) for babies grouped

into four categories depending on both singleton/multiple

pregnancy status and cerclage/no cerclage status demon-

strated that using cerclage in singleton pregnancies decreased

the risk of pregnancy loss or death before discharge from

hospital but that using cerclage in multiple pregnancies

increased the risk substantially.

Absence of neonatal morbiditySingleton pregnancies onlyThe trial-specific odds ratios, together with the pooled odds

ratios and corresponding 95% CI for the two analyses are

displayed in Figure 3. It should be noted, however, that three

trials, representing 66% of randomised women, were ex-

cluded from the analysis of this outcome. These figures sug-

gested no heterogeneity in treatment effect across trials.

Introducing a covariate representing indication for cerclage

in a meta-regression did not have a statistically significant

effect (P value including all babies: 064; P value excludingbabies not alive at discharge: 044).

When fitting two-level logistic regression models, intro-

ducing a treatmentobstetric history term did not have a sta-

tistically significant effect (Table 4). The same applied for

a treatmentcervical length interaction (Table 4).

Multiple gestationsIncluding a treatmentmultiple gestation interaction effect in

a three-level logistic regression model gave a significant result

(Table 4) for the analysis including all babies and this suggests

that cerclage has a detrimental effect on this outcome for

such babies.

Calculating risk scores (data not shown) for babies grouped

into four categories depending on both singleton/multiple

pregnancy status and cerclage/no cerclage status demon-

strated that using cerclage in singleton pregnancies increased

the likelihood of a baby being healthy at discharge but that

using cerclage in multiple pregnancies decreased the likelihood

Figure 2. Forest plot comparing cerclage with no cerclage for outcome of pregnancy loss of death before discharge from hospital.

Jorgensen et al.


substantially. However, the test for an interaction was non-

significant (Table 4) when excluding babies not alive at dis-

charge from the analysis.

Maternal morbiditySingleton pregnancies onlyThe trial-specific odds ratios, together with the pooled odds

ratio and corresponding 95% CI for each outcome are dis-

played in Figure 4. These figures suggested that there was sig-

nificant heterogeneity in treatment effect for the outcomes of

chorioamnionitis and PPROM. On inspecting the forest plots

for these outcomes, treatment effect in the trial of Althuisius

et al.8 is noticeably different to the other trials; however, there

is no immediately apparent reason for this difference.

In a meta-regression model, indication for cerclage did not

have a statistically significant effect for any of the outcomes

(P value 0.36 or greater for all outcomes).

Finally, introducing treatmentobstetric history terms to

a logistic regression model did not have a significant effect

on any of the outcomes (Table 4), with similar nonsignificant

results for a treatmentcervical length interaction (Table 4).

Multiple gestationsThe results from including a treatmentmultiple gestation

interaction term in a two-level logistic regression model are

summarised in Table 4. There was insufficient data on mul-

tiple pregnancies for which the outcome of pyrexia had been

measured to undertake the test for this outcome.

Preterm birthWe were interested in investigating the effect of cervical

cerclage on the timing of preterm births. For cutoffs

between 16 and 37 weeks of gestation, pooled odds ratios

were calculated. An increased confidence level of 99% was

used to calculate the intervals for these multiple pooled

odds ratios (Figure 5). The effect estimates favoured no

cerclage for the earlier cutoff points and cerclage for the

later cutoffs, although the results do not reach statistical

significance.

Statistical significance of the impact of obstetric history and

cervical length on treatment effect for the outcome of preterm

births before all these cutoffs was also assessed by way of

logistic regression models. Neither of these two factors was

Table 4. Results from undertaking logistic regressions

Outcome Treatmentobstetric

history interaction*

Treatmentcervical

length interaction**

Treatmentmultiple gestation interaction

P value P value OR (95% CI) P value

Pregnancy loss

or death before

discharge from hospital

0.92 0.78 588 (1143019)*** 0.03***

Baby healthy

at discharge from

hospital (all babies)

0.69 0.71 012 (002089)*** 0.04***

Baby healthy

at discharge from

hospital

(babies alive at discharge only)

0.69 0.37 0.54 (0.070.48)*** 0.56***

Spontaneous labour 0.83 0.19 108 (022544)**** 0.92****Pyrexia 0.56 0.8 Insufficient data available

Chorioamnionitis 0.6 0.96 365 (0472817)**** 0.21****PPROM 0.44 0.32 157 (034728)**** 0.56****Need for induction/caesarean section 0.68 0.08 0.74 (016342)**** 0.70****

*The P values here are those obtained from undertaking a likelihood ratio test comparing a logistic regression model including treatmentobstetric

history interaction terms to a model without the interaction terms.

**The P values here are those obtained from undertaking a likelihood ratio test comparing a logistic regression model including treatmentcervical

length interaction term to a model without the interaction term.

***The odds ratios here are those obtained from fitting a multilevel logistic regression model with trial as the first level, woman as the second

level and baby as the third level. The model includes indicator variables to represent both treatment group (random effect) and multiple gestation

status (fixed effect) and also a treatmentmultiple gestation interaction variable. The P values are those obtained from undertaking a likelihood

ratio test to compare a model with the interaction variable to one without.

****The odds ratios here are those obtained from fitting a multilevel logistic regression model with trial as the first level and woman as the

second level. The model includes indicator variables to represent both treatment group (random effect) and multiple gestation status (fixed effect)

and also a treatmentmultiple gestation interaction variable. The P values are those obtained from undertaking a likelihood ratio test to compare

a model with the interaction variable to one without.



found to be statistically significant at the 1% level for any of

the gestational age cutoffs investigated.

There was not enough data on women in multiple preg-

nancies delivering before each gestational age cutoff to inves-

tigate the effect of multiple gestation on treatment effect for

this outcome (results not shown).

Comment on studies for which IPD werenot obtainedTwo studies were eligible for inclusion in this meta-analysis for

which the authors confirmed that IPD were no longer available

(Dor et al.18 and Lazar et al.19). The trial of Dor et al.18 included

women with twin pregnancies only and so these women would

not have formed part of our main analysis even if IPD had been

available. For the trial of Lazar et al.,19 the aggregate results for

the outcomes of induced labour or caesarean section, pre-

term delivery before 32 weeks of gestation, preterm delivery

before 36 weeks of gestation and preterm delivery before 37

weeks of gestation were all obtainable from the published

paper and therefore for these four outcomes a comparison

was made between pooled results both excluding and including

the aggregate results from this trial. The results were found to

be almost identical (results not shown).

Discussion

There continues to be considerable controversy about the value

of cervical cerclage in the management of women considered to

be at high risk of PTD. Our IPD review included trials where

main indication for cerclage was based on obstetric history, as

well as trials where the main indication was short cervical

length detected by ultrasound. The availability of IPD enabled

us to standardise outcome definitions across trials, which led to

an increase in the number of women contributing to each

outcome, and hence more precise effect estimates.

Although the overall results suggest that, in singleton preg-

nancies, cervical cerclage may reduce the risk of pregnancy

loss or death before discharge from hospital (OR 0.81), this

result did not reach statistical significance at the 5% level. The

true effect on the outcome of pregnancy loss or death before

discharge from hospital could range from a reduction in odds

of up to 40% in favour of cervical cerclage to an increase in

10% against the intervention. We believe that this trend

towards treatment benefit warrants further study. The confi-

dence intervals for the absence of neonatal morbidity were

much wider since only three trials collected sufficient infor-

mation on this outcome.

Figure 3. Forest plots comparing cerclage with no cerclage for outcome of (A) baby healthy when discharged from hospital (all babies); (B) baby healthy

when discharged from hospital (excluding babies not alive at discharge).

Jorgensen et al.


In terms of maternal morbidity, a statistically significant

increased risk of maternal pyrexia was observed in the cerc-

lage group. It was decided following publication of the pro-

tocol,6 but prior to analysis, that onset of labour was also of

interest since we wished to test the hypothesis that cervical

cerclage could damage the cervix and prevent spontaneous

labour or indeed cause morbidity that would force induction

or caesarean section. There was no significant evidence that

the likelihood of induction or caesarean section was higher in

the cerclage group. The data were quite limited because there

have been some difficulties in classifying women in terms of

this outcome for many of the trials (Table 1).

It is important to note that, although the trials included in

the review contributed data from over 2000 women in total,

the outcomes and covariates of interest were not recorded for

all women.

A previously published meta-analysis5 suggested that the

intervention of cervical cerclage in women with twin preg-

nancies increased the risk of preterm birth before 35 weeks of

gestation, although the number of women for which data was

available was small. Our analysis suggested that cerclage has

a detrimental effect on the outcome of pregnancy loss or

death before discharge from hospital and our composite

outcome of a baby being healthy at discharge, for multiple

Figure 4. Forest plots comparing cerclage with no cerclage for outcomes of maternal morbidity.



gestations. These results must be interpreted with caution

due to the relatively small number of women with a multiple

gestation for which data were available.

The focus in studies to date has been on investigating

whether cervical cerclage has the ability to prevent preterm

birth. However, increasing gestational age at delivery does not

necessarily mean an improvement in the babys outcome. For

example, a baby delivered at 35 weeks of gestation may not

necessarily fare better than a baby delivered a few weeks earlier

if spontaneous delivery was artificially delayed. Care should

always be taken to ensure that the gestational age of preterm

birth is not mistaken as a surrogate outcome for pregnancy

loss or death before discharge from hospital/neonatal mor-

bidity. It is for this reason that we chose pregnancy loss or

death before discharge from hospital and neonatal morbidity

as our primary focus. However, the timing of PTD is impor-

tant in its own right for the purpose of investigating other

hypotheses of interest relating to the use of cervical cerclage.

These hypotheses are as follows:

1 That cervical cerclage delays delivery only for a short period

of time.

2 That cervical cerclage is only effective in improving neo-

natal outcome where the risk of preterm birth is during

a specific time interval.

We undertook an exploratory analysis to investigate

whether the effect of cerclage varied according to gestational

age. On inspecting the point estimates for effect, there was

a suggestion of a change from favouring no cerclage to

favouring cerclage at around the 21-week cutoff point,

although the results did not reach statistical significance.

Due to the small number of events occurring at earlier ges-

tations, the confidence intervals are very wide. The analysis

was limited further by the fact that some women were not

recruited until they had reached a gestational age greater than

some of the earlier cutoff points, which meant that they had

to be excluded from the analysis of those cutoffs. Excluding

such women meant that the balance achieved from random-

isation was potentially disrupted. For these reasons, our

results must be treated with caution.

There is also a possibility that the stage of pregnancy at

which the cervical cerclage is administered may play a part in

how effective it will be. Indeed, the intervention may some-

times occur too late during the pregnancy to have any effect.

Gestational age of the cerclage procedure was recorded for

women in four trials.8,1113 We used these data to investigate

whether there was any association between gestational age

of the procedure and the outcome of pregnancy loss or

neonatal death before discharge from hospital by fitting

Figure 4. Continued from previous page.

Jorgensen et al.


a two-level logistic regression. No significant association was

found (P = 0.26).

Neither obstetric history nor cervical length was found to

have a significant impact on the effect of cerclage on PTD.

Our five obstetric history categories were purposefully chosen

to reflect the subgroup analyses undertaken in the MRC

trial4 since this trial had found a significant treatment effect

(P < 0.05) on the outcome of PTD before 33 weeks of gesta-

tion in a subgroup of women with no previous cervical sur-

gery but three or more previous PTDs or STLs. This result was

not confirmed in our analysis.

Similar analyses looking at the impact of obstetric history

and cervical length on cerclage effect were also undertaken for

the outcomes of pregnancy loss or death before discharge

from hospital, neonatal morbidity and maternal morbidity,

but no significant results were found.

We also found no evidence that the effect of cerclage in

trials where the main indication was short cervical length on

ultrasound was different from the effect in trials where indi-

cation was based on obstetric history alone.

Although it was apparent that some women were entered

into the trials of Rust et al.,12 Rush et al.11 and MRC4 more

than once, it was not always possible to identify them. For the

purpose of this review, it is therefore assumed that pregnancy

outcomes for the same woman are independent, although this

may have introduced a small amount of over-precision into

the results.

Implications for practice

Although the results for the outcome of pregnancy loss or

death before discharge from hospital in singleton pregnancies

appears promising, further research is required before any

conclusive advice can be provided with regard to the benefits

of using cervical cerclage to improve neonatal outcome.

Women should be advised of the increased risk of maternal

pyrexia and treated accordingly. Cerclage in multiple preg-

nancies should be avoided.

Implications for research

There is an urgent need for further large trials to elucidate the

risk-benefit ratio in singleton pregnancies with precision and

to identify groups most likely to benefit.

Conflicts of interest

Z.A. and P.R.W. were authors of a paper that is included in

the IPD meta-analysis.13 Z.A. was an author of the non-IPD

systematic review on this topic.1 The authors declare that they

do not have any other competing interests.

Contribution to authorship

A.L.J. organised, cleaned and checked the individual patient

data sets, contacted the authors with queries, wrote the sta-

tistical analysis plan, performed data validation checks and

statistical analyses and co-wrote the review.

Z.A. assessed eligibility and methodological quality of

trials, liaised with individual trialists, provided clinical guid-

ance and provided comments on the manuscript.

C.T.S. prepared the protocol, assessed eligibility and

methodological quality of the trials and provided comments

on the manuscript.

P.R.W. conceived the idea for undertaking the IPD meta-

analysis, supervised A.L.J. on all aspects of the review, pro-

vided advice on the statistical analysis plan and the statistical

analyses and provided comments on the manuscript.

Cerclage IPD meta-analysis group membersA.L. Jorgensen (Liverpool); Z. Alfirvec (Liverpool); C. Tudur

Smith (Liverpool); P.R. Williamson (Liverpool); S.M.

Althulsivus (William Harvey Hospital, Kent); V. Berghella

(Thomas Jefferson University, Philadelphia; O.C. Ezechi

(Nigerian Institute of Medical Research); MRC/RCOG work-

ing party; R.W. Rush (previously from University of Cape

Town); O.A. Rust (Lehigh Valley Hospital and Health Net-

work, Pennsylvania); M.S.T. (Fetal Medicine Foundation); K.

Nicolaides (Fetal Medicine Foundation).

Acknowledgements

The authors would like to thank the Fetal Medicine Founda-

tion, a registered UK charity, for providing some financial

support for the project and also the cerclage IPD meta-analysis

group for kindly providing the data, responding to the vari-

ous queries raised and providing valuable feedback on the

Figure 5. Odds ratios of preterm delivery comparing cerclage with no

cerclage.



draft paper. They would also like to thank Adrian Grant

(MRC/RCOG working party) who provided helpful com-

ments on an earlier draft. j

References

1 Drakeley AJ, Roberts D, Alfirevic Z. Cervical stitch (cerclage) for pre-

venting pregnancy loss in women. In: The Cochrane Library. Chichester,

UK: John Wiley & Sons Ltd, 2003.

2 Chalmers I. The Cochrane collaboration: preparing, maintaining and

disseminating systematic reviews of the effects of healthcare. Ann N Y

Acad Sci 1993;703:15665.

3 Stewart LA, Parmar MKB. Meta-analysis of the literature or of individ-

ual patient data: is there a difference? Lancet 1993;341:41822.

4 MRC/RCOG Working Party on Cervical Cerclage. Final report of the

Medical Research Council/Royal College of Obstetricians and Gynae-

cologists multicentre randomised trial of cervical cerclage. Br J Obstet

Gynaecol 1993;100:51623.

5 Berghella V, Odibo AO, To MS, Rust OA, Althuisius SM. Cerclage for

short cervix on ultrasonography: meta-analysis of trials using individual

patient-level data. Obstet Gynecol 2005;106:1819.

6 Tudur-Smith C, Jorgensen AL, Alfirevic Z, Williamson PR. Individual

patient data meta-analysis: cervical stitch (cerclage) for preventing

pregnancy loss in women. BMC Pregnancy Childbirth 2005;5:5.

7 Williamson PR, Gamble C. Identification and impact of outcome selec-

tion bias in meta-analysis. Stat Med 2005;24:154761.

8 Althuisius SM, Dekker GA, Hummel P, Bekedam DJ, van Geijn HP. Final

results of the cervical incompetence prevention randomized cerclage

trial (CIPRACT): therapeutic cerclage with bed rest versus bed rest

alone. Am J Obstet Gynecol 2001;185:110612.

9 Berghella V, Odibo AO, Tolosa JE. Cerclage for prevention of preterm

birth in women with a short cervix found on transvaginal ultrasound

examination: a randomised trial. Am J Obstet Gynecol 2004;191:

13111317.

10 Ezechi OC, Kalu BKE, Nwokoro CA. Prophylactic cerclage for the pre-

vention of preterm delivery. Int J Gynecol Obstet 2004;85:2834.

11 Rush RW, Isaacs S, McPherson K, Jones L, Chalmers I, Grant A. A ran-

domized controlled trial of cervical cerclage in women at high risk of

spontaneous preterm delivery. Br J Obstet Gynaecol 1984;91:72430.

12 Rust OA, Atlas RO, Reed J, van Gaalen J, Balducci J. Revisiting the short

cervix detected by transvaginal ultrasound in the second trimester:

why cerclage therapy may not help. Am J Obstet Gynecol 2001;185:

1098105.

13 To MS, Alfirevic Z, Heath VCF, Cicero S, Cacho AM, Williamson PR,

et al. Cervical cerclage for prevention of preterm delivery in women with

short cervix: randomised controlled trial. Lancet 2004;363:184953.

14 [www.ich.org/]. Accessed 1 June 2007.

15 Yusuf S, Peto R, Lewis J, Collins R, Sleight P. Beta blockade during and

after myocardial infarction: an overview of the randomized trials.

Prog Cardiovasc Dis 1985;27:33571.

16 Whitehead A. Meta-Analysis of Controlled Clinical Trials. Chichester,

UK: John Wiley & Sons Ltd, 2002.

17 Kassanos D, Salamalekis E, Vitoratos N, Panayotopoulos N, Loghis C,

Creatsas C. The value of transvaginal ultrasonography in diagnosis and

management of cervical incompetence. Clin Exp Obstet Gynecol

2001;28:2668.

18 Dor J, Shalev J, Mashiach S, Blankstein J, Serr DM. Elective cervical

suture of twin pregnancies diagnosed ultrasonographically in the first

trimester following induced ovulation. Gynecol Obstet Invest

1982;13:5560.

19 Lazar P, Gueguen S, Dreyfus J, Renaud R, Pontonnier G, Papiernik E.

Multicentred controlled trial of cervical cerclage in women at moderate

risk of preterm delivery. Br J Obstet Gynaecol 1984;91:7315.

20 Althuisius SM, Dekker GA, Hummel P, van Geijn HP. Cervical incom-

petence prevention randomized cerclage trial: emergency cerclage

with bed rest versus bed rest alone. Am J Obstet Gynecol 2003;189:

90710.

21 Blair O, Fletcher H, Kulkarni S. A randomised controlled trial of out-

patient versus inpatient cervical cerclage. J Obstet Gynaecol 2002;22:

4937.

22 Caspi E, Schneider DF, Mor Z, Langer R, Weinraub Z, Bukovsky I.

Cervical internal os cerclage: description of a new technique and com-

parison with shirodkar operation. Am J Perinatol 1990;7:3479.

23 von Forster F, During R, Schwarzlos G. Therapie der Zervixinsuffizienz-

Zerklage oder Stutzpessar. [Treatment of cervical incompetence- cerc-

lage or supportive therapy]. Zentralblatt fur Gynakologie 1986;108:

2307.

24 Szeverenyi M, Chalmers I, Grant AM, Nagy T, Nagy J, Balogh I, et al.

Operative treatment of the cervical incompetency during the pregnancy:

judgement of the competence of cerclage operation (translation).

Orvosi Hetilap 1992;133:18236.

25 Beigi A, Zarrinkoub F. Elective versus ultrasound-indicated cervical cerc-

lage in women at risk for cervical incompetence. Med J Islam Repub

Iran 2005;19:1037.

26 The CIRCLE TrialCerclage in Relation to Cervical Length. A clinical

study to determine the effects of cervical scanning and cervical stitches

in preventing preterm labour. ISRCTN52175236. London: Prof Andrew

Shennan, Maternal and Fetal Research Unit, St Thomas Hospital.

27 Owen J. Surgical procedure to prevent premature birth (ongoing trial).

[www.clinicaltrials.gov/]. Accessed 1 June 2007.

28 Silver R. Vaginal ultrasound cerclage trial (ongoing). [www.enh.org/].

Accessed 1 June 2007.

29 Althuisius SM, Dekker GA, van Geijn HP, Bekedam DJ, Hummel P.

Cervical incompetence prevention randomized cerclage trial

(CIPRACT): study design and preliminary results. Am J Obstet Gynecol

2000;183:8239.

Jorgensen et al.


Documents

Cerclage Journal Pubmed- 2007