American Journal of Obstetrics and Gynecology (2005) 193, 141–52

www.ajog.org

Perinatal outcomes of in vitro fertilization twins:A systematic review and meta-analyses

Sarah McDonald, MD,a,* Kellie Murphy, MD, MSc,b,d Joseph Beyene, PhD,d,e

Arne Ohlsson, MD, MScc,d

Departments of Obstetrics and Gynecology and Clinical Epidemiology, Ottawa Hospital, University of Ottawa,Ontario, Canadaa; Departments of Obstetrics and Gynecology,b and Pediatrics,c Mount Sinai Hospital;University of Torontod; Program in Population Health Sciences, Hospital for Sick Children,e

Toronto, Ontario, Canada

Received for publication August 16, 2004; revised November 16, 2004; accepted November 29, 2004

KEY WORDSIn vitro fertilizationTwinsMeta-analysis

Perinatal outcomeCase-control study

Objective: Uncontrolled studies suggest that in vitro fertilization twins have increased rates ofpreterm birth and low birth weight and would warrant increased antenatal monitoring. The

objective of this meta-analysis was to determine whether the incidence of poor obstetric outcomesis higher for in vitro fertilization twins than for spontaneously conceived twins who were matchedfor maternal age.

Study design: Medline and EMBASE were searched with comprehensive search strategies. Case-control and cohort studies of twins who were conceived by in vitro fertilization or in vitrofertilization/intracytoplasmic sperm injection, with the transfer of fresh embryos or cryopreserved

(frozen) in women with infertility, and/or whose partners were subfertile or infertile, comparedwith naturally (spontaneously) conceived twins who were matched for maternal age (case-controlstudies) or which were controlled for it (cohort studies). Two reviewers independently assessedtitles, abstracts, and study quality and extracted the data. Statistical analysis was performed with

commercial statistical software. Dichotomous data were meta-analyzed with odds ratios asmeasures of effect size, and continuous data was meta-analyzed with mean differences. Interstudyvariation was incorporated with the assumption of a random effects model for the treatment

effect.Results: Compared with spontaneously conceived twins who were matched for maternal age, invitro fertilization twins have an increased risk of preterm birth between 32 and 36 weeks of

gestation (odds ratio, 1.48; 95% CI, 1.05-2.10), and an elevated risk of preterm birth at !37weeks of gestation when parity is also matched for an odds ratio of 1.57 (95% CI, 1.01-2.44).There was an increased rate of cesarean delivery among in vitro fertilization twins (odds ratio,

1.33; 95% CI, 1.06-1.67). There were no significant differences in incidences of perinatal death,low birth weight infants, or congenital malformations.

* Reprint requests: Dr S.D. McDonald, OMNI Research Group, Obstetrics and Gynecology and Clinical Epidemiology Program, Ottawa

Hospital, General Campus 501 Smyth Rd, Box 241, Ottawa, ON K1H 8L6.

E-mail: samcdonald@ottawahospital.on.ca

0002-9378/$ - see front matter � 2005 Elsevier Inc. All rights reserved.

doi:10.1016/j.ajog.2004.11.064

142 McDonald et al

Conclusion: In vitro fertilization twins have increased rates of preterm birth compared withspontaneously conceived twins who were matched for maternal age, despite the fact that theiroutcomes would be expected to be better because of the decreased proportion of monochorionictwins.

� 2005 Elsevier Inc. All rights reserved.

In the quarter of a century that has followed the firstsuccessful birth after in vitro fertilization (IVF),1 theproportion of infants who have been born as a result ofIVF has increased dramatically to approximately 1 in 50births in Sweden,2 1 in 60 births in Australia, and 1 in 80births in the United States.3

IVFhas resulted in an increase in the number of pretermbirths (PTB) and low birth weight (LBW) infants,4-6 whichare themost important determinants of neonatalmortalityrates and neonatal, infant, and childhood morbidityrates.7-9 PTB, which accounts for 60% of perinatal deathsand almost one half of long-term neurologic disability(including cerebral palsy), is the most significant obstetricproblem.10 However, patients who undergo IVF aretypically older than the general obstetric population andare more likely to have multiple births, which are 2important confounders of PTB and LBW. In addition,there are growing concerns about increased risks ofcongenital malformations in children who are born afterIVF,11 including cardiac and neural tube defects.12,13

This systematic review seeks to resolve the conflictingresults of published studies in the literature and tocompare the incidence of adverse obstetric outcomes inIVF twins with spontaneously conceived twins who werematched for maternal age. Twin pregnancies were ofparticular interest, given their increasing frequencybecause of IVF. Because of the paucity of studies thataddress long-term follow-up periods, this review willconcentrate on perinatal outcomes.

Methods

The Meta-analysis of Observational Studies in Epide-miology consensus statement was followed.14

Search strategy

Medline (1966-October 28, 2003) and EMBASE (1980-October 28, 2003) were searched with the help of anexperienced librarian who used a comprehensive searchstrategy for MEDLINE and a separate comprehensivesearch strategy for EMBASE (Appendices I and II). Thereference lists of identified articles were searched foradditional references.

Outcome measures

The primary outcome was perinatal death. The mainsecondary outcomes were the incidence of PTB (definedas delivery at !37 weeks of gestation, 32-36 weeks of

gestation, and !32 weeks of gestation) and the in-cidence of LBW (defined as birth weight !2500 g) andvery LBW (defined as birth weight !1500 g).

Other secondary outcomes included birth weight andthe incidence of small for gestational age (SGA; defined asbirth weight !10%), the incidence of severe neonatalmorbidity (respiratory distress syndrome; intraventricularhemorrhage; necrotizing enterocolitis; retinopathy of pre-maturity; transient tachypnea of the newborn infant;congenital malformation; low Apgar score at 1, 5, or 10minutes; lowpHon cord gas), the incidence of intrauterinefetal demise (ie, stillbirth), the incidence of pretermpremature rupture of membranes (defined as prelaborrupture of membranes at!37 weeks gestational age), theincidence of antepartum hemorrhage, the incidence ofplacenta previa, the incidence of chorioamnionitis (mater-nal fever of O38(C, tender abdomen, fetal tachycardia),the length of maternal and neonatal postpartum hospital-ization, the incidence of cesarean delivery, the incidence ofnonvertex lead twin, the incidence of preeclampsia (de-fined as blood pressure O140/90 mm Hg, with O1Cproteinuria on dipstick or O300 mg/d on 24-hour urine,and the incidence of gestational diabetes mellitus.

Study selection

Inclusion criteriaCase-control and cohort studies that examined the out-comes of interest were included if they compared twinswho were conceived either by IVF or IVF/ intracytoplas-mic sperm injection (ICSI) with either fresh or frozenembryos with spontaneously conceived twins. To controlfor the potential confounding that is introduced bymaternal age, we included only the studies that matched(case-control studies) or controlled for it (cohort studies).Similarly, by restricting the focus of this meta-analysis totwin births, we controlled for the potential confoundingthat is introduced by the issue of multiple births. Therewere no language restrictions.

Exclusion criteriaStudies were excluded if they involved surrogacy, oocytedonation, gamete intrafallopian transfer, zygote intra-fallopian transfer, multifetal or selective reduction, severeovarian hyperstimulation syndrome, or studies thatmatched for gestational age or birth weight, because thesewere key outcome measures. The major reason for theexclusion of studies was the lack of either any controlgroup or an appropriate control group that matched formaternal age.

McDonald et al 143

Table I Characteristics of the studies that were included

Study Country Years of study Exclusion criteria Inclusion criteria Control selection

Dhont et al15 Belgium 1991-95 d d Computer selected from registryDhont et al16 Belgium 1992-97 Ovulation induction All cases R500 g in

Dutch-speaking partComputer selected from registryStudy Centre

Isaksson et al32 Finland 1993-99 d O22 Wk or O500 g Medical Birth RegistryKoivurova et al17 Finland 1990-95 d Registry at hospital

C private clinics(all IVF in Finland)

Randomly selected, FinnishMedical Birth Registry, O22wks or R500 g

Koivurova et al18 Finland 1990-95 d Registry at hospital Cprivate clinics (allIVF in Finland)

Randomly selected, FinnishMedical Birth Registry, O22wks or R500 g

Moise et al28 Israel 1990-95 d University hospital,dizygotic cases andcontrol subjects

Do not state

Nassar et al29 Lebanon 1995-2000 Ovulation induction,multifetal reduction,maternal diseasethat could involve ahigher rate ofcomplications (eg,renal disease)

R25 Wk, AmericanUniversity of BeirutMedical Centre

Sequentially 1:2

Tan et al30 England 1978-87 d British residents, O28wk:live/stillborn

2 Public hospitals (1 teaching)sequential deliveries inprimiparas from 1988-89

Wennerholm et al23 Sweden 1990-95 d O28 Wk: alllive/stillborn

Delivery books at 3 hospitals, 1of which was universityaffiliated

Westergaard et al24 Denmark 1994-95 Infer all cases included Random selected, Danish MedicalBirth Registry

Zadori et al25 Hungary 1995-2002 d d Control subjects inferred to befrom University of SzegedHospital

Two independent reviewers (S.M. and K.M.) per-formed the title and abstract screening. Disagreementswere resolved by discussion and consensus. An indepen-dent third adjudicator (A.O.) was available if uncer-tainty or disagreement existed.

Description of studiesEleven case-control studies that involved 2303 IVFtwins and 2326 spontaneously conceived twins wereincluded in this meta-analysis (Tables I-III). Dhont et alpublished 2 studies: One study15 in 1997 that reportedon 115 cases and 115 control subjects from 1991 through1995, and a second study16 in 1999 that reported on1241 cases and 1241 control subjects from 1992 through1997. Both studies were included in the review becausethey each reported on different outcome variables ofinterest. Similarly, Koivurova et al published 2 studies in2002: The first study17 reported on 20 cases and 100control subjects from 1990 through 1995, and a secondstudy18 reported on 103 cases and 103 control subjectsfrom 1990 through 1995. Similar to the 2 studies from

Dhont et al, both studies were kept in the review becausethey did not report any of the same outcomes.

Three cohort studies that involved at least 1509 IVFtwins (the number of twin sets was not stated byLancaster19) were included in a descriptive analysis.The number of spontaneously conceived twins was notstated clearly (Tables IV-VI). It was unclear from thedescription of the study from Australia and NewZealand whether gamete intrafallopian transfer wasincluded.19 Although we had hoped initially to be ableto meta-analyze cohort studies as well, the data were notamenable to this, and most authors did not respond toa request for further information. Hence, the data fromcohort studies are presented in a descriptive paragraphin the ‘‘Results’’ section.

Quality assessment

Two independent reviewers (S.M. and K.M.) performedthe quality assessment. Disagreements were resolved bythe aforementioned consensus process.

144 McDonald et al

Table II Characteristics of case-control studies that were included

Study

IVF twinpregnancies(n)

Control twinpregnancies(n) Intervention

Matchedfor maternalage

Matchedfor parity

Matchedfor deliverydate

Matched forother variables

Dhont et al15 115 115 IVF, IVF/ICSI Yes Yes G1 Y Order in gestation,zygosity

Dhont et al16 1241 1241 IVF, IVF/ICSI G5 Y Yes G1 Y Order in gestation,gender

Isaksson et al32 20 100 IVF, IVF/ICSI Yes Yes Year of delivery Mother’s residenceKoivurova et al17 62 82 IVF, fresh Yes Yes Year of delivery Gender, social class,

areaKoivurova et al18 103 103 IVF, fresh Yes Yes Year of delivery Gender, social class,

areaMoise et al28 20 40 IVF G3 Y Yes No Ethnic origin, all

twins dizygoticNassar et al29 56 112 IVF G1 Y Yes No Ethnic origin (all were

Middle Eastern)Tan et al30 125 21 IVF Yes, by

stratumYes(primiparas)

No d

Wennerholm et al23 98 49 IVF, fresh Cfrozen

G5 Y No Yes d

Westergaard et al24 427 427 IVF, IVF/ICSI Yes No Yes (age of child) dZadori et al25 36 36 IVF Yes Yes No Gravidity, previous

obstetric outcome

Data abstractionTwo independent reviewers (S.M. and K.M.) performedthe extraction of data from full articles. Disagreementswere resolved by discussion and consensus.

Because both reviewers were familiar with the layoutof the relevant journals and with the literature in thearea, reviewers were not masked to the authors, in-stitutions, or journal of publication.

A data extraction form was used to collect raw data.A pilot instrument was tested, and modifications weremade. A neonatologist (A.O.) reviewed the data collec-tion instrument to ensure that all relevant outcomeswere included.

Study information, which included the country oforigin, years under study, intervention, characteristics ofcases and control subjects, potential confounders, andoutcomes, was collected and included mean values andstandard deviations (Tables I-III).

AnalysisStatistical analysis was performed with the ReviewManger software (RevMan 4.2; The Cochrane Collab-oration, Oxford, England).

Categoric or dichotomous data were meta-analyzedwith odds ratios (ORs) as a measure of effect size.Continuous data were analyzed with weighted meandifference. The random effects model, which was describedby DerSimonian and Laird,20 was selected over the fixedeffects model because it incorporates within- and between-study variability into the analysis and, as such, represents 1

method of addressing heterogeneity. The random effectsmodel assumes that the effects are distributed randomly,and the central point of this distribution is the focus of thecombined effect estimate.21 Because this meta-analysisdeals with observational studies that inherently havemore variability than randomized control trials, the ran-dom effects model was felt to be more applicable and isgenerally more conservative, with wider confidence inter-vals. Clinical heterogeneity among the case-control studieswas felt to be small enough to permit statistical pooling.

Assessment of statistical heterogeneityStatistical heterogeneity was assessed with a chi-squaredtest and quantified with the I2 statistic. The I2 statistic,a relatively new statistical measure that was designed toquantify heterogeneity, describes the percentage of totalvariation across studies that is due to heterogeneityrather than chance.22 I2 values of 25%, 50%, and 75%have been suggested as indicators of low, moderate, andhigh heterogeneity, respectively.22

Heterogeneity that was detected in the data wasincorporated in the meta-analysis with a random effectsmodel, and the reasons for heterogeneity were explored.Results were also examined with the fixed effects model;little change was noted, which suggests that statisticalheterogeneity may have been less than anticipated.

Sensitivity analysisA few a priori sensitivity analyses were planned to testthe effect of ICSI, freezing embryos, parity, and study

McDonald et al 145

Table III Characteristics of included case-control studies: Quality assessment based on Cochrane Handbook criteria for observationalstudies31

Study Selection bias Performance biasAttritionbias Detection bias

Dhont et al15 No deliberate case selection, bestmatch on control*

* * MD questionnaires orhospital recordsy

Dhont et al16 All cases included, best matchon control*

* * *

Isaksson et al32 * * * *Koivurova et al17 * * * *Koivurova et al18 * * * *Moise et al28 Does not state where control

subjects from (dizygotic casesand control subjects; allcases included)y

* * Hospital records*

Nassar et al29 All cases included (controlsubjects: best matched,sequentially selected,of higher socioeconomicstatus)*

Only other factormatched for: parity

* *

Tan et al30 Unusual ratio of cases and controlsubjects (cases selected from2 clinics where the authorswork; information on controlsubjects from a computersystem)y

Only other factor matchedfor: parity

* Case information fromMedical ResearchCouncil registry*

Wennerholm et al23 Included total group IVF* Only other factor matchedfor: age of child

* *

Westergaard et al24 9 Private clinics, 6 public IVFclinics*

Only other factor matchedfor: date of delivery

* *

Zadori 200325 Cases and control subjectsinferred to be fromUniversity of SzegedHospital*

Many factors matched for * *

* No significant bias felt to be present.y Bias may exist.

quality. Funnel plots were used to explore the potentialfor publication bias in the reported results.

Results

Amaximum of 8 case-control studies reported on a givenoutcome, with most outcomes reported on by fewerstudies. The 3 studies that examined perinatal death hadresults that varied widely (OR, 1.40; 95% CI, 0.22-9.11;Figure 1; Note, Table VII includes key outcomes thatinclude fixed and random effects summary statistics)

Seven studies that reported on PTB (defined asdelivery at !37 weeks of gestation) resulted in a com-bined OR of 1.44 (95% CI, 0.94-2.22; Figure 2). The I2

value of 55.6% suggested moderate heterogeneity, atleast part of which could be explained by the observa-tional nature of the studies. The 3 studies that reportedon PTB between 32 and 36 weeks of gestation found an

increase in early delivery in the IVF conceived twins,with a combined OR of 1.48 (95% CI, 1.05-2.10; Figure3). Two studies found an increased rate of neonatalintensive care unit admission in IVF twins (OR, 2.22;95% CI, 1.64-3.02).

There was no difference in the proportion of SGA(birth weight !10th percentile) between IVF andspontaneously conceived twins (OR, 0.92; 95% CI,0.62-1.38). Similarly, there was no statistically signifi-cant difference in the rate of LBW (!2500 g; OR, 1.13;95% CI, 0.85-1.51) or birth weight (weighted meandifference, �51.40 g [range, �147.67-44.86 g). There wasalso no difference in the rate of congenital malforma-tions (OR, 1.14; 95% CI, 0.85-1.52).

IVF twins had had a higher rate of cesarean deliverythan spontaneously conceived twins, with an OR of 1.33(95% CI, 1.06-1.67). The OR in the 2 studies thatexamined the rates of elective cesarean deliveries wasOR 1.56 (95% CI, 0.73-3.36), and in the studies that

146 McDonald et al

Table V Characteristics of cohort studies that were included

StudyIVF twinpregnancies (n)

Control twinpregnancies (n) Intervention Factors addressed in analysis

Bergh et al26 1069 Unclear ‘‘general population’’ IVF, ICSI Stratified by maternal age, education,smoking, parity, no. of years of infertility,delivery variables

Gissler and Hemminki27 440 5004 IVF Background (age, county, smoking, maritalstatus, previous pregnancy, previousdeliveries), plurality

Von During et al12 191 2681 IVF Adjusted for maternal age, parity

Table IV Characteristics of cohort studies that were included

Study Country Data source Years of study Inclusion Exclusion

Bergh et al26 Sweden Swedish Registry 1982-95 O28 wk dGissler and Hemminki27 Finland Finnish Medical Birth Registry 1991-93 d dVon During et al12 Norway Norwegian Medical Birth Registry 1988-91 O15 wk d

Table VI Characteristics of cohort studies that were included: Quality assessment was based on Cochrane Handbook criteria forobservational studies

Study Selection bias Performance bias Attrition bias Detection bias

Bergh et al26 Control subjects from registry; no.of control subjects not stated*

y No mention* y

Gissler and Hemminki27 Control subjects from registry y Mentioned that up to 20% ofdata for IVF newborninfants missing*

y

Von During et al12 Control subjects from registry y y y

* Bias may exist.y No significant bias felt to be present.

examined the nonelective cesarean deliveries, the ORwas 1.01 (95% CI, 0.60-1.72). The length of maternalhospitalization in IVF twin pregnancies was reported aslonger in 2 studies (weighted mean difference, 4.86 days;95% CI, 1.16-8.20)

There were no significant differences in the ORs ofstillbirth (OR, 0.94; 95% CI, 0.4-2.19), very LBWinfants (OR, 1.22; 95% CI, 0.5-2.9), placenta previa(OR, 3.0; 95% CI, 0.4-24.5), antepartum hemorrhage(OR, 2.69; 95% CI, 0.62-11.57), nonvertex presentationof the lead twin, (OR, 1.16; 95% CI, 0.96-1.40), PTB at!32 weeks of gestation (OR, 1.03; 95% CI, 0.4-2.9),preterm premature rupture of the membranes (OR, 0.99;95% CI, 0.32-3.10), preeclampsia (OR, 1.07; 95% CI,0.64-1.78), and gestational diabetes mellitus (OR, 1.70;95% CI, 0.66-4.35). Only single studies reported on theincidence of chorioamnionitis (OR, 0.73; 95% CI, 0.73-3.51), respiratory distress syndrome (OR, 3.41; 95% CI,1.43-8.15), intraventricular hemorrhage (OR, 31.86;95% CI, 4.13-245.64), and necrotizing enterocolitis(OR, 10.16; 95% CI, 0.48-213.40). No study reportedon retinopathy of prematurity or transient tachypnea of

the newborn infant. Apgar scores and cord pHs wererarely commented on and not reported in a fashion thatwould permit meta-analysis.

Subgroup analysis was not performed that comparedIVF, IVF/ICSI, and spontaneously conceived twins asoriginally planned because the 4 studies that includedIVF/ICSI did not subdivide outcomes based on thesefactors. For similar reasons, subgroup analyses were notdone that examined fresh embryos, frozen embryos, andspontaneously conceived twins. Sensitivity analysis wasperformed that excluded the 2 studies that did not matchfor parity (Wennerholm et al23 and Westergaard et al24).In only 1 case did it appear to change the outcome:When the 7 studies that matched for parity were meta-analyzed (Figure 4), the OR of PTB at !37 weeks ofgestation in IVF twins was 1.57 (95% CI, 1.01-2.44),and the I2 value decreased slightly to 55.4%.

A post hoc sensitivity analysis was used to evaluatethe effect of excluding Zadori et al,25 which was the onlystudy that matched for previous obstetric outcome;however, it did not significantly change either thedirection or magnitude of any of the outcomes.

McDonald et al 147

Figure 1 Outcomes of IVF twin pregnancies: Perinatal deaths in case-control studies.

Table VII Summary of key perinatal outcomes from case-control studies of IVF twins

OutcomeStudies(n)

IVF twins(n)

Spontaneouslyconceivedtwins (n)

OR of randomeffects (95% CI)

OR of fixed effects(95% CI)

Measure ofheterogeneity,I2 (%)

Perinatal death 3 2562 2762 1.40 (0.22-9.11) 0.80 (0.58-1.11) 59.6PTB, 32-36 wk 3 316 267 1.48 (1.05-2.10) 1.48 (1.05-2.10) 0PTB, ! 37 wk 7 902 954 1.41 (0.96-2.08) 1.18 (0.98-1.43) 64.2PTB, ! 37 wk, matched forparity as well

6 350 506 1.57 (1.01-2.44) 1.51 (1.15-1.99) 55.4

Neonatal intensive care unitadmission

2 342 454 2.22 (1.64-3.02) 2.23 (1.64-3.02) 0

SGA 7 772 798 0.92 (0.62-1.38) 0.87 (0.66-1.14) 42.9LBW (!2500 g) 8 1853 1699 1.13 (0.85-1.51) 1.01 (0.88-1.16) 43.3Congenital malformations 4 2681 2897 1.14 (0.85-1.52) 1.14 (0.85-1.51) 0Cesarean delivery 6 1623 1646 1.33 (1.06-1.67) 1.37 (1.19-1.59) 14.5Elective cesarean delivery 2 154 161 1.56 (0.73-3.36) 1.59 (0.85-2.96) 33.1Length of maternalhospitalization

2 118 194 4.68 (1.16-8.20;weighted meandifference)

4.68 (1.16-8.20;weighted meandifference)

0

Funnel plots of case-control studies that reported onperinatal death, our primary outcome, or one of thesignificant secondary outcomes (such as PTB at 32-36weeks of gestation) each consisted of only 3 data pointsand were less useful than the funnel plot for PTB at!37weeks of gestation, which consisted of 8 data points andsuggested little publication bias, if any existed.

Cohort studies

Similar to the case-control results, Bergh et al26 (Swed-ish Medical Birth Registry) found no significant differ-ences in perinatal death or congenital malformation(relative risk [RR], 1.03; 95% CI, 0.93-1.25) in IVFtwins compared with spontaneously conceived twins.26

However, they also found no difference in the rate ofcesarean deliveries (RR, 1.06; 95% CI, 1.00-1.12).

Gissler and Hemminiki27 (Finnish Medical BirthRegistry cohort from 1991-1993) found that, even afteradjusting for maternal background and plurality, IVF

babies had an OR of 1.54 for LBW (95% CI, 1.30-1.83),although the difference in perinatal mortality rates wasnot significant (OR, 1.17; 95% CI, 0.73-1.86).

In the Norwegian cohort, von During et al,12 whoadjusted for maternal age and parity, found that IVFtwins had a higher cesarean delivery rate (RR, 1.4; 95%CI, 1.2-1.6). There were no significant differences inbirth weight of !1500 g (RR, 0.9; 95% CI, 0.6-1.3),birth weight 1500 to 2499 g (RR, 0.9; 95% CI, 0.8-1.1)or birth weightO2500 g (RR, 1.1; 95% CI, 1.0-1.2), andPTB at !37 weeks of gestation (RR, 1.1; 95% CI,0.9-1.3), early neonatal death (RR, 0.9; 95% CI,0.5-1.7), or stillbirth (RR, 1.4; 95% CI, 0.5-3.7).

Quality assessment of case-control studies

Quality assessment of studies was performed with theCochrane Handbook guidelines for observational stud-ies31 (Table III). If significant bias existed, the study wasexcluded. Performance bias was difficult to assess

148 McDonald et al

Figure 2 Outcomes of IVF twin pregnancies: PTB at !37 weeks of gestation in case-control studies.

Figure 3 Outcomes of IVF twin pregnancies: PTB at 32 to 36 weeks of gestation in case-control studies.

because most of the reports did not address the issue atall. Attrition bias is difficult to detect in case-controlstudies because the authors chose cases and controlsubjects with complete follow-up information. Also,follow-up information was within hospital; hence in-formation was more readily available than if it includedpostnatal follow-up information. Detection bias wasalso felt to have minimal impact, given that the out-comes of interest have standard definitions (for instance,SGA is consistently defined as birth weight !10thpercentile). We attempted to minimize selection biasby ensuring that case-control studies matched for theknown confounder of maternal age. Control subjectsusually were drawn from registries, either national orless commonly from delivery records at specific hospitals(Table II). Only a minority of studies noted thatconsecutive cases were used.16,28,29 Dhont et al15 ob-tained information from physician questionnaires andhospital records, the former possibly being more subjectto selection and detection bias. Moise et al28 did notstate from where control subjects were drawn. Tan

et al30 selected an unusual number of cases to controlsubjects (125-21, respectively). There were only minorconcerns about the quality of 3 of the studies: Dhontet al,15 Moise et al,28 and Tan et al30). Because Dhontet al15 and Moise et al28 also matched for zygosity andbecause the latter also stated that consecutive cases wereused, it was felt that these advantages outweighed anyother potential detractions in quality; therefore, sensi-tivity analysis was performed excluding Tan et al.30

Excluding Tan et al,30 the significance and direction ofthe combined effect estimates did not change, exceptminimally in one case: the 95% CI just crossed 1 forPTB at!37 weeks of gestation in studies that controlledfor parity (OR, 1.47; 95% CI, 0.99-2.19).

Quality assessment of cohort studies

Overall, greater concerns about the quality of the cohortstudies existed, given that basic information (such as thenumber of control subjects) was missing from 1 of the 3studies (Tables IV and VI) and that the number of cases

McDonald et al 149

Figure 4 Outcomes of IVF twin pregnancies: PTB at !37 weeks of gestation in case-control studies that controlled for parity.

was also missing from 1 study.27 Similarly, inclusion andexclusion criteria were absent in most studies. Attritionbias was possibly more of a concern; 1 study mentionedthat up to 20% of IVF newborn infants might have beenmissing from the data. The comments on performancebias and detection bias in case-control studies also holdfor the cohort studies.

Comment

IVF twins have an increase in PTB from 32 to 36 weeksof gestation compared with spontaneously conceivedtwins who were matched for maternal age (OR, 1.48;95% CI, 1.05-2.10). The OR for PTB at !37 weeks ofgestation in studies that also matched for parity wassimilar and approached statistical significance at a 5%level of significance (OR, 1.47; 95% CI, 1.01-2.44). Theorigin of the increase in PTB has yet to be determined.Possible causes include a factor inherent to the IVFtechnology, a history of infertility itself, or physician orpatient anxiety. The risks of extreme prematurity cannotbe assessed in this meta-analysis because of the exclusioncriteria that were established by the original studies.Two studies excluded deliveries at !28 weeks ofgestation, and another study excluded deliveries at!25 weeks of gestation. Four other studies excludeddeliveries of !500 g (Table I). Most studies found anincreased rate of cesarean delivery, although the Swed-ish cohort did not, which perhaps reflects nationaldifferences in physician or patient attitudes towardsthe mode of delivery.

The selection of the appropriate control group posedchallenges. We chose to include case-control studies thatmatched for maternal age and cohort studies thatcontrolled for it. A history of infertility and poorobstetric history have also been associated with in-creased obstetric risks.30 However, most published

studies (except Zadori et al,25 which matched forprevious obstetric outcome) do not include these poten-tial risk factors; therefore, they were not mandatorymatching or regression variables. Of note, post hocsubgroup analysis with the elimination of Zadori et al25

did not change significantly the magnitude or directionof any of the results.

Potential confounders that were unexplored in theoriginal studies are the greatest limitation to this meta-analysis. IVF parents generally have a higher socioeco-nomic status,29,30 although Zadori et al25 found similarlevels of education, and smoke less than those whoconceive spontaneously,32 which again suggests thatthey should have better perinatal outcomes. Compara-ble outcomes between IVF and spontaneously conceivedtwins (such as perinatal death) might actually suggestthat IVF twins are not doing as well as they should.However, unfortunately, few studies reported onsmoking and socioeconomic status. It is also well-established that monozygous twins who are monochor-ionic have a much higher rate of perinatal complica-tions.33-35 The lower (approximately 2%) incidence ofmonozygous twins in IVF,36 compared with the onethird incidence in spontaneously conceived twins,37

suggests that IVF twins should have better outcomesthan spontaneously conceived twins, yet they do not.Similar outcomes between IVF and spontaneouslyconceived twins actually suggest that IVF twins arenot doing as well as they should. The few studies thatmatched for zygosity15,28 or gender16 found that IVFtwins often had worse outcomes, which is consistentwith this hypothesis. An examination of outcomes inIVF singleton births showed that the problem could becircumvented.

A recent meta-analysis on assisted reproductivetechnology singleton and twin outcomes has beenpublished with similar conclusions38; however, differ-ences in our methods are worth highlighting. They

150 McDonald et al

included 1 study that involved multifetal reduction,which has been shown to be associated with worseperinatal outcomes compared with spontaneously con-ceived twins (at least when starting from R5 fetuses andreducing to twins39,40) and was excluded in our meta-analysis. They included 2 studies in which pregnancieswere conceived with intrauterine insemination andovarian stimulation with clomiphene citrate (Clomid)or gonadotropins, not IVF. They included studies thatused gamete intrafallopian transfer, which we excludedbecause we felt that an operation potentially couldimpact negatively on the outcome. They had relativelylittle discussion of the methods issues, which we felt wascritical, given that the field of meta-analysis of observa-tional studies is still being developed.

Finally, the lack of uniformity in reporting outcomeslimited the ability to do meta-analysis and limited theconclusions that could be drawn from the existingstudies in the literature. If reproductive, obstetric, andneonatal societies could reach consensus on defining andreporting outcomes, it would greatly aid research in thefield. Validation studies of the databases also would addcredibility.

Future issues to be addressed include whether freshembryos have different outcomes than frozen embryos,whether IVF has different results than IVF/ICSI, andlong-term follow-up of IVF children.

IVF twins have worse perinatal outcomes thanspontaneously conceived twins who are matched formaternal age, despite the fact that their outcomes shouldbe better because of the decreased proportion ofmonochorionic twins and the likely decreased rate ofsmoking and higher socioeconomic status among themothers. Compared with spontaneously conceived twinswho are matched for maternal age, IVF twins have anincreased risk of PTB between 32 and 36 weeks ofgestation (OR, 1.48; 95% CI, 1.05-2.10) and an elevatedrisk of PTB at !37 weeks of gestation when parity isalso matched for (OR, 1.57; 95% CI 1.01-2.44). Therewas an increased rate of cesarean delivery among IVFtwins (OR, 1.33; 95% CI, 1.06-1.67). In this meta-analysis, there were no significant differences in perina-tal mortality rates, the proportion of SGA infants, orthe rate of congenital malformations.

References

1. Steptoe PC, Edwards RG. Birth after the reimplantation of

a human embryo [Letter]. Lancet 1978;2:366.

2. Anonymous. Assisted reproduction: results of treatment 1994-

1997. Statistics: health and diseases-the National Board of Health

and Welfare Centre for Epidemiology. Stockholm, Sweden:

Norstedts; 2000.

3. Shultz R. Effect of embryo culture on gene expression and

behaviour. Proceedings of the Second World Congress on Fetal

Origins of Adult Disease; June 7-10, 2003; Brighton, UK.

4. McFaul PB, Patel N, Mills J. An audit of the obstetric outcome of

148 consecutive pregnancies from assisted conception: implications

for neonatal services. BJOG 1993;100:820-5.

5. Tough SC, Greene CA, Svenson LW, Belik J. Effects of in vitro

fertilization on low birth weight, preterm delivery, and multiple

birth. J Pediatr 2000;136:618-22.

6. D’Souza SW, Rivlin E, Cadman J, Richards B, Buck P, Lieberman

BA. Children conceived by in vitro fertilisation after fresh embryo

transfer. Arch Dis Childhood Fetal Neonat Ed 1997;76:F70-4.

7. Johnson HC. Report on the health of Albertans. Edmonton,

Alberta, Canada: Health Surveillance, Alberta Health; 1996.

8. Kramer MS. Determinants of low birth weight: methodological

assessment and meta-analysis. Bull World Health Organ

1987;65:663-737.

9. McCormick MC. The contribution of low birth weight to infant

mortality and childhood morbidity. N Engl J Med 1985;312:

82-90.

10. McCormick MC, Brooks-Gunn J, Workman-Daniels K, Turner J,

Peckham GJ. The health and developmental status of very low-

birth-weight children at school age. JAMA 1992;267:2204-8.

11. DeBaun MR, Niemitz EL, Feinberg AP. Association of in vitro

fertilization with Beckwith-Wiedemann syndrome and epigenetic

alterations of LIT1 and H19. Am J Hum Genet 2003;72:156-60.

12. von Durning V, Maltau JM, Forsdahl F, Abyholm T, Kolvik R,

Ertzeid G, et al. Pregnancy, births and infants after in-vitro-

fertilization in Norway, 1988-1991 [in Norwegian] Tidsskr Nor

Laegeforen 1995;115:2054-60.

13. Lancaster PA. Congenital malformations after in-vitro fertilisa-

tion. Lancet 1987;2:1392-3.

14. Stroup DF, Berlin JA, Morton SC, Olkin I, Williamson GD,

Rennie D, et al. Meta-analysis of observational studies in

epidemiology: a proposal for reporting: Meta-analysis of Obser-

vational Studies in Epidemiology (MOOSE) group. JAMA

2000;283:2008-12.

15. Dhont M, De Neubourg F, Van der EJ, De Sutter P. Perinatal

outcome of pregnancies after assisted reproduction: a case-control

study. J Assist Reprod Genet 1997;14:575-80.

16. Dhont M, De Sutter P, Ruyssinck G, Martens G, Bekaert A.

Perinatal outcome of pregnancies after assisted reproduction:

a case-control study. Am J Obstet Gynecol 1999;181:688-95.

17. Koivurova S, Hartikainen AL, Karinen L, Gissler M, Hemminki

E, Martikainen H, et al. The course of pregnancy and delivery and

the use of maternal healthcare services after standard IVF in

Northern Finland 1990-1995. Hum Reprod 2002;17:2897-903.

18. Koivurova S, Hartikainen AL, Gissler M, Hemminki E, Sovio U,

Jarvelin MR. Neonatal outcome and congenital malformations in

children born after in-vitro fertilization. Hum Reprod

2002;17:1391-8.

19. Lancaster PA. Registers of in-vitro fertilization and assisted

conception. Hum Reprod 1996;11(suppl):89-104.

20. DerSimonian R, Laird N. Meta-analysis in clinical trials. Control

Clin Trials 1986;7:177-88.

21. Egger M, Smith GD, Altman DG. Systematic reviews in health

care: meta-analysis in context. London: BMJ Books; 2001.

22. Higgins JP, Thompson SG, Deeks JJ, Altman DG. Measuring

inconsistency in meta-analyses. BMJ 2003;327:557-60.

23. Wennerholm UB, Hamberger L, Nilsson L, Wennergren M,

Wikland M, Bergh C. Obstetric and perinatal outcome of children

conceived from cryopreserved embryos. Hum Reprod

1997;12:1819-25.

24. Westergaard HB, Tranberg Johansen AM, Erb K, Nyboe AA.

Danish National In-Vitro Fertilization Registry 1994 and 1995:

a controlled study of births, malformations and cytogenetic

findings. Hum Reprod 1999;14:1896-902.

25. Zadori J, Kozinszky Z, Orvos H, Katona M, Pal A, Kovacs L.

Dilemma of increased obstetric risk in pregnancies following IVF-

ET. J Assisted Reprod Genet 2003;20:216-21.

McDonald et al 151

26. Bergh T, Ericson A, Hillensjo T, Nygren KG, Wennerholm UB.

Deliveries and children born after in-vitro fertilisation in Sweden

1982-95: a retrospective cohort study. Lancet 1999;354:1579-85.

27. Gissler M, Hemminki E. The danger of overmatching in studies of

the perinatal mortality and birth weight of infants born after

assisted conception. Eur J Obstet Gynecol Reprod Biol 1996;

69:73-5.

28. Moise J, Laor A, Armon Y, Gur I, Gale R. The outcome of twin

pregnancies after IVF. Hum Reprod 1998;13:1702-5.

29. Nassar AH, Usta IM, Rechdan JB, Harb TS, Adra AM, Abu-

Musa AA. Pregnancy outcome in spontaneous twins versus twins

who were conceived through in vitro fertilization. Am J Obstet

Gynecol 2003;189:513-8.

30. Tan SL, Doyle P, Campbell S, Beral V, Rizk B, Brinsden P, et al.

Obstetric outcome of in vitro fertilization pregnancies compared

with normally conceived pregnancies. Am J Obstet Gynecol

1992;167:778-84.

31. Alderson P, Green S, Higgins JPT, editors. Formulating the problem.

Cochrane Reviewers’ Handbook 4.2.2 [updated March 2004];

Section 6.8. Available at: http://www.cochrane.org/resources/

handbook/hbookhtm. Accessed on January 31, 2004.

32. Isaksson R, Gissler M, Tiitinen A. Obstetric outcome among

women with unexplained infertility after IVF: a matched case-

control study. Hum Reprod 2002;17:1755-61.

33. Hoskins RE. Zygosity as a risk factor for complications and

outcomes of twin pregnancy. Acta Genet Med Gemellol (Roma)

1995;44:11-23.

34. Grennert L, Persson PH, Gennser G, Gullberg B. Zygosity and

intrauterine growth of twins. Obstet Gynecol 1980;55:684-7.

35. Layde PM, Erickson JD, Falek A, McCarthy BJ. Congenital

malformation in twins. Am J Hum Genet 1980;32:69-78.

36. Alikani M, Cekleniak NA, Walters E, Cohen J. Monozygotic

twinning following assisted conception: an analysis of 81 consec-

utive cases. Hum Reprod 2003;18:1937-43.

37. Multifetal pregnancy. In: Cunningham FG, MacDonald PC,

Gant NF, Leveno KJ, Gilstrap LC, Hankins GDV, et al, editors.

Williams obstetrics. Stamford (CT): Appleton & Lange; 1997.

p. 861-94.

38. Helmerhorst FM, Perquin DA, Donker D, Keirse MJ. Perinatal

outcome of singletons and twins after assisted conception: a sys-

tematic review of controlled studies. BMJ 2004;328:261-5.

39. Torok O, Lapinski R, Salafia CM, Bernasko J, Berkowitz RL.

Multifetal pregnancy reduction is not associated with an increased

risk of intrauterine growth restriction, except for very-high-order

multiples. Am J Obstet Gynecol 1998;179:221-5.

40. Evans MI, Berkowitz RL, Wapner RJ, Carpenter RJ, Goldberg

JD, Ayoub MA, et al. Improvement in outcomes of multifetal

pregnancy reduction with increased experience. Am J Obstet

Gynecol 2001;184:97-103.

Appendix I

Search strategy for EMBASE from 1966 to October2003

1. Twins (4737 entries)2. Explode congenital malformation (171,243 entries)3. Neonatal morbidity (1398 entries)4. Premature delivery/ or premature labour/ or pre-

term birth/ or preterm labor/ or preterm labour/ orpremature labor (7144 entries)

5. Low birth weight/ or congenital hypotrophy/ orfetal growth retardation/ or fetus growth retarda-tion/ or growth retardation in utero/ or hypo-

trophy, congenital/ or in utero growth retardation/or prenatal growth retardation/ or retarded in-trauterine growth (13,058 entries)

6. (Low birth weight or neonatal underweight).mp(10,610 entries)

7. (Premature delivery or premature labour or pre-term birth or preterm labor or preterm labour orpremature labor).mp (9519 entries)

8. Fetus death/ or stillbirth/ or fetus mortality (6056entries)

9. (Stillbirth or antepartum death or dead fetus ordead fetus syndrome or death fetus or death,intrauterine or death,prenatal or endouterine deathor fetal death or foetal death or foetus death or in-trauterine death or intrauterine fetal death orprenatal death or fetus death or fetus mortality orstillborn baby).mp (7790 entries)

10. (Death neonatal or death newborn or neonataldeath or neonate death or newborn death).mp(2121 entries)

11. (Newborn morbidity or morbidity neonatal orneonatal morbidity or neonatus morbidity).mp(2053 entries)

12. Pregnancy complication/ or diseases complicatingpregnancy/ or gestational complication/ or preg-nancy complications/ or pregnancy complications,cardiovascular/ or pregnancy complications, hema-tologic/ or pregnancy complications, infectious/ orpregnancy complications, neoplastic/ or pregnancycomplications, parasitic/ or pregnancy outcome(8072 entries)

13. (Pregnancy complication or diseases complicatingpregnancy or gestational complication or preg-nancy complications or pregnancy complications,cardiovascular or pregnancy complications, hema-tologic or pregnancy complications, infectious orpregnancy complications, neoplastic or pregnancycomplications, parasitic or pregnancy outco-me).mp. (11,189 entries)

14. (Low birth weight or congenital hypotrophy orfetal growth retardation or fetus growth retarda-tion or growth retardation in utero or hypotrophy,congenital or in utero growth retardation or pre-natal growth retardation or retarded intrauterinegrowth).mp. (11,601 entries)

15. Fertilization in vitro/ or micromanipulation/ orextracorporeal fertilization/ or in vitro fertilisation/or in vitro fertilization/ or testtube baby (13,954entries)

16. (Fertilization in vitro or micromanipulation orextracorporeal fertilization or in vitro fertilisationor in vitro fertilization or testtube baby).mp.(15,837 entries)

17. (Twins or diseases in twins or twin or twin data ortwin studies or twin study).mp. (15,958 entries)

18. congenital malformation:.mp. (17,900 entries)

152 McDonald et al

19. 1 or 17 (15,958 entries)20. 15 or 16 (15,837 entries)21. 19 and 20 (615 entries)22. or/2-14,18 (209,123 entries)23. 21 and 22 (198 entries)

Appendix II

Search strategy for MEDLINE from 1966 to OctoberWeek 4 2003

1. Exp fertilization in vitro (17210 entries)2. Exp FERTILIZATION IN VITRO/ or TEST

TUBE BABIES/ or ECTOGENESIS/ or test-tubefertilization/ or fertilizations in vitro/ or in vitrofertilization (17375 entries)

3. Exp Twins/ (14234 entries)4. Twin:.mp. (26248 entries)5. Pregnancy complications/ or fetal death/ or fetal

diseases/ or fetal anoxia/ or fetal distress/ ormeconium aspiration/ or labor complications/ orabruptio placentae/ or fetal membranes, prematurerupture/ or placenta accreta/ or placenta praevia/ orpostpartum hemorrhage/ or labor, premature/ oroligohydramnios/ or exp placenta diseases/ or preg-nancy complications, cardiovascular/ or pregnancycomplications, hematologic/ or pregnancy compli-cations, infectious/ or pregnancy in diabetics/ orpregnancy toxemias/ or puerperal disorders/ (151800entries)

6. Infant Diseases/ or Newborn Diseases/ or stillbirth/or gestational diabetes mellitus/ or Growth/ orGrowth Disorders/ or PRENATAL INFLUEN-CES/ or Asphyxia/ or Aspiration/ or Meconium/or aspiration Pneumonia/ or Respiratory DistressSyndrome/ or PLACENTA DISEASES/ or PLA-CENTA DISORDERS/ or Pregnancy-induced hy-pertension/ or pre-eclampsia/ or eclampsia/ orPUERPERAL DISORDERs/ or infant, premature/or prematurity.mp. [mp=title, abstract, cas registry/

ec number word, mesh subject heading] (110277entries)

7. Fetal growth retardation/ or infant, low birthweight/ or infant, small for gestational age/ orinfant, very low birth weight/ or Growth/ or GrowthDisorders/ or Birth Weight/ or fetal macrosomia/(67990 entries)

8. (Pregnancy complication: or gestational diabetesmellitus or pregnancy-induced diabetes or gesta-tional diabetes or fet: death or stillbirth: or AS-PHYXIA NEONATORUM or fetal anoxia or FET:DISTRESS or fet: disease: or SMALL FOR GES-TATIONAL AGE or growth retardation or fetalgrowth retardation or iugr or intrauterine growthretardation or growth restriction or fetal growthrestriction or intrauterine growth restriction orGrowth or Growth Disorders or Birth Weight orlow birth weight or Fetal Diseases or Infant Disease:or Newborn Disease: or stillbirth or FETALDEATH or PRENATAL INFLUENCES or As-phyxia or Aspiration or Meconium or aspirationPneumonia or Respiratory Distress Syndrome orlabor complication: or abuptio: or Fetal Membranesor prom or placenta: accreta or Placenta Praevia orPlacental Insufficiency or PLACENTA: DISEASE:or PLACENTA: DISORDERS or cardiovascularpregnancy complications or hematologic pregnancycomplication: or infectious complication: orPREGNANCY TOXEMIA: or Pregnancy-inducedhypertens: or pre-eclampsia or eclampsia or PUER-PERAL DISORDER: or infant, premature or pre-mature: or preterm deliver: or premature deliver: orpreterm labo?r or premature labo?r).mp. (790519entries)

9. 3 or 4 (26248 entries)10. 1 or 2 (17375 entries)11. 9 and 10 (770 entries)12. 5 or 6 or 7 or 8 (804151 entries)13. 11 and 12 (280 entries)14. limit 13 to human (264 entries)