Methods/designs
Criteria for inclusion of studies in IPD
We propose an IPD meta-analysis of randomized controlled trials of 17- OHPC or vaginal progesterone versus placebo in women with twin pregnancies.
Participants
Inclusion criteria will be women with twin pregnancies, with chorionicity and gestation confirmed by ultrasound, who were included in a RCT comparing progestogens with placebo for the prevention of preterm birth. Women with congenital abnormalities, contraindications to progestogen treatment and cervical cerclage will be excluded.
Although studies have been performed in women with triplet pregnancies [
12,
21,
22] these will not be included in the IPD meta-analysis. There appear to be differences in the response to progestogens in women with singleton and twin pregnancies, therefore further differences might be anticipated between women with twins and those with higher order multiples. As the number of trial participants with triplet pregnancies is comparatively small compared to the number with twin pregnancies, excluding these women should not negatively affect the power of the meta-analysis, whilst ensuring the group is as homogenous as possible. Exclusion of triplet pregnancies from the analyses will not compromise the validity of the study since randomization was stratified for twin or higher order multiple in the studies that included both twin and triplet pregnancies [
7,
8,
12,
21,
22].
Intervention
The intervention will be either weekly intramuscular injection of 17-hydroxyprogesterone caproate (17-OHPC) or daily vaginal progesterone. As these treatments may act differently and have different distribution profiles we will analyze the results of the two types of treatment separately.
Outcome measures
The primary outcome will be adverse perinatal outcome, a composite outcome of perinatal death, defined as death before discharge from the hospital, and significant neonatal morbidity at discharge, defined as one or more of respiratory distress syndrome (RDS) requiring ventilation for ≥ 24 hr, bronchopulmonary dysplasia (BPD), intraventricular hemorrhage (IVH) grade III or IV), periventricular leucomalacia (PVL), necrotizing enterocolitis (NEC) grade II or more) culture proven sepsis, retinopathy of prematurity (ROP) requiring treatment.
Secondary outcomes will be intrauterine death < 32 weeks or preterm birth < 32 weeks; intrauterine death < 35 weeks or preterm birth < 35 weeks; intrauterine death < 37 weeks or preterm birth < 37 weeks; intrauterine death; fetal loss < 28 weeks or early preterm birth < 28 weeks gestation; time to delivery or death. If data is detailed enough, preterm birth will be analyzed separately for spontaneous preterm birth and indicated preterm birth.
Subgroup analyses
Subgroup analyses will be performed for the primary outcome only, in the following groups
-
each of ultrasonographically diagnosed monochorionic and dichorionic twins
-
women who completed ≥ 90% of treatment
-
women with a cervical length < 25 mm on baseline assessment (in studies where transvaginal cervical length measurement was specified in protocol)
-
women with a prior spontaneous preterm birth < 37 weeks
-
ethnicity
-
each dose of vaginal progesterone, e.g. ≤ 100 mg versus ≥ 200 mg.
Identification of studies
We will perform an electronic search of the Cochrane Central Register of Controlled Trials (CENTRAL), PubMed, MEDLINE and ClinicalTrials.gov for published or registered randomized controlled trials including women with twin pregnancy that were randomly allocated to treatment with progestogens (including vaginal progesterone and 17-hydroxyprogesterone caproate) or placebo in the second or third trimester with the intention to prevent preterm birth. We will use the search terms "preterm birth" AND ["progesterone" OR "17 hydroxyprogesterone caproate" OR "progestogen"] AND ["pregnancy multiple" OR "pregnancy twin"] AND "randomized controlled trial" AND "human".
Two review authors (ES and SS) will independently assess inclusion criteria and study quality and risk of bias. A third author (BWJM) will review studies in which there is any disagreement about study quality. Risks of bias will be assessed in all of the identified studies based on [
23]:
-
sequence generation (i.e. computer generated random number, use of random number table or other truly random process)
-
allocation concealment (i.e. web-based or telephone central randomization or consecutively numbered sealed opaque envelopes)
-
blinding for participants, study personnel and outcome assessors
-
Incomplete outcome data
-
Selective outcome reporting
-
Other sources of bias
In cases where study quality is not clear from trial protocols or publications, then the authors will be contacted for clarification.
The corresponding authors of eligible studies will be approached to take part in the IPD meta-analysis. They will be invited to take part if the study is complete and data available by 1st July 2011.
Data quality will be independently assessed by two review authors (ES and SS). A third author (BWJM) will review data in which there is any disagreement about quality. Only studies with adequate outcome data (< 10% participant attrition or exclusion, with full reporting of reasons for withdrawals and protocol violations and no imbalance in drop-outs across groups) and adequate reporting (all of the study's pre-specified outcomes and all expected outcomes of interest made available) will be included.
Analysis
Overall effects of each treatment (17-OHPC and vaginal progesterone) in women with twin pregnancies will be estimated in the pooled IPD. Descriptive comparisons between studies will be conducted to assess between-study differences. We assume the data to be missing at random (MAR), therefore observed patient characteristics will be used to impute missing data, by means of multiple imputation [
24]. Missing data will be imputed within each original study, before data of the individual studies are pooled. Treatment effects will be estimated by means of a random effects log-binomial model and, hence, the measure of association is the risk ratio. The presence of heterogeneity of outcomes across trials will be assessed using the I
2 measure [
25]. Heterogeneity across studies and dependency between data originating from the same study will be taken into account by fitting a random intercept for each original study. If necessary, analyses will be adjusted for variables used in stratified randomization. Furthermore, dependency between children born from the same pregnancy will be accounted for by means of generalized estimating equations (GEE) [
26]. To investigate subgroup effects, the treatment effects will also be estimated within strata based on single subgrouping variables, as well as using an interaction term in the regression model.
Time-to-delivery analysis will be performed with Kaplan-Meier analysis and Cox proportional hazards regression analysis. Again, dependency between data originating from the same study will be taken into account by conducting a stratified analysis (stratified by study) [
27].
We will perform subgroup analysis with pooled individual datasets of women stratified by the pre-specified criteria outlined above. Where available we will plot cervical length against gestational age of cervical length measurement. When differences in gestational age explain differences in length, we will apply standardization for gestational age. Differences in cervical length between the studies will also be explored. We will assess the absolute value of cervical length (corrected for gestational age differences) as well as the percentiles of cervical length (5th, 10th and 25th) in each dataset. We will assess interaction between the treatment effect of progestogens and cervical length, using both time to delivery and the primary and secondary endpoints. To ensure that subgroup effects are not confounded by between-trial differences, dependency between data originating from the same trial will be taken into account using a random intercept for every study in the regression model [
18].
Competing interests
The authors declare that they have no competing interests.
Authors' contributions
ES, ACL and BWJM were involved in the concept and the design of the study. ES and SS planned and wrote the initial protocol with significant contributions by RHHG, AK, JEN, KGMM, DJR and BWJM. ES, SS, RHHG, KM, CAC, TG, CYS, EAT, DJR, SNC, GRS, JMZ, JEN, LR, AT, JCM, EFM, VS, AP, JM, ACL, KGMM, AK and BWJM participated in face-to-face meetings and/or teleconferences to discuss the protocol, the design of the meta-analysis, the choice of outcome measures and analysis strategies. KM, CAC, TG, CYS, EAT, DJR, SNC, JEN, LR, KK, AT, EC, PR, JCM, EFM, CMB, VS, AP, JM, AHN, ACL, BWJM were involved in the construction and design of one of the trials included in the meta-analysis. All authors critically reviewed the subsequent versions of the manuscript and approved the final manuscript.