Tocolysis for inhibiting preterm birth in extremely preterm birth, multiple gestations and in growth-restricted fetuses: a systematic review and meta-analysis

A comprehensive search of MEDLINE, Embase, the Cochrane Library, CINAHL, POPLINE and the WHO Global Health Library databases was conducted on 14 February 2014 to identify studies that contained information about tocolysis for extremely preterm births, tocolysis in multiple pregnancies, and tocolysis in growth-restricted fetuses. No language restriction was imposed. The search was developed using related thesaurus terms and a wide range of free subject categories and/or keywords that met the objective of this review concept. Search terms included “tocolytic agents,” “prolonged pregnancy,” “premature,” “fetal development,” “maternal death,” “low birth weight,” “multiple pregnancy” and “restricted growth.” Further detail on the search strategy is provided in an additional file (see Additional file 1). All identified bibliography and reference studies were managed by using the EndNote® version 6 and Reference Manager® (Thomson Reuters, USA).

Titles and abstracts retrieved electronically were screened for potential eligibility by two researchers (CM and RMG) independently and consulted with technical advisory experts when necessary. Irrelevant or duplicated reports were removed and multiple reports for the same studies were link together as one study. Selected articles were then evaluated fully by their relevance in addressing one of the following health care questions for this systematic review: (1) Is tocolysis effective and safe for inhibiting extremely preterm birth?; (2) Is tocolysis effective and safe for inhibiting preterm birth in multiple pregnancy?; and (3) Is tocolysis effective and safe for inhibiting preterm birth in growth-restricted fetuses?. Disagreements on which studies should be included were resolved by discussions with other review authors or people in the advisory group. All potential full-text articles were examined by the compliance with the criteria as follow: (1) For extremely preterm birth, defined as birth before 28 weeks of gestation, pregnant women with threatened extreme premature labour (with uterine activity and/or cervical changes) were considered for inclusion and studies that recruited women of all gestational ages and provided a proper stratified analysis for extremely preterm births were considered eligible [3, 14]. (2) For multiple pregnancies, women who were carrying twins, triplets or more fetuses with high risk of preterm delivery were eligible for inclusion. The term “high risk of preterm delivery” encompasses preterm uterine activity and/or cervical changes. (3) Studies in which pregnant women with growth-restricted fetuses had high risk of preterm delivery and preterm uterine activity and/or cervical changes. The definition of growth-restricted fetuses was referred as the measurement by small for gestational age (SGA), which is defined by a birth weight less than the 10th percentile for gestational age, and is associated with preterm birth and other pregnancy complications related to preeclampsia or infections [15‐17]. Studies involving the administrations of corticosteroids were considered eligible and studies that used any type of tocolysis (i.e. calcium channel-blockers, b-sympathomimetic, oxytocin inhibitors, among others) were included. Study designs, such as individual, cluster or quasi randomized controlled trials (RCTs), controlled before-after studies, prospective or retrospective cohorts with control groups, and, if necessary, case–control studies featuring one treatment group and a comparison group were considered for inclusion. Head-to-head studies were not included (e.g. atosiban vs. salbutamol).

Two researchers independently conducted data collection from eligible studies by using a data extraction form, which was developed with the experts’ recommendation and in reference to the data collecting approach in Cochrane Collaboration Handbook [12]. Any discrepancy on specific data was resolved by consulting with a third researcher. To assess the internal quality of the studies, the Cochrane Collaboration’s tool for assessing risk of bias was used for each RCT study [12]. The risk of bias assessment tool was a domain-based evaluation: sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcomes, incomplete outcome data, selective reporting and other bias. The judgment for risk of bias was made according to the criteria and the judgments were reported by assigning low risk, high risk and unclear risk to each domain. For non-randomized studies (non-RCTs), the Risk of Bias Assessment Tool for Nonrandomized Studies (RoBANS) was used [18]. The RoBANS tool composed of six domains: selection of participants, confounding variables, measurement of exposure, blinding of outcomes, incomplete outcome data and selective reporting. The assessment for risk of bias was made according to RoBANS criteria and the judgments were reported by assigning low risk, high risk and unclear risk to each domain.

Data analyses were conducted by using a statistical software, Review Manager Version 5.3 (RevMan 5.3) [19]. Data from RCTs were assessed by meta-analyses, and data from non-RCTs (e.g. prospective and retrospective cohorts, case–control studies) were analyzed separately from RCTs and described narratively. Relative risk in terms of risk ratio (RR) was calculated from 2 by 2 table to measure the effect estimate for binary outcomes. All statistical analyses used a 95 % confidence interval and a p-value with a cut-off point of 0.05. Fixed-effect modeling was carried out to determine the effect estimates and a Chi² statistic with a cut-off point of 0.10 was used to determine heterogeneity. The I² statistic was used to assess the inconsistency among the studies, in which to detect the variability in the effect estimates due to heterogeneity. When the fixed-effects assumption could not provide the true effect of the intervention, a random-effects model was used. When treatment results showed statistical effectiveness, number needed to treat (NNT) was calculated from risk difference. To synthesize the data from the non-RCTs, the unadjusted relative risks were generated from 2 by 2 table, and when the data were considered to be appropriate, an estimate of the effect size was made. When unadjusted relative risk could not be obtained from the non-RCTs, their published results were presented accordingly and separately with either adjusted odds ratio (aOR), adjusted risk ratio (aRR) or adjusted hazard ratio (aHR) in this review.

To rate quality of evidence, the Guideline Development Tool (Copyright © 2014, McMaster University and Evidence Prime Inc.) template was utilized. The assessment was made in compliance with the Grades of Recommendation, Assessment, Development and Evaluation Working Group (GRADE) guidelines, and the rating of evidence were determined by GRADE guidelines on study limitations criteria [20]. Quality rating was made for each outcome and was presented in four levels of quality recommended by The GRADE approach: high, then moderate, low and very low quality of body evidence. In terms of non-RCTs, quality of evidence rating began from low quality and could either be judged with very low, moderate or high quality of evidence as recommended in GRADE guidelines.

Prolongation of pregnancy more than 24 h in women with extremely preterm birth was found in one trial [21] but there was no significant difference in the rate of prolonging the pregnancy between the tocolysis (atosiban) group and the placebo group (RR 1.15, 95 % CI 0.81 to 1.63, 77 women) (Table 2). For prolongation of pregnancy more than 48 h, there were two trials [21, 23] and the meta-analysis showed that there was no relative risk difference found between the tocolysis group and the placebo group (RR 1.04, 95 % CI 0.83 to 1.31, I² = 0 %, 117 women). The effect estimate and the confidence interval crossed the line of no effect (Fig. 2a). There were no evident heterogeneity and inconsistency indicated among the studies.

For prolongation of pregnancy more than 7 days, there were two trials [21, 23] and the pooled relative risk indicated no difference between the tocolysis group and the placebo group (RR 1.05, 95 % CI 0.75 to 1.48, I² = 48 %, 117 women). The effect estimate and the confidence interval crossed the line of no effect (Fig. 2b). There was a substantial heterogeneity in this comparison and a moderate inconsistency across the studies.

Two RCTs [21, 22] reported on outcome of perinatal deaths and their combined perinatal deaths were 50 out of a total of 265 fetuses. The relative risk from these combined trials was RR 2.22, 95 % CI 0.26 to 19.24, I² = 77 %, 265 fetuses (Fig. 2c). This estimate suggested that there was an increase risk of perinatal death with tocolytic treatment relative to placebo group, but the estimate showed statistical uncertainty due to wide confidence interval that crossed the line of no effect. A substantial heterogeneity and inconsistency were indicated among the studies.

The quality of evidence for prolongation of pregnancy more than 24 h was rated to be moderate. However, the evidence for prolonging pregnancy more than 48 h or more than 7 days were rated very low due to small sample size, which contributed to imprecision, and one study [23] did not mention allocation concealment, which the limitation of the trial decrease the confidence in the estimated results. The quality of evidence for outcome of perinatal death was rated very low as well because of significant heterogeneity attributed by small sample size and the overall estimate had wide confidence interval, which indicated substantial uncertainly. The summary of quality of evidence is presented in GRADE tables (see Additional file 4).

One non-RCTs [25] reported there was a relative different in rate between the tocolytic treatment group and the no treatment group for women remaining pregnant for 7 days or more after admission (RR 2.13, 95 % CI 1.12 to 4.06, 148 women; NNT = 6, 95 % CI 3.2 to 23.5) (Table 4).

One study [24] reported that there was no difference in the rate of prolonging pregnancy for more than 24 weeks (RR 0.91, 95 % CI 0.76 to 1.09, 101 women); meanwhile, another study [27] also reported no significant difference was found in the rate of prolonging pregnancy for more than 28 weeks (RR 0.91, 95 % CI 0.69 to 1.20, 222 women) between the tocolytic treatment group and the no treatment group. Two non-RCTs reported on prolongation of pregnancy for more than 32 weeks: one study [27] showed that there was no difference found in the tocolytic treatment group relative to that in the no treatment group (RR 0.94, 95 % CI 0.68 to 1.30, 222 women) and another study [24] also showed a similar findings (RR 0.95, 95 % CI 0.73 to 1.24, 101 women). When the relative effect from these two non-RCTs [24, 27] were pooled, the overall estimate showed that there was no significant difference between the tocolysis (indomethacin) group and the no treatment group (RR 0.94, 95 % CI 0.76 to 1.17, I² = 0 %, 323 women) (Table 4). Two non-RCTs [24, 27] reported on prolongation of pregnancy for more than 35 weeks. One study [27] showed that there was no significant difference found between the two comparison groups (RR 1.01, 95 % CI 0.68 to 1.48, 222 women), and similar result was reported in the other study [24] (RR 0.92, 95 % CI 0.68 to 1.24, 101 women). The overall effect estimate from these two non-RCTs [24, 27] indicated that there was no relative effect difference found between the tocolytic treatment group and the no treatment group (RR 0.96, 95 % CI 0.75 to 1.23, I² = 0 %, 323 women) (Table 4).

Perinatal deaths were reported in two studies [25, 27] (Table 4). One study [27] showed that there was no significant difference found in the rate of neonatal survival (RR 1.12, 95 % CI 0.92 to 1.37, 222 fetuses) and the same indicated that there was no significant difference in the rate of perinatal deaths as well (RR 0.81, 95 % CI 0.54 to 1.21, 222 fetuses). The other study [25], in which several tocolytic were used, showed that there was a significance difference found in the rate of perinatal deaths and reported that the tocolysis group had 36.6 % of perinatal deaths as opposed to 62.5 % in the no treatment group (RR 0.65, 96%CI 0.45 to 0.94,148 fetuses; NNT = 6, 95 % CI 2.9 to 3.28, 148 fetuses). The overall effect estimate from the two non-RCTs [25, 27] showed a significant difference in reducing the rate of perinatal deaths (RR 0.73, 95 % CI 0.55 to 0.95, I² = 0 %, 370 fetuses) and the pooled estimate implied that tocolytic treatment decreases the risk of perinatal deaths by 27 % (Table 4).

The quality of evidence for prolongation of pregnancy for more than 7 days or 24, 28,32 and 35 weeks was rated very low because majority of the information were from studies with unclear risk of bias in the incomplete outcome data and high risk of bias in selective reporting and confounding variable. The quality of evidence for perinatal death was rated very low since the information was from two studies with unclear risk of bias for incomplete outcome data and selective reporting and one study was with high risk of bias for confounding variables (see Additional file 4).