Types of studies

We included parallel‐group randomised controlled trials (RCTs) that compared automated mandatory bolus (AMB) with basal infusion (BI) for the maintenance of labour epidural analgesia, irrespective of language, publication date, or publication type. 

We excluded non‐randomised studies such as cohort studies due to their increased risk of bias, and cross‐over studies as this methodology was not appropriate to evaluate interventions administered at specific time points. 

Types of participants

We included studies involving term, pregnant women who requested for labour epidural analgesia. Studies in which a subset of participants met our eligibility criteria were included if these participants comprised at least 65% of the study population, and only data relevant to the eligible participants were analysed.

We excluded studies of women in preterm labour, with multiple pregnancies, or with fetal malposition such as breech presentations.

Types of interventions

We included studies that compared AMB with BI to maintain epidural labour analgesia. AMB was defined as automated, intermittent bolus administration of local anaesthetic into the epidural space at set time intervals. Conversely, BI was defined as continuous administration of local anaesthetic into the epidural space without interruption. All forms and doses of local anaesthetics with the addition of opioids administered during labour epidural analgesia were included. Studies that utilised patient‐controlled epidural analgesia (PCEA) were included, as long as the intervention groups compared AMB with BI.

We excluded interventions involving intrathecal or spinal catheters, those that did not use automated delivery or which utilised manual delivery of local anaesthetics to maintain labour analgesia, and interventions where AMB and BI were combined.

Types of outcome measures

Outcomes were dichotomous (breakthrough pain, caesarean delivery, and instrumental delivery), continuous (duration of labour analgesia, and local anaesthetic consumption), or ordinal (maternal satisfaction and Apgar score). 

Outcomes were measured from the start of labour analgesia to immediately after childbirth, as reported by the individual studies. Outcomes were not used as eligibility criteria for study selection.

Electronic searches

Following the Cochrane guidelines for searching and identification of relevant studies (Lefebvre 2021), the databases of Cochrane Central Register of Controlled Trials (CENTRAL, Wiley Cochrane Library); MEDLINE (National Library of Medicine); Embase (Elseiver), Web of Science (Clarivate), the WHO‐ICTRP (World Health Organization) and ClinicalTrials.gov (National Library of Medicine)  were searched from inception to 31 December 2020, with our search strategies detailed in Appendix 1. Updated searches were performed from 1 January 2021 to 31 December 2021, and 1 January 2022 to 31 December 2022. Collections used for the databases were: CENTRAL ‐ all, MEDLINE ‐ Ovid Medline (R) 1946‐2022, Embase ‐ Biomedica, Web of Science ‐ Core Collection, WHO‐ICTRP ‐ all, ClinicalTrials.gov ‐ all.

No language restrictions were placed on our searches. We used free‐text terms in all databases and subject headings in combination when thesauri were a component of a database. 

Searching other resources

We reviewed the 'Related articles' feature of PubMed for all eligible trials and reviews. We screened the reference lists of all eligible trials, reviews, and systematic reviews for potentially eligible studies. We also contacted authors of included studies in this field in order to identify unpublished research and trials still underway. Reference lists of the included articles were screened for potentially relevant articles.

Selection of studies

Titles, abstracts, or records identified by our search criteria (Criteria for considering studies for this review) were uploaded into Covidence, a systematic review screening tool (Covidence). A minimum of two review authors (HST, ZYZ, YYQ, FJS) independently reviewed each title and abstract, followed by an examination of the full‐text documents to identify studies meeting the inclusion criteria. Conflicts were resolved by discussion, or with arbitration by a third review author (BLS). 

Data extraction and management

A minimum of two review authors (HST, ZYZ, YYQ, FJS) independently extracted the data using a standardised form that was pilot‐tested prior to use (see Appendix 1). We extracted information pertaining to the study design, method of randomisation, use of allocation concealment, blinding of caregivers and outcome assessors, reporting of the study setting and participants, inclusion and exclusion criteria, sample size, interventions, outcomes, and loss to follow‐up. Two review authors (HST, ZYZ) entered and checked the data independently, and a third review author (BLS) resolved disagreements. The included studies were checked for errata, comments and retractions. The outcomes of included studies were compared with the ones reported in ClinicalTrials.gov protocols.

Non‐English studies were professionally translated prior to data collection.

Assessment of risk of bias in included studies

A minimum of two review authors (HST, ZYZ, YYQ, FJS, RS) independently assessed trial quality and risk of bias for each study using the criteria outlined in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011), and a third review author (BLS) resolved any disagreements.

Based on the Cochrane risk of bias tool in Review Manager Web (RevMan Web), we considered the following domains (Higgins 2011): random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, selective reporting, and other bias. Appendix 2 presents the details on the Cochrane risk of bias tool and criteria for judgement.

We graded each of the above dimensions of trial quality as being at low, high or unclear risk of bias. In this review, stratified analysis based on study quality was not performed given the lack of included studies with high risk of bias.

Measures of treatment effect

Unit of analysis issues

The woman was the unit of analysis in all of the studies. In the case of multi‐arm studies, only the relevant groups were to be included. In the event that data from multi‐arm studies was used in single meta‐analysis, we divided the number of participants in the control group by the number of arms.

Dealing with missing data

Wherever possible, we contacted the authors of the original articles for missing data via the provided contact information in the original paper.

Assessment of heterogeneity

We evaluated clinical heterogeneity by qualitatively appraising differences in study characteristics such as participants, interventions, outcomes assessed, and study methodology. Quantitative pooling of the data was first justified by a consensus of clinical judgement of sufficient clinical homogeneity. We informally evaluated and investigated the degree of statistical heterogeneity by visual inspection of forest plots and more formally by using the Tau², I², and Chi² statistics. We regarded heterogeneity as considerable if the I² value was greater than 75%, substantial if the I² value was between 50% and 75%, moderate if the I² value was between 30% and 50%, and low if the I² value was less than 30%. In future updates of this review outcomes with substantial or considerable heterogeneity (I² greater than 50%) will be further evaluated for sources of heterogeneity and if found, subgroup analysis or meta‐regression analysis will be considered.

Assessment of reporting biases

We checked the methodology and study protocols of the primary studies where available. We assessed publication bias and other small‐study effects in a qualitative manner using a funnel plot.

Funnel plot asymmetry was tested using weighted linear regression of effect estimates on their standard error (SE) if more than 10 trials were included in an analysis (Egger 1997).

Data synthesis

Statistical analyses were performed using RevMan Web. Data synthesis of dichotomous outcomes was performed using the Mantel‐Haenszel method, with the results presented as  RRs and 95%CIs. The inverse variance method was used for continuous outcomes, and reported as MD with 95%CI. We analysed maternal satisfaction as a continuous outcome, even if measured on an ordinal scale. Some studies administered ropivacaine or levobupivacaine local anaesthetics in place of bupivacaine. For such studies we assumed 60% potency of bupivacaine based on a similar systematic review and meta‐analysis (George 2013), and the means and standard deviations (SDs) in our results were multiplied by 0.6.

In the case of data presented in the included studies as median and range, we attempted to obtain data in the form of mean and standard deviation (SD) from the respective authors. If this was not possible, we converted the median and range to mean and SD using the formula by Hozo 2005. Data presented as 95%CI were converted to SD (Cochrane Handbook 7.7.3.2 Obtaining standard deviations from standard errors and confidence intervals for group means).

We expected both clinical and statistical heterogeneity, and therefore we used the the random‐effects model for all analyses.

Subgroup analysis and investigation of heterogeneity

The following subgroup analyses were chosen a priori based on prior evidence of association with the outcomes in this review. 

1. Epidural technique: epidural alone versus combined spinal‐epidural technique (the dural‐puncture epidural technique was not included). Rationale: prior evidence suggests that combined spinal‐epidural technique may reduce breakthrough pain and local anaesthetic consumption (Tan 2019).

2. PCEA: regimens that used PCEA versus those that did not. Rationale: prior studies demonstrated that PCEA use was associated with reduced local anaesthetic consumption and breakthrough pain, while other studies reported that PCEA increased local anaesthetic consumption without improving analgesia (Tan 2019).

3. Nulliparous versus combination of nulli‐ and multi‐parous women. Rationale: nulliparity has been associated with increased risk of breakthrough pain in several studies (Tan 2019; Tan 2021).

Subgroup differences were analysed by testing for heterogeneity across subgroup results (Borenstein 2013).

In addition, outcomes with substantial or considerable heterogeneity (I² greater than 50%) were evaluated for sources of heterogeneity and if found, subgroup analysis or meta‐regression analysis were considered.

Sensitivity analysis

We did not perform sensitivity analyses on the quality of the studies because the quality of the studies was consistent across the different studies (Risk of bias in included studies). We will consider performing sensitivity analyses in future updates of this review if required.

Sensitivity analysis for trial quality involves analysis based on the rating of selection bias and attrition bias. We excluded studies of poor quality from the analysis (those rated as unclear or high risk of bias) in order to assess for any substantive difference to the overall result. The sensitivity analysis for compliance were based on trials where women did not receive their allocated treatment, combination therapy, or intervention, or if they received an additional form of analgesia to the one allocated. If required, these sensitivity analyses will be performed on the primary outcomes only.

Summary of findings and assessment of the certainty of the evidence

We used the principles of the GRADE system in order to assess the certainty of evidence associated with the following specific outcomes (Guyatt 2008).

1. Incidence of breakthrough pain requiring anaesthesiologist intervention during labour epidural analgesia

2. Incidence of caesarean delivery during labour epidural analgesia

3. Incidence of instrumental delivery during labour epidural analgesia

4. Duration of labour analgesia 

5. Hourly dose of local anaesthetic during labour epidural analgesia

6. Maternal satisfaction following fetal delivery

7. Apgar score at 1‐ and 5‐minutes following fetal delivery

We constructed a summary of findings table comparing programmed AMB versus BI using GRADEpro software (GRADEpro GDT 2015). The GRADE approach is a transparent and structured system of assessing the certainty of evidence based on the confidence that an estimate of effect reflects the true value. Evidence from randomised trials were assigned high certainty, but can be downgraded based on risk of bias, inconsistency of results, indirectness of evidence, imprecision, or publication bias.

In this review, the risk of bias was considered present if there was a high risk of lack of allocation concealment or assessor blinding, significant loss to follow‐up, or selective reporting that may affect interpretation of results. The GRADE level was downgraded one level for inconsistency if I² 50% to 90%, and two levels if I² >90%. Imprecision was considered present if the upper or lower 95% CI extended from the line of equality by >5%. Publication bias was considered significant with P < 0.05 in the Egger’s test.

Assumed comparator risks for dichotomous outcomes were derived from the median outcome incidence in patients receiving basal infusion within the studies included in this systematic review.