Methods

Parts of this chapter have been reproduced from Daniels et al.1 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: https://creativecommons.org/licenses/by/4.0/. The text below includes minor additions and formatting changes to the original text.

Trial design

The GBS2 trial was a multicentre, prospective, unblinded, parallel-cluster, randomised controlled trial (RCT), with a nested test accuracy study and a nested microbiological substudy, with economic evaluation.

The GBS2 trial had a favourable ethics opinion from the West Midlands – Edgbaston Research Ethics Committee (reference number 16/WM/0036), including waiver of individual research consent.

There were three substantial amendments to the protocol. The first amendment was approved on 18 April 2017, before the first unit started identifying women, and aligned eligibility criteria to those anticipated to be in the revised 2017 RCOG guidelines,4 which redefined the maternal risk factors for early-onset GBS infection. The second amendment was approved on 7 March 2018 and updated the recommended antibiotics, following the publication of the RCOG’s guidelines. The third amendment was approved on 3 December 2018 and formally capped each unit’s recruitment target at 86 women.

Cluster randomised controlled trial

The GBS2 trial was designed as a cluster RCT that was randomised at the level of the maternity unit. Sites were randomised to the rapid test strategy (i.e. IAP administration based on the rapid test results) or to usual care (i.e. the standard risk-based screening strategy and IAP offered to all women with risk factors).

Nested test accuracy study

To establish the real-time accuracy of the Cepheid GeneXpert system rapid POC test for GBS colonisation in women presenting to a labour ward with risk factors for early-onset GBS infection, the results of the rapid test were compared with the reference standard of selective enrichment culture in a prospective cohort study. The results of the rapid test preceded those of the culture test and interpretation of the reference standard was performed blind to the rapid test.

Process evaluation in rapid test sites

In the units that were allocated to rapid testing in the cluster randomised trial, we undertook an observational study to determine in real time the timings between rapid test results and commencement of IAP, and to identify if there were any failures in producing results once the tests were initiated on the machine.

Economic evaluation

The economic analysis is based on primary clinical data collected in the cluster RCT, adopting the UK NHS perspective. Full methods of the health economics analysis are described in Chapter 7.

Microbiological substudy

A substudy embedded in the GBS2 trial estimated the relatedness of certain bacterial species that are of public health concern that may colonise the mother during the birth of her baby and her child when they reach 6 weeks of age. The substudy also aimed to detect the antibiotic susceptibility of these strains. This substudy took place in participating units in London and the south-east of England (LSE) that were randomised to receive a rapid test system. The methods are described in Chapter 8.

Centres and participants

Eligibility of centres

Maternity units were eligible to participate in the GBS2 trial if they were prepared to accept a policy of rapid test-directed IAP administration to all women with risk factors for the duration of the trial period when allocated to the rapid test strategy. Maternity units also had to be prepared to include all preterm labours considered as high risk, irrespective of the implementation date of the RCOG’s guidelines and its current local policy. The trusts hosting the maternity units were also required to have access to microbiology facilities that were able to perform a selective enrichment bacteriological culture to detect GBS.

Randomisation

Randomisation of clusters was performed at the Birmingham Clinical Trials Unit using a minimisation algorithm programed in a Microsoft Excel^® spreadsheet (Microsoft Corporation, Redmond, WA, USA) incorporating the following factors:

• region [the Midlands (MID) or LSE]
• pre-trial IAP rate (above or below the median)
• the number of vaginal or emergency caesarean births (above or below the median).

The pre-trial IAP rates and the birth rates over a period of 12 months were determined from all potential recruiting units prior to randomisation. This estimate provided the trial with the size of the population eligible to be tested, and served as a minimisation variable and as the denominator for the IAP use calculation. Antibiotic data were derived from interrogation of hospital records and pharmacy prescribing databases at the level of the maternity unit, not at the individual level. Using the standard dosing regimen of antibiotics, the size of the eligible population and the assumptions regarding the average duration of antibiotic administration, we estimated the number of women who received IAP prior to the trial. The median value of the eligible population sizes and pre-trial IAP rates were calculated for the first 20 maternity units that were intending to participate in the GBS2 trial, and these were used as thresholds for dichotomising the minimisation variable data.

Blinding

Owing to the different study procedures under the two testing strategies, it was not possible to blind maternity unit staff to their allocation. A summary of the blinding status along the trial pathway is shown in Figure 3.

FIGURE 3 Cluster randomised trial blinding timeline for the GBS2 trial. EOGBS, early-onset group B Streptococcus (disease/infection). This figure was produced using the Timeline Cluster Tool.53 (Figure not reproduced here; see the original article.)

Setting

The cluster was defined as the maternity unit. Women were identified and screened for eligibility by clinical midwives and doctors in various locations, including the delivery suite, the maternity triage unit and the induction ward. No research-specific consent was obtained, although women in the rapid test units provided verbal clinical assent to have the vaginal/rectal swab.

Participant eligibility criteria

The eligibility pathway for the GBS2 trial is shown in Figure 4.

FIGURE 4 The GBS2 trial eligibility criteria flow chart. (Figure not reproduced here; see the original article.)

Women were eligible for inclusion in the GBS2 trial if they met one or more of the following criteria:

• a previous baby with early- or late-onset neonatal GBS disease, as reported by the mother and documented in the maternal notes
• GBS bacteriuria during the current pregnancy, as documented in the maternal notes, irrespective of whether or not the GBS bacteriuria was treated at the time of diagnosis with antibiotics
• GBS colonisation of the vagina and/or rectum (determined from a vaginal/rectal swab) in current pregnancy, as documented in the maternal notes
• preterm labour (< 37 weeks’ gestation) whether suspected, diagnosed or established, regardless of whether membranes were intact or there was PROM of any duration
• maternal pyrexia (≥ 38 °C) observed at any point in labour, including clinically suspected/confirmed chorioamnionitis.

Women were screened for eligibility to participate in the GBS2 trial if they presented to the maternity unit in preterm labour (suspected, diagnosed or established), regardless of rupture of membranes, in term labour (latent or established) or if they were about to be induced.

Participant exclusion criteria

Women were ineligible for inclusion in the GBS2 trial if they were aged < 16 years; their pregnancy was at < 24 weeks’ gestation; they were in the second stage of labour at admission or considered likely to give birth to their baby imminently; they had planned an elective caesarean birth; or their baby was known to have died in utero or had a congenital anomaly incompatible with survival at birth.

Study procedures

The study procedures at each site varied according to the testing strategy randomly allocated to the participating maternity unit. The recommended antibiotic regimen for preventing early-onset GBS infection in both maternity unit groups of the trial were identical: intravenous administration of 3 g of benzylpenicillin as soon as possible after the onset of labour, and half of that dose at 4-hourly intervals until birth. If the woman was known to be allergic to penicillin she was offered a cephalosporin, and if she had a history of serious reactions to beta-lactams, vancomycin was indicated.

The protocol stipulated that all women in the usual-care group should have received IAP, whereas only those who tested positive or for whom a test result was not available should be offered IAP in the rapid test group (unless there was a clinical reason for prescribing IAP or the women had already had a previous baby with GBS infection and requested IAP).

Usual-care units (risk factor-based provision of intrapartum antibiotic prophylaxis)

The study procedure for usual-care sites, in which IAP is offered to women with clinical risk factors, is outlined in Figure 5. All women with risk factors for early-onset GBS infection, whether initially apparent or emerging, were considered eligible for the study.

FIGURE 5 Study procedures for usual practice maternity units. (Figure not reproduced here; see the original article.)

Rapid test strategy

The units that were randomised to the rapid test received a GeneXpert^® Dx IV GBS rapid testing system (Cepheid), which was installed and commissioned by a field specialist from the manufacturers. Each unit allocated to the rapid test screening policy was required to house the GeneXpert machine and computer centrally in the unit and have it operational at all times. Units were provided with a sufficient supply of Xpert GBS test cartridges (Cepheid) and double-headed swabs in transport tubes containing Stuart transport medium. In units participating in the substudy, additional single-headed swabs were also supplied. The use of the rapid test for screening was restricted to women who presented to the labour ward and who were eligible according to the participant eligibility criteria. A summary of the pathway for rapid test units is shown in Figure 6.

FIGURE 6 Study procedures for rapid test maternity units. a, Failed tests include, but are not limited to, machine failure and failure to process swab in required time frame; and b, unless otherwise clinically indicated. This was at the discretion of local clinical care. (Figure not reproduced here; see the original article.)

Obtaining a vaginal/rectal swab

Swabs were taken using a double-headed swab. Depending on the stage of labour, the swabs were obtained by either the woman herself or a suitably trained member of the woman’s care team. This could be on admission to the labour or induction ward before a vaginal examination was performed, or after a risk factor was detected (e.g. when maternal fever was observed). Swabs were first taken by gently rotating the swabs across the mucosa of the lower vagina. The same swab was then used to take a sample from the rectum, inserting the swab through the anal sphincter then gently rotating. The shafts of the double-headed swabs were then separated carefully. One was returned to the transport tube and sent to the local microbiology laboratory for selective enrichment culture to detect GBS (see Enrichment culture method) and the other was immediately used for the rapid test. For sites taking part in the microbiological substudy, and for participants who had consented to take part, an additional single-headed swab was taken and sent to the substudy laboratory (see Chapter 8).

In practice, a number of substances containing antimicrobial compounds administered vaginally may interfere with the results of the rapid test. The use of lubricant that contains antimicrobial substances, such as K-Y Jelly (Reckitt Benckiser, Slough, UK), for vaginal examination or for taking swabs was discouraged. It was suggested that sterile non-bacteriostatic fluid (e.g. sterile water or saline) was used when lubrication was necessary. Other procedures that may have involved antimicrobial substances include placing a pessary to induce labour, or chlorhexidine to cleanse the perineum or as a vaginal douche (although this is not currently indicated in NICE guidelines).54 Nevertheless, as the GBS2 trial aimed to establish the use of a rapid test in a real-world setting and, indeed, vaginal examinations are usually undertaken to establish labour, women for whom antimicrobial substances were used were not excluded from the trial. Instead, the use of any antimicrobial substances after admission to the maternity and prior to the swab being taken was recorded.

Delayed labour

If more than 48 hours had elapsed since the test result became available and the woman had yet to give birth, the test result was regarded as invalid. In this situation, it was advised that the woman should be re-swabbed and the presence of GBS tested for again using both the rapid test and the microbiological techniques.

Conducting the rapid intrapartum test

After sampling, one swab was used to inoculate a test cartridge. Should the inoculated cartridge not have been loaded into the rapid test system and the test commenced within 15 minutes of the swab being introduced into the cartridge, the test was be deemed to have failed. The NHS or hospital number was entered into the GeneXpert software alongside the woman’s unique trial number. This trial number was identical on each item of study material associated with one woman and her child (or children). The test was conducted as per the operator manual.55 It takes, on average, 35 minutes to get a result if GBS is present, 55 minutes to confirm if there is no GBS present and an error message is presented if the test has failed. Used test cartridges and swabs were disposed of in accordance with local policies for clinical waste. Instructions provided to sites for conducting the rapid test are available on request from the manufacturer.

Enrichment culture method

The second vaginal/rectal swab was sent to the local microbiology department where it was used to inoculate a selective enrichment medium prior to plating to detect the presence of any GBS, as per the current Public Health England recommendations.56 Results from the microbiological cultures were returned to the care team using the usual reporting pathways and were recorded in the woman’s notes. Swabs, sent to the microbiology laboratory for the determination of the GBS colonisation status, were inoculated into Todd–Hewitt broth for overnight enrichment at 37 °C. This enriched broth was subcultured on chromogenic GBS agar and incubated aerobically overnight at 37 °C. GBS was identified by the presence of pink/red colonies. Microbiological data were transcribed to the GBS2 trial database by either a member of the microbiology department or a local research nurse. When manual data entry was required, the data entry screen did not allow review of the rapid test results by those outside the study office, therefore reducing the risk of review bias.

Neonatal ear canal swab

For eligible women in rapid test units, a single swab was taken from the baby’s ear canal as soon as convenient after birth. This swab was put into a transport tube, labelled with a numbered sticker and sent to microbiology for culture using the hospital’s usual request system to detect the presence of GBS, as per the mother’s swab (see Enrichment culture method). The neonatal swab was not used in the rapid test machine.

Antibiotic regimen

Subsequent clinical management of mother and baby was the responsibility of the local health-care team and was not directed by the GBS2 trial. Recommendations for appropriate antibiotic care are included in the NICE guidelines57 and RCOG’s Green-top GBS guidelines.4

Outcome measures

Owing to the difference in the strategies for testing women and for directing IAP, it was not possible to blind women or their care team to the randomised allocation. Data were extracted from maternity and neonatal notes by research midwives within each unit who were involved in the implementation of the study, and therefore it was not possible to blind them to the randomised allocation.

Research midwives regularly collected data from consecutive eligible women regarding the use of antibiotics from the women’s health-care records, including information on whether or not any antibiotic was given, along with the date and time of IAP initiation.

Primary outcomes

• The proportion of women who received only IAP for GBS prophylaxis, of all women identified with one or more risk factors for neonatal GBS infection.
• Measures of test accuracy (i.e. the sensitivity and specificity of the GeneXpert GBS rapid test).

The indication for any intrapartum antibiotic administration was ascertained by making available the following options: GBS prophylaxis (in usual-care units only), a positive rapid test and a failed rapid test (both in rapid test units), pyrexia during labour, a caesarean birth, maternal request and other reason (to be specified). Multiple reasons could be recorded.

Secondary outcomes

Cluster randomised trial

Maternal

Intrapartum maternal antibiotic use for any indication

• The proportion of women receiving any intrapartum antibiotic that has been indicated as being for GBS prophylaxis for a maternal clinical indication, such as pyrexia, on maternal request, prior to a caesarean or for any other reason, as a proportion of those women identified as having one or more risk factors for early-onset GBS infection.

Intrapartum maternal antibiotic use for any indication other than for a caesarean

• The proportion of women receiving any intrapartum antibiotic that has been indicated as being for GBS prophylaxis for a maternal clinical indication, such as pyrexia, on maternal request or for any other reason other than for a caesarean, as a proportion of those women identified as having one or more risk factors for early-onset GBS infection.

Postpartum maternal antibiotic use for any indication

• This was defined as those women receiving any postpartum antibiotic that had been indicated as being a maternal clinical indication, such as pyrexia, on maternal request or for any other reason, as a proportion of those women identified by the delivery suite midwives as having one or more risk factors for early-onset GBS infection. The period in which these data were collected was from birth until the mother’s discharge from either the hospital she gave birth or any hospital to which she was immediately transferred. Data on antibiotic use following any readmittance or prescribed by her general practitioner were not collected.

Time of intrapartum antibiotic prophylaxis exposure

• This was defined as the duration between the start time of the first dose of IAP and the birth of the baby. Sufficient exposure was considered as an interval of either > 2 hours or > 4 hours before birth.

Neonatal

Neonatal antibiotic use for prophylaxis or treatment

• The proportion of babies receiving antibiotic prophylaxis because of maternal GBS status or antibiotic treatment for suspected or confirmed neonatal infection, as a proportion of babies born to women identified as having risk factors for early-onset GBS infection.

Suspected neonatal infection

• The number of babies prescribed antibiotics for presumed neonatal infection, as a proportion of all live born babies.

Neonatal group B Streptococcus colonisation rates

• The rate of GBS-positive selective enrichment cultures from the neonatal ear swabs, as a proportion of all neonatal ear swabs cultured.

Neonatal mortality

• Includes stillbirth rate and early neonatal death (before 7 days), and these are combined as the perinatal mortality rate.

Test accuracy secondary outcomes

Maternal and neonatal colonisation rates, and the mother-to-baby transmission rate.

Process outcomes

A number of process outcomes were also evaluated, for the rapid test units only, including the:

• duration between a positive test becoming available on the GeneXpert machine and the time the result is collected by a midwife, and the duration between that point and the start of IAP
• proportion of the cartridges on which the tests were not commenced within 15 minutes of inoculation, which is defined as an invalid test
• proportion of tests initiated on the Cepheid GeneXpert machine that failed to produce a result within 55 minutes, which is defined as a failed test, or were reported as failed by the system.

Serious adverse event outcomes

Serious adverse events are expected to be extremely rare. There may be significant consequences of the failure to offer IAP to a woman with GBS risk factors and to women identified by rapid test as being colonised with GBS in terms of the increased risk of their baby developing early-onset GBS infection. Conversely, there is a risk of overtreatment if IAP is administered to women without risk factors or with a negative rapid test. However, these instances were considered outcomes of interest within the study, and not adverse events. The risk-based approach involves the noting of historical risk factors and the monitoring of women for emerging risk factors, such as chorioamnionitis. This presents no risk to the women other than a failure to identify and act on these risk factors. The rapid test requires a vaginal/rectal swab to be collected during labour, which is benign and presents no foreseeable risk of harm. As with almost any diagnostic test, there was a risk that testers may suffer an inoculation injury (most likely mucous membrane exposure) with clinical material. A proportion of women will receive antibiotics, particularly benzylpenicillin for IAP, which carry a very small risk of anaphylaxis.58

The outcomes of economic evaluation and for microbiological substudy are provided in Chapters 7 and 8.

Statistical methods

For baseline characteristics, categorical data are summarised by frequencies and percentages. Continuous data are summarised by the number of responses, mean and standard deviation (SD) if deemed to be normally distributed, and number of responses, median and interquartile range (IQR) if data appear skewed. Tests of statistical significance were not undertaken. In accordance with the Consolidated Standards of Reporting Trials (CONSORT) reporting guideline, both absolute and relative measures of treatment effects are reported, and so for the primary and secondary outcomes in the cluster randomised trial component both risk differences (RDs) and relative risks (RDs) are presented where possible. Estimates of differences between groups are presented with 95% two-sided CIs. p-values are reported from two-sided tests at the 5% significance level and no corrections are made for multiple testing.

In the primary analysis for the cluster randomised trial, a mixed-effects binomial regression with a log-link was used to estimate the RR, and a binomial model with identity link was used to estimate the RD. In the case of non-convergence of the binomial model with a log-link, a Poisson model with robust standard errors was fitted. If the binomial model with the identity link did not converge, then only a RR was reported. Both models allowed for clustering by maternity unit as a random effect. To correct the potential inflation of the type I error rate due to small number of clusters, the Kenward and Roger method59 was used.

Comparative estimates of differences between groups were adjusted for the hospital and the minimisation variables of region (i.e. LSE or MID), baseline IAP rate (i.e. equal to or greater than the median or less than the median) and the number of vaginal or emergency caesarean births (i.e. equal to or greater than the median or less than the median). These effects were all be assumed to be fixed in the regression models used. A secondary analysis for the primary outcome additionally adjusted for maternal temperature of ≥ 38 °C observed while in labour, any previous baby with GBS disease, GBS bacterium detected in current pregnancy, and woman in suspected/diagnosed/established preterm labour. When covariate adjustment was not practical, unadjusted estimates were produced and it was made clear in the output why this occurred (e.g. not possible because of low event rate, lack of model convergence or poor recording accuracy of covariates). Multiple imputation analysis, accounting for clustering, was planned if the proportion of observations with missing complete information on these covariates was > 5%.

Prespecified subgroup analyses were limited to the primary outcome. The following subgroup effects were investigated:

• maternal temperature of ≥ 38 °C observed while in labour
• previous baby with GBS disease
• GBS bacterium detected in current pregnancy
• preterm labour (< 37 weeks’ gestation) with intact membranes or rupture of membranes of any duration, whether suspected, diagnosed or established.

Treatment effects were summarised within each subgroup separately, and an interaction test was performed between each subgroup variable and the received allocation.

The diagnostic accuracy of the rapid test was estimated through the standard calculations of sensitivity and specificity. Estimates are presented with 95% CIs calculated using binomial exact methods.60 We also undertook a binomial proportions test to compare the observed sensitivity with the hypothesised minimal performance value of 90%.

Sample size for the cluster randomised trial

The focus of sample size calculation was on the effectiveness of the rapid test. We aimed to recruit a minimum of eight clusters per randomisation group (i.e. a total of 16 maternity units). This was a realistic recruitment goal and was deemed sufficient to allow estimation of model parameters. We estimated that the sample size per cluster would be approximately 83 (this is informed by routinely collected data on number of eligible women giving birth). This equated to a total sample size of approximately 664 per strategy group to allow recruitment of the required number of individuals for the test accuracy study. As there were two strategy groups (i.e. usual care and rapid test), each with at least 664 participants, it was expected to provide a sample size of 1328 women. This was rounded up to a target sample size of 1340 women. To allow for dropouts at the level of the cluster, anticipating implementation issues in maternity units, the number of clusters was increased from 16 to 20.

The proportion of women receiving antibiotics to prevent vertical GBS transmission (i.e. the primary outcome in the cluster RCT) was expected to be in the region of 50–75% (based on the previous study51 and assuming better compliance with RCOG guidelines since that study). This primary outcome was a process outcome and so the within-cluster correlation of this outcome [the intracluster coefficient (ICC)] was expected to be higher than it would be for a clinical outcome. We therefore considered sensitivity of our calculations to a range of proportions in the usual-care units and a range of ICC values, which we believe to be quite conservative. All of our calculations allowed for 90% power and 5% significance. We did not acknowledge varying cluster sizes in our calculations, as we considered the coefficient of variation of unit birth sizes to be small (approximately 0.21 based on annual birth rates for 2013) and could be altered by varying the study duration at maternity units. However, given our conservative assumptions, and that the impact of the testing strategies on the primary outcome was expected to be large (i.e. a priori we expect the effect size to be large), we expected the impact of any varying cluster sizes to be minimal. For a range of values for the important parameters, as described above, we worked out the detectable differences and equivalent relative RR (Table 1).

This means that this trial would have around 90% power to detect a reduction in proportion of women prescribed antibiotics to prevent GBS transmission from 75% to 63% (i.e. a RR of about 20%) for a low value of the ICC to a reduction from 75% to 38% for a very conservative value of the ICC (0.2), equating to a RR reduction of 50%.

Sample size for the test accuracy study

The sample size of the test accuracy study was dependent on the sensitivity of the rapid test. For the test to be proven useful, we needed to show that it would detect a higher proportion of GBS colonisation than other tests, but not at the cost of low specificity and/or unnecessary administration of IAP. Results from the Group B Streptococcus 1 (GBS1) study51 suggested that the most cost-effective test (if untargeted universal IAP was excluded) was antenatal culture for GBS at 35–37 weeks’ gestation. In the GBS1 study51 the sensitivity of this test was 75.8% (95% CI 47.2% to 91.5%). Therefore, if we could prove that the sensitivity of rapid test was > 90% then the results of the GBS2 study51 would be convincing. This was a stringent test and a lower threshold might also be adequate. We did not compare the rapid test with antenatal GBS culture testing within the study, but compared with this result from external literature. Therefore, we undertook a ‘one sample, sample size’ computation comparing against a fixed value.

We had data from a systematic review on the performance of the GeneXpert GBS test (J Daniels, personal communication). From the meta-analysis of nine studies, the pooled accuracy of the test was estimated, giving a sensitivity of 96.4% (95% CI 90.8% to 98.6%) and specificity of 98.9% (95% CI 97.5% to 99.5%). Sample size calculations are therefore based on showing that a test with sensitivity of 96.4% is greater than a fixed value of 90%. With a power of 90% to demonstrate this sensitivity, 167 cases of maternal GBS colonisation are required (or 136 at 80% power).

A sample size of 676 women would provide 90% chance of us accruing enough GBS-colonised women to have 90% power to show the sensitivity of the rapid intrapartum test to be statistically significantly (with p < 0.05) > 90% should the meta-analytical estimate of its performance (96.4%) be correct, while allowing for 10% loss from failed tests, based on the GBS prevalence observed in the GBS1 study,51 which was 29.8% (89/299, 95% CI 24.6% to 35.2%). Of the 606 participants with data, we would expect 167 to be GBS carriers and 439 to be negative for GBS colonisation. If the prevalence of GBS colonisation was actually at the lower end of the 95% CI from GBS1 study,51 namely 24.6%, then a total of 673 women would give a 90% chance of observing 136 cases of GBS colonisation, including 10% lost tests. The 95% CIs we would observe on sensitivities and specificities of 85%, 90%, 95% and 98% with a sample size of 676 women (606 women with data) are shown in Table 2, and these have adequate precision (sensitivity within 10% and specificity within 6%) for modelling.