Main findings
The GBS trial did not find any evidence that use of a rapid POC intrapartum test for maternal GBS colonisation in women whose babies are considered to be at higher risk reduced the rates of intrapartum antibiotics administered to prevent early-onset GBS infection, compared with usual care where IAP is offered to women based on only risk factors. There were no differences between the two strategies in maternal antibiotics use for any indication. However, use of the rapid test reduced the neonatal exposure to antibiotics and fewer babies were administered with antibiotics for suspicion of early-onset infection than in usual care. The rapid intrapartum test, when administered in a real-life scenario, showed good sensitivity and specificity. Overall, 4 in 10 mothers with risk factors, and 1 in 10 of their babies, were colonised with GBS. For the outcomes of cost per case of maternal antibiotics avoided, rapid test was not more cost-effective than usual care.
There is some evidence to suggest an association between multidrug-resistant Gram-negative bacterial colonisation in the mother and similar colonisation in the newborn at 6 weeks of age.
Strengths and weaknesses of the cluster randomised trial
The cluster randomised trial with the embedded test accuracy study successfully implemented the rapid test strategy to enable robust comparison with usual practice. Although two units withdrew soon after randomisation, we ensured that these were replaced with two other units. We recruited more than the required number of clusters to take into account potential dropout of sites after recruitment started, although that was not so in our case and the required sample size was reached. The usual-care units all started and completed their accruals ahead of the rapid test units, although this lack of synchronicity did not appear to influence the outcomes. However, it is possible that eligible women may have been missed in the rapid test arm as, unlike the usual-care arm where the data were collected from recorded notes, lack of trained staff availability, both during the day and night, may have influenced recruitment to the intervention arm.
This primary outcome was a process outcome and so the within-cluster correlation of this outcome, the ICC (i.e. the proportion of the total variance in the IAP rate that can be explained by variation between maternity units), was expected to be higher than it would be for a clinical outcome. We, therefore, considered the sensitivity of our calculations to a range of proportions in the usual-care group and a range of ICC values that we believed to be quite conservative. The ICC was calculated at the end of the trial at 0.06 (95% CI 0.03 to 0.12). This was in the range anticipated at the start of the trial and used in the sample size calculation. Analysis of the primary outcome allowed for clustering at a maternity unit level as a random effect and corrected for the small number of clusters.
Owing to the obvious differences in the two strategies, blinding of the clinical staff and women was not possible. The primary outcome was an objective measure that was recorded in the women’s health-care record, and so adjudication was not required. It is, however, entirely possible that awareness of the maternal GBS colonisation modified the threshold of clinicians for suspecting early-onset infection and prescribing antibiotics to babies. We did not define clinically suspected infection in the protocol. Knowledge of their unit’s allocation could influence the research midwives’ selection of reasons for antibiotic administration from the woman’s health-care records, but under the data protection restrictions of the ethics approval, source data verification by trial staff was not possible.
To compare strategies, a randomised comparison provides the most reliable data. We considered and rejected individual randomisation because of the risk of contamination (i.e. once the GeneXpert system was made available on a maternity unit, it would be difficult for midwifery staff not to offer this testing strategy to all consenting women, given the presumed high accuracy, thereby removing the benefit of randomisation).74
In cluster trials, if participants cannot be identified in advance, it is important that of all those who are eligible follow their cluster’s allocated strategy, without overt or unintentional selection. The trial showed some evidence of differential ascertainment of participants across rapid test and usual-care units, both with respect to the number of participants and some of the characteristics of the participants.
There were imbalances in the number of women in whom GBS was detected earlier in pregnancy, with more cases detected in the rapid test units. These women might have been more aware of the test procedure and actively requested the intrapartum test. Conversely, there were more women with a diagnosis of a raised maternal temperature in the usual-care units, although closer inspection shows a wide range of rates in the rapid test units (1–36%), excluding the two units that did not reach their targets. This is further evidence of midwives approaching women for the rapid test for selective inclusion in the study. Adjusting the primary comparison for the risk factor rates did not substantially alter the result.
The proportion of women receiving IAP in the usual-care group was considerably lower than expected at 36%, compared with a projected 50–75%, given that all women should be considered for IAP by virtue of maternal risk factors being observed. There were very few missing study data on the outcome of antibiotics prescribed, but there are fewer sources of routinely collected data on IAP rates against which to compare our observation. In the previous UK study51 of rapid test accuracy, only 50% of women with a risk factor received IAP. The RCOG GBS audits found that most maternity units had local policies aligned with the RCOG guidelines, but these audits were undertaken before the 2017 revision of the guidelines.75 The audits highlighted the occasional misinterpretation of the definition of IAP, antibiotics for other maternal reasons and antibiotics used as prophylaxis against wound infection in caesarean deliveries. It is possible that was some miscoding of the need for IAP in the usual-care arm. Notwithstanding this, two-thirds of women received intrapartum antibiotics of any kind. The survey of units prior to randomisation illustrated the difficultly in collecting IAP administration rate routinely, without resorting to audit of individual health-care records. Instead, a crude estimated rate was obtained from the number of vials of benzylpenicillin and clindamycin dispensed on each delivery suite, although this gave rates of IAP use ranging from 7% to 87%.
Strengths and weaknesses of the test accuracy study
The robust design and execution of our test accuracy study allows us to be confident that the estimates of accuracy are valid and identifies specific aspects of the testing process that maximise the accuracy. We minimised methodological bias by ensuring that the rapid tests and selective enrichment culture were performed independently and interpreted blind to each other. In addition, the double-headed swab used in this study was designed to ensure that the index and reference tests were undertaken on contemporaneous samples. Maternal colonisation was significantly higher than in the previous GBS test accuracy study (at 41% vs. 29% in the presence of risk factors).51
There was also a good proportion of participants providing rapid test data that were verified by selective enrichment culture at 74%. The missing test data were equally distributed between missing rapid test and missing selective enrichment culture test results, with little overlap, and appeared to be due to systemic problems in conducting the tests, in different hospitals and for limited periods of time, rather than evidence of systematic partial verification based on rapid test results. The number of paired test data, at 534, were smaller than the 606 required by the sample size calculation, which was due in part to the number of failed rapid tests and the need to curtail accrual of data at the end of the study recruitment period. However, this was counterbalanced by the higher than anticipated maternal GBS colonisation rate, and so, although it was anticipated that there would be 167 women with GBS colonisation in a cohort of 606 women, we observed 229 positive selective enrichment cultures. This number of true-positive cases provided 96% post hoc power to determine whether or not the observed accuracy was less than the minimally acceptable threshold of 90% sensitivity; therefore, we can be confident that our observed results can be interpreted as clinically acceptable. The design of the trial had inherent spectrum bias, as only those women presenting with risk factors for early-onset GBS infection were included; however, previous work has shown that post-test probability of maternal colonisation did not vary substantially between women with risk factors present and women with risk factors absent.51
Strengths and weaknesses of the economic evaluation
To our knowledge, ours is the first economic evaluation to be conducted alongside a RCT comparing a strategy of exploring rapid test with screening based on risk factors alone for higher risk of early-onset GBS infection. The resource use and outcome data were prospectively collected at different points in the trial. Unit costs were obtained from standard and recognised sources. The cost-effectiveness results also benefited from the robustness of the main analyses and sensitivity analyses. One of the key principles of health economic analysis is to maximise the health benefits from, and ensure the most efficient allocation of, scarce resources. It is plausible to incur analyses that suggest that a potentially small decrement in an outcome is acceptable on cost-effectiveness grounds if the potential cost saving is great enough to more than offset the loss in health outcome, and if the saved resources can be used to better effect elsewhere. However, this interpretation does not apply in the current analysis because of uncertainty in a number of areas. The principal issue in the current analysis is that the rapid test has not managed to achieve the hypothesised reduction in the use of prescribed antibiotics to reduce GBS transmission. Another limitation relates to the intermediate outcome on which the economic analysis is based. The outcome of ‘cost per additional antibiotic prescription avoided’ is severely limiting. It is an intermediate outcome, as it provides no indication of the pathways that would be followed for the infant based on any decision to prescribe antibiotics to the mother for preventing GBS. The key issue is the pathway and wellness of the baby in this study, for which there are no data. Such an intermediate outcome is credible only when the hypothesis is upheld by the results. What we do not know with any certainty is whether or not the antibiotics prescribed in the rapid test units helped to avoid cases of GBS that might otherwise have occurred. There was very limited swabbing of neonates to establish whether or not they incurred colonisation with GBS. This was recorded for only 40% of infants in the rapid tests units and, therefore, the results are potentially biased by the large number of missing data. Furthermore, these data were available only for the rapid test units as no swabbing of infants for colonisation was undertaken in the usual-care units.
It has been neither possible nor appropriate to conduct a model-based economic evaluation in this study. When analyses are based on an intermediate outcome, it can be appropriate for the health economic analysis to model beyond the outcome recorded in the trial to ascertain the likely impact on longer-term health outcomes. However, it is inappropriate and infeasible to attempt to model beyond the intermediate outcome in this case, as any assumptions required cannot be supported by any data. If it is assumed that women appropriately received more antibiotics based on the results of the rapid test, compared with usual care, then the assumptions in the model would be in the direction of preventing more cases of GBS colonisation. In the absence of data on GBS colonisation in the usual-care units, such an assumption is not supportable. Furthermore, such an assumption directly opposes the hypothesis that was predetermined in the trial design (i.e. that the rapid test will reduce the number of women who are prescribed antibiotics). Therefore, modelling beyond the intermediate outcome collected in the trial is not appropriate in these circumstances and would produce misleading results.
Public and parent involvement
We have been supported throughout the project by the charity Group B Strep Support (GBSS) (Haywards Heath, UK) and, in particular, its chief executive. Public and parent involvement was crucial in improving the acceptability of the GBS2 trial and promoting engagement of maternity units. We engaged with GBSS chief executive throughout, improving our understanding of the opinions and uncertainty surrounding testing for GBS among pregnant women, doctors, microbiologists and midwives, and the barriers to accessing testing that women encounter. This prompted us to provide a study-specific testing information leaflet to all eligible women. Members of the GBSS charity and a small group of parents on the GBSS mailing list commented on public-facing materials and posters to ensure that they were clear and comprehensive. The GBSS helpline took many telephone calls from women booked to give birth in maternity units allocated the rapid test strategy, supporting their decisions whether or not to consent to the swab and the implications of the test.
The substudy public-facing materials and approach for consent pathway were discussed and reviewed by Katie’s Team (London, UK), an East London-based patient and public involvement group that includes members of the public with lived experiences of pregnancy.
We will engage with GBSS regarding the dissemination of our findings, providing a plain English summary of the findings and the uncertainties around the evidence we have discussed here. This will be distributed via GBSS’s website, e-newsletter and social media channels. Any future research groups taking forward the research recommendations from this project would benefit from engaging with GBSS.
Interpretation of findings
Use of the rapid test did not appear to lower the use of IAP administered to the mother to prevent vertical transmission of GBS to the offspring. This could be for the following reasons. First, the rate of IAP for GBS in the usual-care arm, in which every participant should have been offered the IAP, was less than what was expected (at < 40%) and this may have influenced the findings. However, our work reflects current clinical practice and we would expect similar findings if rapid test was implemented. Second, when the outcome of maternal antibiotics for any indication was considered, although the overall rates of antibiotic use increased, it still does not show a reduction with rapid test. Third, clinicians still decided to administer IAP to prevent GBS in 17% of test-negative women, indicating reluctance in some instances to follow the test result, which may have affected the primary outcome. Conversely, 22% of test-positive women did not receive IAP. Furthermore, the test results were available in only four-fifths of all women and IAP was administered to fewer than half of these women.
However, we found a significant reduction in neonatal exposure to antibiotics in the rapid test units. Antibiotics are often started prophylactically in newborns when the clinicians suspected a high risk of neonatal sepsis based on symptoms in the newborn, maternal history and any treatment given to the mother. In our trial, fewer babies in the rapid test arm than in the usual care were suspected to have sepsis, a major indication for starting antibiotics. The high rates of women administered antibiotics 4 hours prior to birth in the rapid test (compared with the usual-care arm) and because neonatologists were less likely to commence antibiotics in rapid test-negative women could have influenced these findings.
The accuracy of the GeneXpert rapid test was marginally better than that of the previous generation of GBS test from Cepheid, the Smart GBS test, which had a reported sensitivity of 84% (95% CI 79% to 84%) and specificity of 87% (95% CI 85% to 89%). The first-generation test was performed in batches by a laboratory assistant, whereas in the GBS2 trial the test was performed in real time by clinical midwives, as a true POC test. Removing data from 244 women who had used an antimicrobial lubricant or cleanser prior to the swab being taken marginally increased the accuracy of the rapid test, compared with the complete population. Conversely, considering only those who were exposed to an antimicrobial solution gave a reduced sensitivity of the rapid test [61 rapid test positive/75 microbiological culture positive, 81% (95% CI 71% to 89%)]. This confirms that the accuracy of the rapid test can be influenced by the environment from which the swabs are taken, and that lubricant or cleanser containing antimicrobial substances, such as chlorhexidine, should be avoided. The NICE guidelines recommend using just water for hygienic cleaning,76 as there is no strong evidence for efficacy of antimicrobial solutions in reducing neonatal infection.30
The results of the economic evaluation show dominance: the strategy of adding the rapid test to usual care is dominated by usual care alone in achieving the predetermined outcome. As dominance exists, an ICER is not appropriate. The results of the stochastic CEA and the presentation in the cost-effectiveness planes support the interpretation that the strategy of adding the rapid test to usual care is less effective than usual care alone in achieving the objective of reducing antibiotic use. Almost all of the scatterplot dots are to the west (less-effective quadrant) of the vertical axis (see Figures 9, 12 and 14); this is because more women in the rapid test arm are prescribed antibiotics than women in the usual-care arm. Whether this is ultimately more or less costly in the short term is more uncertain, as the scatterplot spanned the north- and south-west quadrants of the cost-effectiveness plane in the base-case analysis.
The sensitivity analyses broadly supports these base-case results, although there are some slight impacts on costs. In sensitivity analysis 1, where the length of stay for mothers is based on trial data, the impact is to increase the overall costs of the rapid test strategy compared with usual care. In sensitivity analysis 2, where the inpatient days for the baby, by level of care, are added to the calculation of costs, the impact is to reduce the overall cost of the rapid test strategy compared with usual care. This latter sensitivity analysis result may imply that some benefit was inferred onto the infants from the mothers in the rapid test units who received more antibiotics than those in the usual-care units. This may have led to fewer inpatient days for infants in the rapid test units and a reduction in overall costs. However, such an implication must be interpreted with caution.
Interpretation of the substudy
Seale and Millar77 reviewed the literature on perinatal transmission of antibiotic resistance from mothers to infants and found a paucity of relevant literature. A more recent systematic review78 reports evidence of carriage of indistinguishable antibiotic-resistant Gram-negative bacteria by mothers and infants, and also concludes that there is a paucity of relevant research and that ‘none of the included studies reported on risk factors for multidrug resistant Gram-negative bacteria transmission from colonised mothers to infants’. There remains little information on the contribution that maternal carriage of resistant bacteria or associated genes makes to carriage by their infants. Recent reports have suggested a weak (if any) association between mode of birth, maternal and infant antibiotic use or feeding practices on the carriage of antibiotic-resistant Gram-negative bacteria by infants.79,80 Despite collecting data on parental carriage of antibiotic resistance, Hetzer et al.81 did not include information in the analysis of risk factors for acquisition of antibiotic-resistant intestinal E. coli during the first year of life.
The identification of risk factors for perinatal transmission has the potential to allow the development of preventative strategies. A study published in 2010 reported that, among 114,999 women who gave birth between 1992 and 2007, approximately one-third had received an antibiotic during pregnancy.82 The extent to which antibiotic prescribing in pregnancy is inappropriate is unknown. Antibiotic stewardship is a potential remediable route to reducing the selection and transmission of antibiotic-resistant bacteria during this critical period of life; however, the extent to which antibiotic usage in pregnancy can be constrained without prejudicing maternal or neonatal outcomes must be determined.
Implications for practice and future research
The GBS2 trial was unable to demonstrate if rapid test-based IAP in women with risk factors for early-onset GBS infection in newborn babies reduces the antibiotic usage for GBS prophylaxis in labour. CIs were wide and included both increases and decreases in the use of antibiotics, which might be clinically important (from a 7% decrease to an 18% increase). The GBS2 trial also highlights the low adherence to the current UK guidelines for GBS prophylaxis and also the widespread use of antibiotics in labour for other maternal indications.
There is work to be carried out to improve the current rate of IAP provision among women with risk factors for early-onset GBS infection. The rapid test did appear to reduce neonatal antibiotic administration, suggesting that neonatologists are willing to take the results of the test into account when considering whether or not to start antibiotic treatment. Intrapartum GBS colonisation status is a variable in the neonatal sepsis risk calculator83,84 that is gaining traction in the UK and has been shown to substantially reduce neonatal antibiotic use.85 The accuracy of the rapid test shows that it has potential to inform both maternal and neonatal antibiotic use.
Use of the rapid test is feasible in clinical practice, with turnaround times compatible with clinical care. The requirements for installation of the rapid test and training of all clinical midwives to allow a unit to provide round-the-clock testing should not be underestimated. Previous research has suggested that midwives want to take ownership of the testing process51 and, anecdotally, once trained, midwives could incorporate the rapid test into their practice. Clinicians will need to be reassured regarding the performance of test, if IAP administration is to be based on rapid test results.
The rapid test is accurate in the diagnosis of maternal GBS colonisation, but there needs to be improvements in reducing the numbers of failed tests. Clinicians should be advised to administer IAP in the event of the failed test. Routine use of the rapid test is not a cost-effective strategy for avoiding maternal antibiotic administration. Any future studies should endeavour to collect data on neonatal clinical outcomes, such as babies colonised in all trial groups and complications of infection to enable economic modelling.
The GBS2 trial is limited to women with risk factors for early-onset GBS infection in their baby. Risk factors are associated only with about one-third of UK early-onset GBS infection. As the GBS2 trial demonstrates, the proportion of women approached for intrapartum testing was lower than previously reported risk factors rates, suggesting imperfect implementation of the testing strategy. The National Institute for Health Research has commissioned a further trial called GBS3 (HTA 17/86/06). This is a cluster randomised trial in 80 maternity units, comparing routine testing of all women intending vaginal birth with the same usual-care strategy. The routine testing units will be further randomised between the same intrapartum rapid test system and antenatal testing by selective enrichment culture. The primary outcome will be all-cause early-onset all-cause neonatal sepsis, both culture proven and clinically suspected, whereas the main outcome for the subrandomisation will be the proportion of women for whom a test result is available in time for adequate IAP to be administered. Approximately 320,000 women and their babies will be included in the study, which will use data from routinely collected sources. A full economic evaluation of the three strategies will be conducted with a long-term perspective. Interviews with parents and all cadres of maternity unit staff will describe qualitatively the barriers to implementation of the testing strategies.