Rapid testing for group B streptococcus during labour: a test accuracy study with evaluation of acceptability and cost-effectiveness

Epidemiology of neonatal GBS disease

Understanding the epidemiology of EOGBS requires consideration of maternal colonisation with GBS, the risks and timing of vertical transmission, and the pattern of neonatal colonisation and disease.

Transmission

Neonates with early-onset infection show initial colonisation mainly in the mucous membranes of the respiratory tract, and the major route of vertical transmission at the time of delivery is thought to be through aspiration of vaginal and amniotic fluid. Vertical transmission in utero is also thought to occur as a consequence of prolonged rupture of membranes (PROM) and is regarded as one of the causes of stillbirth. 16 Colonisation of the mother is less predictive for late-onset GBS infection, with prematurity the major risk factor. 17

The association between the rates of maternal colonisation, transmission and infection has been established. A meta-analysis12 of six studies of the maternal and baby colonisation rates in an untreated general population showed a transmission rate of 36.4% (95% CI 28.1–45.0%). A further meta-analysis in the same report of EOGBS disease in colonised babies of untreated mothers gave an average incidence of 3.0% (95% CI 1.6–4.7%).

These rates, as illustrated in Figure 1, project an overall incidence of 1.5 cases per 1000 deliveries, with a potential range of 0.4–3.9 cases per 1000. This is consistent with a reported incidence from a surveillance population of 0.47 (95%CI 0.42–0.52) cases of EOGBS per 1000 live births. 18

FIGURE 1.

Model of colonisation and transmission of group B streptococcus (GBS) and early-onset GBS disease.

Risk factors for neonatal GBS disease

Various factors at the time of birth have been shown epidemiologically to increase the risk of GBS disease, presenting either as early- or late-onset infection. A recent systematic review12 estimated that 71% of deliveries had no recognised maternal risk factors for GBS disease.

Bacteriological culture

GBS grows on blood agar plates, forming characteristic glossy white colonies surrounded by areas of β-haemolysis after 24–48 hours. The use of a selective enrichment broth before plating is widely recommended to optimise the recovery of GBS from genital and anorectal samples, increasing the recovery rate by over 50%. 31 Lim broth, comprising a Todd–Hewitt base with nalidixic acid and colistin to suppress gram-negative bacteria, is the most widely used enrichment media before plating on to blood agar, although the necessity of selective enrichment has been questioned. 32 Obtaining swab specimens from both vaginal and rectal sites increases the incidence of maternal GBS colonisation by 40% over vaginal swabs only. 33,34

Rapid tests with timely screening potential

Several non-culture-based tests are available that could be developed into rapid point-of-care diagnostic tools for intrapartum screening. This would allow optimal targeting of IAP to women carrying GBS during labour. As well as being accurate, the ideal test would need to be rapid enough to allow adequate time for IAP to be effective, and would require minimal preparatory steps and be easily interpretable to enable routine use on busy delivery suites. Each of the tests currently available has advantages and disadvantages. Even laboratory-based use of these tests is limited, and there has been no proper evaluation of any of these tests in the point-of-care setting.

Accuracy of rapid tests for GBS

To determine the accuracy and rapidity of various intrapartum maternal GBS colonisation tests we conducted a systematic review of the literature. 35 A systematic search for test accuracy studies, identified without language restriction, was performed on the MEDLINE and Cochrane databases, bibliographies of known primary and review articles, and through contact with authors, experts and manufacturers. From a total of 1296 citations, 23 relevant papers of 29 test accuracy studies assessing a total of six tests were identified. The majority of excluded papers assessed tests antenatally.

The quality of the included studies was assessed using a standard checklist of indicators of methodological bias. Studies were considered to be of high quality if they reported a prospective design, consecutive patient enrolment, an adequate test description, blinding of the test results and use of selective medium for incubation of specimen for gold standard culture. Unfortunately, as shown in Figure 2, the quality of the studies and reporting thereof was poor, and only three were large enough (sample size > 1000) to give reasonable precision.

FIGURE 2.

Methodological quality of studies included in the systematic review of intrapartum tests for maternal group B streptococcus (GBS) colonisation.

The prevalence of maternal GBS colonisation in these studies varied from 5% to 32%,36,37 yet none explored the variation in the test’s performance across a spectrum of population subgroups. Most studies obtained vaginal swabs for tests and gold standards without speculum examination. Only four studies assessed the accuracy of rectal swabs. None of the studies considered the acceptability of testing during labour from the women’s or health-care providers’ perspectives, nor the costs or cost-effectiveness of the testing strategies.

The review shows that many of the GBS tests, with the exception of real-time PCR and OIA, either took too long to produce a result or were not of sufficient accuracy to be feasible for maternal intrapartum testing. The review focused on studies in which selective media were used for gold standard culture. Exploration of the reasons for heterogeneity in the primary study results was hampered by the small number of studies included in the review. Pooled data produced summary sensitivities and specificities, shown for PCR in Figure 3 and for OIA in Figure 4. PCR had a pooled mean sensitivity of 97% (95% CI 94–98%) and a pooled mean specificity of 97% (95% CI 96–98%). Pooled sensitivity for OIA was 55% (95% CI 48–61%) and pooled specificity was 96% (95% CI 95–97%). Metaregression analysis showed that test accuracy did not vary according to overall quality (p = 0.58) or the addition of anorectal swab (p = 0.064). Funnel plot analysis did not show any evidence of asymmetry to indicate the presence of publication or related bias for the largest subgroup of studies.

FIGURE 3.

Summary of pooled sensitivity and specificity for intrapartum polymerase chain reaction (PCR) testing for maternal group B streptococcus (GBS) colonisation (studies using selective enrichment culture as reference only). PCR, conventional PCR; real-time, real-time PCR; R, rectal sample; V, vaginal sample.

FIGURE 4.

Summary of pooled sensitivity and specificity for intrapartum optical immunoassay (OIA) testing for maternal group B streptococcus (GBS) colonisation (studies using selective enrichment culture as reference only). V, vaginal sample.

Although OIA seems less accurate than PCR, it was more rapid and less complex to perform, making it more feasible as a near-patient test. The relative accuracy of the two tests remains unclear because of the poor quality of previous accuracy studies and the absence of any direct comparison against a common standard. This systematic review highlighted a clear need for a well-designed accuracy study comparing PCR and OIA tests, the subject of this Health Technology Assessment (HTA) report.

Risk factor-based screening

Risk factor-directed screening is based on the assessment of women at labour for the presence of one or more of the risk factors described in the previous section on risk factors for neonatal GBS disease. IAP may then be offered to those with risk factors.

This approach was introduced in the UK in 2001 when the GBS Working Group of the Public Health Laboratory Service produced interim guidelines. 59,60 These recommended IAP specifically for women who had had a previous GBS-infected baby or when GBS was found incidentally in the vagina or urine during the current pregnancy. It was also recommended that IAP should be considered for preterm births, when the mother has intrapartum fever and/or when there is PROM in labour. For mothers requiring broad-spectrum antibiotic therapy for chorioamnionitis it was recommended that the chosen regimen should include activity against GBS.

This strategy was endorsed by the Green-top Guideline of the Royal College of Obstetrics and Gynaecology (RCOG) in 2003. 61 The guidelines were revised slightly in 2004 when the Health Protection Agency advised that IAP should be considered only when GBS is detected incidentally in pregnancy. 62

Bacteriological screening at 35–37 weeks’ gestation

This approach involves the culturing of vaginal and rectal swabs from all women between 35 and 37 weeks of gestation and offering IAP to those in whom GBS is detected. Inevitably, results will not be available in time for all women screened this way, either because of prematurity or delivery before the results are available. Women with missing culture results can be either treated on the basis of risk factors or automatically offered IAP. However, culture-based approaches rely on colonisation at the time of swab collection being predictive of colonisation status at the onset of labour.

In the USA interim guidelines produced in 1996 suggested that either a risk factor- or a culture-based approach should be adopted but that IAP should be offered to all those considered at risk. 56,63 The risk factors considered as definite indicators for IAP were preterm delivery, PROM (> 18 hours) or intrapartum fever. Under both strategies women with GBS bacteriuria during their current pregnancy, or who previously gave birth to an infant with EOGBS disease, were candidates for IAP.

These guidelines were revised in 2002 to recommend universal screening by culture of 35- to 37-week swabs. Women in labour for whom the culture status was not known would be screened on the basis of risk factors. Further refinements included advice for penicillin-hypersensitive women and an algorithm for threatened preterm delivery, otherwise recommendations from 1996 were retained.

A meta-analysis31 of four studies, involving nearly 4000 women, assessing the accuracy of culture at 35–37 weeks, with colonisation status at delivery as the reference standard, has recently been performed. The mean sensitivity was 76% (95% CI 47–92%) and mean specificity 95% (95% CI 89–99%). This sensitivity estimate is 21% lower than that for PCR in the review by Honest et al. ,35 suggesting that PCR could be a promising alternative to culture. This would require confirmation through an RCT.

Vaccination

Vaccination of pregnant women offers the opportunity for primary prevention of GBS disease by two mechanisms. First, a vaccine that induces mucosal immunity would decrease maternal colonisation and consequently the risk of transmission to the baby. Potentially more important would be the transplacental transmission of protective antibodies to the baby. Babies with low concentrations of antibodies to GBS proteins have an OR of 0.002 of developing EOGBS disease compared with those with greater levels. 64 Protective maternal antibodies are believed to persist in the baby for about 3 months after birth,65 affording additional passive protection against late-onset disease.

The search for a suitable candidate molecule for vaccination has been ongoing for two decades, but a vaccine has yet to be licensed for use and evaluated for effectiveness in reducing neonatal GBS disease. Initial developments involved carbohydrate-based vaccines, and immunogenic efficacy has been demonstrated in women. 66 The major disadvantage to this approach is that there are five major, and several minor, serotypes of GBS, each with a different outer carbohydrate, and so any vaccine would have to be multivalent and appropriate to the serotype prevalence within the population. Focus therefore shifted to a ubiquitous protein that is present on the outer surface of all GBS serotypes. 67 Protein-based antigens are inherently more immunogenic than carbohydrates, are less likely to cross-react with human tissues and can more readily be manipulated by molecular techniques. Phase 1 trials have been conducted for one protein candidate. 68

With all candidate vaccines, to definitively demonstrate effectiveness would ultimately require a placebo-controlled trial with a reduction in the incidence of GBS disease. Given the low incidence of disease, this would require an unfeasible number of participants. Surrogate outcomes, such as neonatal antibody titre, would need to be accepted by licensing authorities before a vaccine could be approved. For the purpose of this study, and the modelling of cost-effectiveness, the data and assumptions used in the cost-effectiveness analysis of Colbourn et al. 69 are used.

The evidence base for screening strategies

In the USA the incidence of EOGBS disease fell from 1.7 per 1000 live births in the early 1990s to 0.5 per 1000 a decade later and 0.34 per 1000 in 2004. 70,71 Implementation of the 1996 policy was widespread, with over 90% of women being screened by culture at 35–37 weeks’ gestion. 63 Both risk factor- and culture-based strategies can lead to a reduction in the incidence of EOGBS disease, although, because of the large sample size required, the strategies have not been directly compared in a randomised trial. 63 A direct non-randomised comparison of the strategies,72 incorporating population-based surveillance for EOGBS disease into a sample survey of a population of over 600,000 live births, found that the culture-based screening approach was > 50% more effective than the risk-based approach at preventing perinatal GBS disease. Amongst the cohort who underwent screening by culture, 18% of all deliveries were to GBS colonised mothers who did not have risk factors. A greater proportion of GBS colonised women received IAP than those with obstetric risk factors, although overall 24% of women in each cohort received IAP. Had implementation been perfect, 31% of all women screened by culture and 29% screened under the risk factor strategy would have received IAP.

The UK has not adopted a similar culture-based policy, despite pressure to do so, as there is no evidence of its clinical effectiveness and cost-effectiveness in the UK setting. The incidence of EOGBS disease in the UK in the absence of systematic screening or widespread IAP is 0.5 per 1000 births,18 which is similar to that seen in the USA after universal screening and IAP, despite comparable maternal colonisation rates. It is estimated that, in the UK, assuming that 30% of women receive IAP because of GBS colonisation and that IAP is 80% effective at preventing GBS infection, 714 colonised women would need to receive IAP to prevent one case of GBS disease and over 7000 would need to receive IAP to prevent one neonatal death. 61 Actual efficacy rates may be lower than this estimate. A national swab-based screening programme would require a substantial reorganisation of the provision of antenatal care in the UK.

Neither the risk factor approach nor culture-based screening at 35–37 weeks is ideal. The risk factor-based approach is inherently crude, with only 60% of EOGBS disease cases having risk factors apparent at labour. 18,20 Screening at 35–37 weeks’ gestation may result in those pregnancies at highest risk of GBS disease being missed: in the UK, 7.4% of births occur before 37 weeks, whereas 32–38% of GBS disease occurs in these neonates. 20,22 Additionally, antenatal screening and treatment has not been shown to have an effect on all-cause neonatal mortality43 and may carry disadvantages for the mother and baby. These include potentially fatal anaphylaxis,73 medicalisation of labour and the neonatal period, and neonatal infection with antibiotic-resistant bacteria. The risk of neonatal infection has still not been fully elucidated. One US study showed an increase in the incidence of Escherichia coli infection in low-weight neonates, comparable to the decline in GBS. 74 Other studies have not confirmed this finding, and it is also possible that the use of benzylpenicillin as IAP in the UK may exert less selective pressure than the broader-spectrum antibiotics often used in the USA.

Acceptability of GBS screening

Very little research has been reported on the acceptability to parents or health professionals of any of the means of preventing neonatal GBS infection. And, in contrast to other fields of antenatal screening, we know little about the psychological impact of GBS screening on mothers. One study,75 conducted in Taiwan, measured attitudes to testing and the anxiety levels of women routinely tested for GBS between 35 and 37 weeks’ gestation. This study found that, although state anxiety rose in women who screened positive, 1 week after delivery anxiety levels had returned to normal, and all mothers supported screening. An Australian study of women’s attitudes to GBS and screening76 showed that most women knew very little about GBS and those who underwent screening did so because it was part of routine care and was perceived to be better for the baby. A qualitative study77 in Canada found that women did not feel that the low risk associated with GBS warranted the use of antibiotics, and they were also wary of taking a vaccine during pregnancy to prevent infection, should this become available.

Two surveys of health professionals’ attitudes, knowledge and practice in relation to preventing neonatal GBS infection have been conducted. An Australian study78 reported that only 56% of the obstetricians and 61% of the neonatologists who responded supported universal antenatal screening. Nearly all of those who rejected this approach supported antenatal screening by risk factors. However, a study79 conducted in New Zealand reported that, although the lead maternity carers they surveyed believed that the prevention of neonatal infection was important and that they were confident in determining risk factors, none correctly identified all high-risk factors and only 26% regularly discussed GBS with their clients.

The technology to allow rapid testing for GBS during labour has only recently been developed and so the acceptability of this form of screening for women and health professionals has yet to be assessed. However, there is a need to better understand both pregnant women’s and health professionals’ perceptions of GBS screening and the prevention of neonatal GBS infection during labour more generally. Any screening and infection prevention regime relies on the adherence of participants and health professionals.

Cost-effectiveness of GBS screening

Despite the contrasting guidelines for screening and prevention of neonatal GBS disease that exist for the UK and the USA, with the former recommending screening based on risk factors and the latter screening based on culture at 35–37 weeks’ gestation, there was no cost-effectiveness evidence available to support either.

Recently, Colbourn et al. 69 carried out a study to determine the cost-effectiveness of prenatal strategies for preventing GBS and other bacterial infections in early infancy and to establish the expected value of further information. Their evaluation was carried out using a decision tree based largely on secondary data. They compared strategies of screening based on culture; screening based on the PCR rapid test; screening based on risk factors; IAP; and vaccination. Their results suggested that screening based on risk factors or on the rapid test PCR was not a cost-effective strategy, and that vaccination was the most cost-effective option.

Given the recent developments in technology that have resulted in the development of the rapid test, a full economic evaluation based on data from the first primary study to evaluate its use in practice is required to ensure that decision-makers use available resources wisely.

Study sample

All pregnant women booked for delivery at one of the two participating units, other than those electing for a Caesarean delivery, were approached for consent to be recruited into the study. Given the need to fully inform each woman about the study, to provide time for her to consider participation and to avoid burdening her with information at the time of labour, a two-stage informed consent strategy was employed. Study information leaflets were given to all mothers at the time of their antenatal booking visit or at the mid-trimester scanning visit. Community midwives in the catchment areas for the hospitals were trained to reinforce the information provided and answer any questions that the women may have had.

Between 20 and 24 weeks, the provisional consent of the women was sought. Those agreeing in principle to provide swabs at labour were registered on a database, recording identifying and demographic details and expected date of delivery. Women’s hand-held notes were marked with coloured stickers to identify whether they had provisionally consented or declined, thereby aiding identification of women to be reapproached for consent at presentation in labour. All women who had agreed in principle were asked to consent to participation when they presented to the labour ward with anticipated delivery or when they were admitted to the antenatal ward for induction of labour at later than 24 weeks’ gestation. Women who had previously declined, who were undergoing elective Caesarean section or who were delivering very prematurely (≤ 24 weeks) were not approached for consent at labour.

Dilatation of the cervix, together with regular, progressively more frequent and painful contractions, indicates the onset of labour. Accuracy of tests for GBS colonisation may be altered by vaginal examinations performed before obtaining swabs for testing,82,83 therefore swabs were taken from the lower vagina and rectum before examination. Women who first presented in false labour, defined as no delivery within 7 days, were reapproached when they returned in labour. Swabs from false labours were not included in the accuracy analysis.

To minimise variation in the amount of bacteria on individual swabs, and to increase patient acceptance, a triple-headed swab was used to collect three samples simultaneously. This consisted of three separate swabs bound together by two plastic bands (Medical Wire and Equipment, Corsham, Wiltshire). The swabs were obtained by the midwife or doctor assessing the woman at the time of admission to the antenatal or labour ward. A numbered testing kit was taken from stocks kept on the wards. These contained the swabs, the reagents for the rapid tests and a transport medium for the reference test, a data collection form and numbered stickers to identify the samples.

Separate triple swabs were used to sample first the lower vagina and then the rectum, when possible before any manual vaginal examination. Vaginal specimens for testing were obtained by gently rotating the swabs across the mucosa of the lower vagina. Rectal swabs were obtained by inserting the swabs through the anal sphincter and then gently rotating. The triple swab was then separated and each swab used for either one of the rapid tests or the reference test.

Index tests

The PCR and OIA rapid tests were performed separately on rectal and vaginal swabs using a standard operating procedure for point-of-care testing, using the Cepheid Smart GBS^® kit and SmartCycler^® system and Inverness Medical BioStarOIA STREP B^® kits, respectively, described in Appendix 1. The swabs were tested on the antenatal or labour ward by trained midwifery assistants or by research staff (the majority of samples).

Reference tests

Laboratory-based selective enrichment culture of maternal and neonatal swabs was the gold or reference standard for verification of index test results. Swabs were inoculated into Lim broth with subculture onto tripticase soy agar after overnight incubation at 37°C. 84 The reference swab cultures were interpreted independently of the index tests by biomedical scientists. A password-restricted database prevented laboratory staff accessing index test results or neonatal outcomes and, as culture results were not available until later, maternity staff could not access laboratory data.

A swab from the neonate was also collected to determine transmission rates. The external ear canal was chosen on the basis of published studies showing it to be the most sensitive indicator of neonatal colonisation. 85–87

Management of labour and neonates

Apart from collection of swabs from mothers and their babies, all other aspects of patient management were entirely at the discretion of the local doctors. The PCR and OIA tests were performed on the labour ward, but the results were not provided to the midwife responsible for the woman. Treatment decisions were made solely according to established local guidelines, based on the presence or absence of risk factors:61 an incidental finding of GBS colonisation or GBS bacteriuria during pregnancy (GBS in the midstream urine specimen or vaginal swab tested opportunistically), previous baby with GBS disease, maternal fever (> 38°C) or chorioamnionitis, PROM (≥ 18 hours at term) and prematurity (< 37 weeks).

When IAP was given, the agents and dosage regimens were in accordance with the Green-top Guideline of RCOG. 61 Briefly, intravenous penicillin, (3 g) given as soon as possible after the onset of labour and 1.5 g given 4-hourly until delivery, was the treatment of choice. Clindamycin 900 mg 8-hourly was used for women who were allergic to penicillin. Of the 122 women included in the final analysis who received antibiotics, none had any adverse reactions. Recent antibiotic therapy was also documented but was not a contraindication to inclusion in the study.

Sample size and power estimation

The literature that exists on sample size calculations for diagnostic studies provides a number of different approaches. 88,89 Given the consequences of GBS infection it was felt that sensitivity below 75% would be unacceptable. Assuming a ‘true’ sensitivity of 90%, 70 cases would need to be recruited to refute reliably a sensitivity of less than 75% with 90% power at p = 0.05. This would require a total of 1400 participants assuming a 5% prevalence of maternal GBS colonisation, as shown in Table 1. The specificity could be estimated with greater reliability because of the large number of mothers without GBS in the study. This approach also produced a sample size estimate that was consistent with the sample size estimate of Alonzo et al. 89 for comparing the accuracy of tests. The assumption regarding disease prevalence was deliberately conservative, predicting rates of 5% or 10% compared with the pooled estimate of 13.6%,12 so as not to compromise the power for estimating sensitivity. 90 A higher prevalence would increase the power for subgroup analysis for maternal colonisation.

Characteristics of participants

The first women were approached for participation in the study in March 2005 and the first woman was swabbed in June 2005. Recruitment to the study closed in January 2007 with 1418 women swabbed, 945 from Birmingham and 473 from Essex, as shown in Figure 6.

FIGURE 6.

Recruitment of women into the diagnostic accuracy study (STARD flowchart). *No reason recorded 827 (44%), objected to rectal swab 73 (3.9%), objected to swabbing 241 (12.8%), planned Caesarean 132 (7.0%), delivery planned elsewhere/at home 17 (0.9%), insufficient English 111 (5.9%), did not want to participate in research 199 (10.6%), wanted no added intervention 77 (4.1%), family refused 38 (2.0%), worried if things went wrong/anxious about pregnancy 28 (1.5%), other reasons 137 (7.3%). Non-participants did not have to give a reason nor were midwives expected to record one. **No time to collect swabs 86 (5.5%), Caesarean section 94 (6.0%), delivered elsewhere or at home 56 (3.6%), language barrier 2 (0.1%), delivered after end of study 12 (0.8%), midwife not trained 8 (0.5%), no sticker on notes/denied giving provisional consent 314 (19.9%), pregnancy loss 8 (0.5%), mother too distressed 4 (0.3%), clinical complications at admission 58 (3.7%), missed for various unknown reasons 630 (40.0%), no reason identified 303 (19.2%). Midwives were not required to record reason. †Throughout study, colonisation or test results from either site were reported only if both vaginal and rectal swab results were available.

Results from 18 women were excluded from the analysis as delivery occurred later than 7 days after collection. A total of 10 women had false labours and were reswabbed on later presentation in labour; the second set of swabs were included in the analysis.

The characteristics of the women who participated in the study are shown in Table 2. Of the 1400 women recruited into the study, 308 (22.1%) had risk factors.

The mean gestational age was 40 weeks (standard deviation of 11.42 days for spontaneous deliveries and 10.84 days for inductions). About one-third of inductions were at gestational age ≤ 40 weeks. The average length of labour was 8 hours 15 minutes for spontaneous deliveries and 7 hours 20 minutes for induced deliveries.

Maternal GBS colonisation, as defined by a positive enriched culture result, was 15.5% from vaginal swabs, 19.2% from rectal swabs and 21.2% if either result was positive. Colonisation rates varied depending on the presence or absence of risk factors, as shown in Table 3.

A total of 1291 baby ear cultures provided colonisation status, out of which 109 were culture positive. Of these, 99 were born from GBS-colonised mothers (as determined by either vaginal or rectal positive culture results), a neonatal colonisation rate of 8.4% of all deliveries and transmission rate of 36.3% of colonised women. There were 15 babies reported to have had an infection immediately postpartum, six of which were invasive infections. Of the invasive infections, three were diagnosed as EOGBS disease, all of whom recovered.

Accuracy of PCR and OIA for maternal GBS colonisation

The accuracy of the rapid tests, compared against enriched culture, is shown in Table 4. PCR performed significantly better than OIA in all combinations of index and reference standard. Considering the PCR results, rectal PCR provided the most sensitive test for maternal GBS colonisation with a sensitivity of 71% (95% CI 66–76%), whereas specificity was 92% (95% CI 90–93%), identical to the specificity of vaginal PCR. Combining the PCR results from both sites increased the sensitivity to 84% (95%CI 79–88%) but with a lower specificity of 87% (95% CI 85–89%). This reflects the fact that, in the pooled test–reference accuracy measures, either test being positive defines GBS colonisation whereas both tests have to be negative for the participant to be considered negative for GBS colonisation.

To statistically compare the relative accuracy of OIA and PCR, the results of OIA and PCR compared with the reference standards were tabulated and McNemar’s test applied. All five combinations of rapid test and reference standard were significantly different, in favour of PCR being more accurate.

Variation in index test accuracy according to the presence or absence of maternal risk factors for GBS colonisation

There was variation in the specificity of the PCR rapid test with the presence or absence of risk factors for every combination of site and reference standard, except for vaginal PCR compared against either the vaginal or rectal reference standard. Variability was also observed in the sensitivity of vaginal OIA. In general, tests were more sensitive in those with risk factors and more specific in those without risk factors (Table 5). Other features of the population and tests that could potentially influence the sensitivity and specificity of the test were considered and, although not prespecified, subgroup analyses were performed to investigate variability. Comparing the accuracy of the rapid tests with maternal colonisation between parous and nulliparous women showed no heterogeneity in the test characteristics (data not shown).

Predictive post-test probabilities

The logistic regression model used maternal GBS colonisation from either vaginal or rectal culture as the binary dependent variable. As shown in Table 6, the prior probability of maternal colonisation for women in this study was 21.7%, the prevalence according to vaginal or rectal culture. This increased to a post-test probability of 64.5% with a positive maternal PCR result from either the vaginal or the rectal swab, and reduced to 4.8% with a negative PCR result. When the presence or absence of risk factors was taken into account these post-test probabilities did not change substantially, despite a slight change in the prevalence in these groups.

Determinants of neonatal GBS colonisation

The multivariable logistic regression model that evaluated the determinants of neonatal colonisation (Table 7) showed that the odds of having a colonised baby increased when an intrapartum rapid vaginal or rectal PCR test was positive but not when maternal risk factors were present. The odds were reduced by 78% with use of sufficient antibiotics (OR 0.22, 95% CI 0.07–0.62, p = 0.004), defined here as at least 4 hours between first dose and delivery. When any antibiotic provision was considered in the model the odds of neonatal colonisation were not reduced significantly.

Summary of main findings

This is the largest test accuracy study comparing PCR and OIA rapid intrapartum tests for maternal GBS colonisation. Rapid PCR was more accurate than rapid OIA for the diagnosis of maternal GBS colonisation. Index test results derived by using vaginal or rectal swab results were more sensitive than either test considered individually. The best accuracy was obtained using vaginal or rectal PCR rapid test results, giving a sensitivity of 84% (95% CI 79–84%) and specificity of 87% (95% CI 85–89%). The sensitivity and specificity of rapid tests varied according to the presence or absence of maternal risk factors. The PCR test results significantly altered the post-test probability of maternal colonisation and were determinants of neonatal colonisation. Knowledge of risk factor status did not alter this substantially.

Strengths and limitations of methods

The validity of our findings relies on the quality of the study. We complied with and reported all criteria for a high-quality test accuracy evaluation. 100 We minimised methodological bias and explored for variability in estimation of test accuracy. We ensured that the index tests and reference standard were performed independently and interpreted blind to each other. There was a very high proportion of index test verification by reference standard (over 96% for PCR). We investigated the effect of spectrum bias by collecting data on characteristics of the sample and assessing variability across various spectra. A sample size calculation was performed to ensure that the study had sufficient power to exclude a clinically unacceptable accuracy and the study recruited to that target. We are therefore confident that our findings concerning superiority of PCR over OIA are robust.

With regard to the generalisability of the findings of our study, one needs to explore the extent to which the observed results can be expected to be replicated in routine care. To begin with, the study sample may not be truly representative of a typical UK maternity population. Several differences were noted compared with other studies on the subject and compared with findings in the general population, as outlined in the following section. Test results may not be as consistently and rapidly obtained when the tests are employed in a point-of-care situation and conducted by staff providing usual care to labouring women. Indeed, the median times taken to obtain rapid test results within the study were 80 minutes and 35 minutes for PCR and OIA, respectively, excluding collection of the swab, compared with 65 minutes and 18 minutes in the time and motion study (see Appendix 6). Although in our study research staff performed the majority of the PCR tests, they did so in batches, with results usually obtained after delivery, as the test technology precluded samples being processed as they were received in real time. New developments in PCR platforms will eliminate most of the preparatory steps and allow multiple samples to be tested on demand, although the accuracy of such systems will need to be validated. Caution is needed when considering the applicability of our findings, which may be moderated by local circumstances such as the availability of staff and their capability to perform tests on demand (see Chapter 5, Strengths and limitations, and Appendix 2 for further discussion).

Interpretation of findings

In this section we seek to interpret our data and to explore the clinical relevance of the findings. We have briefly discussed the validity of the evidence and the extent to which the results can be applied to usual care. Here we compare our findings with other evidence, to put them into context.

Concerning the representativeness of our study sample, the patients recruited comprised only about 10% of the total number of women delivering in the two centres and there is evidence to suggest that the sample was not representative. Proportionally more white women were recruited (e.g. 62% in the study compared with 55% of the total deliveries at Birmingham Women’s Hospital), following a national trend in research in which South Asians are under-represented101 but which may also reflect a lack of acceptability of the procedure or other barriers to participation. Variations in the reasons for non-participation between ethnic groups will be discussed in Chapter 3 (see Declining to take part in the study during pregnancy). Only 2.7% of pregnancies included in the study delivered prematurely, lower than the national average of 7.1% of all live births in England and Wales. 102 This may be because staff tended to avoid asking this group for consent, to avoid additional anxiety for the mother. Conversely, a greater proportion of women in the study were undergoing induction of labour than the population average, 45% compared with approximately 18% (Birmingham Women’s Healthcare NHS Trust, 2007, personal communication), as midwives had more time to obtain consent and take samples on the antenatal ward. Emergency Caesarean sections are also over-represented in the study sample, making up 21.6% of the deliveries compared with a national rate of 13.5%;103 although this excess will not influence maternal GBS colonisation it might have an effect on neonatal colonisation. A lower proportion of study participants had risk factors than in other reported studies – 22% compared with 28.9% in the study of Colbourn and Gilbert12 – possibly because fewer premature births were recruited.

The rate of maternal GBS colonisation in this study, at 21.2% for combined vaginal or rectal colonisation, is higher than that reported in a recent meta-analysis of UK studies (14%; 95% CI 10–18%),12 although the latter includes studies reporting vaginal colonisation only. The risk of neonatal GBS colonisation given maternal colonisation is similar to the pooled estimate from a meta-analysis of six UK studies. 12 The study sample was too small to compare EOGBS disease rates with those in previous reports.

Women were not offered IAP on the basis of the index test results but were treated according to RCOG and Health Protection Agency guidelines,61,62 although not comprehensively. The administration of antibiotics to the mother may eliminate GBS colonisation and prevent transmission, but only if a sufficient dose is received before delivery. Multiple logistic regression demonstrated that a positive rapid PCR test does significantly increase the odds of neonatal colonisation, whereas IAP significantly reduces the odds if given for at least 4 hours before delivery. There were insufficient cases of EOGBS disease to determine predictors of disease.

We found that the accuracy of the rapid tests was not as high as that previously reported. The highest levels of accuracy came from combining the results from vaginal and rectal swabs, for both PCR and OIA. However, this was considerably lower than the pooled estimates from a meta-analysis of all previous studies. 35 That reported a pooled sensitivity and specificity of 97% for PCR and a pooled sensitivity of 55% and a pooled specificity of 96% for OIA. This may be because, with more robust methodology, we avoided an overestimation of accuracy associated with review bias, a feature prevalent in previous studies. 104

Updating the meta-analysis to include our study reduces the pooled estimates for PCR, with a new pooled sensitivity of 90% (95% CI 88–93%) and a pooled specificity of 92% (95% CI 91–94%). For OIA the pooled sensitivity increased to 63% (95% CI 57–97%) and the pooled specificity decreased to 79% (95% CI 78–80%). Although meta-analysis weights the individual study data according to the precision of the study, it does not take into account the methodological weaknesses of the included studies. With PCR, for example, six of the seven previous studies did not blind the results of the index test from the interpreter of the reference standard, which may have biased verification. The accuracy of PCR, when considering samples from both sites, compares favourably with that of screening by culture of swabs taken at 35–37 weeks’ gestation. A meta-analysis demonstrated a pooled sensitivity of 76% and a pooled specificity of 95% for culture-based tests. 69

The prevalence of maternal GBS colonisation varies with the site of swabbing, with rates highest when both vaginal and rectal swabs are considered. The is mirrored by increasing sensitivities, with the lowest sensitivity being observed using vaginal swabs and the highest observed when vaginal and rectal swabs are combined. Specificities do not vary as much with prevalence. It is unclear as to why there is this variation. The mucosal environment of the vagina and rectum are different, especially with regards to pH, and GBS has obviously adapted to both sites, but perhaps colonisation is heavier in the rectum and therefore more likely to be collected by swabbing. It is apparent that rectal swabbing is necessary to maximise sensitivity but the acceptability of this procedure is an issue, which will be discussed in detail in Chapter 3.

There is also variation in accuracy in the presence or absence of maternal risk factors, with sensitivities generally higher in the presence of risk factors and specificities higher in their absence. Differences are generally small and there is no general pattern as to which test–reference combinations are statistically different. So, although there is some variation in test characteristics according to presenting characteristics, the impact is not significant. This is also observed in the logistic regression model to calculate post-test probabilities, in which the presence or absence of risk factors did not significantly alter the outcomes.

Neonatal GBS infection is the reference standard of real interest, although it was not possible to use the neonatal ear swab as a reference standard here as antibiotics were prescribed to mothers on the basis of maternal risk factors, leading to a reduction in the rates of transmission from mother to baby. Any treatment that alters the verification of the reference standard after the index test inherently biases the study. Risk factors were not a significant factor in the prediction of neonatal colonisation, whereas a positive result from PCR was predictive, again reflecting their respective sensitivities.

Recommendations for the economic model

The decision-analytic model needs to consider estimates of probabilities obtained in this study, other estimates available for other screening tests or strategies and patient preferences. One of the key issues concerning screening tests in this project is that, if available, effective interventions are relatively simple, inexpensive and without a high risk of harm or side effects (to both mother and child). In this situation high test sensitivity is more important than high specificity, because the cost of false-negative results is likely to be high in relation to the costs incurred in treating all index test-positive results. In this regard, PCR is superior to OIA in our study, although both tests tend to have higher specificity than sensitivity. It seems unlikely that consideration of maternal risk factors will improve the cost-effectiveness of rapid test-based screening as it adds little to the post-test predictive values. It appears that index tests obtained from a combination of vaginal and rectal swabs would perform best, but the acceptability of rectal swabbing needs to be evaluated (see Chapter 3). A threshold analysis (see Chapter 4, Deterministic sensitivity analysis) will be required to determine the levels of sensitivity and specificity required to make rapid testing cost-effective in the prevention of EOGBS disease with currently available preventative treatment.

Implications for practice

Given the generally low sensitivity of the OIA test system evaluated, a practical recommendation for clinicians when making a decision about the use of intrapartum antibiotics is not to use OIA as a test for maternal GBS colonisation. If an intrapartum rapid test is to be considered for practice, PCR appears currently to be the most accurate test, although the current commercially available test may not have sufficiently high sensitivity to be cost-effective (see Chapter 4).

Recommendations for research

The evaluation of a test system should follow a robust design for test accuracy studies with sufficient power to estimate test sensitivity over and above that achieved by the PCR tests used in this study.

The extent to which cost-effectiveness threshold analysis can be developed to inform sample size estimation in test accuracy studies needs to be investigated. The minimum duration of IAP during labour and the relative importance of individual risk factors in predicting maternal colonisation can be estimated from this data set and validated in future larger studies. More work and discussion is required to establish the models of independent monitoring of test accuracy studies and stopping rules.

Participation in the study

Data were collected on the mothers who declined to take part in the study when they were first approached, during pregnancy. This consisted of basic demographic details and reasons for declining to take part. The reasons for declining were entered in free text and subsequently coded for analysis.

Acceptability of rapid testing

Declining to take part in the study during pregnancy

Table 9 indicates that the proportions of different ethnic groups and age groups among those declining to participate in the study when first approached during pregnancy differed from the proportions among those who eventually participated in the study. There was a significant association between ethnic group and declining to take part (χ² = 249.90, p < 0.001) and between age and declining to take part (χ² = 149.04, p < 0.001). There was a higher proportion of South Asian women, particularly those of Pakistani origin, in the women who declined. (Here, and in the remainder of this chapter, women are referred to by their self-reported ethnic origin for the sake of brevity, for example women whose ethnic origins are Pakistani may be referred to as Pakistani women. This is not intended to imply anything about their status or citizenship, which was not recorded.) There was also a higher proportion of younger women in the sample who declined to take part.

Reasons for declining to take part in the study during pregnancy

Table 10 indicates that, although no reason was recorded for the majority of participants declining to participate in the study, those who did express a reason declined to take part for both a general unwillingness to participate in the research and specific reasons relating to procedural aspects of participating (swabs) and other worries concerning their pregnancy. The reasons for declining during pregnancy varied by ethnicity (χ² = 296.93, p < 0.001) and age group (χ² = 146.73, p < 0.001). No pattern was discernable in the proportions of each group giving no reason for declining. However, white British and Irish women were over-represented in those who declined because they objected to the swabs, particularly rectal swabs, and amongst those who planned to have a Caesarean section or deliver somewhere other than in the participating hospital. Pakistani women and ‘other white’ women were over-represented amongst those who had insufficient English to participate. South Asian, particularly Pakistani, women were over-represented amongst those who objected to the vaginal swab, but this was a very small number overall. 19 Younger women were more likely to decline to take part because they objected to either or both of the swabs.

Acceptability of rapid testing for GBS during labour

Overall acceptability with information, testing procedure and rapid testing as routine care

Table 11 shows that the average ratings of the study and rapid testing for GBS are high. However, although scores are still above the midpoint on the scale, women who participated found vaginal swabbing more comfortable and less embarrassing than rectal swabbing. There is also some indication that some women would prefer a doctor or midwife to treat them on the basis of their professional judgement of the risk, rather than undergoing a rapid test.

Satisfaction with information

One-way ANOVA revealed no significant differences in the mean scores for acceptability of information between women of different parity, ethnicity or age. Table 11 shows that the participants’ evaluation of the information given to them was quite high. However, this needs to be considered with some caution as women reported in the open comments at the end of the questionnaire that the information had been given to them so long before that they could not remember it well:

Not entirely clear about study because have forgotten what midwife explained. She did explain but I have forgotten a lot of it.

Felt it was a long time ago I was given information on study and had forgotten most of it when it came to filling out form.

Some women interpreted the illness perception questionnaire as a test of knowledge rather than a test of perceptions of GBS, and gave that as a reason for not completing it: ‘Sorry that I can’t complete the detailed GBS info – I haven’t got the leaflet with me and can’t remember.’

Acceptability of the testing procedure

There were five questions on the testing procedure relating to the samples taken for testing. As there was not sufficiently high internal consistency in the responses to the five questions to assume that they were measuring the same construct, they have each been analysed separately. Again, as Table 11 shows, the acceptability of the swabs was generally high. As one participant said:

The last thing I was worried about was the swabs. I’d just started contractions and since everyone’s heads were going to be peeking down there anyway . . . I didn’t worry.

Acceptability of rapid testing as part of routine care

A series of questions was designed to measure women’s perceptions of using rapid testing as part of routine care. For the women in this study these questions were hypothetical as they were not treated on the basis of the results of the tests and, indeed, were not for the most part made aware of the results. Table 13 summarises the results on the acceptability of rapid testing for GBS as part of routine care.

Acceptability of rapid testing by risk status

Analyses of demographic and psychological predictors of acceptability

As indicated above, six psychological variables were measured in the questionnaire: state anxiety, health anxiety, health anxiety for the baby, negative illness perceptions of GBS in the baby, perceived likelihood of symptoms of GBS (in mother) and concern about GBS (in mother). These were included, together with the demographic variables analysed above, in a series of multiple regression analyses to see whether the psychological variables explained variation in the acceptability measures. The results are reported in the following sections for completeness, with the caveat that, although several of the regression models were statistically significant, they explained very little of the variance. Also, many cases had to be dropped from these analyses because of missing data on one or more of the variables.

Acceptability of rapid testing by salience of GBS

The final set of analyses on the questionnaire data looks at the acceptability of rapid testing by the salience of GBS for women. Women were classified either in the high-salience category (previous baby with GBS disease and/or GBS colonisation in current pregnancy and/or GBS colonisation during a previous pregnancy and/or received antibiotics during labour for GBS) or in the low-salience category (all other cases including all those with delivery-related risk factors but who were not given antibiotics). Note that this classification is conservative in that not all women who had GBS during a previous pregnancy would have had this noted, and there would be other reasons for GBS being more salient, for example having a friend with experience of GBS.

Acceptability to midwives

Several inter-related themes emerged from the two focus groups. Although there was some discussion of the experience of taking part in the study per se, prompted by the questions posed by the facilitator, the participants mainly used their experiences of the study to illustrate and validate their reflections on testing and treating GBS more broadly.

Summary of the findings on acceptability

There is no evidence that taking part in the study raised anxiety levels, as anxiety levels were similar to or lower than those recorded in other studies of antenatal and perinatal screening. The responses of women who took part in the study were generally very positive. They were satisfied with the provision of information, although there is some doubt that they had retained the information. They did not find the process of swabbing unpleasant, although vaginal swabs were more acceptable than rectal swabs. This was echoed in the midwives’ comments; they did not see rectal swabs as part of their normal role and saw it as unpleasant. The taking of swabs was also a major reason why women declined to take part when first approached.

The responses to questions about rapid testing and treatment for GBS during labour becoming part of routine care were again generally positive. Participants were confident that the test would be carried out properly and were happy to be treated on the basis of results. There was a marginal preference for rapid testing over treatment on the basis of risk assessment. Most women felt that the test was important and would recommend it to others. There is little evidence that the perceptions of participants were related to whether or not they fell into a high-risk category, except that high-risk participants felt that the test was more important and would be happier to be treated on the basis of the test result.

The ethnic group from which women came was related to the acceptability of testing and the uptake of the study. This was mainly due to differences between white British and Irish women and South Asian, particularly Pakistani, women. South Asian women were less happy with the sampling procedure and with the prospect of rapid testing as part of routine care; they were more likely to prefer professional judgement as the basis for treatment. South Asian women were also over-represented in the group of women who declined to take part in the study at all. Age was also related to perceptions of the test, with the younger age groups (below 25 years) being less positive than older participants, regardless of parity level.

Analyses of the links between acceptability and the psychological variables measured showed that none accounted for a large percentage of the variance in the data. State anxiety was the most predictive in that women who were more anxious were less satisfied with the information given and less happy with the way that samples were taken. Women’s perceptions of GBS in themselves were related to the importance that they placed on the test, using the test as a basis for treatment and whether they would recommend it to others. However, the psychological predictors of whether women said that they would prefer to be treated on the basis of professional judgement than the test were more negative perceptions of GBS in the baby and lower health anxiety for their own health.

There were some differences found between women for whom GBS should be salient and those who had no recorded experience of GBS. The high-salience group was more anxious and they reported lower levels of embarrassment with the rectal swabbing. However, the main differences came in relation to rapid testing as part of routine care. The high-salience group was less confident that the test would perform correctly and be confidential, and that they would be free to make treatment choices. These participants, however, reported that they felt that the test was more important than those for whom GBS was not salient.

The focus groups with midwives found that they were also generally positive about rapid testing but had some reservations from their experiences of the study. They feared that the high prevalence of GBS colonisation in women may lead to overtreatment and unnecessary interference in birth and overprescription of antibiotics if the test was routine, despite the low incidence of GBS infection in neonates. For similar reasons they were opposed to the universal administration of prophylactic antibiotics. The practical problems associated with testing were mainly the time it took to get the results, the availability of staff to use the current test equipment and interpret the results and the availability of the equipment itself. If rapid testing was shown to be accurate and is introduced into routine care, they would see it as part of their professional responsibilities. They could also see it having a role in a mixed methods strategy alongside screening in late pregnancy and professional risk assessment.

Strengths of the study

This is the first large-scale study to be published on rapid testing for GBS during labour that has sought the views of the women being tested themselves. The participants who agreed to take part actually had experience of the test procedure rather than being asked to imagine what the experience would be like. A range of data was collected on and from the women and this allowed the analysis of influences of demographic and other variables on satisfaction with and acceptability of different aspects of testing to be carried out. Rather than asking generic questions about satisfaction and acceptability, various aspects of the process were explored, as well as views on the use of testing during labour as part of routine care. The study also included an analysis of those women who declined to take part in the study and incorporated the views of midwives who had worked on the study.

Limitations of the study

The main limitation of this study is that it was not a treatment trial and so women only experienced the test procedure. No woman was treated on the basis of the test result and the participants may not have realised what treatment would have involved. Other studies have shown that high false-positives and missed cases affect the acceptability of screening procedures. 105

Alternative methods of preventing GBS transmission were not discussed with the women involved in the study so they could not give their views on their preferences for screening in late pregnancy, the administration of universal prophylactic antibiotics or the possibility of a vaccination being developed to be taken during pregnancy. A small qualitative study77 found that women in Canada did not feel that the low risk associated with GBS warranted the use of antibiotics but were also wary of taking a vaccine during pregnancy, should this become available. Only those who had direct experience of delivering a baby with GBS infection endorsed the vaccine.

Similarly, details of treatment with intravenous antibiotics during labour were not discussed with all women. Only those who were being given antibiotics anyway would have been made aware of what this involved. Lack of knowledge of the treatment regime may have influenced the views of women on satisfaction with information and acceptability of rapid testing in routine care.

The study involved women opting in rather than opting out. Many women who were approached did not agree to take part. This leads to inevitable biases in the sample. Women from South Asian ethnic origins were under-represented in the sample of those swabbed, for various reasons, including language barriers. As this group was generally more negative about rapid testing than white British and Irish and black Caribbean women, the overall acceptability of the test is likely to be overestimated. Women of any ethnicity who objected strongly to the swabs also did not take part. Again this may have led to an overestimation of the acceptability of the procedure.

As this study was not a treatment trial, midwives reported that they were less concerned about ensuring that all women who could be were swabbed. Therefore, a woman being in pain, being very young or not seeming to have a good level of English were all given as reasons by the midwives for not raising participation in the study. There is no way of knowing what effect missing out these possible participants had on the estimations of women’s evaluations of the test.

Findings in the light of limitations

Women’s views on rapid testing in practice have to be interpreted in the light of the fact that their deliveries were not affected by the study. They only experienced the sampling procedure. However, those who underwent this procedure were positive and their views on the introduction of testing as part of routine care are valuable.

A qualitative study of women’s attitudes to GBS and screening76 showed that most women knew very little about GBS and those who underwent screening did so because it was part of routine care and perceived to be better for the baby. The Canadian study cited above showed that how positive women were towards treatment for GBS was dependent on their experience and level of knowledge. 77 In this study those for whom GBS was salient, through their own experience of it, would be expected to have more knowledge. This group was more suspicious that the test could be carried out properly in practice as well as less confident that they would be free to make treatment choices, but, interestingly, this group also reported that they felt that the test was more important than the group for whom GBS was not salient.

It is important that those who have risk factors for GBS transmission find any screening procedure acceptable. Studies of HIV screening have shown that women at risk may be less likely to be tested, although this may be because of the impact of the results on women themselves and lifestyle characteristics, which influence the uptake of maternity care generally. In this study those in the higher risk category did not differ in their views of testing from those in the lowest risk category, except that they believed that the rapid test was more important for them to have. What is not known is how many of the women who declined to take part in the study initially were in the higher risk categories.

However, if those falling into the higher risk categories are likely to receive treatment regardless of their GBS status being known, which is essentially what current standard care in the participating hospitals offers, it is more important that those not falling into one of the higher risk categories find screening acceptable. These would be the people who would be missed under this regime. As we have seen, there is no evidence that the lower risk group finds the test less acceptable.

If testing for and treating GBS is to be introduced successfully, it is important that midwives are positive about it in their roles both in the community and in the delivery suite. The focus groups consisted of midwives with wide experience including those with current or recent experience of working in the community. Generally the midwives were positive, and this is further supported by the recruitment rates to the study. However, they did emphasise the practical problems of introducing the system, especially as it is now configured. They also had some doubts that introducing routine screening during labour would be cost-effective or justifiable given the incidence of GBS infection in neonates.

Ramifications for the economic model

The midwives in this study were opposed to the overuse of antibiotics, and felt that the women they saw would endorse this view. Therefore any strategy that involved the widespread unnecessary use of antibiotics would be unlikely to be successful in practice. The midwives also reported that they were able to take samples and process them because the timings were not important: the swabs could be taken and then analysed at a later time. If this was to become routine care, then ensuring that the test was carried out in a timely fashion and the results reported back to the midwife would be crucial. Midwives would expect to carry out the tests themselves, as they have professional responsibility for the mother, increasing the demands on their time rather than the time of midwifery assistants. They could not see the current diagnostic procedure, i.e. having the testing hardware in a separate room, fitting into their normal practice. They favoured testing hardware that would be available by the bedside for ease of use and confidentiality.

Both the women tested and the midwives found rectal testing more embarrassing and unpleasant than vaginal testing. If more rectal swabs are missed as a result of this, then this may impact on how many cases are missed.

Recommendations for practice

There are some features of current care that could be improved and some recommendations about the introduction of rapid testing if the technology continues to develop. The first is that midwives need to be better informed about GBS and the treatment for it; this will enable them to better inform women affected and better follow current guidelines.

For women to be able to make informed choices about their care, information needs to be given about GBS closer to the time at which testing and treatment can be carried out effectively. Giving information early in pregnancy, when women have many decisions to make, means that it is likely to be forgotten. Routine information giving later in pregnancy would be more beneficial.

In general, women from South Asian groups and younger women were less positive about the results of the test being used as the basis for treatment in routine care. They would prefer to be treated on the basis of the judgment of their doctors or midwives. This is essentially the status quo. This may reflect a greater confidence in professionals in these groups; however, it may also be linked to the acceptability of the test procedure and a lack of confidence that the test would be carried out properly at that stage. Testing during pregnancy may be more acceptable to these groups for these reasons, but further education on the accuracy of the test would be necessary in any case. If testing for GBS during the late stages of pregnancy is introduced with or without testing during labour, processes for communication of results need to be improved.

If testing during labour does become available, careful consideration needs to be made of the testing procedure so that it fits into midwifery practice. Staffing concerns also need to be overcome.

Recommendation for research

This study suggests that there is enough support for GBS testing during labour for the tests to be developed further. The procedure for collecting samples is largely acceptable to women and midwives although vaginal swabs were preferred by both groups to rectal swabs. To be suitable for routine care the devices need to be not only accurate but also easy to use and genuinely rapid. More research is needed on the views of midwives on routine testing during labour.

Many women did not take part in this study because of language barriers. It is important that groups of women are not disadvantaged through non-inclusion in research. In future studies midwives could be encouraged to assess the communication skills of potential participants and other barriers to participation more formally.

More research is also needed on the views of pregnant women and mothers who would be affected by GBS screening during pregnancy and labour. Compared with other areas of antenatal and perinatal screening, very little research has been undertaken on women’s beliefs about GBS and their own risks, and whether they would welcome testing during pregnancy, including home testing, testing during labour or indeed universal prophylactic treatment.

More work also needs to be carried out on the development of the intervention, including not only the development of the equipment but also the procedure, and such work needs to incorporate the views of the women and health professionals involved.

Model structure

In this primary study no action in terms of treatment is taken on the basis of the PCR or OIA rapid test that is carried out on consenting women because it is carried out in addition to risk factor-based screening, which they should receive as part of current practice. The only action that is taken is on the basis of the risk factor-based screening. Thus there is no comparator in terms of outcomes for women whose rapid test results are contrary to those predicted by risk factor-based screening. Hence, a model is required to carry out the analysis. A decision tree was considered to be the most appropriate model for this study, given the short-term nature of the decision problem. 122 The model was constructed in DATA TreeAge (TreeAge Software, 2005; Williamstown, MA, USA).

Figure 7 presents an overview of the decision tree and the 10 policy alternatives that have been modelled. Separate subtrees are presented for the strategies listed in Figure 7. For each strategy in turn the model considers the number of EOGBS-associated deaths, the number of EOGBS disease cases and the associated costs. Space constraints do not allow the entire tree to be presented but subtrees 1 (Figure 8) and 4 (Figure 9) are presented as an illustration and the remainder of the subtrees are presented in Appendix 7.

FIGURE 7.

Model structure. OIA, optical immunoassay; PCR, polymerase chain reaction.

FIGURE 8.

Subtree 1. AB, antibiotic.

FIGURE 9.

Subtree 4.

Figure 8 (subtree 1) represents the ‘routine untargeted IAP to all’ option. It presents the possible paths that might be followed under such a regimen (i.e. no systematic testing but untargeted treatment with antibiotics of women in labour). The assumption here is that all women would be given antibiotics intravenously as soon as they are admitted in labour. The analysis assumes that clindamycin and penicillin are the antibiotics offered in line with current practice at Birmingham Women’s Hospital and as per the Green-top Guideline of RCOG. 61

Subtree 2 (Appendix 7, Figure 20) represents the ‘do nothing’ strategy. Here, women are not screened for GBS and are also not given antibiotics, regardless of risk factors. In subtree 3 (Appendix 7, Figure 21) the use of a culture test at 35–37 weeks’ gestation by culture of vaginal and rectal swabs is introduced. This strategy is based on the US Centers for Disease Control and Prevention recommendations that all pregnant women undergo bacteriological screening,123 with all women testing positive for GBS then given IAP. However, because of the high risk of EOGBS disease associated with preterm birth, the assumption is made that all mothers presenting in labour before 35–37 weeks’ gestation receive IAP.

Subtree 4 (Figure 9) and subtree 5 (Appendix 7, Figure 23) present the key alternatives that the current study aimed to compare, namely the rapid tests, PCR and OIA. Figure 9 presents the subtree for PCR; the subtree for OIA is exactly the same and is presented in Appendix 7, Figure 23. For the strategy of the rapid test based on PCR or OIA, all women presenting in labour are tested for GBS and those who test positive are treated with antibiotics.

In subtree 6 (Appendix 7, Figure 24), women who have at least one of five risk factors are treated with antibiotics. The five risk factors are previous EOGBS-affected baby, preterm labour, GBS bacteriuria detected during the current pregnancy, PROM and fever in labour.

Subtrees 7–10, presented in Appendix 7, outline the strategies that are a combination of the risk factors and rapid tests. In Figure 25 women who possess one or more of the five risk factors are further tested for GBS using the PCR test and are only treated if the test result is positive. Figure 27 presents a similar strategy but with the exception that the rapid test used in this case is the OIA. In Figure 26 women who possess one or more of the five risk factors are treated with antibiotics whereas those who do not exhibit any of the risk factors are subjected to a PCR test and treated if the result of this test is positive. In Figure 28 the strategy is repeated but for the rapid test based on OIA.

Model data

For each strategy in the model the underlying maternal colonisation rate is required. The colonisation rates are presented in Table 18. The parameter ‘overall maternal colonisation’ represents the overall maternal colonisation rate expressed as a probability (p = 0.2128). This implies that 0.7872 is the underlying probability of overall maternal non-colonisation. Maternal colonisation was further characterised by the site of colonisation (i.e. rectal only, vaginal only and both rectal and vaginal combined). Again the parameters ‘rectal only’ and ‘vaginal only’ represent probabilities associated with rectal only and vaginal only colonisation respectively. Because probabilities need to sum to one, a formula ensures that the probability for both rectal and vaginal colonisation is the difference, defined as [1 – (rectal only + vaginal only)]. There is evidence to suggest that the risk of GBS transmission from mother to baby will differ by type of delivery. 124 Therefore, in the model, vaginal delivery was distinguished from Caesarean delivery.

The overall neonatal colonisation rate (weighted by mode of delivery and maternal colonisation), which was estimated in the current study to be 0.0921, is presented in Table 19. The prevalence of EOGBS disease given neonatal colonisation has been estimated elsewhere to be 0.027. 69 However, the product of this estimate and the overall neonatal colonisation rate should result in the observed incidence of EOGBS disease in the absence of systematic screening or widespread IAP in the entire population, which is approximately 0.5 per 1000. 61 As the overall neonatal colonisation rate was calculated from the new empirical data estimated by the current study it is considered more accurate than estimates from other sources. Thus, it was necessary to calibrate the prevalence of EOGBS disease, given neonatal colonisation, to obtain the ‘correct’ value of the population incidence of EOGBS disease in the absence of systematic screening or widespread IAP. This resulted in a value of 0.00518 being obtained for the prevalence of EOGBS disease given neonatal colonisation.

Table 19 also shows the neonatal colonisation prevalence, calculated from the data in the current study. The details of the calculation are given in Appendix 6. The treatment effect of maternal antibiotics associated with a baby developing EOGBS disease, given maternal colonisation, was obtained from other sources. 69 When a mother received intravenous antibiotics, the OR for the effect of antibiotic therapy on EOGBS disease was used together with the prevalence of neonatal EOGBS disease in the calculation of the outcome. When no antibiotics were given to the mother, the prevalence alone was used. The terminal nodes indicate the main outcomes, which are EOGBS-related deaths avoided and cases of EOGBS disease avoided.

Probabilistic sensitivity analysis

A probabilistic sensitivity analysis was carried out for the base-case analysis only, for analyses 1 and 2, to explore the effects on the ICERs of the uncertainty in the model input data.

In probabilistic analysis, each model parameter is assigned a distribution reflecting the amount and pattern of its variation, and cost-effectiveness results are calculated by simultaneously selecting random values from each distribution. The process is repeated many times in a Monte Carlo simulation of the model to give an indication of how variation in the model parameters leads to variation in the ICERs for a given combination of a test and treatment pairing. In total, 2000 replications were carried out. This number is arbitrary but typical standard practice. The appropriate distribution for the data on test accuracy (sensitivity and specificity) is either a beta or normal distribution depending on the statistical characteristics of the parameters. The appropriate distribution for data on intervention effectiveness is a beta distribution. These are standard distributions for these data and are deemed to be appropriate in this case.

Unit cost uncertainty is excluded from the probabilistic analysis here because any variations that might exist for unit costs are of a different nature from the data-driven uncertainty in the patient flow parameters.

Deterministic sensitivity analysis

The following one-way deterministic sensitivity analyses were conducted.

Analysis 1

The results from the deterministic analyses are virtually identical to those from the probabilistic analyses and so, in this case, it was considered acceptable to report deterministic results as the primary analysis. The main results for the deterministic analysis for the base-case model are presented in Table 23, with the strategies ordered from least costly to most costly. The cost and effectiveness of each strategy are presented incrementally to the one presented directly before it, except when a strategy presented is referred to as dominated, in which case the incremental costs and effectiveness are in comparison to the preceding strategy that is not dominated.

The results show that the ‘do nothing’ strategy, which implies no screening and no IAP for all women, is the least costly strategy but also the least effective. The average cost per woman for this strategy is £1055. Under a strategy of ‘do nothing’, 476 per million infants will develop the disease and 999,524 per million infants will not develop the disease. This is shown in Table 23 under effectiveness of the intervention, which shows that 99.9524% of infants will not develop the disease and therefore 1 – 0.999524 = 0.000476 or 476 per million infants will develop the disease. This means that a higher effectiveness represents a more favourable outcome. This result is an important cross-checking reference point for the model, because it confirms that, in the absence of screening for GBS, the effect of ‘do nothing’ is that 476 per million infants develop the disease, which concurs with the observed population rate, which is approximately 0.5 per 1000 women. 61

The next least costly strategy is screening based solely on the presence of the risk factors. The average cost per woman for this strategy is £1061, an average of £6.12 more per woman than the ‘do nothing’ strategy. It is also more effective than the ‘do nothing’ strategy. Spending £6.12 million (£6.12 × 1,000,000) would mean that 107 cases (0.000107 × 1,000,000) of EOGBS disease would be avoided. The ICER of this strategy compared with the ‘do nothing’ strategy is £57,038 per case of EOGBS disease avoided.

The next least costly strategies presented in Table 23, ‘risk factors positive, OIA positive’ and ‘risk factors positive, PCR positive’, are both dominated by the strategy of risk factors alone in that both of these strategies are more expensive and less effective than a strategy of risk factors alone.

Screening based on a culture test at 35–37 weeks’ gestation is the next least costly strategy. At £1068 per woman it costs on average £7.31 more per woman than the strategy of screening based on risk factors but it is also more effective. For an additional £7.31 million (£7.31 × 1,000,000) an additional 147 cases of EOGBS disease could be avoided, compared with the strategy of screening based on risk factors. The ICER for screening based on the culture test at 35–37 weeks’ gestation compared with screening based on risk factors is £49,726 per case of EOGBS avoided.

The strategy of ‘routine untargeted IAP to all’ costs £0.63 more per woman than the strategy of culture at 35–37 weeks’ gestation but is also more effective. For an additional £0.63 million (£0.63 × 1,000,000), 99 cases (0.000099 × 1000,000) of EOGBS could be avoided. The ICER for the strategy of routine untargeted IAP to all compared with the strategy of culture at 35–37 weeks’ gestation is £6414 per case of EOGBS avoided.

The results presented in Table 23 suggest that, if an ICER of £57,038 per additional case of EOGBS avoided is considered acceptable, then adopting a strategy of screening based on risk factors, as opposed to doing nothing, would be the preferred strategy. If this is the case, then the ICER of £49,726 per additional case of EOGBS disease avoided to adopt the strategy of screening based on culture at 35–37 weeks’ gestation would also be acceptable. Following this logic the ICER of £6414 per case of EOGBS disease avoided would also be acceptable and therefore the strategy of providing routine untargeted antibiotics to all would be the preferred strategy because there is considered to be extended dominance.

If the ICER of £57,038 per additional case of EOGBS disease avoided is not considered acceptable, the strategy of screening based on risk factors would not be adopted and other strategies would have to be considered. The strategy of routine untargeted antibiotics to all is presented incrementally in Table 23 in a comparison with screening based on culture at 35–37 weeks’ gestation. If this strategy, routine untargeted antibiotics to all, is compared with a strategy of doing nothing, as presented in Table 24, the ICER is £39,813 per case of EOGBS disease avoided, which is a more favourable ICER compared with the strategy of doing nothing than exists for the strategy of risk factors. It is therefore more cost-effective to move from the strategy of doing nothing to the strategy of routine untargeted IAP to all. This is shown in Figure 10. However, the acceptance of this strategy depends on whether or not the ICER of £39,813 per case of EOGBS disease avoided is considered acceptable.

FIGURE 10.

Cost-effectiveness analysis based on early-onset group B streptococcus (EOGBS) cases avoided. AB, antibiotic; OIA, optical immunoassay; PCR, polymerase chain reaction.

In Table 25 the results of analysis 1 are reported with the outcome expressed as EOGBS-associated infant deaths avoided. Under a strategy of doing nothing approximately 35 per million infants will die as a result of EOGBS and 999,964 per million infants will not die. This is shown in Table 25 under effectiveness of the intervention, which shows that 99.9964% (0.999964 in the table) of infants will not die of the disease and therefore 1 – 0.999964 = 0.000035 or 35 per million infants will develop the disease. This result concurs with the observed incidence of EOGBS-associated deaths in the population. 18

When the outcome is EOGBS deaths avoided, the ICER for the strategy of screening based on risk factors compared with a strategy of doing nothing is £764,579 per death avoided (Table 25). However, under extended dominance, the ICER of £533,683 per EOGBS death avoided for the strategy of routine untargeted IAP to all compared with the strategy of doing nothing is more favourable and thus would be the preferred strategy if the ICER is deemed to be within an acceptable threshold. This is presented in Table 26 and Figure 11.

FIGURE 11.

Cost-effectiveness analysis based on early-onset group B streptococcus (EOGBS) deaths avoided. AB, antibiotic; OIA, optical immunoassay; PCR, polymerase chain reaction.

If this ICER of £533,683 per EOGBS-associated death avoided is converted to a utility on the basis that a life in full health, discounted at the rate recommended by NICE of 3.5%, is worth approximately 27 discounted QALYs, the conversion (£533,683/27 QALYs) produces an ICER of £19,766 per QALY.

Analysis 2

In analysis 2, the analysis is repeated but the strategy of providing routine untargeted IAP to all is removed. Again, the results from the deterministic analyses are virtually identical to those from the probabilistic analyses and so it was considered acceptable to report deterministic results as the primary analysis.

In Tables 27 and 28 the results are presented based on the outcome of cases of EOGBS disease avoided. In Tables 29 and 30 the results presented are based on the outcome of deaths associated with EOGBS avoided.

The ICER for the strategy of screening based on risk factors compared with a strategy of doing nothing when the outcome is in terms of cases of disease avoided is £57,038 per case of disease avoided (Table 27). The ICER for the strategy of culture at 35–37 weeks’ gestation compared with the strategy of risk factors is £49,726 per case of disease avoided. Under extended dominance the ICER for the strategy of culture at 35–37 weeks’ gestation compared with the strategy of doing nothing is the most favourable at £52,810 per case of disease avoided (Table 28). This result is presented diagrammatically in Figure 12. This would be the preferred strategy only if the ICER is deemed to be within an acceptable threshold.

FIGURE 12.

Cost-effectiveness analysis based on early-onset group B streptococcus (EOGBS) cases avoided – routine intravenous antibiotic prophylaxis excluded. AB, antibiotic; OIA, optical immunoassay; PCR, polymerase chain reaction.

In Table 29 the results of analysis 2 are presented with effectiveness measured in terms of deaths avoided. The ICER for the strategy of risk factors compared with a strategy of doing nothing is £764,579 per infant death avoided. The ICER for the strategy of culture test at 35–37 weeks’ gestation compared with the strategy of risk factors is £729,623 per infant death avoided. Under extended dominance the comparison of the strategies of culture test at 35–37 weeks’ gestation and doing nothing provides a slightly more favourable ICER of £745,142 per death avoided, compared with the comparison of the strategies of screening based on risk factors and doing nothing (£764,579) (Table 30 and Figure 13). However, this would be the preferred strategy only if the ICER of £745,142 per death avoided is deemed to be within an acceptable threshold.

FIGURE 13.

Cost-effectiveness analysis based on early-onset group B streptococcus (EOGBS) deaths avoided – routine intravenous antibiotic prophylaxis excluded. AB, antibiotic; OIA, optical immunoassay; PCR, polymerase chain reaction.

If this ICER of £745,142 per EOGBS-associated death avoided is compared with the current NICE threshold and on the basis that a life in full health, discounted at the discount rate recommended by NICE of 3.5%, is worth approximately 27 discounted QALYs, the appropriate conversion (£745,142/27 QALYs) produces an ICER of £27,500 per QALY.

Probabilistic sensitivity analysis

Probabilistic sensitivity analysis was carried out on the base case in terms of pairwise comparisons for the following strategies:

• routine untargeted IAP to all versus no screening and no IAP

• screening based on culture test at 35–37 weeks versus no screening and no IAP

• rapid test 1 (PCR) versus no screening and no IAP

• routine untargeted IAP to all versus screening based on risk factors.

Figure 14 presents the cost-effectiveness acceptability curve (CEAC) for the two strategies routine untargeted IAP to all and no screening and no IAP. A CEAC is a method of illustrating the uncertainty in estimates of cost-effectiveness. 131 The diagram illustrates that, in choosing between these two strategies, at a threshold below £470,000 per EOGBS-associated infant death avoided, the strategy of routine untargeted IAP to all would not be considered cost-effective and the choice of strategy would be to do nothing (i.e. no screening and no IAP). However, at a threshold above £630,000 the decision would be to choose the strategy of routine untargeted IAP to all with certainty. In between these two thresholds the choice could be either strategy if the parameters were known with certainty.

FIGURE 14.

CEAC 1 – the cost-effectiveness acceptability curve for the two strategies routine untargeted intravenous antibiotic prophylaxis (IAP) to all and no screening and no IAP. ICER, incremental cost-effectiveness ratio.

None of the other comparisons can present a strategy that would be accepted with certainty at a threshold lower than £630,000 with the exception of the final pairwise comparison between routine untargeted IAP to all and screening based on risk factors (Figures 15–17). In this final comparison, the strategy of routine untargeted IAP would be accepted with certainty at a willingness to pay threshold of just over £500,000 (Figure 17).

Figure 15.

CEAC 2 – the cost-effectiveness acceptability curve for the two strategies culture test at 35–37 weeks and no screening and no intravenous antibiotic prophylaxis (IAP). ICER, incremental cost-effectiveness ratio.

FIGURE 16.

CEAC 3 – the cost-effectiveness acceptability curve for the two strategies rapid test PCR and no screening and no intravenous antibiotic prophylaxis (IAP). ICER, incremental cost-effectiveness ratio.

FIGURE 17.

CEAC 4 – the cost-effectiveness acceptability curve for the two strategies routine untargeted intravenous antibiotic prophylaxis (IAP) to all and risk factors. ICER, incremental cost-effectiveness ratio.

Figure 16 illustrates that, in choosing between the strategies of the rapid test based on PCR and doing nothing, at a threshold below £2,000,000 per EOGBS-associated infant death avoided the strategy of the rapid test based on PCR would not be considered cost-effective and the choice of strategy would be to do nothing (i.e. no screening/no antibiotics) with certainty. However, at a threshold above £3,200,000 the decision would be to choose the strategy of rapid testing based on PCR with certainty.

Deterministic sensitivity analysis

In Tables 31 and 32 the summarised results of the deterministic sensitivity analyses are presented for analyses 1 and 2 respectively. For both analyses the results are based on an outcome of EOGBS-associated infant death avoided. A summarised set of similar sensitivity analyses based on the outcome of EOGBS disease cases avoided is presented in Appendix 8.

Summary of main findings

The results of this economic evaluation suggest that the strategy of routine untargeted IAP prophylaxis to all is the most cost-effective strategy compared with doing nothing and relative to all other strategies that have been compared in this analysis. This conclusion is drawn from the ICER based on the outcome of EOGBS-associated infant deaths avoided, which has been compared with the NICE threshold of £20,000 per QALY. The results of the full modelling analysis (based on analysis 1) estimated that the ICER for routine untargeted IAP to all compared with doing nothing was approximately £533,000 per EOGBS-associated infant death avoided. Assuming that a surviving infant would be in full health and that a life in full health, discounted at the discount rate recommended by NICE of 3.5%, is worth approximately 27 discounted QALYs, the estimated ICER is £19,766 per QALY. Thus, this strategy is likely to be accepted by NICE on cost-effectiveness grounds alone.

When the results of the model were evaluated based on an outcome of EOGBS disease avoided, routine untargeted IAP to all was again the most cost-effective strategy compared with doing nothing and relative to all other strategies. The corresponding ICER was estimated to be £39,813 per case of EOGBS disease avoided compared with a strategy of doing nothing, but this ICER is less easy to interpret for decision-making.

When the strategy of providing antibiotics to all is removed from the model in analysis 2, the most cost-effective strategy was the culture test at 35–37 weeks’ gestation compared with doing nothing and relative to all other remaining strategies. Based on the assumptions of full health and discounted QALYs as described previously, the ICER for culture at 35–37 weeks’ gestation compared with doing nothing was approximately £745,000 per EOGBS-associated infant death avoided, which translated to £27,620 per QALY. This exceeds the acceptable threshold set by NICE of £20,000 per QALY and thus would not be automatically accepted on cost-effectiveness grounds alone. Therefore the uncertainty surrounding this estimate would require greater scrutiny. If the assumption that surviving infants would survive in full health was considered plausible, then the strategy is more likely to be accepted. If the assumption that survivors would experience full health was thought not to be plausible, then the strategy is likely to be rejected and doing nothing would be the recommended strategy based on the current evidence.

These results, for both analysis 1 and analysis 2, were shown to be robust for the majority of sensitivity analyses. However, there were two main exceptions. First, it was notable that changing the estimated effect of intravenous antibiotic therapy on EOGBS disease given maternal colonisation, resulted in a significant effect on the ICERs in both analyses, which served to make the most cost-effective ICER for each analysis even more favourable relative to the NICE threshold. Second, in analysis 2, when either the cost of the culture test was increased by a small amount, or the base-case assumption that women who gave birth before receiving the screening test based on culture at 35–37 weeks’ gestation were given intrapartum antibiotics was removed, and it was instead assumed that none of these mothers received intrapartum antibiotics, screening based on risk factors became the most cost-effective strategy and the strategy of screening based on the culture test at 35–37 weeks’ gestation was no longer a contender in the results at all.

Strengths of the methods used

The strength of this analysis is that the majority of the data used to populate the options in the model is based on the latest empirical data from the current primary study in particular cost and resource data associated with the strategies of the rapid test, providing antibiotics and screening based on risk factors were carefully monitored prospectively as part of the primary study. The costs and resources associated with performing the screening test based on culture were also recorded, although in this analysis we used culture at delivery as a proxy for the costs and resources associated with the culture test at 35–37 weeks’ gestation. The effectiveness data for the accuracy of the rapid test, risk factor-based screening and culture were also empirically evaluated during the study.

The model used was comprehensive and compared all of the options that are available to date. Although some studies have evaluated the use of vaccination for GBS, there is currently no licensed vaccine available.

Limitations of the study

In the main analysis, providing routine untargeted IAP to all is shown to be the most cost-effective option. However, the full cost associated with this strategy is likely to have been underestimated. This is primarily because the model has not included any costs associated with potential resistance to antibiotics or side effects in this population, both of which could lead to complications for the woman or baby later on. Furthermore, if policy were to change in favour of this strategy, then the capacity available in labour wards would have to be expanded as more women would be required to give birth in hospitals or birthing centres that are equipped to provide intravenous antibiotics. The additional demand and its impact on costs to hospitals and delivery units have not been incorporated into the current analysis. Added to this is the likelihood that such a strategy would not be acceptable to the majority of women who are anxious and resist the further medicalisation of childbirth.

In analysis 2, which specifically excluded the strategy of routine untargeted IAP to all, the strategy of a culture test at 35–37 weeks’ gestation for all women was shown to be the most cost-effective option. This strategy included the assumption that women who went into labour before undergoing this test would receive routine antibiotics; when this assumption was removed from the analysis the strategy of screening based on risk factors became the most cost-effective option. Furthermore, it is likely that the costs associated with the strategy of a culture test at 35–37 weeks’ gestation will also have been underestimated, and it too is a strategy that will impinge on the way that women have traditionally been cared for. Similar to the strategy of providing routine untargeted IAP to all, it will be necessary for large numbers of midwives to be trained in the prescribing and administration of intravenous antibiotics. There are also likely to be late changes to women’s plans for delivery, which will in turn have capacity issues.

Although there exist some limited data on survival and quality of life for infants who experience this disease, given the current available lifesaving technology that can assist infants who are born preterm and/or who experience EOGBS disease, there is no concurrent evidence on the quality of life experienced by the infants who do survive. 21,132 The outcomes presented in terms of cost per QALY in the current study have been estimated based on available guidelines on discounting and known acceptable thresholds for decision-makers and not on quality of life data.

Finally, in terms of limitations, it is not clear or possible to determine exactly what constitutes current practice for the prevention of EOGBS disease in infants in the UK. The main comparator for the economic evaluation has been to do nothing, and the model has been calibrated to the population prevalence of neonatal EOGBS disease and EOGBS-associated infant mortality. These population morbidity and mortality rates are based on current practice, but current practice is likely to be heterogeneous with regard to the application of risk factors, in terms of whether screening is based on one risk factor or more than one or any at all. 133

Interpretation of the findings

The results show that the strategy of routine untargeted antibiotics to all is the most-cost-effective strategy overall. When this is removed from the analysis on the grounds that it is likely to be unacceptable to women, screening based on culture at 35–37 weeks’ gestation, with antibiotics given to all those women who deliver before 35 weeks, becomes the most cost-effective option.

The results of the analysis have shown that the rapid test in its current form, either PCR or OIA based on vaginal or rectal swabs, does not offer a cost-effective alternative option for screening women to assess colonisation by GBS.

In analysis 1, for the rapid test to become a more favourable strategy than antibiotics, given its current accuracy, it would have to cost as little as £6.00 based on the combined rectal and vaginal swabs and as little as £4.50 based on the vaginal swab alone. In analysis 2, the corresponding costs required of the rapid test in order for it to become a more favourable option than screening based on culture at 35–37 weeks’ gestation are £10.50 and £7.50. These costs contrast sharply with the current cost of the rapid test, which is approximately £29.50.

Comparison with other studies

In the study by Colbourn et al. 69 a model was developed for use in the economic evaluation of prenatal screening and treatment strategies to prevent GBS disease in early infancy. Key differences between the results of the study by Colbourn et al. and those of the current study include that their results were reported in cost per QALYs and their analysis attempted to incorporate the adverse effects of antibiotic use. They also used a different estimated effect of intravenous antibiotic therapy on EOGBS disease and death given maternal colonisation – OR 0.028 (95% CI 0.0015–0.12) – which was much lower than the one reported in the Cochrane library. 43 The justification for the use of the alternative estimate by the authors is based on opposing schools of thought regarding the use of a fixed-effects estimator in meta-analysis. Colbourn et al. argue that the Cochrane estimate is based on the Peto one-step method,134 which is a fixed-effect estimator, which they assert produces ‘seriously biased estimates when the true treatment effects are large (as in the present analysis)’ (references 124 and 125 in their report). Our sensitivity analysis showed that when the alternative estimate was used there was a significant impact on the results, which made the relative ICERs more favourable.

The strategies compared were broadly very similar, although Colbourn et al. divided the population into specific risk groups and also included vaccination against GBS as a strategy for comparison in their model, which was not included in the current analysis.

Overall, the results of the analysis by Colbourn et al. were broadly similar to those reported here. They showed that screening based on risk factors and screening based on the rapid test were not cost-effective strategies. Vaccination was shown to be the most cost-effective option, a strategy not considered in the current analysis because there is currently no available vaccine to prevent EOGBS disease. In the absence of vaccination their study concluded that treatment with antibiotics for high-risk groups and for women who delivered preterm was cost-effective, whereas screening based on culture at 35–37 weeks’ gestation was found to be cost-effective for low-risk women.

Recommendations for practice

The results show that, based on current evidence, neither rapid test, as evaluated in the current study, should be used in practice as it is clearly not a cost-effective method of screening women for GBS colonisation.

There is some evidence to support the serious consideration of the strategies of providing routine untargeted antibiotics prophylactically to women and of testing women at 35–37 weeks’ gestation based on culture of vaginal and rectal swabs, with antibiotics provided to women who deliver before 35 weeks.

Recommendations for research

There is a clear need to develop a simple point-of-care test that has a high level of accuracy compared with the gold standard of culture. Based on their current accuracy performance the rapid tests, as evaluated in this study, need to be both cheaper and more accurate. The costs of screening by culture at 35–37 weeks’ gestation have been based on the costs of the reference standard and may not reflect the true costs of this strategy. These could only be derived with more accuracy by conducting a feasibility and costing study in a hospital setting. Future modelling should consider this.

There is also a clear need for studies to explore the quality of life of infants who have experienced EOGBS disease and survived. Value for money of antenatal screening and testing programmes crucially depends on the values attributed to the adverse outcomes averted by testing and these should be the subject of further research and explicit public debate.

Polymerase chain reaction test

The Smart GBS^® kit and SmartCycler^® system (Cepheid; Sunnyvale, CA) is a PCR kit system that can be run using Lim broth cultures or directly from vaginal/rectal specimens, using real-time PCR technology to detect GBS even in lightly colonised specimens (www.cepheid.com/Sites/cepheid/content.cfm?id=188).

Optical immunoassay test

The BioStarOIA STREP B^® test (Inverness Medical; Princetown, NJ ) provides rapid detection of group B streptococcal antigen directly from cervical or vaginal swabs by a simple optical immunoassay (www.invernessmedicalpd.com/poc/products/oia_strepb.html)

Staff selection

We elected to make midwifery assistants responsible for testing. These are unqualified staff, with little or no scientific background, who support the day-to-day running of the delivery suite. This staff group was chosen because it was considered that midwifery assistants would be more readily available to undertake testing on demand, as well as to minimise the salary component of the overall cost of tests, which require considerable hands-on time. Moreover, some midwifery assistants were already undertaking POC tests for haemoglobin and plasma glucose concentrations.

Training needs analysis

Because of the complexity of the tests being evaluated and the special infection control considerations involved in microbiological testing, a full training needs analysis was undertaken. This identified a number of basic elements of laboratory practice in which it was considered essential that midwifery assistants achieved competence before receiving any hands-on training with the diagnostic kits. These areas were general good laboratory practice, health and safety and quality assurance. The results generated from the rapid tests used in this study were not used to influence patient management and therefore comprehensive training on the interpretation of results generated by the POC tests was not given. However, the training needs analysis identified that those undertaking testing would have to understand the importance of generating timely and accurate results and be competent in identifying patterns of results that might indicate a problem with testing.

Training and competency assessment

A series of three seminars was provided for midwifery assistants to introduce them to the study and to provide training on the basic elements of laboratory practice outlined above. Training was provided by a biomedical scientist. Each seminar lasted approximately 45 minutes and consisted of an informal presentation with discussion allowed around each point made during the presentation. At the end of each seminar each midwifery assistant was given a questionnaire designed to assess the level of understanding and to highlight areas of uncertainty. When difficulties were identified these were dealt with discreetly by further tuition on a one-to-one basis.

Only when we were satisfied that individuals had demonstrated basic competency were they allowed to progress to hands-on training in the use of the kits. Individuals conducted the tests under the supervision of the biomedical scientist, as many times as necessary, until they felt comfortable in what they were doing and the assessor was satisfied that they could achieve a satisfactory and consistent level of competency.

Feedback

Midwifery assistants were given feedback questionnaires in which they were asked their opinions on the quality of the seminars and their overall views of the study. They were free to make any suggestions on how they thought that training could be improved. Finally, it was made clear that they could telephone either the biomedical scientist or the research midwife at any point during the study if there was a problem.

Ongoing competency and quality assurance

Each participating midwifery assistant was regularly asked if any issues or problems had arisen. Results were regularly monitored by the biomedical scientist to establish if there were any patterns of invalid results or test failures occurring. The PCR kit has its own integral controls that have to be incorporated with each run; if either, or both, of these are invalid then the whole run is invalidated. The OIA kit contained internal controls that were used with each batch.

Results

In total, 12 midwifery assistants were trained in POC testing. All easily attained competency in the basic elements of laboratory practice. Responses in the feedback questionnaires also indicated that they were happy with the method of training and that they felt confident and enthusiastic about the study and their new role in POC testing.

All midwifery assistants readily become competent and confident at undertaking the OIA test after an average of 2 hours of hands-on training. The only aspect to cause concern was the reading of the final result, which was open to a degree of subjective interpretation. Trained midwifery assistants were able to process a swab and achieve a result within 35 minutes, with 20 minutes of actual hands-on time. The PCR test, which entails over 20 separate manipulations during sample preparation, proved more daunting and, although all midwifery assistants achieved competency after the initial training, few had sufficient confidence to feel comfortable about testing. Training also took much longer than for the OIA test, averaging 7 hours per midwifery assistant. It took an average of 90 minutes to obtain a PCR result, including 40 minutes hands-on time. An unanticipated problem that arose during the study, related to the complexity of sample preparation for both tests, meant that staff were unable to begin processing a further sample until they had completed work on the first sample.

At no time during the study were we able to establish testing on demand on a 24/7 basis. In the early stages of the study, midwifery assistants were available to undertake POC testing during daytime hours, 7 days per week. However, over a period of time, fewer tests were being performed in real time by fewer midwifery assistants. There were several reasons for this, with the main ones being the impossibility of maintaining adequate levels of fully trained staff because of high staff turnover, compounded by delays in recruiting replacements, and the fact that midwifery assistants were often called away to undertake other duties for which they were responsible as part of their daily work schedule. Over the course of the study many midwifery assistants became disenchanted with the study as they saw it as extra work with no extra reward and they found it was easier to leave the swabs for the research staff to complete.

Discussion

A US survey145 showed that nurses and midwives were the commonest staff group undertaking POC testing (46%), followed by medical assistants (25%) and physicians (9%). For the complex tests used in this study we did not consider that it was feasible to employ midwives to perform tests that would take them away from patient care for over 1 hour per test. We found that it was possible to train midwifery assistants to conduct the POC tests, although initial and ongoing training required a significant time commitment on the part of midwifery assistants and the trainer. However, with the resources available to us we encountered unexpected practical difficulties in delivering results within a clinically useful time frame and throughout the day and night. Staff shortages, created by vacancy management and used as a component of financial recovery plans, are common in NHS hospitals and we did not appreciate beforehand the impact that this could have on the viability of a POC testing programme requiring a significant amount of labour. A survey145 of US sites undertaking POC testing highlighted the difficulties in maintaining adequate staffing even for simple waived tests. The impact of staff shortages in our study would probably have been less had the testing requirement been limited to a single test of moderate complexity.

We do believe that, in principle, both the OIA and PCR tests are feasible at the POC level; however, to achieve the degree of reliability in the generation of timely results that would be required to underpin a screening programme a very significant additional staff resource would be required, especially for the PCR test. Moreover, because of the complexity of sample preparation we consider that testing would have to be undertaken at defined times (e.g. hourly or 2-hourly) rather than strictly on demand. This would have an impact on the proportion of tests that deliver results sufficiently early in labour to enable adequate IAP to be provided to colonised mothers.

Neonatal outcomes and adverse events

In total, 15 babies were reported to have had an infection immediately postpartum, of which six were invasive. Of the invasive infections three were diagnosed as EOGBS disease; all of these babies recovered. Four babies with infection were not given antibiotics, three of whom had superficial infection only. The fourth died from hypoxia and infection, although the cause of the infection is unknown as the postmortem was performed elsewhere. Four babies died from other causes.

In total, 82 babies were given antibiotics within 7 days of delivery. Of these, nine had an infection, eight had no information about infection and the remaining 65 received antibiotics prophylactically.

There were no adverse events associated with taking the swabs, the testing procedure or the administration of antibiotics to the mother or baby.

Background

The NIHR Health Technology Assessment (HTA) programme funded this project on the accuracy and cost-effectiveness of the rapid diagnosis of group B streptococcus during labour (project number 02/38/04) in October 2004. The issue of independent monitoring of the test accuracy study within this project was raised at the outset. The HTA programme had commissioned few primary test accuracy studies previously, but the requirements for independently monitoring these types of study had not necessarily been considered to be distinct from existing guidance on independent monitoring of RCTs. The investigators took it upon themselves to develop a model for independent monitoring that would provide external input and supervision throughout the course of their study, which was accepted by the HTA programme. Based on our experience we report some recommendations for the monitoring of test accuracy studies.

Preamble

All clinical research that involves the prospective study of patients needs some form of independent monitoring. 146 In the UK, public funders of research, including the HTA programme and the Medical Research Council, require independent monitoring committees to be established for all RCTs,147,148 although the approach employed in the UK may not be the model used elsewhere. 149 In RCTs of effects of interventions, an independent trial steering committee (TSC) oversees the management and financial aspects of the study, and a data monitoring committee (DMC) (also sometimes known as a data monitoring and ethics committee or a data monitoring and safety board) oversees the accumulating data. There is no legal requirement by regulatory authorities for RCTs to convene DMCs, although guidance exists from the US Food and Drug Administration as to when and how they should be employed. 150

The TSC and DMC are each composed of a small number of individuals who have pertinent expertise and are independent of the study investigators and host institutions, generally having no role in the initial design or conduct of the study. Their primary role is to provide overall supervision of the study, to ensure that the study is conducted to good research standards, to scrutinise interim data and to advise the sponsor regarding the continuing integrity of the study and the safety of current and potential future participants in the trial. By having access to the accruing data, the DMC is able to make an objective assessment as to whether the primary research question has been answered. In the process, the TSC and DMC also play a valuable role in overseeing protocol compliance and study validity so that reliable and objective evidence is produced. There have been numerous recommendations regarding the establishment and operation of DMCs for RCTs, most notably the UK DAMOCLES project. 151 Following a comprehensive review of the literature151 this project provided detailed guidance on the conduct of DMCs. Other groups, including the Society for Clinical Trials, have considered monitoring arrangements for trials that typically have not employed DMCs, including early phase exploratory trials. 152 Unfortunately, none of these provide guidance as to how and when independent monitoring should be employed in primary test accuracy studies.

Here we propose guidance for monitoring test accuracy studies, detailing the main issues to be considered and providing operational guidance for investigators and independent monitoring committee members. Our objective is to encourage researchers and sponsors currently engaged in diagnostic research to consider employing external monitoring of accuracy studies, and to initiate a discussion about formalising the independent monitoring of such research in the future.

Study design considerations for independent monitoring

In the evaluation of tests there are many possible designs. 81,153 The fundamental difference between test accuracy studies and RCTs of effectiveness of interventions is that, in the former, randomisation to difference tests is generally not necessary; typically all participants suspected of having the target condition are subject to both the test of interest (index test) and a confirmatory test or observation (reference standard). The aim, therefore, is to determine the consistency of the index test relative to the reference standard (as in Chapter 2). Accuracy is described in terms of sensitivity and specificity of the index test, amongst other parameters. Inevitably test accuracy studies will vary in complexity depending on the numbers and types of index tests and reference standards, the numbers and types of participants and centres, the need to embed these within an HTA, etc. Our guidance largely relates to relatively large studies conducted over a period of time such that monitoring will influence their ongoing progress.

In RCTs comparing the effectiveness of interventions, the benefits and harms of which are unknown, there is an over-riding concern about safety and stopping rules that underpins the desire to have independent monitoring. Usually, DMCs for RCTs review accruing data in confidence and make a recommendation on whether to stop the trial, continue recruiting or modify the trial protocol. The DMC may prespecify criteria or thresholds of difference between the treatment groups, with important and statistically significant differences dictating closure of the trial. 154,155 Any DMC recommendation is considered by the TSC, which makes a decision on behalf of the sponsor, based on the DMC’s recommendation and other external evidence and circumstances. The TSC does not have access to the interim data seen by the DMC, allowing them to make objective decisions without being influenced by the emerging results. Figure 18a demonstrates the relationship between various committees in a typical RCT of interventions.

FIGURE 18.

Independent monitoring of randomised controlled trials of effects of interventions and of test accuracy studies. (a) Model for independent monitoring of randomised controlled trials. (b) Model for independent monitoring of test accuracy studies.

Can the same data monitoring model be applied to primary test accuracy studies? There are some important considerations that may require the model to be adapted. These centre around the need for confidential assessment by the independent monitors, the relationship that the DMC has with the TSC and the criteria used to decide whether to recommend halting the study. Difficulties arise in the area of stopping criteria, in which there is virtually no work, and in the area of confidentiality of the interim data. The question to ask is whether access to the interim data would influence the decision of the monitors regarding study continuation. If it is considered likely that it might, as is often the case in RCTs of interventions, then the TSC and DMC should be established to function separately. This guidance does not seek to dictate when analyses should be kept confidential or not. The consequences for each study should be assessed before the start and the most appropriate model of monitoring employed depending on the perceived risks to the study.

In general, it can be argued that the observational design of test accuracy studies does not require the same levels of independence as those employed in RCTs. In test accuracy studies the main concern is harm to participants from the tests, which might also affect the completeness of verification. Another consideration is whether clinical behaviour might be influenced by early release of interim results, leading to early adoption of the test or loss of the opportunity to verify the index test by the reference standard. If so, the monitors should review the accuracy data in confidence. However, when it is unlikely that access to accuracy data will influence the decision-making processes there is the potential for a single entity to monitor data and oversee the study, or for the independent members of the monitoring committee to be convened as a subcommittee for data monitoring (Figure 18b). This would facilitate easy decision-making and avoid duplication of effort, simplifying the monitoring process. Further work to guide study investigators and, in particular, the development of criteria for stopping test accuracy studies is needed.

Aspects for consideration by independent monitors

Independent monitoring should consider the rate of recruitment, the disease prevalence, study quality, accuracy estimates, safety and the need for additional analyses. We believe that most of this information, except for accuracy estimates, can always be shared openly.

Organisation and conduct of independent monitoring committees

Many of the useful recommendations on the practical aspects of independent monitoring of RCTs148,149,151 can be easily translated to the conduct of monitoring committees for test accuracy studies.

Conclusion

Monitoring should be considered in all primary test accuracy studies. The specific role of independent monitors in test accuracy studies differs from their role in RCTs, although no formal recommendations pertaining specifically to independent monitoring in test accuracy studies exist. The monitors should ensure that the aim of the study to be undertaken is explicit from the outset and that the design chosen and power estimations made are cognate with the aim. They should ensure that the disease prevalence and spectrum are monitored along with study quality, compliance with the protocol, safety and recruitment rates. Interim test accuracy estimates should be examined confidentially by independent monitors if there is a potential for clinical practice to be influenced by knowledge of interim results. Further work is required to establish the role of statistical stopping rules in the monitoring of test accuracy studies.

Contributors to this appendix

Khalid S Khan,¹ Jane Daniels,^1,2 Shakila Thangaratinam,¹ Ruth Gilbert,³ Carole Cummins,⁴ Richard Gray,² Jos Kleijnen,⁵ Gerben ter Riet⁶ and Patrick Bossuyt⁷

Prevalence and relative risks

The following neonatal prevalence calculations are based on Tables 38 and 39.

Time and motion study

A time and motion study was conducted to ascertain the time that it would take to conduct the PCR and OIA rapid tests (Tables 40–42). We followed an experienced biomedical scientist from the time that he received the maternal vaginal and rectal swabs from the maternity ward to the time that he produced a set of results after running the rapid tests. This was carried out on 3 and 2 non-consecutive days for the PCR and OIA tests respectively. Following the observation period the times that it took for each process were logged and the average of the three or two periods was calculated. It was assumed that it took on average 15 minutes for a midwife to collect swabs from a woman. This time was added to the total time taken for a test to be carried out. The steps described in the instruction manuals for the PCR and OIA tests were adhered to.

Cost inputs

The cost of each test and each intervention was estimated from different sources, described in more detail below. All costs are presented in UK pounds and 2006 prices and are summarised in Table 43.

Discounting life-years

The life expectancy at birth in the UK is 76 years for men and 81 years for women (http://www.gad.gov.uk/Demography%20data/Life%20tables/Interim_life_tables.html). However, even at full health these estimates cannot be used as the true yard stick for calculating an individual’s stream of future health benefits as they do not take into account time preferences. A discounted life expectancy that incorporates these preferences would be a more accurate reflection. The recommended discount rate in the UK is 3.5%. 129

Table 51 shows the discounted survival curves that have been used to calculate the discounted life expectancy at full health. The first column (years) is the age of an individual and this has been capped at 100 years. The sum of the second and third columns (males and females, respectively) is the life expectancy at birth for an individual. This is shown to be 76 years for men and 81 years for women. Considering that a whole year is assumed for anyone who survives their first birthday, these values are an overestimate. The sum of the fifth and sixth columns (males discounted and females discounted) shows the discounted equivalent life at full health for men and women respectively. This means that the discounted equivalent life cannot be more than 27.0 years for men or 27.4 years for women.