Economic evaluation

This chapter reports the economic evaluation conducted alongside the GBS2 clinical trial. The aim of the economic evaluation was to explore whether or not a rapid intrapartum test for GBS was cost-effective in reducing maternal and neonatal antibiotic usage, compared with usual care, where the administration of IAP was directed based on maternal risk factors alone.

Methods

The economic analysis conducted alongside the GBS2 trial was based on primary clinical data collected in the cluster RCT, which was reported in earlier chapters. The principal clinical data of relevance to the economic evaluation included the use of IAP, maternal length of stay and neonatal length of stay. In the RCT, clusters were allocated to testing strategies of either usual care, based on the assessment of risk factors for GBS, or a rapid intrapartum test, which involved the use of the GeneXpert test and the results of the strategy informing IAP use. All resource use by the testing strategy group was recorded and unit costs applied.

The outcome of interest was the incidence of the use of maternal IAP that was avoided by a more targeted approach using the rapid test; hereafter this is referred to as the case of IAP avoided. The base-case cost-effectiveness analysis (CEA) results and the results of the sensitivity analyses were expressed in terms of the cost per case of IAP avoided. In the base-case analyses, costs and outcomes were based on antibiotics given specifically to reduce the risk of GBS transmission to the baby. As set out in the protocol, this included IAP for preventing vertical transmission of GBS from mother to baby only and IAP in combination with antibiotics for any indication. The length of hospital stay in the base case related only to the mother and the cost was derived from NHS reference costs for an indicative length of stay according to the mode of birth, and was not based on the trial data.

Sensitivity analyses were carried out:

• Sensitivity analysis 1 – the length of stay for only the mother was included and was based on trial data, and the NHS reference costs for each mode of birth were removed from the analyses for both trial strategy groups.
• Sensitivity analysis 2 – as sensitivity analysis 1, but with the costs of inpatient stay for infants added to the costs for both testing strategy groups of the trial based on the trial data.

The focus of the trial was on the period of labour and birth and, therefore, within a 12-month time frame, and so the discounting of costs and outcomes was not applicable. The results of the economic evaluation are reported in accordance with the Consolidated Health Economic Evaluation Reporting Standards (CHEERS).61

Resource use and costs

The CEA adopted the UK NHS perspective. In alignment with recommended practice, only direct costs to the health service were included in this analysis.62 The data from the trial were at an individual woman level, as resource use data were collected via electronic health records. The health-care resource utilisation data were obtained on respondents’ usage from hospital admission until the discharge date. The main resource use monitored included the use of intrapartum antibiotics, and resource use associated with the midwife monitoring the woman, admission to labour and delivery units and the length of labour. In terms of undertaking the rapid test during labour, the resource use was based on the listed resource use from a previous time-and-motion study51 and was varied in sensitivity analysis.

Valuing resource use

Antibiotic prices were obtained from the British National Formulary63 and secondary care costs from the National Schedule of Reference Costs (2017–2018).64 All resource use was multiplied by the unit cost to estimate the overall costs for each treatment allocation. All items, their relevant unit costs and the sources for the cost data are presented in Tables 25–27. All unit costs were valued at 2017/18 prices in UK pounds sterling. When necessary, costs from prior years were inflated to the current price year using the Hospital and Community Health Service index.64

Economic analysis

All statistical analyses were conducted using Stata^® version 15 (StataCorp LP, College Station, TX, USA). Initial analyses included conducting a cost–consequence analysis, detailing all costs and outcomes incurred in each trial strategy in a disaggregated form. This analysis was based on an intention-to-treat basis, meaning that within the analysis all women were analysed according to their maternity unit’s allocated testing strategy, regardless of whether the individual adhered to the site’s testing strategy or IAP was administered appropriately.68

The total costs incurred by each woman during labour were calculated by multiplying the resource use volume by the unit costs. Moreover, mean costs and outcomes were generated for each trial testing strategy. The principal outcome on which the economic evaluation was based is the number of instances of IAP avoided. The timeliness of the test, in terms of the rapid test producing a result in time to administer intrapartum antibiotics before labour begins, was not factored in, as the cost of the rapid test is incurred regardless of whether or not the result was presented in time to be acted on. An incremental analysis was conducted if appropriately indicated by the cost–consequence analysis. An incremental analysis is estimated by comparing the differences in mean costs and outcomes between the two strategies using seemingly unrelated regression estimates. By dividing the cost differences by the outcome differences, the incremental cost-effectiveness ratio (ICER) is calculated. To account for the inherent skewness of the cost data, 95% CIs around the mean differences were calculated using the bias-corrected and accelerated bootstrap method.69

Stochastic cost-effectiveness analysis

A stochastic CEA was conducted to examine the sensitivity of the results. The approach taken in the probabilistic sensitivity analysis was that all important variables relating to costs and clinical outcomes were given a distribution that describes the uncertainty surrounding the mean. The distributions were simulated 10,000 times. Each time, random numbers were drawn from the appropriate distributions. After each simulation, the incremental costs and effects were plotted in a cost-effectiveness plane that comprised four quadrants: (1) north-east, (2) north-west, (3) south-east and (4) south-west. The scatterplot that was produced represented the simulations. If dots from the scatterplot were in the north-east quadrant, then this indicated that the addition of the rapid test was more costly and more effective than usual care, based on risk factors. Dots in the north-west quadrant indicated that the rapid test was more costly and less effective than the usual care. Based on these simulations, the probability that the rapid test strategy would be cost-effective was presented. This was the standard approach for health economics following accepted guidelines (i.e. CHEERS)70 and was a presentation of results of cost-effectiveness studies that would be required by decision-makers, such as NICE.

Results were also presented using cost-effectiveness acceptability curves (CEACs) to reflect sampling variation and uncertainties in the cost-effectiveness value, when appropriate. The CEACs show the probability of the rapid test being cost-effective, compared with usual care, based on risk factors and observed data for a range of maximum monetary values (thresholds) that decision-makers might be willing to pay for a particular unit change in outcome.71 We examine cost-effectiveness across a range of monetary willingness-to-pay (WTP) thresholds.

Additional sensitivity analyses were carried out:

• Sensitivity analysis 3 – the proportion of midwife time to conduct the rapid test was reduced from 40 minutes in the base case to 10 minutes.
• Sensitivity analysis 4 – only the women for whom IAP was given specifically for GBS and no other reason were analysed.

Results

Rapid intrapartum test

A total of 721 women gave birth in units randomised to the rapid test (one woman delivered elsewhere) and 906 women gave birth in usual-care units. Almost all women in the rapid test arm had at least one swab taken and reported (for three women it is unknown whether or not the swab was taken and no test results are available). However, more tests were issued within the rapid test strategy, depending on the length of labour and delivery stage. The data identified 26 women who had a repeat test performed because either the test was not started within 15 minutes of taking the swab or labour had not progressed within 48 hours of initial admission.

Risk factors

Of the total population, the most common risk factor was preterm labour, identified in 757 (46%) women. This was followed by a history of temperatures of ≥ 38 °C during labour, identified in 662 (41%) women (see Table 8). It should be noted that, for this analysis, any costs associated in assessing risk factors were assumed to be included within the ‘usual’ birth costs.

Vaginal deliveries

Vaginal deliveries were grouped into four categories: (1) spontaneous vaginal delivery following spontaneous onset, (2) vaginal delivery following induced labour, (3) instrumental delivery following spontaneous onset of labour and (4) instrumental delivery following induction of labour. These categories are based on the trial data set and the available cost data for these types of birth in the NHS reference costs.64 Across both strategy groups, 584 (48.1%) women had spontaneous deliveries, 397 (32.7%) had a vaginal birth with induced labour, 124 (10.2%) had an assisted vaginal birth and 109 (9.0%) had an assisted and induced birth.

Caesarean deliveries

Elective caesareans were not included in this analysis, as they were part of the exclusion criteria. Women who gave birth by emergency caesarean section were included. In total, 406 (25.0%) women underwent an emergency caesarean section.

Resource use

The following resource use and costs were reported by category. The estimation of resource use and associated cost was carried out on complete-case data.

Antibiotics provided

The average doses of antibiotics administered per woman during labour are presented by testing strategy in Appendix 3, Table 33. The antibiotic usage was categorised in terms of loading doses and maintenance doses. Loading doses were high initial doses issued at the beginning of the treatment given prior to reduced maintenance dosages.72 Overall, on average, women in the rapid test units received a slightly higher level of antibiotic in terms of loading doses. There was little variation in antibiotic for maintenance dose overall.

Average costs

To calculate the mean costs for each testing strategy, mean resource use was multiplied by the relevant unit cost. The mean cost varied between strategies for individual antibiotics (see Appendix 3,Table 34), but the overall the mean of the total antibiotic cost was higher in the rapid test arm.

Loading dose costs

There was variation in the costs regarding the different types of antibiotics issued. In terms of the costs for loading dosages, sizeable differences in costs were noted for benzylpenicillin, cephalosporins and metronidazole, as noted by the bootstrapped difference 95% CIs. With the exception of cephalosporins, the rapid test units showed a higher mean loading dose cost for both benzylpenicillin and metronidazole.

Maintenance dose costs

There was less variation in the mean costs for the maintenance dose of IAPs. Notable differences were observed for cephalosporins and metronidazole. A difference was noted for the benzylpenicillin maintenance dosage; here, the rapid test has a mean cost of £11.43, whereas the usual-care strategy has a mean cost of £9.06. Table 28 presents the combined costs for both the loading and the maintenance doses. There was a variation in the types of antibiotic issued. Clear differences between the two strategies existed for almost all antibiotics, with the data showing that typically the rapid test incurs higher mean total costs than usual care. The only exception being other antibiotics, namely amikacin, amoxicillin, ampicillin, erythromycin and gentamicin, outside the main group issued to women during labour. The other antibiotic average cost in the rapid test strategy was lower than the usual-care strategy (£12.65 vs. £19.63), with an adjusted mean difference of £6.98 less than the usual-care arm.

Cost–consequences analysis

Table 29 presents the costs and outcomes assessed within the GBS2 trial. In the rapid test strategy, the costs included the following: the mean cost of IAP, the mean costs of birth/deliveries and the mean cost of issuing the rapid test. The mean cost of issuing the rapid test included the additional midwifery time per woman, the costs of vaginal/rectal swabs, the costs of a cartridge for the GeneXpert machine and the machine running time. By contrast, the costs of the usual-care strategy comprised the mean cost of IAP and the mean costs of the baby being born. The assumption was made that assessing for risk factors is an integral component of the overall cost of the baby being born. The mean cost per woman in the rapid testing strategy was approximately £4128, whereas under usual care the mean cost was approximately £4003 per woman.

Outcomes

The primary outcome of the trial was the proportion of women for whom IAP was used as prophylaxis against GBS infection in her baby. In the rapid test sites, IAP for GBS prophylaxis was administered to 297 women [proportion 0.41 (297/722)], compared with 328 women in the usual-care units [proportion 0.36 (328/906)]. A higher proportion of women in the rapid test units received IAP to prevent neonatal GBS disease, but there was no statistically significant difference from those in the usual-care units (see Table 12).

Base-case analysis

The results in Table 29 show that the mean cost per woman in the rapid test units was approximately £4128 and the mean proportion of women treated with antibiotics was 0.41. This compared with the mean cost per woman in usual-care units being approximately £4003 and a mean proportion of women treated with antibiotics of 0.36. Therefore, to achieve the outcome of averting antibiotic use, the rapid test was more costly and less effective than usual care, as it cost more and antibiotics use to prevent GBS infection was greater. The use of the rapid test in this analysis was said to be dominated by usual care. Therefore, producing an ICER is not applicable for the base case, as the cost–consequence analyses revealed the situation of dominance.

The cost-effectiveness plane for the base-case analysis is presented in Figure 9. The majority of the scatterplot dots (depicting paired incremental costs and outcomes) are in the north-west and south-west quadrants. Scatterplot dots falling in the north-west quadrant represent higher costs and worse outcome than the comparator, whereas scatterplot dots in the south-west quadrant represent worse outcome and lower costs. Therefore, Figure 9 suggests that the rapid test is less effective at reducing the use of IAP. However, it is uncertain whether the rapid test is likely to be more costly (north-west quadrant) or less costly (south-west quadrant) compared with usual care (represented by the origin). The CEAC (Figure 10) shows the probability of the rapid test being cost-effective at various values of decision-makers’ WTP per case of antibiotic avoided. The CEAC (see Figure 10) suggests that the probability of the rapid test being cost-effective is > 40% for all WTP values, tending to 0% as the WTP increases. Therefore, the probability of cost-effectiveness decreases as the WTP increases.

FIGURE 9 Cost-effectiveness plane: the base case. (Figure not reproduced here; see the original article.)

FIGURE 10 Cost-effectiveness acceptability curve: the base case. (Figure not reproduced here; see the original article.)

Sensitivity analyses were carried out:

• Sensitivity analysis 1 – the length of stay for the mother was based on trial data, and the mode of birth (which includes costs of an indicative length of stay that was used in the base case) was removed from the analyses for both trial testing strategy groups.
• Sensitivity analysis 2 – as for sensitivity analysis 1, but the costs of inpatient stay for newborns were added to the costs for both trial testing strategy groups based on the trial data.

The descriptive statistics for inpatient days for the mother and the babies on which these two sensitivity analyses are based are presented in the Appendix 3, Tables 35 and 36, respectively.

Sensitivity analysis 1

The mean length of stay and mean costs per strategy are presented in Appendix 3, Table 35. The cost-effectiveness plane for sensitivity analysis 1 is presented in Figure 11. The majority of the scatterplot dots now resided in the north-west quadrant and there were fewer in the south-west quadrant. This suggested that the true length of stay, based on the length of stay recorded in the trial, was actually more costly than the reference cost data implied. The test remained ineffective, but some uncertainty around the cost data was reduced, as the rapid test strategy was shown to be more costly than usual care. Figure 11 broadly supported the base-case analysis and showed that the rapid test strategy was more costly and less effective than usual care. The CEAC (Figure 12) was similar to that of the base case.

FIGURE 11 Cost-effectiveness plane: mothers’ length of stay (sensitivity analysis 1). (Figure not reproduced here; see the original article.)

FIGURE 12 Cost-effectiveness acceptability curve: mothers’ length of stay (sensitivity analysis 1). (Figure not reproduced here; see the original article.)

Sensitivity analysis 2

This was similar to sensitivity analysis 1, but the costs of inpatient stay for newborns by their level of care were added to the costs in both strategy groups of the trial, based on the trial data. When needed, assumptions were required as indicated in the Appendix 3, Table 36.

The cost-effectiveness plane for sensitivity analysis 2 is presented in Figure 13. The majority of the scatterplot dots have now shifted to the south-west quadrant, showing that although, overall, more antibiotics were prescribed in the rapid test units than in the usual-care units, the overall costs were now less for the rapid test strategy. This meant that the length of stay of babies and their corresponding costs were higher in the usual-care units. The implication was that these babies required more care than the babies in the rapid tests units, whose mothers had received IAP. Therefore, Figure 13 differed from the base-case analysis and suggested that the addition of the babies’ length of stay changes the costs of the rapid test strategy to be less costly than usual care overall. The CEAC (Figure 14) showed that the probability of being cost-effective declined as the WTP increased, primarily because more antibiotics were prescribed in the rapid test group and so the rapid test had failed to meet the objective of reducing the prescribing of antibiotics, despite being slightly cheaper overall.

FIGURE 13 Cost-effectiveness plane: mothers’ and babies’ length of stay (sensitivity analysis 2). (Figure not reproduced here; see the original article.)

FIGURE 14 Cost-effectiveness acceptability curve: mothers’ and babies’ length of stay (sensitivity analysis 2). (Figure not reproduced here; see the original article.)

Additional probability sensitivity analyses were also conducted. Sensitivity analysis 3 reduced the length of midwife time in the rapid test strategy from 40 minutes to 10 minutes and showed no substantial change on the base case. Sensitivity analysis 4 was the same as the base case, but woman who were prescribed GBS prophylaxis only and not antibiotics for any other reason were included. No substantial change in the base case was observed, with the cost-effectiveness plane focusing on the origin.