Extracorporeal carbon dioxide removal compared to ventilation alone in patients with acute hypoxaemic respiratory failure: cost-utility analysis of the REST RCT

Measurement of health resource use and costs

Hospital resource use data were collected prospectively using the case report form during the participants’ primary admission. Length of stay for the primary critical care admission was calculated from the date of randomisation to the date of critical care discharge or date of death if this occurred within critical care. General hospital ward length of stay was calculated from the date of critical care discharge to the date of hospital discharge or date of death if this occurred on the ward. For ICUs that participate in the case mix programme (CMP), additional information was obtained from the Intensive Care National Audit and Research Centre (ICNARC) on any readmissions to critical care. For the four Scottish, non-CMP ICUs the intention had been to obtain this data from the Scottish Intensive Care Society Audit Group (SICSAG) via the electronic Data Research and Innovation Service (eDRIS) but unfortunately the application to obtain this data was delayed and data linkage could not be obtained prior to the early closure of the study. This meant data on readmission to critical care was unavailable for some participants.

To facilitate the costing of the critical care admissions the Healthcare Resource Group (HRG) codes corresponding to each critical care admission and any readmission during the primary hospital admission were provided by ICNARC for the CMP sites. The HRG codes represented the highest level of complexity, based on the total number of organs supported during the admission. Scotland has not fully adopted the HRG methodology, so for a consistent costing approach we applied the modal HRG code observed for the critical care admissions of participants at the CMP sites to the critical care admissions for the Scottish participants. Critical care costs were calculated by multiplying the unit cost associated with the HRG by the length of stay for that admission. The unit costs associated with the HRGs were obtained from the National Schedule of NHS Costs 2018/194 (see Appendix 1, Table 5). Ward stay costs were calculated by multiplying the number of ward days during the primary hospital admission by the unit cost associated with rehabilitation for respiratory disorders.

Participants’ use of health and social care services from hospital discharge to 12 months was captured via a postal questionnaire completed at 6 and 12 months post randomisation. Telephone completion was also used for non-responders. Participants were provided with a health service log booklet at hospital discharge and again at 6 months to encourage them to keep track of their health resource use and facilitate questionnaire completion. Mortality status was confirmed prior to participant contact by contacting general practitioners (GPs).

Individual-level resource use was combined with unit costs to estimate costs for each participant. Unit costs were obtained from publicly available sources; National Schedule of NHS Costs4 and Unit Cost of Health and Social Care from PSS Research Unit. 13 The cost of supplying oxygen at home was provided by the Northern Ireland Health and Social Care Board and costs for intervention consumables were obtained directly from Alung, the device provider. The last follow-up data was collected in 2019 and the price year was set at 2018/2019 (see Appendix 1, Table 5).

Measurement of health outcomes

The outcome of interest in the cost-effectiveness analysis was the QALY, a generic HRQoL measure. Utilities for the calculation of the QALYs were obtained using the EuroQol-5 Dimensions, five-level version (EQ-5D-5L)14 administered at 6 and 12 months post randomisation via a postal questionnaire. Telephone completion was also used for non-responders. The EQ-5D-5L is a generic preference-based measure of HRQoL, which provides a description of health using five dimensions (mobility, self-care, usual activities, pain/discomfort and anxiety/depression) each with five levels of severity. Responses were converted into utility scores using the Crosswalk Value Set15 for the UK population. This tariff maps the EQ-5D-5L responses on to the EuroQol-5 Dimensions, three-level version (EQ-5D-3L) and is currently the approach recommended by NICE. 16 QALYs were calculated using the area under the curve method. As patients were critically ill at baseline, an EQ-5D-5L utility score of zero was assumed, in keeping with other studies in the critical care setting. 5,17,18

Analysis of health resource use, costs and outcomes

The descriptive statistics were used to summarise (by treatment arm) the resource use (during primary hospital admission and after discharge until 12 months), the associated costs, EQ-5D-5L scores and QALYs. Death was not treated as a censoring event and periods after death were counted as observations with known outcome19 an approach used previously in similar patient populations5,18 This meant that for participants who had died in hospital, resource use and EQ-5D-5L scores after hospital discharge until 12-month follow-up were considered to be zero. For patients who died between hospital discharge and 6 months resource use and EQ-5D-5L scores from 6 to 12 months were considered to be zero. Total costs and QALYs were analysed using linear regression. Significance (p < 0.05) was judged where the confidence intervals (CIs) excluded zero. Analyses were undertaken using Stata 15/IC for Windows^® (StataCorp LP, College Station, TX, USA).

Cost-effectiveness analysis

Trial-based economic evaluations tend to measure participants’ health resource use and health outcomes at multiple time over the duration of the study using self-complete questionnaires. As a result, missing data is a common problem due to reasons such as non-returns or loss to follow up. This has the potential to introduce bias into the results as participants with incomplete health economic data may be systematically different from those with complete data. 20 Therefore for the cost-effectiveness analysis we imputed missing total cost and QALY data using multiple imputation with chained equations and predictive mean matching using the ‘mi impute chained command’ in Stata. This assumes that data are missing at random (MAR). This involved a regression model being specified to predict the missing total cost and QALY data: treatment group, baseline acute physiology and chronic health evaluation II (APACHE II) score, age, mortality at 28 days and primary hospital admission costs were entered into the model as predictors. Forty imputed data sets were generated, which was similar to the maximum percentage of incomplete cases observed (40%) in the data as recommended by White et al. 21 The ‘mi estimate’ Stata command was used to facilitate the analysis of each of the imputed data sets and then combine the results. Linear regression was used to estimate the incremental (differential) mean costs and QALYs. The ICER was calculated by dividing the incremental mean costs by the incremental mean QALYs to estimate the cost per QALY. As negative ICERs are not meaningful, if this occurred we stated whether the intervention was dominant (i.e. more effective and less costly than standard care) or was dominated (less effective and more costly than standard care).

Uncertainty in the cost and QALY data was explored by the non-parametric bootstrapping of the linear regression cost and QALY models simultaneously, drawing 1000 samples of the same size as the original sample with replacement. 22 The resulting 1000 ICER replicates were plotted on the cost-effectiveness plane,23 and used to construct a cost-effectiveness acceptability curve (CEAC). 24 This showed the probability of ECCO₂R being cost-effective compared to ventilation alone at various willingness to pay (WTP) per QALY thresholds. In general NICE considers interventions with an ICER of ˂£20,000 to be cost-effective. 7

The incremental net monetary benefit (INMB) was also used to aid interpretation. The INMB is a summary statistic representing the value of an intervention in monetary terms when a WTP threshold for a unit of benefit is known. This was calculated by multiplying the incremental mean QALY by NICE’s threshold of £20,000 and then subtracting the incremental mean costs. A positive INMB indicates the intervention is cost-effective. 10

Sensitivity analysis for the cost-effectiveness analysis

The robustness of the results from the cost-effectiveness analysis was explored via the following sensitivity analyses: adjusting for baseline age and APACHE II score via multiple regression; reducing the time horizon of the cost-effectiveness analysis to 6 months, and scenarios of plausible departures from the MAR assumption. The latter was done via pattern-mixture models implemented using multiple imputation. 25 The impact of the following scenarios was explored: participants with missing QALY data were assumed to have worse HRQoL (10%) than those with observed QALY data; participants with missing cost data were assumed to have higher costs (10%) than those with complete cost data; and those with missing QALY and cost data were assumed to have both lower QALYs and higher costs.

Health resource use and costs

Resource use during the primary admission is presented in Appendix 1, Table 6 and self-reported health service use from hospital discharge to 12 months is presented in Appendix 1, Tables 7 and 8. Data is presented for all patients with available data without imputation of missing cases, by treatment arm. The costs (Great British pounds) associated with resource use are presented in Table 2. There was a trend for patients receiving ECCO₂R to require marginally longer primary ICU stay, more ICU readmission days and more ward days than those receiving ventilation alone; however, none of these differences were statistically significantly different and the overall difference in primary admission costs (excluding intervention costs) was also not significantly different (mean difference £2666.97; 95% CI –2886.42 to 8220.35). The cost difference for the period from discharge to 6 months was relatively small (mean difference £172.70; 95% CI –871.81 to 1217.21), but a larger and statistically significant difference for the 6- to 12-month period (mean difference £858.52; 95% CI 125.83 to 1591.22) was observed. The resource use and costs associated with ECCO₂R were also considered. These consisted of a cartridge for the Alung device (£3000 per patient) and a catheter (£650 per patient) and this total cost of £3650 was added to the total for each patient in the ECCO₂R arm. Total health service use costs (including intervention costs) over the full 12 months were calculated for those patients with complete cost data (125 ECCO₂R patients and 126 ventilation alone patients). The difference was large and statistically significantly different (£7668.76, 95% CI 159.75 to 15,177.77).

Health utility scores and QALYs

Mean EQ-5D-5L utilities and QALYs over the study period are presented in Table 3 for patients with available data. Overall mean utility scores were low and similar for each group at both 6 and 12 months with no discernible difference in QALYs (mean difference –0.01; 95% CI –0.06 to 0.05).

Cost-effectiveness analysis

The results of the cost-effectiveness analysis for the base-case analysis and the sensitivity analyses are presented in Table 4. For all scenarios ECCO₂R was associated with lower QALYs compared to ventilation alone, but the differences were small and not statistically significant indicating broad equivalence in terms of HRQoL impact. However, statistically significantly higher health-care costs were observed with ECCO₂R compared to ventilation alone and so ECCO₂R can be described as being dominated by ventilation alone and not cost-effective given the data. Uncertainty in the cost and QALYs estimates was explored by displaying the results of the non-parametric bootstrapping on the cost-effectiveness plane for the base-case analysis (see Figure 1). It can be seen that ICER replicates straddle the north-west and north-east quadrants reflecting the consistently higher costs associated with ECCO₂R in the data, and similarity in QALYs. The CEAC (see Figure 2) is in fact a line running along the x-axis indicating that there was 0% probability of ECCO₂R being cost-effective compared to ventilation alone for any of the WTP threshold per QALY we considered (£0–50,000) in the analysis. The negative INMBs reflect that the intervention is not cost-effective compared to standard care at a WTP threshold of £20,000.

FIGURE 1.

Cost-effectiveness plane for the base-case cost-utility analysis showing 1000 bootstrapped replications of incremental mean costs and QALYs and the WTP threshold of £20,000/QALY.

FIGURE 2.

Cost-effectiveness acceptability curve showing the probability of intervention being cost-effective compared to standard care (base-case analysis).