Cost-effectiveness of craniotomy versus decompressive craniectomy for UK patients with traumatic acute subdural haematoma

Participants

The RESCUE-ASDH trial5 is a multicentre, international, pragmatic, parallel-group, randomised trial that compared craniotomy with DC. Patients were eligible if they were ≥16 years, had an ASDH on CT scan and the admitting surgeon felt that the haematoma needed evacuating either by craniotomy or DC. The economic evaluation was nested within the RESCUE-ASDH trial and based on UK participants only.

Treatment and randomisation

Enrolled patients had their ASDH evacuated in the operating room under general anaesthesia. The bone flap was raised, the dura opened and the haematoma evacuated, after which patients were randomly assigned to receive either craniotomy (bone flap restored before skin closure) or DC (bone flap removed prior to skin closure with a view to being restored later). Patients were only randomised if either treatment was feasible, those patients whose brain was too swollen to allow replacement of the bone flap were not randomised. These patients would have the bone flap left out and were not included in the intention-to-treat (ITT) analysis presented in this paper. As a pragmatic study, management of patients preoperatively, intraoperatively and postoperatively was undertaken according to each centre’s standard of care.

Blocked randomisation (block size 4) with 1:1 ratio was used, with allocation stratified by geographical region, age group, severity of injury and CT findings.5 Patients randomised to craniotomy could have a DC at a later stage if their condition deteriorated and at the discretion of the treating clinician. It was not possible to blind patients, relatives and treating clinicians but the primary outcome (see below) was adjudicated centrally by blinded investigators.

Measuring costs

Costs were estimated from a UK National Health Service (NHS) and Personal Social Services (PSS) perspective.7 Resource use data were collected via two methods: hospital-recorded data and a Patient Self-Report (12-month follow-up) Questionnaire (PSRQ). Both methods of data collection were developed in consultation with hospital staff/patients and focused on big cost drivers/resources that were expected to differ between arms.8 All resource use items that were costed (see below) were estimated in £ Sterling for the 2018/2019 financial year, resource use items undertaken for research purposes were not costed.

The hospital-recorded data included the following resource use items: details of the intervention (craniotomy or DC) including length of operation and graft details; time spent in the intensive care unit (ICU) and neurosurgical unit (NSU) during initial (index) admission; cranioplasties and shunt placements (these could be received as part of the index admission and/or after discharge from the NSU); any further neurosurgical procedures received during index admission.

The 12-month follow-up PSRQ could be completed by a relative/friend/carer if the participant was unable to complete it and referred only to the time since discharge from NSU. Information requested included the following resource use items: overnight stays in a hospital or other healthcare facility (length of stay, ward type, any associated skull/brain operation); healthcare professional visits (professional seen, frequency and most common location); head/brain scans (MRI, CT or ‘other’); time in a care home; help received from a family member/friend or carer.

After assigning unit costs to the resource use items (see online supplemental table S1 for unit costs), the costs associated with both the hospital-recorded data and PSRQ resource use items, excluding wider societal costs (care home and help/carer costs), were summed to estimate the total NHS and PSS cost per participant. For each group, the mean total costs were estimated over the 12-month follow-up period, along with the associated p value for the mean cost difference between groups. An exception to the above was that, to avoid double-counting, patient self-reported postdischarge overnight stays with an associated skull/brain operation would not be costed if the total reported number was less than the total reported number of hospital-recorded postdischarge cranioplasty and shunt procedures (including revisions). In line with previous work,9 the higher of the two values was considered the most accurate. It should also be noted that patients who were known to have died postdischarge (mortality is collected as part of the primary outcome, see below) were not sent the PSRQ. As such, postdischarge costs for these participants would have been treated as missing and estimated via imputation (see below for details of the imputation methods, where time postdischarge was included in the multiple imputation (MI) model). In contrast, cost data for participants who died before discharge from their index admission would not have been considered missing as hospital-recorded data would still have been available for such participants (postdischarge costs were set as equal to zero for such participants).

Measuring outcomes

To estimate health-related quality-of-life and conduct a cost–utility analysis,6 in line with UK National Institute for Health and Care Excellence (NICE) guidance,4 7 the EuroQoL 5-Dimension 5-Level questionnaire (EQ-5D-5L)10 was combined with mortality data to estimate quality-adjusted life-year (QALY)6 scores. Participants completed the EQ-5D-5L at discharge from NSU (assumed baseline score), 6 and 12 months follow-up (if discharged from NSU by these time points). As recommended at the time of analysis,11 the cross-walk mapping function12 was used to convert responses into utility scores (range: −0.594 (worse than death) to 1 (full health)). Participants who died were assigned a utility score of 0 on their date of death (death was collected as part of the hospital-recorded data as it was required for the primary outcome, see below). Utility values were used to estimate QALYs over 12 months, based on the total area under the curve method and linear interpolation.13

For ease of interpretation, as is convention,14 the trial primary outcome measure, the (ordinal) extended Glasgow Outcome Scale (GOSE),15–18 at 12 months, was converted into a binary scale using a fixed dichotomy analysis (ie, favourable vs unfavourable)5 to enable a cost-effectiveness analysis6 to be undertaken. Favourable outcomes were defined as upper severe disability or better while unfavourable outcomes included death, vegetative state and lower severe disability. A sliding dichotomy analysis5 was also undertaken and is described in online supplemental appendix 2.

Missing data

Missing data are common in randomised trials and can lead to bias and lack of precision.19 As recommended, patterns of missing data were examined to explore the mechanism of missingness.19 Accordingly, MI with chained equations (MICE) under missing at random was used to impute missing costs and outcome data, by treatment group. The ‘mi impute chained’ command (Stata V.17.0 (StataCorp)) was used to create 30 data sets (based on recommendations in relation to the level of missing data19) that were then pooled using Rubin’s rules.20

Due to the way data were collected/different levels of missing data, missing data for costs were imputed for total index admission costs (hospital-recorded data collection), total cranioplasty and shunt costs (hospital-recorded data collection over 12-month trial period) and total postdischarge costs (patient self-report questionnaire data collection at 12 months). These three costs were then combined to estimate total NHS and PSS costs. For outcomes, missing data were imputed for utility scores (EQ-5D-5L) at baseline, 6 and 12 months, Glasgow Coma Scale (GCS) score21 22 at baseline and GOSE score at 12 months. In addition to these costs and outcomes, the MI model also included age (years), sex and time postdischarge (the number of days from discharge to the 12-month point or death).

Incremental analyses

For both the cost-utility and cost-effectiveness analyses, a 12-month within-trial, ITT approach was adopted. In this base-case analysis, patients were analysed according to the treatment to which they were randomised, regardless of the treatment received. No discounting was undertaken.

For the cost–utility analysis, to estimate the mean incremental cost and incremental effect (QALY gain) associated with craniotomy compared with DC, a seemingly unrelated regression (SUR) analysis was undertaken.23 Regressions included those baseline variables expected to be predictive of total costs and outcomes: age (years), sex and baseline utility score. Assuming dominance,6 where an intervention was both more costly and less effective, did not occur the incremental cost-effectiveness ratio (ICER=mean incremental cost/mean incremental QALY),7 for craniotomy versus DC, would be estimated.6 In the UK, NICE refers to a cost-effectiveness threshold of £20 000–£30 000 per QALY.7 As such, if craniotomy had an ICER below this level, this would suggest it is cost-effective, compared with DC. It should be noted that economists have argued that decisions about treatment adoption should be made based on mean estimates, irrespective of whether such differences are statistically significant.24 Therefore, the treatment option which is estimated to be most cost-effective should be provided.25 This approach is consistent with the objective of maximising benefits from a given budget.

For the cost-effectiveness analysis, in terms of the incremental effect, the outcome (based on the GOSE) had a binary scale (favourable/unfavourable) and logistic regression26 was undertaken to estimate the OR (95% CI) of a favourable outcome for craniotomy compared with DC. Separately, the mean incremental cost associated with craniotomy compared with DC was estimated using linear regression. Both regressions included variables age (years) and sex, which were expected to be predictive of total costs and GOSE outcomes. Together, in the absence of dominance, the incremental cost and incremental effect would enable the ICER to be estimated in terms of the cost per percentage increase in the odds of a favourable outcome.

Decision uncertainty

To estimate the level of uncertainty associated with the decision,19 the probability of craniotomy being cost-effective, compared with DC, at a threshold of £20 000/QALY on the cost-effectiveness acceptability curve (CEAC)25 was calculated. This was estimated by combining the mean coefficients and covariance matrix from the SUR model, as described in Faria et al.19 The CEAC was only estimated in relation to the cost–utility analysis.

Sensitivity analyses

The above analysis constituted the base-case analysis6 and was carried out in accordance with a prespecified health economic analysis plan (HEAP) (see: https://www.rescueasdh.org/trial-documents). To assess the robustness of conclusions, sensitivity analyses (SA) were undertaken.6 To analyse the data from a wider cost perspective the care home and carer costs (which were excluded from the base-case analysis) were added to the total NHS and PSS costs (SA wider cost perspective). A further sensitivity analysis (for the cost-utility analysis only) tested the use of the EQ-5D-5L score at discharge from NSU as the baseline for QALY calculations. As any benefits could already have been partially/wholly achieved by discharge, QALY scores were re-estimated with the assumption that given the grave nature of the condition and following expert advice, participants had the lowest possible EQ-5D-5L score at baseline (date of index surgery): −0.594 (SA lowest EQ-5D-5L baseline score). Four further sensitivity analyses (including a per-protocol analysis) were conducted and are presented in online supplemental appendix 3. ‘SA wider cost perspective’ deviated from the HEAP, for reasons explained in online supplemental appendix 4.

Patient and public involvement

The aforementioned patient self-report questionnaire was developed in consultation with non-trial patients.

Participants

Between September 2014 and April 2019, 248 UK patients were recruited, 126 in the craniotomy arm and 122 in the DC arm. Compared with the 450 patients recruited to the full (international) trial (the baseline characteristics of which are summarised in table 1 of Hutchinson et al
5), these UK patients are slightly older (3.5 years on average) and more likely to be on antithrombotic medication (table 1).

Levels of missing data were slightly lower in the craniotomy group compared with the DC group for cost variables and outcome variables (except at baseline) (online supplemental table S2).

Costs

Levels of resource use by intervention arm are summarised in table 2, under three main categories: (1) hospital-recorded index admission; (2) hospital-recorded cranioplasties and shunts and (3) patient-reported (PSRQ) postdischarge.

The hospital-recorded index admission data show that the length of stay in ICU and NSU was slightly lower in the craniotomy group compared with the DC group, but not significantly so. Only small numbers of other neurosurgical operations were reported. With regard to cranioplasties and shunts (index admission and postdischarge), as expected, more patients in the DC group had cranioplasties than in the craniotomy group (DC is prerequisite to a cranioplasty). There were, however, patients who were randomised to craniotomy that went on to have a DC, 21 of which had a cranioplasty in the 12-month follow-up period. Most cranioplasties use a synthetic material. Shunts were uncommon and occurred at a similar frequency between the groups. In terms of the patient-reported (PSRQ) postdischarge resource use, there were no significant differences between the groups for any of the parameters measured.

Mean cost estimates are summarised in table 3 and divided into the same three main categories. As expected, given the procedure complexity and recovery time, total NHS and PSS costs are high in both groups. High index admission costs particularly accounted for this, largely due to the high cost of ICU stays, along with postdischarge costs, largely due to the high cost of overnight stays in a rehabilitation unit. There were, however, few significant differences between groups, the only notable one being the cost of cranioplasty procedures which, for the aforementioned reasons, was significantly higher in the DC group. As the number of postdischarge hospital-recorded cranioplasty/shunt procedures exceeded patient-reported overnight stays with an associated skull/brain operation (table 2), the latter has not been costed.

Outcomes

Outcomes are summarised in table 4. Follow-up mean EQ-5D-5L scores were higher in the craniotomy group compared with the DC group, significantly so at 12 months. Furthermore, the change (increase) in EQ-5D-5L score from baseline was significantly higher at both 6 and 12 months in the craniotomy group compared with the DC group. There was no significant difference between groups for the total QALY score, based on available data.

At 12 months, the percentage of favourable GOSE scores was higher, but not significantly, in the craniotomy group compared with the DC group.

Analyses

Main findings

In this UK population of patients with traumatic ASDH that warrants surgical evacuation, based on the results of the cost-utility and cost-effectiveness analyses, craniotomy dominated DC as it was estimated to have a lower mean cost, a higher mean QALY gain/higher probability of a more favourable outcome on the GOSE. Craniotomy was, therefore, estimated to be cost-effective, on the basis that the associated level of significance is considered to be irrelevant.24 25 In the cost-utility analysis (QALY outcome), there was only an estimated 13% probability (at a threshold of £20 000/QALY) of making the wrong decision by choosing craniotomy. The results of the sensitivity analyses are in keeping with this result.

Within this study, it is important to highlight that costs were estimated from the viewpoint of the UK NHS and PSS and that associated resource use and outcome data were based only on participants from UK sites. As, for example, unit costs may differ outside the UK it is important to note that it is unclear whether these results are generalisable to sites outside the UK. Further associated research may, therefore, be warranted in relation to this and that ≥20% of patients who were randomised to craniotomy went on to have a DC (as an ITT approach was adopted these patients were included in the craniotomy arm in the base-case analysis).

Strengths and limitations

Regarding health-related quality of life, QALY scores (EQ-5D-5L recorded at all time points) were available for 53% of participants only and the amount of missing data was greater at discharge than at 6 and 12 months (online supplemental table S2). Some missing EQ-5D-5L baseline (NSU discharge) data may be due to participants being discharged at short notice or at the weekend when a research nurse was not available. As some patients had not yet been discharged from the hospital by 6 months, this may explain the higher rates of EQ-5D-5L missing data at this time point compared with 12 months. Postdischarge costs (based on patient self-report data) were also missing for 27.4% of patients at 12 months (online supplemental table S2). Such missing data are a limitation, but we did impute missing data and take an ITT approach, which meant that all patients were still included in the analysis.

A further limitation is that, for ethical reasons, baseline EQ-5D-5L scores were taken at discharge from NSU, rather than at randomisation. Therefore, any benefits could be underestimated by assuming this score is the baseline score. To test the potential impact of this, a sensitivity analysis (SA lowest EQ-5D-5L baseline score) assumed the baseline EQ-5D-5L score to be that of the worst possible health state (−0.594). The results differ little from the base case (table 5) with craniotomy still dominating DC. It should also be noted that, in the cost-effectiveness analysis, as the cost and outcome regressions are performed separately any correlation between the cost and outcome variables would not be accounted for. A final limitation is that the 12-month follow-up period may not be sufficient to capture all expected cranioplasties. For example, of those randomised to DC (122), only 62 had received a cranioplasty within the 12 month follow-up period. As such, further cranioplasties (aside from those who were randomised but did not receive DC (8/122) and those who had died (31/122)) could take place beyond the 12-month period. Though this is a limitation, the inclusion of such costs would only be expected to increase the long-term incremental cost of DC, and therefore, not change the conclusion that craniotomy dominated DC.

Comparisons with other studies

We are not aware of any previous economic evaluations that have specifically compared craniotomy with DC for patients with ASDH. Previous economic evaluations of DC have been undertaken,27–31 but these have had different comparators and used a variety of different populations/methods (most developed a decision analytical model to estimate costs and benefits,29–31 and the two papers27 28 that used actual patient data were not based on randomised data and were of a smaller sample size than used here, with different cost perspectives and timeframes). Thus, it is difficult to make direct comparisons to our study, and the use of different methods may explain why there were differences in the results as to whether DC was estimated to be cost-effective or not.27–31

Implications

In a UK population of patients with traumatic ASDH, craniotomy was estimated to have a lower mean cost, a higher mean QALY gain and a higher probability of a more favourable outcome on the GOSE, dominating DC. Based on the QALY, there was a high probability that craniotomy, compared with DC, was cost-effective (at a threshold of £20 000/QALY). When sensitivity analyses were conducted, the main conclusion (that craniotomy was, therefore, estimated to be cost-effective) remained unchanged. Consequently, the health economic analysis supports the recommendation, based on the primary outcome,5 that a craniotomy should be undertaken, rather than a DC, if it is operatively feasible to replace the bone flap.