Economic evaluation results

Model analysis and model population characteristics

The cost-effectiveness analysis has been published in BMJ Global Health.41 Parts of this chapter have been reproduced from Williams et al.41 in accordance with © Williams et al.41 [or their employer(s)] 2020. [Re-use permitted under CC BY. Published by BMJ. https://creativecommons.org/licenses/by/4.0/. This is an open access article distributed in accordance with the Creative Commons Attribution 4.0 Unported (CC BY 4.0) license, which permits others to copy, redistribute, remix, transform and build upon this work for any purpose, provided the original work is properly cited, a link to the licence is given, and indication of whether changes were made. See: https://creativecommons.org/licenses/by/4.0/. The text below includes minor additions and formatting changes to the original text.]

The economic analysis assessed the cost-effectiveness of treating TBI patients with TXA and without TXA, as per the trial treatment arms. The following health economic section has been reported to meet the criteria of the CHEERS (Consolidated Health Economic Evaluation Reporting Standards) checklist.42

As stated above, the trial included patients treated within 3 hours of their injury with either a GCS score of ≤ 12 or any intracranial bleeding on their CT scan, and without extracranial bleeding. The trial found that TXA reduced head injury deaths among those with TBI, with a RR of 0.94 (95% CI 0.86 to 1.02). However, there was evidence that people with mild or moderate TBI (baseline GCS score of 9–15) had a greater benefit from TXA treatment, in terms of reduction in head injury death (RR 0.78, 95% CI 0.64 to 0.95), than those with a severe head injury (GCS score of 3–8; RR 0.99, 95% CI 0.91 to 1.07). For this reason, the mild and moderate population was used as the base-case population, excluding those with severe head injury. In addition to considering the cost-effectiveness of TXA based on baseline GCS score, we also evaluated the cost-effectiveness of TXA for an alternative subgroup of patients: those with head injury of any severity with both pupils reactive (RR 0.87, 95% CI 0.77 to 0.98), based on the clinical results presented in Figure 3. We excluded those with either pupil unreactive (RR 1.03, 95% CI 0.94 to 1.13), as there was no evidence of a reduction in head injury deaths for this subgroup.

The model was analysed over a lifetime time horizon with costs presented in Great British pounds, and outcomes presented as life-years (LYs) and quality-adjusted life-years (QALYs). The analysis was performed from a UK NHS and personal social services perspective. The model estimates the incremental cost-effectiveness ratio (ICER) by dividing the incremental costs of TXA by the incremental health outcomes associated with TXA treatment, to give a cost per LY or QALY gained. We used the lower bound of the £20,000 to £30,000 per QALY cost-effectiveness threshold stated by NICE to estimate the cost-effectiveness of TXA.40 If the ICER falls below the cost-effectiveness threshold, then that intervention can be considered cost-effective. Both costs and outcomes were discounted at a rate of 3.5%, in accordance with NICE guidelines,40 to capture the higher value of current costs and outcomes compared with those occurring in the future. The mean age of individuals entering the model was derived directly from the CRASH-3 trial (41.7 years for patients with mild and moderate injury and 41.6 years for those with both pupils reactive). Deterministic sensitivity analyses were performed, in which alternative discount rates (0% and 6%) were evaluated. The cost-effectiveness model was developed in Microsoft Excel^® (Microsoft Corporation, Redmond, WA, USA), with the analysis of trial data performed in Stata^® 16 (StataCorp LP, College Station, TX, USA).

Model structure

A Markov model captured the long-term outcomes associated with head injury, and is shown in Figure 6. It consists of two health states, alive and dead, and includes the risk of death during the first 28 days of the trial from both head injuries and non-head injuries along with estimates of longer-term mortality. The model uses a daily cycle length for the first year, to allow the events during the trial period to be accurately modelled, followed by an annual cycle length thereafter.

FIGURE 6 Model structure. (Figure not reproduced here; see the original article.)

Clinical outcomes

The 28-day risk of head injury and non-head injury death for the placebo group were derived from the CRASH-3 trial, with the risk in high-income countries used to estimate the risk in the UK. A RR of head injury death was applied for patients receiving TXA, as derived directly from the CRASH-3 trial. The risk of non-head injury death was equal for placebo and TXA groups in the model, based on the CRASH-3 trial. The risk of head injury and non-head injury death within the 28-day follow-up period, and the head injury rate ratio associated with TXA, are presented for the mild and moderate CRASH-3 population in Table 6 and for patients with both pupils reactive in Table 7.

Following the 28-day trial follow-up period, the risk of death was assumed equal for people treated with and people treated without TXA. Standardised mortality ratios (SMRs) were used to account for the higher risk of death post TBI compared with the general population. SMRs were derived from a Scottish study that included a variety of head injury severities. It estimated a SMR of 4 for the first year following injury, and 2.26 thereafter, compared with a group of matched community controls.43 These SMRs were applied relative to age-based, UK general population mortality estimates, and were assumed to be the same for those with mild or moderate TBI and those with both pupils reactive.44 It was assumed that the additional long-term risk of death continued throughout the duration of the model; however, a sensitivity analysis that excluded this long-term risk of death was performed, to assess the impact of this parameter.

Health status, utility and quality-adjusted life-years

In the CRASH-3 trial, there was little difference between the DRS scores for each treatment arm reported for those with mild or moderate TBI [TXA 3.12 (SD 5.6) vs. placebo 2.91 (SD 5.1), with lower scores representing better outcomes] and those with both pupils reactive [TXA 4.38 (SD 7) vs. placebo 4.33 (SD 6.9)].

To capture the quality of life for patients post TBI, utility values for the ‘alive’ health state were derived from a systematic review and EuroQol-5 Dimensions utility mapping study, which identified five studies reporting utility values stratified by the severity of TBI outcomes.45 This mapping studies then estimated utility by Glasgow Outcome Scale (GOS) outcomes, using a UK value set.45 The utilities associated with each GOS outcome is shown in Table 8.45 In our analysis, we estimated the overall utility by estimating the corresponding GOS outcome for each patient by using the DRS outcomes reported in the CRASH-3 trial. The mapping of each DRS outcome to the GOS outcome is presented in Appendix 6, Table 19. Once this mapping was performed, a weighted average of GOS outcomes was used to estimate the average utility for each population. The average utility was 0.74 for the mild and moderate population and 0.70 for those with both pupils reactive (Table 9). It was assumed that individuals who died within the 28-day study period had a utility of 0 between their injury and death.

Owing to the uncertainty around the utility estimates used in the base-case analysis, three sensitivity analyses were performed to consider the impact of alternative utility values on the cost-effectiveness of TXA. First, a sensitivity analysis was performed in which the DRS scores for those receiving TXA and placebo were modelled independently, and independently mapped to utility scores, which resulted in a marginally lower utility among those receiving TXA. This resulted in utility values of 0.74 for TXA and 0.75 for placebo in the mild and moderate group, and 0.69 for TXA and 0.7 for placebo for patients with both pupils reactive. Second, a sensitivity analysis considered an alternative method to estimate GOS outcomes among CRASH-3 patients, using a previous study reporting the correlation between GCS score at injury and GOS outcomes (see Appendix 6, Table 20, for additional details).45,46 This allowed for the distribution of GCS scores for patients in the CRASH-3 trial to be used to estimate the distribution of GOS outcomes, to which utility values could be applied. This produced higher estimated utilities of 0.79 for the mild and moderate population and 0.76 for patients with both pupils reactive (see Appendix 6, Table 21). Last, a sensitivity analysis considered the impact of a lower utility value, of 0.63, for both treatment groups, and for both model populations (patients with mild and moderate injury and patients with both pupils reactive), to assess the impact of a lower utility estimate on cost-effectiveness. This was an average estimate derived from a Swiss study of trauma patients reporting utility values in mild (0.7) and moderate (0.56) TBI patients, with GCS score of 9–15, and with an abbreviated injury score of 0–2, representing mild or no TBI.47

Age-based utility estimates for the UK general population were used to account for the decline in utility with age (Table 10).48 The post-TBI utility estimates (see Table 9) were derived from a cohort with a median age of 50 years. Therefore, a utility decrement model population for those reaching the age of ≥ 55 years.

For example, the average utility for patients with mild or moderate TBI after discharge would be 0.75 until they reach 55 years, when the utility would decrease to 0.70 (0.75 minus 0.05). The utility would then decrease to 0.68 at 65 years (0.75 minus 0.07). The utility estimates were not inflated between 42 (average starting age in the model) and 44 years, to remain conservative.

Costs

Treatment costs

The model captured the costs of TXA treatment, including the cost of TXA, needle and syringe, and nurse administration time, which were applied to the TXA intervention only (see Table 14). The total cost of TXA was derived from the British National Formulary49 (£6 per person), as were the costs of the infusion bags (£3.25 for a 100-ml and 500-ml bag). The costs of needles and syringes were derived from a NICE costing template for the UK.50

The nurse time required to administer TXA was assumed to be 21 minutes (as per the CRASH-218 economic analysis), and the hourly cost of a nurse was derived from UK social service costs, based on the hourly cost of a band 5, hospital-based NHS nurse.51,52 This gave a total cost of £22.25 for treatment, equipment and treatment administration.

Hospital costs

There was little difference in hospital length of stay for those treated with and those treated without TXA in high-income countries [TXA 14 days (SD 9.8 days), placebo 13.3 days (SD 9.3 days), overall 13.7 days (SD 9.6 days)], and, therefore, this was assumed to be the same for both arms. A sensitivity analysis was performed to assess this assumption, in which the trial data for hospital length of stay were modelled specifically for each treatment arm. Inpatient hospital costs were derived from UK NHS reference costs,53 using the cost associated with head injury admissions. A weighted average of all head injury admission costs was calculated, based on the severity of the head injury (case mix adjusted). As the length of stay was assumed to be the same for those treated with TXA and those treated without TXA, hospital costs did not affect the incremental cost-effectiveness, except in the sensitivity analysis to assess this assumption.

Monitoring costs

Patients were assumed to incur additional health-care resources post discharge. These long-term monitoring costs include the increased use of health services, such as outpatient clinic visits and more frequent visits to GPs. It also includes rehabilitation and physiotherapy, and community care, such as formal carers. These costs were assumed to differ between the first year post injury and after 12 months. First-year monitoring costs were derived from a UK costing study,54 for those with good recovery, moderate disability and severe disability. These costs were mapped from DRS scores (using the same method described above to map from DRS to GOS outcomes) to estimate the average annual monitoring costs (Table 11). These costs have also been used in a previous HTA analysis.55

The average first-year monitoring cost was estimated to be £11,662 for those with mild or moderate head injury and £14,259 for those with both pupils reactive.54 Long-term monitoring costs (applied after the first year post injury) were estimated by expert opinion in a previous HTA.55 The average cost was £2505 per year for patients with mild or moderate TBI and £3405 for patients with both pupils reactive, and was assumed to be incurred until the patient died. We explored the impact of excluding monitoring costs beyond the first year post injury and applying monitoring costs until 5 years post injury in sensitivity analyses, owing to the uncertainty in these estimates.

The average monitoring costs for the UK were estimated by combining the annual cost by GOS status (see Table 11) with the proportion of patients across each GOS outcome (see Appendix 6, Table 19). A weighted average was used to provide the average annual monitoring cost for each population, as displayed in Table 12.

All costs for the mild and moderate TBI population are shown in Table 13, and for the both pupils reactive population in Table 14. All costs were inflated to 2018 prices using a UK hospitals and community service index.52

Sensitivity analyses

The main analysis was performed using probabilistic sensitivity analyses to simultaneously capture the uncertainty in model parameters. Distributions were assigned to each probabilistic parameter, with each sampled simultaneously across 1000 Monte Carlo simulations. One-way deterministic sensitivity analyses were also performed to assess the sensitivity of specific parameters on the cost-effectiveness estimates, and are presented relative to the base case as a tornado diagram.

Primary analysis of base-case incremental costs, quality-adjusted life-years and incremental cost-effectiveness ratio: mild and moderate traumatic brain injury patients

The costs, LYs and QALYs associated with TXA treatment and without TXA treatment are presented in Table 15. In the base-case analysis, TXA is highly cost-effective in the UK for those with mild and those with moderate TBI, at £4288 per QALY gained. When considering LYs only, the ICER was £3078 per LY gained.

The cost of purchasing and administering TXA represented a very small proportion of the incremental costs (3%), with long-term monitoring costs contributing to most of the incremental costs for the TXA group (97%). These higher costs are due to a higher proportion of patients surviving when given TXA, as monitoring costs per person were the same in both treatment groups.

The long-term survival projections of the model for patients with mild or moderate TBI receiving TXA or placebo are presented in Appendix 6, Figures 13 and 14.

Sensitivity analyses of base-case population: mild and moderate traumatic brain injury

Probabilistic sensivity analysis

Tranexamic acid was highly likely to be cost-effective in the probabilistic sensitivity analysis (PSA), with a 99% probability of being cost-effective at the NICE £20,000 per QALY willingness-to-pay threshold (Figure 7).

FIGURE 7 Cost-effectiveness acceptability curve for TXA for patients with mild or moderate TBI. (Figure not reproduced here; see the original article.)

Deterministic sensitivity analysis

A number of sensitivity analyses were performed, but none increased the ICER above the cost-effectiveness threshold, meaning that TXA remained cost-effective in all deterministic sensitivity analyses (Figure 8). Assuming a lower utility among those receiving TXA than among those receiving placebo increased the ICER the most, to £14,465 per QALY. Restricting monitoring costs to only the first year or first 5 years post injury reduces the ICER to £979 and £1646 per QALY, respectively. When considering a longer length of hospital stay for those receiving TXA than for those receiving placebo, the ICER increased to £5567, whereas assuming a lower utility (0.63 for both arms) increased the ICER to £5112 per QALY. TXA remained cost-effective when the RR increased to 0.95, with the ICER increasing to £4721 per QALY. The discount rate, and excluding excess mortality after the trial period, had little impact on the ICER.

FIGURE 8 Tornado diagram showing deterministic sensitivity analyses and the impact on the ICER per QALY gained, for those with mild or moderate TBI. (Figure not reproduced here; see the original article.)

Analyses for patients with both pupils reactive: incremental costs, quality-adjusted life-years and incremental cost-effectiveness ratio

Deterministic results

When considering individuals with both pupils reactive, treatment remained highly cost-effective with an ICER of £6097 per QALY in the UK. When considering LYs only, the ICER was £4066 per LY gained (Table 16).

Probabilistic sensitivity analysis

At the UK cost-effectiveness threshold of £20,000 per QALY, TXA is 99% likely to be cost-effective in the PSA (Figure 9).

FIGURE 9 Cost-effectiveness acceptability curve for TXA treatment for patients with both pupils reactive. (Figure not reproduced here; see the original article.)

Deterministic sensitivity analysis

The deterministic results for patients with both pupils reactive show that, for all sensitivity analyses in the UK, TXA remained highly cost-effective (Figure 10). Similarly to the results for the mild and moderate TBI population, a reduction in monitoring costs being applied for either 1 year only or 5 years reduced the ICER to £1257 and £2233 per QALY, respectively. Assuming a lower utility among those receiving TXA than among those receiving placebo increased the ICER to £9512 per QALY. When assuming a longer length of hospital stay for TXA based on the CRASH-3 trial, the ICER increased to £7804. At a head injury treatment effect rate ratio of 0.98 (representing the upper 95% CI), TXA remained cost-effective, with the ICER increasing to £6949 per QALY. The ICER also increased when considering a lower utility (0.63), to £6753 per QALY. The discount rate, and excluding excess mortality after the trial period, had little impact on the ICER.

FIGURE 10 Tornado diagram showing deterministic sensitivity analyses and the impact on the ICER per QALY gained for patients with both pupils reactive. (Figure not reproduced here; see the original article.)