A systematic review of economic evaluations of advanced therapy medicinal products

Aims Advanced therapy medicinal products (ATMPs) represent a new category of medicinal products with a potential for transformative improvements in health outcomes but at exceptionally high prices. Routine adoption of ATMPs requires robust evidence of their cost‐effectiveness. Methods A systematic literature review of economic evaluations of ATMPs, including gene therapies, somatic cell therapies and tissue‐engineered products, was conducted. Literature was searched using MedLine, Embase, PubMed, Cochrane Register, the NHS Economic Evaluation Database and the grey literature of health technology assessment organisations with search terms relating to ATMPs and economic evaluations. Titles were screened independently by 2 reviewers. Articles deemed to meet the inclusion criteria were screened independently on abstract, and full texts reviewed. Study findings were appraised critically. Results 4514 articles were identified, of which 23 met the inclusion criteria. There was some evidence supporting the cost‐effectiveness of: chimeric antigen receptor T‐cell therapy axicabtagene–ciloleucel (Yescarta), embryonic neural stem cells, tumour infiltrating lymphocytes, in vitro expanded myoblast, autologous chondrocyte implantation, ex vivo gene therapy (Strimvelis) and voretigene neparvovec (Luxturna). However, estimates of cost‐effectiveness were associated with significant uncertainty and high likelihood of bias, resulting from largely unknown long‐term outcomes, a paucity of evidence on health state utilities and extensive modelling assumptions. Conclusion There are critical limitations to the economic evidence for ATMPs, most notably in relation to evidence on the durability of treatment effect, and the reliability of opinion‐based assumptions necessary when evidence is absent.

is frequently referred to as regenerative medicine. The number of ATMPs being approved is rising 3 and, given their high cost, there is a pressing need for robust economic evidence of these therapies in order to inform decisions made by healthcare payers.
ATMPs pose specific challenges in evidence generation, health technology assessment (HTA) and financing. 4 A key feature of ATMPs is their price, which can in some instances exceed £1m per patient. Such high (often up-front) costs make ATMPs particularly problematic in terms of meeting usual thresholds of cost-effectiveness and being affordable to healthcare payers.
Moreover, there may be methodological challenges, such as in relation to uncertainty in the evidence of the effectiveness of newly approved ATMPs; the nature of the distribution of costs in relation to the accrual of benefits, and how these are affected by choice of discount rates; whether curative treatments may be considered differently to treatments that create smaller incremental benefits; and consideration of value attributes that may not be captured adequately in the quality-adjusted life year (QALY).
The National Institute for Health and Care Excellence (NICE) in the UK suggested that a completely new reference case is not needed. Their mock economic evaluation of a chimeric antigen receptor (CAR) T-cell therapy accepted existing methods of economic evaluation as being fit for purpose in the evaluation of ATMPs. 5 More recently, the independent US-based Institute for Clinical and Economic Review following a review in collaboration with NICE and the Canadian Agency for Drugs and Technologies in Health, published adaptations to its value assessment framework for potential cures and other treatments that qualify as high-impact single or short-term therapies. 6     Only full economic evaluations were included (i.e. cost effectiveness, cost utility or cost benefit analyses). Partial economic evaluations (e.g. cost minimisation or cost consequence analyses) were excluded, as were studies only reporting the burden of disease or cost of illness.

| Eligibility criteria/study selection
We excluded editorials, letters, historical articles, discussion or commentary articles, and evaluations published only as abstracts.

| Data extraction
Identified articles were screened by 2 reviewers independently according to the exclusion and inclusion criteria; first by title, followed by abstract, and finally by full article text. Any discrepancies were resolved in discussion with the third reviewer. Extracted data included year and country of publication, clinical indication, ATMP and comparator, method of economic evaluation, time horizon, total intervention and comparator costs, QALY gain, incremental cost-effectiveness ratios (ICERs), results of sensitivity analyses, principal study findings, issues of generalisability, study limitations and key methodological challenges as reported by the authors of each study.

| Quality of reporting assessment
Articles were assessed for their quality of reporting by their compliance with the Consolidated Health Economic Evaluation Reporting Standards. 10 Studies were scored against each of the 24 checklist items according to whether reporting fully satisfied or did not satisfy the item requirements. The overall quality of reporting was presented as a percentage score of applicable items. Studies scoring above an arbitrary threshold of 75% were considered to be of higher reporting quality. The quality of reporting of individual items from the checklist is expanded further in the narrative.

| Narrative synthesis
A narrative synthesis of the methodological challenges associated with economic evaluations of ATMPs was carried out following the methods of Nagpal et al. (2019), 11 and based on the information extracted and judgements made on study quality. This approach synthesises findings from multiple studies and uses the words and text from these studies to produce a summary and explanation of the findings therein.

| Search results
In total, 4514 studies were identified following the initial search.
Removal of duplicates resulted in 3358 potentially relevant articles.

| Gene therapy medicines
The cost effectiveness of Strimvelis was examined in 2 analyses, of which 1 was deemed to be cost effective. South   Main purpose was to report the potential cost-effectiveness of CAR T-cell therapy; and to highlight key uncertainties surrounding these results.

| Quality of reporting
Not reported. This exercise was conducted on theoretical data and assumed costs, and may not capture the problems associated with real-world data.
Although evidence about ATMPs is expected to be associated with uncertainty in determining the long-term costs and benefits to patients and the NHS, existing methods available to estimate the implications of this uncertainty are sufficient. Challenges include: the potential curative nature and claims of long-term/lifetime benefits; the potentially rapid changes that may arise in product characteristics over time; potential longer-term patient safety issues because of persistence; organisational and scaling issues; and the potentially significant upfront costs that may arise. 12 Long-term cost savings in most instances in early onset Parkinson's disease patients in HY stages III-IV.
The model was based on the Swedish health care system, but devised to be applicable to available data on treatment costs and health state utilities for different HY stages. Such data are now available from a variety of countries.

Small number of patient-level data; clinical effectiveness data based on open-label transplantation trials
The frequent use of placebo as a comparator, together with the extra attention given to randomised control trial patients may contribute to nonrepresentative outcomes. Use of real-life observations claimed to be less restricting to allow hypothetical comparisons between standard therapy and a range of different alternatives. 13 Tumour infiltrating lymphocytes are expected to generate more QALYs than its comparator at a lower cost and so dominates.
The prices of treatments vary substantially between countries. This reduces the generalisability of the results.
No clinical trial data available and therefore data on the effectiveness of tumour infiltrating lymphocytes had to be drawn from various sources. Longer-term data are required to confirm whether axicabtageneciloleucel is a curative treatment. 16 The intervention produced an ICER of £49 975 per QALY gained when compared to chemotherapy regimen gen-ox, which is under the NICE £50 000 threshold Not reported. Haematological malignancy research network data were used to estimate overall survival for chemotherapy patients meaning that a naïve indirect comparison was used as the basis of the estimation of clinical outcomes in the economic model.
An assumption was made that that patients who were alive at 24 months were effectively cured. 17 A potential for long-term cost savings by reducing the disability after stroke; societal value up to US $166 500 (US $184 567), particularly in younger patients with stroke with moderate disability, with possible cost effectiveness estimated down to relative efficacy of 14%.

Enables cost-benefit analysis for patients with stroke under a wide range of assumptions
Effectiveness of SCT was based on expert opinion; did not include differential costs of early vs late administration poststroke; limited standard care data reflecting survival, treatment patterns, and transition probabilities for mRS.
Ideally health economic analyses are based on long-term data. If not available, and for most treatments only short-term data are available, disease modelling provides a way of estimating long term effects. 18 A potential to achieve a level of cost effectiveness that is acceptable to policymakers and health care purchasers, but is largely determined by the interpretation of available clinical effectiveness data and the duration over which such effects may be observed.
The focus of the analysis was on the potential cost effectiveness of autologous HSCT in the management of secondary progressive multiple sclerosis only.
The absence of direct randomised controlled trial evidence to input into the model.
Modelling cannot be considered a substitute for good quality clinical trial evidence. 19 Total cost for tisagenlecleucel was double that of clofarabine, while the gains in QALYs of tisagenlecleucel was 4× that of clofarabine. The probability of cost-effectiveness at $50 000 per QALY was about 0.7.
Cost perspective specific to US payer which may not be generalisable to other settings.
This analysis was limited primarily by the lack of comparative evidence available for these therapies. Evidence on long-term effectiveness is still unknown, which resulted in assumptions being made related to trial survival curve extrapolation and the time point at which long-term survivors would be considered effectively cured.
The authors closely followed the methodology used in the 'curative intent' mock evaluation of CAR T-cell therapy. 5 The differences in estimates between the 2 models are probably due to the use of 2 different approaches to curve extrapolation. 20 IVM is dominated by MUS treatment but as costs of cell expansion are likely to reduce in the future this may reduce the cost of the IVM procedure.
Using QALYs based on the same multiattribute health status classification system internationally would aid generalisability.
Lack of uniform reporting tools to define outcome of stress urinary incontinence interventions. When robust evidence was not available, Concerns about the sensitivity of generic multiattribute health outcomes measures in the context of urinary incontinence. (Continues)

Study findings Generalisability Limitations
Key methodological difficulties the estimates relied on expert opinions. 21 The cost-effectiveness is probably between $37 000 and $78 000 per QALY gained over a patient's lifetime horizon.
Not reported. Lack of evidence for the comparator, which affects the calculation of the ICER. Due to limited follow up, assumptions had to be made about long-term survival and when a patient is effectively cured.
Flattening in the tail of the survival curves was observed for both tisagenlecleucel and clofarabine. Standard parametric models probably underestimate survival when flattening in the tail exists; therefore, they used a flexible parametric model to account for this flattening. 22 Reduction of the price of tisagenlecleucel to $200 000 or $350 000 would allow it to meet a $100 000 or $150 000 per QALY willingness-to-pay threshold in all scenarios.

Not reported
No high-quality long-term clinical outcomes data exist for tisagenlecleucel The authors addressed the main limitation by modelling multiple long-term effectiveness scenarios, including 1 where all patients eventually experience relapse. 23 The likelihood that axicabtagene-ciloleucel is cost-effective was 95% at a willingness to pay of $100 000 per QALY.

Not reported
The current data of the ZUMA-1 trial are limited at a median follow up of 15.4 months.
As this analysis used axicabtagene-ciloleucel 1-year follow-up data, the authors find it prudent to re-examine cost effectiveness after additional follow-up. 24 For all measures, a 1-point increase in clinical scores had lower costs for microfracture than for ACI at 5 years.

Tissue engineered medicine
Unit prices came from a single orthopaedic hospital, which may limit the generalisability of the findings.
Small study population leading to bias. MF group had slightly smaller lesions meaning that they are more responsive to physiotherapy.
Clinical uncertainty limits robustness of economic analysis. 25 ChondroCelect shown to be a cost-effective strategy compared with microfracture and the ICER is below the NICE threshold.
Not reported. Absence of firm data on the probability and time to occurrence of osteoarthritis TKR. Therefore, a Markov model was not possible.
When the need for TKR increases, ICER expected to decrease in favour of ChondroCelect. Due to higher discount rates for costs rather than effects, the procedure resulting in more TKR patients would also generate more QALYs. However, for the patient the optimal treatment is 1 that minimises pain and discomfort and avoids the need for TKR. Long-term data are needed to characterise specific events. 26 The headroom for tissue-engineered bladder was estimated at around Not reported. Not reported. The headroom method is claimed to inform decisions without the T A B L E 3 (Continued)

Reference Study findings Generalisability Limitations
Key methodological difficulties £16 268. However, the market size is limited reducing potential profitability.
need for complex modelling, which may have very wide parameter uncertainty. In the case of a technology yet to be developed, or in early stages of development, the very nature of the product is uncertain, leading to difficulties in its economic evaluation; although the method proposed is a simple cost utility analysis. 27 If the decision-maker is willing to pay £20 000 for a QALY, ACI is 56-59% more likely to be cost-effective than microfracture.
Not reported. The length of clinical trial follow-up was too short and hence, there are no long-term data on the success and failure rates.
Because of the paucity of data from clinical studies, transition probabilities were not available for each transition in the model.
There is a clear lack of evidence on health state utility values for patients that have had cartilage defects of the knee. 29 IMPACT can be dominant to ACI over a 5-year horizon in terms of cost effectiveness All costs were derived from the hospital administration data and/or from other Dutch data resources, which may limit its transferability to other settings.
Patients included in these models, who reflect randomised controlled trial populations, are not always typical of patients seen in orthopaedic sports practice.
Included only a small number of patients from a randomised controlled trial with a follow-up of 5 years. Greater patient numbers and a longer follow-up period will make such an early analysis more reliable.
Gene therapy 27 Treatment with gene therapy is likely to be cost saving for the treatment of severe haemophilia A compared with the current standard of care with factor VIII prophylaxis.
Age is an important variable in potentially curative treatments. The results are generalisable to different age groups because altering the probability of death, a good approximation for changes in age, did not significantly alter the cost-effectiveness of gene therapy.
The assumption that successful gene therapy results in full quality of life could potentially bias results toward gene therapy. The lack of commercially available gene therapy for haemophilia A. limiting the time frame to 10 years reduces the cost-effectiveness of gene therapy significantly.
The assumption that gene therapy leads to full quality of life could potentially bias the results towards gene therapy. 13 The ICER for Strimvelis is below the £100 000 per QALY cost-effectiveness threshold for highly specialised technologies.
Not reported. Quality of life data had to be collected from the literature.

| Generalisability
The main themes in terms of generalisability relate to costs. Costs of ATMPs obtained from specific hospitals in specific countries, for instance, might limit generalisability to other jurisdictions. 12,13,24,28,29 This may be due to different methods of production, pricing and service delivery in different settings. Other issues of generalisability highlighted in the reviewed studies, include the transferability of results from a US to a UK setting, 19 the importance of age as a variable in potentially curative treatments 30 and using QALYs based on the same multi-attribute health status classification system internationally. 20

| Analysts' resolution of methodological challenges
The main methodological challenge was the lack of clinical data with which to inform any modelling or economic evaluation attempted. 12 29 The narrative overview of the methodological challenges encountered in the identified papers revealed as the principal difficulties, the paucity of trial data to inform economic analysis, the lack of long-term data on outcomes and costs, and dependence on critical and often unsubstantiated assumptions. The clinical evidence was insufficient in many (if not most) instances to support claims that treatment was curative, which has a major bearing on estimates of survival and quality-adjusted life expectancy required for calculating cost-effectiveness.

| Strengths and weaknesses of this review study
The main strength of this review is that it brings together an array of The differential timing in the costs and accrual of benefits associated with ATMPs suggests that time preference, and the choice of discount rate, is likely to be more impactful on their cost-effectiveness compared to many other conventional health technologies. NICE (2017) 31 applied a discount rate of 1.5% per annum for costs and benefits, in accordance with its guidance for treatments that restore people to full or near-full health when they would otherwise die. 37