Axicabtagene ciloleucel compared to standard of care in Swedish patients with large B-cell lymphoma: a cost-effectiveness analysis of the ZUMA-7 trial

Abstract Aim Our study aimed to evaluate the cost-effectiveness of the chimeric antigen receptor (CAR) T-cell therapy, axicabtagene ciloleucel (axi-cel), compared to standard of care (SOC) in Sweden for second-line (2L) treatment of adult transplant-intended diffuse large B-cell lymphoma (DLBCL) patients who relapse within 12 months from completion of, or are refractory to (early r/r), first-line (1L) chemoimmunotherapy. Methods Cost-effectiveness was assessed using a three-state partitioned survival model. Mixture cure models were used to extrapolate time-to-event data from the ZUMA-7 trial (NCT03391466) beyond the observational period. Sensitivity and scenario analyses were performed to test the robustness of the base case results, including an analysis that assumed no switching to off-protocol CAR T-cell therapy in subsequent lines in the SOC arm. Results The model estimated an incremental cost-effectiveness ratio (ICER) of SEK 534,704 (EUR 50,303) per quality-adjusted life year (QALY) gained over a lifetime horizon of 50 years, with an incremental cost of SEK 812,944 (EUR 76,479) and incremental QALY of 1.52 for axi-cel compared with SOC. The probabilistic sensitivity analysis showed that axi-cel was cost-effective in 73% of the simulations when assuming a willingness-to-pay threshold of SEK 1,000,000 (EUR 94,077) per QALY. The ICER was SEK 694,351 (EUR 65,313) in the scenario analysis where the costs and effects of treatment switching were not included. Conclusion 2L treatment with axi-cel in transplant-intended DLBCL patients with early r/r after completing 1L chemoimmunotherapy was cost-effective compared to SOC in a Swedish setting. Administering axi-cel in 2L is cost-effective as it enhances the possibility of curing more patients, resulting in not just a survival advantage, but also a reduction in the burden on quality of life and cost of subsequent therapy. This will be advantageous to both patients and society.


Introduction
Diffuse large B-cell lymphoma (DLBCL) is an aggressive and common subtype of lymphoma with an annual incidence of approximately 700 people in Sweden 1 .The primary treatment in the first-line (1L) setting comprises chemoimmunotherapy consisting of rituximab and anthracycline-based chemotherapy (normally R-CHOP), which cures three in four patients 2 .For the quarter of patients who are not cured with 1L therapy, prognosis is determined by a range of factors, including response to 1L chemoimmunotherapy and timing of relapse 2 .
Patients who relapse within 12 months from completion of, or are refractory to (early r/r), 1L chemoimmunotherapy have particularly poor survival outcomes.A Swedish study of 736 patients with relapsed or refractory (r/r) DLBCL treated during the period 2007-2018 found that patients with primary refractory disease, which was defined as stable or progressive disease as best response to primary therapy (28% of 736), had a median overall survival (OS) of 4.4 months and a two-year OS of 14% 2 .Early relapse, which was defined as relapse within 12 months of primary diagnosis (62% of 736), was strongly associated with the selection of less intensive treatments, and the two-year OS was less than 20% in this patient group, compared to 55% for patients relapsing after more than two years 2 .
For patients with r/r disease after 1L chemoimmunotherapy, the standard of care (SOC) is platinum-based chemoimmunotherapy followed by consolidation with highdose therapy (HDT) and autologous stem cell transplantation (ASCT) for patients who are treated with curative intent 2,3 .However, among patients with early relapse after 1L chemoimmunotherapy, fewer than two-thirds of Swedish patients up to 70 years of age meet these criteria and initiate SOC therapy (i.e. are transplant-intended), and among those who start less than one in three complete ASCT 2 .Overall, 22% of Swedish patients up to 70 years of age with early relapse completed ASCT compared to 57% for those with late relapse 2 .
Axicabtagene ciloleucel (axi-cel) is a CAR T-cell therapy that received market authorization from the European Medicines Agency (EMA) in 2022 indicated for the treatment of adult patients with DLBCL and high-grade B-cell lymphoma with early r/r after completing 1L chemoimmunotherapy.Approval was granted based on results from the pivotal phase 3 trial ZUMA-7 (NCT03391466).The ZUMA-7 trial evaluated the efficacy and safety of axi-cel versus SOC (salvage chemoimmunotherapy with HDT followed by ASCT for responders) in transplant-intended DLBCL patients with early r/r after completing 1L chemoimmunotherapy 4 .At a median follow-up of 24.9 months, the primary analysis found that axicel demonstrated a 60% improvement in centrally assessed event-free survival (EFS) compared to SOC (hazard ratio (HR): 0.40; 95% CI: 0.31, 0.51; p< .0001) 4 .In an interim analysis (including survival data on eight patients that were lost to follow-up at the data cut-off for the primary analysis), the median OS was not reached in the axi-cel group and was 25.7 months in the SOC group (HR: 0.71; 95% CI: 0.52, 0.97) 5 .Off-protocol treatment switching occurred in the ZUMA-7 trial and 56% of patients, i.e. 100 out of the 179 patients in the SOC arm received subsequent cellular therapy 4 .A prespecified sensitivity analysis was performed in ZUMA-7, which adjusted OS for the confounding effect of treatment switching using the rank-preserving structural failure time model (RPSFTM) or the inverse probability of censoring weights (IPCW) method 6 .These analyses showed a greater OS benefit in favour of axi-cel 4 .
Axi-cel is currently recommended and funded in Denmark, Finland and Norway for the treatment of 2L DLBCL, while a decision is pending in Sweden.In third-line or later (3L+) DLBCL, axi-cel is recommended and funded in Norway, Finland and Sweden based on data from the ZUMA-1 trial.5-year follow-up data from the ZUMA-1 trial showed that 43% of patients treated with axi-cel were alive after five years and a 5-year EFS rate of 30.3% (95% CI: 21.5%, 39.6%) was estimated 7 .58% of the patients achieved a complete response and, in these patients, the 5-year OS rate was 64.4% (95% CI: 50.8%, 75.1%) 7 .These findings support axi-cel as a curative treatment option for many DLBCL patients 7 .
Our study is the first study in the Nordic countries to evaluate the cost-effectiveness of CAR T-cell therapy using randomized clinical trial evidence, a gap that has been highlighted by all Nordic HTA bodies 8-11 .In addition, little is known about the cost-effectiveness of treating patients with CAR T-cell therapy when moving to an earlier line of treatment (from 3Lþ to 2L).Therefore, the objective of this study was to evaluate the cost-effectiveness of axi-cel compared to the SOC and the long-term economic implications of introducing axi-cel in Sweden for the treatment of transplant-intended DLBCL patients with early r/r after completing 1L chemoimmunotherapy.

Model overview
A three-state partitioned survival model was developed in Microsoft Excel and populated with clinical data from ZUMA-7 (NCT03391466) to reflect the target patient population of transplant-intended DLBCL patients with early r/r after completing 1L chemoimmunotherapy.The model simulated the proportions of a theoretical cohort transitioning between each of the three mutually exclusive health states (eventfree, post-event and death) over time and assigned corresponding costs and utilities.The patient cohort entered the model in an event-free state.If patients experienced an event, defined as disease progression or initiation of the next line of therapy, they transitioned to the post-event state.To estimate treatment-related and adverse event (AE) costs, sub-states were included in the event-free and post-event health states to reflect whether patients were on or off treatment.Figure 1 illustrates the model structure.The proportion of patients in each health state at a given point in time was determined based on independently fitted parametric extrapolations of centrally-assessed EFS, OS and time-to-next treatment (TTNT), as illustrated in Figure 2.
Based on previous studies investigating mortality related to DLBCL 12,13 , it was assumed that patients who remain event-free after five years are considered long-term responders with resource utilization and utility equivalent to the general population.Furthermore, studies have shown that these patients have a similar or slightly higher mortality rate than the general population 14,15 .Thus, a standardized mortality ratio (SMR) of 1.09 was applied to long-term survivors in the model based on the findings by Maurer et al. 15 .The analysis was conducted from a Swedish healthcare perspective with costs and health effects estimated over a lifetime horizon of 50 years and discounted by 3% per annum.The model had a monthly cycle length and half-cycle corrections were applied to account for within-cycle transitions.Finally, the model estimated total costs, life years (LYs), and quality-adjusted life years (QALYs) as well as the incremental cost-effectiveness ratio (ICER).

Time-to-event outcomes
Time-to-event data from the ZUMA-7 intention to treat (ITT) population (data cut-off date: 18 March 2021) was used to populate efficacy inputs in the model.A prior validation study using long-term data from ZUMA-1 determined that Mixture Cure Models (MCMs) were the most accurate at predicting long-term survival for DLBCL patients treated with axi-cel after at least two prior lines of therapy 16 .Hence, MCMs were used to extrapolate time-to-event data (centrally-assessed EFS, OS, and TTNT) from the ZUMA-7 trial beyond the observational period.Statistical fit and clinical plausibility 17 were used to determine the most appropriate extrapolations for the base case analysis.Independent survival models were fitted to individual patient level data from ZUMA-7 (diagnostic plots can be found in Supplementary Figure S4 and Figure S5).The Kaplan-Meier and the extrapolated survival curves applied in the model base case are illustrated in Figure 3.More details on the extrapolation of time-to-event data can be found in the Supplementary Materials.

Health-related quality of life
Effects on health-related quality of life (HRQoL) was modelled using state-specific utility values derived from EuroQoL fivedimensions five-levels (EQ-5D-5L) instrument responses collected in ZUMA-7 (event-free state) and ZUMA-1 (post-event state) 18,19 .Health state utility values were age-adjusted based on Swedish general population utilities by Burstr€ om et al. 20 who used the UK EQ-5D index tariff by Dolan 21 .

Event-free utilities
In ZUMA-7, patient-reported outcomes (PROs) were routinely collected during the first 24 months after randomization.82% (296 of 359) of patients enrolled in ZUMA-7 had reported at least two EQ-5D-5L instrument responses.EQ-5D-5L responses were cross-walked to EQ-5D-3L index values using the van Hout algorithm 22 .The algorithm uses the UK value set developed by Dolan 21 , in line with TLV standard practice 23 .A mixed model with repeated measures (MMRM) was used to estimate state-specific utilities, accounting for multiple observations per patient.
The utility value for the event-free state was disaggregated into the periods on-treatment and off-treatment.This approach has been used in previous models for other CAR Tcell therapies in the 3Lþ setting 24,25 .The on-treatment utility was applied for the first month for axi-cel and the first three months for the SOC.Patients who were still event-free after five years were reverted to the age-and gender-matched Swedish general population utilities 20 .Utility decrements due to AEs were not applied in the model, as the potential influence of AEs was assumed to be captured by the event-free on-treatment utility values.

Post-event utilities
The collection of PROs post-event was not mandated in the ZUMA-7 trial, and data collection after switching to subsequent therapy did not usually include PRO reports 26 .Although some sites continued to collect PROs after EFS events, these comprised a minority of observations (less than 11% of total PRO observations).Because of these limitations, the post-event health state utility value was informed by the ZUMA-1 event-free utility value 27 .Utility values for patients treated with axi-cel in 3Lþ represent a similar patient group, e.g.due to the high amount of subsequent CAR T-cell therapy use post-event in ZUMA-7.Finally, we made the conservative assumption of not including utility decrements due to subsequent therapy use.

Costs
Treatment specific costs (i.e.costs related to axi-cel treatment and treatment with SOC including subsequent treatments and management of AEs) and state-specific costs (e.g.costs related to monitoring) were included in the model.Table 1 provides an overview of the resource use and unit costs applied in the model.All costs were reported in 2022 Swedish krona (SEK).

Treatment costs
The cost of 2L treatment with axi-cel included the costs of leukapheresis, conditioning chemotherapy, CAR T-cell infusion and hospitalization following infusion in addition to the Implementation method of the partitioned survival model estimating health state occupancy through the disaggregation of sequential event curves (illustrative example).The health state occupancy in the event-free and death health states were derived from the centrally-assessed event-free survival (EFS) curve and overall survival (OS) curve, respectively.The health state occupancy in the post-event health state was calculated as the difference between the OS curve and the EFS curve (the proportion of patients who are still alive but are no longer event-free).The post-event subsequent therapy was determined by the time-to-next therapy (TTNT) curve.
drug acquisition cost for axi-cel.The share of patients in the ZUMA-7 trial who underwent leukapheresis, received conditioning chemotherapy and completed the axi-cel infusion are presented in Table 1.The cost of 2L treatment with SOC included salvage chemoimmunotherapy followed by HDT and ASCT for patients with complete or partial response.In ZUMA-7, salvage chemoimmunotherapy consisted of R-ICE, R-GDP, R-DHAP and R-ESHAP regimens.According to the Swedish clinical guidelines for treatment of DLBCL 3 , R-ESHAP is not used as salvage chemoimmunotherapy in Sweden,  For axi-cel, the pharmacy purchasing price was used (derived from pharmacy selling price 29 ).Unit costs were sourced from price lists from Region € Osterg€ otland 31 and S€ odra Regionvårdsn€ amnden 32 .The Swedish Regions' Price Index (LPIK) 1980-2021 36 and forecast for LPIK in 2022 37 (excluding pharmaceutical costs) were used to adjust the cost of leukapheresis to a 2022 price level.Additional information on 2L and subsequent therapies is provided in the Supplementary Materials.

Health state costs
Health state resource utilization was assumed to be independent of treatment arm and included outpatient visits, nurse visits, inpatient stays, CT scans, PET-CT scans, and blood tests.Resource use was stratified into event-free and post-event and adjusted to reflect Swedish clinical practice.The event-free resource use included the resource use for follow-up after ASCT and axi-cel treatment.The resource use for patients who were still event-free after five years was reverted to zero based on the assumption that patients who are still event-free after five years are effectively considered long-term responders with minimal healthcare resource utilization.Patients transitioning to the death health state incurred a one-off endof-life cost.The number of visits and tests included per month are presented in Table 1.Unit costs were sourced from price lists from S€ odra Regionvårdsn€ amnden 32 and Region Skåne 34 .The Swedish Regions' Price index (LPIK) 1980-2021 36 and forecast for LPIK in 2022 37 (excluding pharmaceutical costs) were used to adjust end-of-life cost to a 2022 price level.

Cost of adverse events
The model included the following Grade 3 and 4 AEs observed in ZUMA-7 that required treatment: cytokine release syndrome (CRS), neurologic events (i.e.immune effector cell-associated neurotoxicity syndrome (ICANS)), thrombocytopenia, anaemia and febrile neutropenia.Cost of managing AEs was not included for subsequent therapies.Unit costs were sourced from price lists from S€ odra Regionvårdsn€ amnden 32 .More details can be found in the Supplementary Materials.

Sensitivity and scenario analyses
Deterministic sensitivity analyses (DSA) and a probabilistic sensitivity analysis (PSA) with 1,000 iterations were conducted to assess robustness of the results.In the DSA, baseline characteristics (proportion of females, mean age, and mean body weight and body surface area), resource use, utilities, AE rates, proportion of patients receiving each type of subsequent therapy and unit costs were varied.The PSA probability distributions are presented in the Supplementary Table S7.
Several scenario analyses were conducted to assess the impact of key inputs and assumptions, including the time horizon (5, 10, 20, 30 and 40 years), the hazard of death compared to the general Swedish population for those who were assumed to achieve long-term remission (SMRs of 1.18 and 1.00), the utility values for patients who were still event-free after five years (reverting to general population vs not reverting) and the perspective of the analysis (healthcare vs societal).The societal perspective included costs of productivity loss, patient and caregiver time spent on treatment-specific activities, and transportation costs to and from the hospital.The patient time ascribed to each treatment-related activity is presented in Table 2.The applied hourly wages by age in Sweden are presented in Table 3.For the cost of transportation, it was assumed that patients have a one-hour drive each way to and from the hospital for any inpatient or outpatient visits.The transportation cost per visit was assumed to be SEK 300.For multi-day inpatient stays, the cost was only triggered once.AEs did not trigger additional transportation costs, as these were assumed to be treated during the inpatient stay at follow-up.As per TLV guidelines, the human capital approach was used to calculate the productivity loss 38 .The productivity loss for patients in the pre-and postevent health states was included together with the productivity loss caused by premature death.The share of patients who are able to work per month is presented in Table 4.
It was expected that the off-protocol treatment switching in ZUMA-7 heavily confounded the OS and costs.Therefore, a scenario analysis was conducted that accounted for the treatment switching from SOC to CAR T-cell therapy by not including the cost and effects of treatment switching in the SOC arm.Established statistical methods such as the RPSFTM  and IPCW are generally regarded as appropriate for handling treatment switching 41 and were applied to model the OS benefit in the absence of switching 6 .The two methods are described further in the Supplementary Materials.Both IPCW and RPSFTM were pre-specified in the ZUMA-7 protocol, and for the RPSFTM model full, partial or no re-censoring were performed following the guidance in the National Institute for Health and Care Excellence technical support document 16 6 .Clinical validation and external controls were used to assess the plausibility of resultant curves.The resultant curve was expected to fall in between the OS curves from the ORCHARRD 42 and SCHOLAR-1 43 as these two trials represent pre-CAR T-cell therapy use in a similar population.The RPSFTM with full re-censoring (HR: 0.416) gave the most clinically plausible estimates and has been accepted by the National Institute for Health and Care Excellence in their assessment of axi-cel in 2L DLBCL 44 .The resultant HR (0.416) was applied to the modelled axi-cel arm to estimate the OS curve for SOC in the absence of subsequent cellular therapy (Figure 4).The proportion of patients who received subsequent cellular therapy (including axi-cel) were redistributed to the chemotherapy regimens to account for costs.As in the base case, the impact on HRQoL of subsequent therapy was not accounted for in the scenario analysis.

Cost-effectiveness analysis
Over a lifetime horizon of 50 years, the model estimated an ICER of SEK 534,704 per QALY gained with an incremental discounted cost of SEK 812,944 and incremental discounted QALY gain of 1.52 for axi-cel compared to SOC.The total discounted cost per patient was SEK 4,137,997 and SEK 3,325,053 for axi-cel and SOC, respectively, and the total discounted QALY gain was 7.51 and 5.99 for axi-cel and SOC, respectively.The model estimated an incremental undiscounted LY gain of 2.56 years for axi-cel compared to SOC, with an average of 14.24 and 11.68 LYs for axi-cel and SOC, respectively.A detailed overview of the results is provided in Table 5.Furthermore, the model estimated 5-year survival rates of 51.3% and 42.9% and 10-year survival rates of 48.9% and 39.6% for patients treated with axi-cel and SOC, respectively.The median OS gain for axi-cel versus SOC was 70 months.The costs included both 2L and subsequent therapyrelated costs as well as costs related to disease management, AEs and end-of-life care.The cost of 2L therapy was highest for patients receiving axi-cel, whereas the subsequent therapy costs were highest for patients receiving SOC due to the high use of CAR T-cell therapy in subsequent treatment lines.

Sensitivity and scenario analyses
Figure 5 presents a tornado diagram with the 10 most influential model parameters from the DSA.The top three influential parameters were identified to be the post-event utility value, mean age and the number of specialist nurse visits in the post-event state for SOC.
The PSA resulted in an incremental discounted cost of SEK 801,930 and an incremental QALY gain of 1.33.The estimated ICER from the PSA was SEK 491,868 per QALY gained.Figure 6 presents the PSA cost-effectiveness plane, and Figure 7 illustrates the probability of axi-cel being cost-effective at different willingness-to-pay thresholds.If assuming a willingness-to-pay threshold of SEK 1,000,000, axi-cel was cost-effective versus SOC in 73% of the simulations.The PSA results are presented in more detail in the Supplementary Table S8.
The results of the scenario analyses are presented in Table 6.Most scenario analyses had a minor impact on the results, and the robustness of the base case result was confirmed.However, shortening the time horizon greatly  49 and SCHOLAR-1 45 .
increased the ICER, especially for time horizons of five or 10 years.The scenario analysis, where costs and effects for SOC were adjusted for off-protocol treatment switching, increased the ICER to SEK 694,351 per QALY gained.

Discussion
This study is the first study in the Nordic countries to evaluate the cost-effectiveness of axi-cel.The results of our analysis indicate that axi-cel is a cost-effective treatment option in 2L for transplant-intended DLBCL patients with early r/r after completing 1L chemoimmunotherapy.
In Sweden, the New Therapies Council, representing regional payers, has outlined criteria to determine the willingness to pay for new medicines.These criteria consider disease severity, clinical benefit, rarity of the disease and uncertainty in the clinical evidence.Based on these criteria and assuming a willingness to pay of SEK 1,000,000, axi-cel can be considered a cost-effective 2L treatment alternative for this patient population in Sweden.Axi-cel is currently recommended and funded by regional healthcare providers for the 3Lþ treatment of r/r DLBCL in Sweden based on the ZUMA-1 trial.In the cost-effectiveness analysis of axi-cel compared to SOC for 3Lþ r/r DLBCL, the estimated ICER was on par with the ICER estimated in the scenario analysis accounting for treatment switching in our study 29 .This is a conservative scenario, as 3Lþ use of axi-cel is approved in Sweden but indicates that axi-cel is a cost-effective 2L treatment alternative and that the use should not be restricted to 3Lþ.
Efforts were made in our cost-effectiveness analysis to reflect Swedish clinical practice.Therefore, the patient characteristics from the ZUMA-7 trial and other parameters that informed the model were aligned with a Swedish clinical expert's inputs as part of the model validation process.Several scenario analyses  were undertaken to investigate the implication of using alternative model inputs and assumptions.The robustness of the base case result was confirmed as most scenario analyses had a minor impact on the result.The scenario analyses using different time horizons stood out, as the ICER was greatly impacted by the shorter (five-and 10-year) time horizons.As axi-cel is a one-off treatment, treatment costs are incurred upfront, whereas the benefits of receiving treatment are accumulated over the patients' lifetime.Axi-cel has the potential to be a curative therapy for approximately 50% of DLBCL patients and shortening the time horizon will therefore significantly reduce the number of QALYs and LYs gained while most of the costs remain,  leading to a higher incremental cost per QALY.The scenario analysis accounting for treatment switching was conducted because 56% of patients randomized to the SOC arm in ZUMA-7 received off-protocol CAR T-cell therapy as subsequent therapy.This is possibly higher than the use in Swedish clinical practice, meaning that the ICER could potentially be higher in a Swedish setting than the base case ICER reported in our study, due to the lower use of subsequent therapy in the SOC arm in Sweden.It is likely that the ICER for axi-cel compared to SOC in Sweden lies between the result of our base case analysis and the scenario analysis, where the OS for SOC was adjusted for treatment switching.However, the ICER is still relatively contained even in the extreme scenario of 0% CAR T-cell therapy use in subsequent treatment lines in the SOC arm.According to the Swedish HTA agency, TLV, cost-effectiveness analyses should include scenarios with and without indirect costs 47 .We therefore performed a scenario analysis, changing the perspective of the analysis from a Swedish healthcare perspective to a social perspective.The analysis showed that changing the perspective reduces the ICER by SEK −81,732 compared to the base case.Tisagenlecleucel, lisocabtagene maraleucel, tafasitamab, and glofitamab are all recently EU-approved treatments for use in r/r DLBCL 38,[48][49][50] .However, none of these drugs are currently recommended or reimbursed in Sweden for r/r DLBCL.The New Therapies Council has issued negative reimbursement recommendations for tisagenlecleucel and tafasitamab, and recommendations are pending for lisocabtagene maraleucel and glofitamab 51 .At the time of the analysis, these treatments were not available to patients in Sweden and were therefore not seen as relevant to the decision problem.In addition, these drugs are approved for different indications, and only lisocabtagene maraleucel and tafasitamab have EU approval in 2L DLBCL (tafasitamab only for transplant-ineligible patients) 33 .
Using the ZUMA-1 event-free utility value as input for the post-event utility value in our analysis may overestimate the utility in the post-event health state.To address this, the post-event utility was included in the DSA, where the base case value was increased/decreased by 20%.The DSA showed that varying the post-event utility value impacted the ICER by less than SEK 51,000 (Figure 5).Applying ZUMA-1 event-free utility values in the post-event state is further supported by the fact that CAR T-cell therapy is reimbursed and used in 3L in Sweden as well as by the fact that around 80% of ZUMA-7 SOC patients receiving 3L treatment ended up receiving CAR T-cell therapy.
The robustness of the results of our study was also supported by Perales et al. 2022 52 , which evaluated the costeffectiveness of axi-cel compared to SOC for 2L DLBCL in the US.Despite the application of different resource use and other costs for treating DLBCL due to the US setting as well as different utility values, Perales et al. 2022 reported comparable results: an ICER of USD 66,381 per QALY (approx.SEK 700,000 [converted to SEK in November 2022]) compared to the SEK 534,704 per QALY reported in our study.Furthermore, the US study demonstrated a discounted QALY gain for axi-cel of 7.08, which is close to the QALY gain of 7.51 in our study.
This cost-effectiveness analysis utilized data from the randomized controlled trial ZUMA-7.The ZUMA-7 trial is the first and largest (N ¼ 359) head-to-head randomized trial comparing CAR T-cell therapy to SOC with a long follow-up duration.Previous CAR T-cell therapies have been investigated in clinical studies using a single-arm study design.The comparative and randomized trial design of the ZUMA-7 trial also supports the validity of the findings of this study.We explored various approaches to extrapolating survival outcomes beyond the ZUMA-7 trial data.These suggest that the existing OS trend observed in the ZUMA-7 trial will develop into substantial long-term benefits for patients.Furthermore, prolonged EFS as observed for treatment with axi-cel, is a key factor for the improvement in quality of life (QoL).These results are supported by the recently published primary PRO analysis of ZUMA-7, which shows a quicker return to baseline QoL in the axi-cel arm compared to SOC 53 .
In this cost-effectiveness analysis, a comparison was made between the use of 2L treatment with axi-cel and SOC in transplant-intended DLBCL patients with early r/r after completing 1L chemoimmunotherapy in Sweden.The results show that introduction of axi-cel comes at a higher total cost, but that this cost is offset by the clear clinical benefit of axi-cel compared to SOC.Treating patients in 2L instead of 3Lþ is cost-effective as it increases the possibility of curing more patients leading to not only a gain in survival, but also a reduction in the burden on QoL and cost of subsequent therapy.This will benefit both patients and society.

Declaration of funding
This study was sponsored by Gilead Sciences.

Declaration of financial/other interests
Axicabtagene ciloleucel is a product of Gilead Sciences (funder).Gilead Sciences provided support in the form of payments to Incentive Denmark.Authors O.E, S.V and V.K are employees at Gilead Sciences and Kite Pharma, a Gilead company.Incentive Denmark was a paid vendor to Gilead Sciences, and authors A.L, E.M and A.D were paid employees of Incentive Denmark.

Figure 1 .
Figure 1.Model structure.The circles represent mutually exclusive health states, and the arrows represent transitions between or within states.

Figure 2 .
Figure2.Health state occupancy.Implementation method of the partitioned survival model estimating health state occupancy through the disaggregation of sequential event curves (illustrative example).The health state occupancy in the event-free and death health states were derived from the centrally-assessed event-free survival (EFS) curve and overall survival (OS) curve, respectively.The health state occupancy in the post-event health state was calculated as the difference between the OS curve and the EFS curve (the proportion of patients who are still alive but are no longer event-free).The post-event subsequent therapy was determined by the time-to-next therapy (TTNT) curve.

Figure 3 .
Figure 3. Survival plots.A) Shows the Kaplan Meier curves and extrapolations for axi-cel, B) Shows the Kaplan Meier curves and extrapolations for standard of care (SOC); and C) Shows the modelled extrapolation curves applied in the model.

Figure 4 .
Figure 4. Validation of the modelled treatment switching adjusted SOC OS curve.ZUMA-7 modelled SOC OS curve adjusted for treatment switching using the RPSFTM model with full re-censoring and HR approach, ZUMA-7 modelled axi-cel OS curve and Kaplan Meier curves from ORCHARRD49 and SCHOLAR-145 .

Figure 5 .
Figure 5. Tornado diagram.The figure illustrates the 10 most influential parameters of the DSA.The base case parameters were varied based on their standard errors.If the standard errors were not reported, the parameters were varied by an arbitrary ±20% around the mean.

Figure 6 .
Figure 6.PSA cost-effectiveness plane (SEK).The figure illustrates that the majority of the alternative ICERs in the cost-effectiveness plane are located in the Northeast quadrant of the graph, where axi-cel is more effective and more costly compared to SOC.A small proportion is placed in the Northwest quadrant, where axi-cel is more costly and less effective compared to SOC.

Figure 7 .
Figure 7. Cost-effectiveness acceptability curve (SEK).The figure illustrates the cost-effectiveness probability at different willingness-to-pay thresholds.At a willingness to pay threshold of SEK 1,000,000, axi-cel is cost-effective in 73% of the simulations.

Table 1 .
Overview of model input values and sources.
thus, patients receiving R-ESHAP in the trial were uniformly allocated to the other regimens in the model.The share of patients who received salvage chemoimmunotherapy (R-ICE, R-GDP and R-DHAP), stem cell harvest, HDT and who completed ASCT in the model are presented in Table1.A weighted average cost of subsequent therapy (3Lþ) was applied in the model when a patient experienced a TTNT event.The proportion of patients receiving each subsequent therapy was collected in ZUMA-7 and applied in the model.Tender prices from Region Stockholm 35 were applied in the model to calculate the drug acquisition costs.Pharmacy selling prices were used in cases where tender prices were not available (see Supplementary Materials).
a Total cost per cycle.b Total treatment cost per event.

Table 2 .
Patient time included in the scenario analysis, applying a societal perspective.

related activities in the axi-cel arm
Leukapheresis 18 h per cycleBased on one inpatient day (an inpatient day was assumed to be 16 h) þ two hours of transportation.In the model, the patient time spent on treatment-related activities in the axi-cel arm was weighted by survival.

Table 3 .
Average age-adjusted hourly wage and workforce.

Table 4 .
Proportion (percentage) of patients being absent from work per month in the pre-event and post-event states in the model.

Table 5 .
Base case cost-effectiveness results.

Table 6 .
Results of scenario analysis.Including costs of productivity loss, patient and caregiver time spent on treatment-specific activities as well as transportation costs to and from the hospital.Abbreviations.QALYs, Quality-adjusted life years; SEK, Swedish krona; ICER, Incremental cost-effectiveness ratio; SMR, Standardized mortality ratio; HR, Hazard ratio; axi-cel: axicabtagene ciloleucel, SOC, Standard of care.