Long-term Survival and Cost-effectiveness Associated With Axicabtagene Ciloleucel vs Chemotherapy for Treatment of B-Cell Lymphoma

Key Points Question What are the incremental survival gains and cost-effectiveness of axicabtagene ciloleucel vs chemotherapy for treatment of B-cell lymphoma in adults? Findings In this economic evaluation study, a survival and cost-effectiveness analysis of ZUMA-1 trial data found that at the end of the trial (24 months), treatment with axicabtagene ciloleucel resulted in more quality-adjusted life-years than chemotherapy, producing a cost-effectiveness estimate between $896 600 and $1 615 000 per quality-adjusted life-year gained. Lifetime survival gains for treatment with axicabtagene ciloleucel ranged from 1.52 to 4.90 more quality-adjusted life-years, and a cost-effectiveness estimate ranged from $82 400 to $289 000 per quality-adjusted life-year gained over a lifetime horizon. Meaning Treatment with axicabtagene ciloleucel appears to be associated with positive, yet uncertain, gains in survival compared with chemotherapy, and its cost-effectiveness is associated with long-term survival.


eAppendix. Model Structure
Methodology was adapted from the Institute for Clinical and Economic Review (ICER) final report on CAR-T therapies. 1 The decision analytic model structure included a short-term decision tree and a long-term semi-Markov partitioned-survival model. The decision tree calculated the costs and consequences from treatment initiation to assessment of response, which was approximately one month. 2 From the decision tree, patients moved to the semi-Markov partitioned-survival model where they were then tracked for a trial-based time horizon of 24 months and a lifetime time horizon. The purpose of the decision tree was to stratify the cohort by which treatment they ended up receiving, because the model starts at treatment initiation (considered leukapheresis for axicabtagene ciloleucel). Further, the decision tree allowed for allocation of upfront costs by treatment and the stratification of the cohort by response status.
For the decision tree, the CAR-T arm included patients who were eligible for axicabtagene ciloleucel and underwent leukapheresis. At the first decision tree event node of the CAR-T arm, patients had three possibilities: 1) continue with axicabtagene ciloleucel after undergoing leukapheresis to receive the infusion; 2) discontinue axicabtagene ciloleucel (before infusion but after leukapheresis) because of adverse events or manufacturing failures; or 3) die before receiving the infusion. Those who discontinued axicabtagene ciloleucel due to adverse events were assumed to not be able to tolerate other active therapies and therefore transitioned to receive no further antilymphomic therapy (i.e., palliative care only). Those who discontinued axicabtagene ciloleucel due to manufacturing failures were assumed to receive chemotherapy. Responses were assessed for patients who received the axicabtagene ciloleucel infusion (second event node of decision tree), which could be: alive and responding to treatment; alive and not responding to treatment; or dead before assessment of response. The model was flexible enough to allow for patients to receive or not receive stem cell transplantation (third event node of decision tree) based on percentages reported in available evidence. The decision tree's comparator arm followed a similar pathway to the CAR-T arm, tracking the patient from chemotherapy treatment initiation through assessment of response and receipt of stem cell transplantation.
From the decision tree, the cohort was assigned to three mutually exclusive health states in a semi-Markov partitioned survival model that followed patients for the remainder of their lifetime using survival curve evidence. The three health states included: 1) alive and responding to treatment, 2) alive and not responding to treatment, and 3) death from DLBCL or other causes.
Patients transitioned between states during predetermined cycles (one month) over a lifetime time horizon. The "alive and responding to treatment" health state included all patients who were alive and responding to treatment (complete or partial responders). The "alive and not responding to treatment" health state included all patients who were alive that did not respond to treatment or relapsed after previously responding to treatment. Patients in the "alive and not responding to treatment" health state remained in this health state until they died from their modeled B-cell malignancy or other causes. Patients not responding to treatment received palliative chemotherapy. End-of-life hospice care costs were assigned to each death event. Health state occupancy was derived using five different partitioned survival techniques involving the direct extrapolation of progression-free survival (PFS) and overall survival (OS) Kaplan-Meier curves: alive and responding to treatment (t)=P(PFS, t) alive and not responding to treatment (t)=(P(OS, t)-P(PFS, t) Although the decision tree separated the cohort based on response status, survival curves were not available stratified by response status for all treatments. Further, definitions of response may vary between treatments; thus, survival curves were based on aggregated cohort data and not stratified by response status. Thus, in our models, there is no structural link between response status and survival. The models were developed in Microsoft Excel.
The analysis provided results from both the public and commercial payer perspective and focused on direct medical care costs only. Outcomes were estimated over a trial-based time horizon of 24 months and a lifetime time horizon using a monthly cycle. Costs and outcomes were discounted at 3% per year. Based on mean time from CAR-T therapy to stem cell transplantation estimated by Lee et al. 3,11 Patients received a single full course of CAR-T therapy.
CAR-T therapies are considered an end-of-line treatment with no clinical evidence on re-treatment. All patients who transitioned to the alive and not responding to treatment health state received palliative chemotherapy.
The intervention and comparator therapies are considered end-of-line treatments.
Patients who discontinued CAR-T due to an AE before receiving the infusion received no further antilymphomic therapy.
Those who experienced a severe AE would be unable to tolerate further active therapy.
Patients who did not receive CAR-T therapy due to a manufacturing failure received the active comparator.
Those who experienced a manufacturer failure would be able to tolerate further active therapy.

The model included costs and outcomes associated with grade 3/4 AEs.
Less severe adverse events are not expected to significantly impact patient health or costs. The cost of a hospital admission for treatment administration included the per diem cost for hospital days and the costs of therapies administered during the hospitalization.
Future bundled payments were assumed to approximate the cost of the resources used under a fee-for-service framework. The denominator is the number of people who received a CAR-T infusion for CAR-T therapies and the number of people who initiated the chemotherapy regimen for comparator therapies.

eTable 4. Source of Kaplan-Meier Curves to Calculate Transition Probabilities B-cell Lymphoma Axicabtagene Ciloleucel Chemotherapy Progression-Free Survival
Progression-free survival curve ( Figure 2B) for ZUMA-1 6 No published progression-free survival curve; therefore, the progression-free survival curve was derived from available overall survival data for SCHOLAR-1 chemotherapies, by assuming the proportional relationship from a published progression-free survival and overall survival curve for R-DHAP in the same disease state. 13