Cost-Effectiveness of Tepotinib Versus Capmatinib for the Treatment of Adult Patients With Metastatic Non–Small Cell Lung Cancer Harboring Mesenchymal–Epithelial Transition Exon 14 Skipping

Objectives: From the US Medicare perspective, this study compared the cost-effectiveness of tepotinib and capmatinib for treating metastatic non–small cell lung cancer with tumors harboring mesenchymal–epithelial transition factor gene exon 14 skipping. Methods: A 3-state partitioned survival model assessed outcomes over a lifetime horizon. Parametric survival analysis of the phase 2 VISION trial informed clinical inputs for tepotinib. Capmatinib inputs were captured using hazard ratios derived from an unanchored matching-adjusted indirect comparison study and published literature. National cost databases, trial data, and literature furnished drug, treatment monitoring, and disease/adverse event management expenditures (2021 US dollars) and utility inputs. Outcomes were discounted at 3% annually. Results: In the base case, tepotinib dominated capmatinib in frontline settings (incremental discounted quality-adjusted life-years [QALYs] and costs of 0.2127 and −$47 756, respectively) while realizing an incremental cost-effectiveness ratio of $274 514/QALY in subsequent lines (incremental QALYs and costs of 0.3330 and $91401, respectively). In a line agnostic context, tepotinib produced an incremental cost-effectiveness ratio of $105 383/QALY (incremental QALYs and costs of 0.2794 and $29 447, respectively). Sensitivity and scenarios analyses for individual lines typically supported the base case, whereas those for the line agnostic setting suggested sensitivity to drug acquisition costs and efficacy inputs. Conclusions: Tepotinib could be cost-effective versus capmatinib in frontline and line agnostic contexts, considering the range of willingness-to-pay thresholds recommended by the Institute for Clinical and Economic Review ($100 000-$150 000/QALY). Tepotinib could be cost-effective in subsequent lines at higher willingness-to-pay levels. These results are to be interpreted cautiously, considering uncertainty in key model inputs.


Introduction
Lung cancer is a leading cause of cancer-related mortality in the United States, accounting for an estimated 21.7% of all cancer deaths in the United States in 2021. 1 Non-small cell lung cancer (NSCLC) accounts for approximately 80% to 85% of cases. 2 Despite innovative therapies, NSCLC survival rates remain low, with the 5-year survival rate in patients diagnosed of distant metastatic (stage IV) cancer being approximately 6.3% in the United States. 1 Approximately 3% to 4% of NSCLCs harbor mesenchymal-epithelial transition factor gene exon 14 (METex14) skipping, which has been recognized as an oncogenic driver. [3][4][5][6][7] Patients with NSCLC harboring METex14 skipping tend to be older, 8 to exhibit a nonsquamous histology, 9,10 and to have poorer prognoses because of higher rates of brain, bone, and liver metastases. [11][12][13] Until tyrosine kinase inhibitors (TKIs) were approved, there were no therapies approved in the United States to specifically treat patients with NSCLC harboring METex14 skipping, generally leaving this patient subgroup with recourse only to approved nontargeted therapies.
In May 2020, capmatinib was the first TKI approved by the US Food and Drug Administration (FDA) for adult patients with metastatic NSCLC (mNSCLC) whose tumors have a mutation that leads to METex14 skipping, as detected by an FDA-approved test. 11,14 In February 2021, tepotinib was approved in the United States, based on the results of a phase 2, single-arm, open-label trial (VISION; NCT02864992) that evaluated its safety, efficacy, and tolerability in patients, including those with locally advanced or metastatic stage IIIb/IV NSCLC harboring METex14 skipping. 9 Other targeted therapies-such as crizotinib (administered off label in some contexts) and savolitinib (conditionally approved in China based on favorable phase 2 results)-have yet to be approved for this indication in the United States. 8,[15][16][17] The burden of mNSCLC on patients, caregivers, payers, and society and the recent emergence of targeted therapies that may prolong survival in the subset of patients whose tumors harbor METex14 skipping but may also result in additional treatment costs underscore the importance of developing evidence to assess both the health and cost outcomes of products within this class of treatments. This study aimed to support this endeavor by evaluating the cost-effectiveness of tepotinib from the US Medicare perspective, compared with capmatinib, for adults with mNSCLC harboring METex14 skipping.

Study Design and Scope
Overview of study design and scope-This cost-effectiveness analysis was based on the efficacy and safety results of the VISION study (cohort A; February 2021 data cutoff) for patients who were treatment naïve (1L) or previously treated (2L+), as per tepotinib's FDA-approved indication. 9 Capmatinib was included as the sole comparator to tepotinib, given that it is currently the only other targeted therapy approved for this indication in the United States. Although crizotinib may in certain contexts be administered for this indication, it was excluded from this analysis because it is not presently FDA approved, is not considered a preferred frontline or subsequent treatment option in current National Comprehensive Cancer Network Clinical Practice Guidelines, 18 and, according to key opinion leaders consulted for this study, is unlikely to experience widespread utilization in the United States because of comparatively poor outcomes in clinical trials. 8,16 The base-case analysis was performed from the US Medicare perspective and evaluated cost-effectiveness in 1L, 2L+, and line agnostic (1L and 2L+) settings. Deterministic sensitivity analysis (DSA) and probabilistic sensitivity analysis (PSA) and scenario analyses were performed to investigate the impact of parameter and structural uncertainty on model outcomes, including the effect of adopting commercial or Medicaid payer perspectives. undergoing treatment. A PSM rather than a Markov state transition model was developed because the PSM readily aligned with key secondary endpoints from VISION, simplified accounting for time dependency (achievable in state transition models through inclusion of tunnel states that contribute to structural complexity), and reduced reliance on unclassified endpoints to model state transitions.
Given that FDA approvals for tepotinib and capmatinib are not contingent upon treatment history, the cost-effectiveness model (CEM) was designed to capture costs and outcomes accrued by those who were 1L, 2L+, or either (ie, line agnostic). In the latter case, the CEM calculated the weighted average of outcomes for 1L and 2L+, with reference to the observed baseline distribution of patients in VISION (ie, 44.5% 1L, 55.5% 2L+), 9 as opposed to explicitly modeling the patient journey across multiple lines of therapy. This approach provides greater flexibility in considering line-specific differences in treatment effects than building the model around the line agnostic intent-to-treat outcomes from the VISION study.
A 10-year time horizon was sufficient for nearly all patients to reach the death state. A monthly model cycle was selected, in alignment with dosing cycles, and a 3% annual discount rate was applied to health and cost outcomes. 19

Data Sources, Inputs, and Modeling
The CEM incorporated a wide range of clinical, utility, and cost/resource utilization inputs. A summary of key inputs used in the model and their respective sources is presented in Table 1. 9,10,[20][21][22][23][24][25][26][27][28][29][30][31][32][33][34][35] Clinical efficacy-Efficacy inputs for tepotinib were derived from the February 2021 data cutoff using the entire cohort A (n = 152) from VISION, 9 consisting of patients who were 18 years of age or older with histologically or cytologically confirmed, locally advanced, or mNSCLC with METex14 skipping. Nearly all VISION participants (98%) had stage IV disease at study entry 9 ; clinical experts advised that the remaining participants (ie, with stage IIIb disease who could not be treated with radiotherapy) were prognostically similar to patients with metastatic disease (of note, 99% of the GEOMETRY mono-1 efficacy population consisted of patients with stage IV disease). 36 All patients had measurable disease, were able to perform most or all of the activities of daily living (ie, Eastern Cooperative Oncology Group score of 0 or 1), and had a negative test result for epidermal growth factor receptor mutations or anaplastic lymphoma kinase rearrangements. 9 Standard parametric survival analysis techniques were applied to patient-level data from VISION to extrapolate PFS, OS, and TTD beyond the trial's follow-up duration, in accordance with guidance issued by the Decision Support Unit for the National Institute for Health and Care Excellence. 37 Briefly, this entailed identifying statistical distributions that exemplified goodness-of-fit (eg, evaluation of Akaike and Bayesian information criteria statistics and graphical assessment of fit versus observed data) and generated clinically plausible extrapolations (assessed via structured interviews with clinical experts). On this basis, exponential distributions were selected to model long-term OS and PFS for tepotinib, both for 1L and 2L+ patients. Details regarding the methodology and results of these statistical analyses are presented in the Supplemental Materials ( Table 1 in Appendix 1 in Supplemental Materials found at https://doi.org/10.1016/j.jval.2022.11.018), whereas the impact of using alternative distributions was evaluated in scenario analyses.
Meanwhile, OS and PFS for capmatinib (Table 1 9,10,20-35 ) were estimated using hazard ratios derived from an unanchored matching-adjusted indirect comparison (MAIC) study, 20 using patient-level data from VISION and baseline variables and outcomes from GEOMETRY mono-1 (NCT02414139), a multi-cohort, open-label, phase 2 study designed to evaluate the safety and efficacy of capmatinib in patients with advanced NSCLC with METex14 skipping or MET amplification. 10 To align with GEOMETRY mono-1 eligibility criteria and the efficacy population referenced in the current tepotinib US prescribing information, the MAIC focused on VISION cohort A participants identified by tissue biopsy (n = 88).
Treatment discontinuation-TTD in VISION was defined as months elapsed between the first and last doses of treatment and was extrapolated beyond the follow-up period by applying an exponential distribution (Figures 1-6 in Appendix 1 in the Supplemental Materials found at https://doi.org/10.1016/j.jval.2022.11.018). TTD for capmatinib was assumed to be the median duration of exposure reported in the GEOMETRY mono-1 study. 10 This approach was applied to account for varying strategies around discontinuation, given that in clinical practice some patients discontinue treatment before progression, whereas others may continue until or past progression.
Costs and resource utilization-The CEM incorporated expenditures attributable to drug acquisition and administration, adverse event (AE) and disease management, treatment monitoring, and subsequent treatments (Appendices 2-5 in the Supplemental Materials found at https://doi.org/10.1016/j.jval.2022.11.018 for full sources and details). Biomarker testing costs were excluded from consideration in the base case, given that all patients entering the model were assumed to have confirmed METex14 skipping before treatment.
All costs were inflated to 2021. 38 Drug acquisition costs were based on dosing schedules in treatment labels and wholesale acquisition costs sourced from RED BOOK. 21 Given that tepotinib and capmatinib are orally administered, recipients were assumed not to incur any drug administration costs. Subsequent treatment costs were applied to patients discontinuing either comparator and consisted of a one-off cost that accounted for post-tepotinib therapies administered in VISION, including TKIs (capmatinib or crizotinib if 2L+ with tepotinib, crizotinib if 2L+ with capmatinib), immuno-oncology monotherapy, immuno-oncology + chemotherapy ± antivascular endothelial growth factor, antivascular endothelial growth factor ± chemotherapy, and chemotherapy alone. Subsequent treatment costs were sourced like primary comparators and accrued for an interval estimated from mean PFS for subsequent therapy observed in VISION (3.0 months). 25 For AE management costs, event incidence was drawn from product labels, 22 Treatment monitoring was limited to laboratory costs, because it was assumed additional cost components were subsumed under disease management; based on input from clinical experts, monitoring was assumed similar for both treatments and consisted of monthly hematology and coagulation, liver function, electrolyte, and urine analysis testing, costs for which were sourced from the Centers for Medicare & Medicaid Services. 24 Resources required for disease management were assumed to vary according to progression status. Estimates of medical resource utilization for NSCLC were based largely upon values obtained from Graham et al, 2016, 28 and Dalal et al, 2018, 27 whereas accompanying costs were sourced from the Centers for Medicare & Medicaid Services. 24,26 Utilization of selected resources for patients with progressed disease (ie, specialist visits, computerized tomography scans, magnetic resonance imaging, ultrasounds, and X-rays) was derived from key opinion leader input. Unit costs and monthly frequencies were combined to obtain monthly disease management costs ($874 and $5462 per month for progression-free and progressed patients, respectively). Finally, one-off costs from published sources were accrued for disease progression ($1079) and terminal care ($4063). 29,30 Health state utility values-Statistical analyses of EuroQol Five Dimension Five Level Scale data from VISION (cohort A) were undertaken to derive baseline, preprogression, and progressed utility values 31 and were assumed equivalent for tepotinib and capmatinib.
The negative effects of AEs on health-related quality of life were captured through utility decrements extracted from publicly available sources and estimated episode duration for each AE. Additional information regarding the health state utilities and utility decrements is supplied in Appendix 6 in the Supplemental Materials found at https://doi.org/10.1016/ j.jval.2022.11.018.

Model Validation and Analysis
A targeted literature review of economic models for mNSCLC was conducted at the model conceptualization phase to ensure the current CEM design would align with previous studies. [40][41][42][43][44][45][46] The model validation process followed current guidelines from the ISPOR-Society of Medical Decision Making. 47 The model structure, assumptions, and inputs were validated by clinical experts and experienced health economists. A technical validation of the model programming was performed by a modeler not previously involved in the study.
The base-case analysis assessed the cost-effectiveness of tepotinib versus capmatinib in terms of the incremental cost-effectiveness ratio (ICER), with health benefits quantified as quality-adjusted life-years (QALYs). Given that no explicit willingness-to-pay (WTP) threshold has been defined for the United States, 48 results were interpreted with reference to the range of WTP thresholds recommended by the Institute for Clinical and Economic Review ($100 000-$150 000/QALY), 49 as in other recently published health economic evaluations in NSCLC. 50 DSA was conducted to probe uncertainty in model parameters by systematically adjusting the upper and lower bounds of individual parameters or groups thereof (using 95% confidence intervals, where available; otherwise, ±15% ranges around the point estimate) and examining the effect on model outcomes, presenting the results as a tornado diagram that ranks them in descending order according to impact. DSA results are expressed in terms of incremental net monetary benefit (INMB) and reflect the difference between incremental costs and the value of incremental QALYs (evaluated using a $150 000/QALY WTP threshold), with positive INMB estimates denoting cost-effectiveness of tepotinib relative to its comparator.
PSA was conducted to further examine the implications of parameter uncertainty by jointly and repeatedly sampling from suitable probability distributions defined for each parameter (summarized in Table 1 9,10,20-35 ) and storing model outputs; 2500 iterations of the model were conducted in this fashion and convergence confirmed, and the results visualized through scatterplots as well as cost-effectiveness acceptability curves that illustrate the likelihood of achieving cost-effectiveness over a range of WTP thresholds.
Scenario analysis was conducted to quantify the cost-effectiveness of tepotinib versus capmatinib when key base-case assumptions were varied. A summary of each scenario and justification for its inclusion is presented in

Base-Case Analysis
In the 1L setting (

Sensitivity Analysis
Deterministic sensitivity analysis-Results of the DSA (Fig. 2) show that, in all patient populations, estimated INMB is most sensitive to monthly drug acquisition costs for both tepotinib and capmatinib, the parameters of the exponential distribution used to extrapolate TTD with tepotinib beyond trial follow-up, capmatinib median treatment duration, and hazard ratios for PFS and OS with capmatinib (detailed tabular results are presented in Tables 8, 9 Table 3 (detailed tabular results are presented in Table 7  Differences in outcomes across treatment lines are largely attributable to estimates of treatment duration. In 1L, mean treatment durations for tepotinib and capmatinib were 13.2 months and 16.5 months, respectively, whereas in 2L+, the average treatment durations were 11.8 months and 7.9 months, respectively. With capmatinib having a longer time on treatment in 1L, additional monthly drug acquisition costs are accrued, and these account for a significant portion (approximately 70%) of the total costs associated with each treatment.

Scenario analysis-Scenario analyses are presented in
The DSA, PSA, and scenario analyses for the 1L and 2L+ settings tended to align with their respective base-case results. By contrast, the cost-effectiveness of tepotinib in the line agnostic context was highly sensitive to uncertainty in drug acquisition costs and efficacy inputs in the DSA and to analytical perspective (eg, higher ICERs with a commercial perspective due to increased disease management unit costs, coupled with increased OS with tepotinib). It was also sensitive to the composition of the patient population (which produced favorable ICERs when applying population weighting from Flatiron, 51  tepotinib and capmatinib in this patient population. We used a modeling methodology that is common in oncology and consistent with best practice as laid out by ISPOR-Society of Medical Decision Making, 47 and the model was thoroughly reviewed and validated by clinical experts.
It is essential to recognize this study's limitations and short-comings in interpreting its results. First, no head-to-head studies of tepotinib versus capmatinib have been undertaken to date; accordingly, an unanchored MAIC was conducted to account for differences in trial populations, focusing on key prognostic differences to preserve sample size. 52 This approach is preferable to naïve side-by-side comparisons of outcomes between clinical trials because it reduces bias that may occur because of differences in baseline demographic and clinical characteristics between the trial populations. 52 Nevertheless, the absence of a common comparator in the VISION and GEOMETRY mono-1 trials necessitates reliance on assumptions that may not hold in practice; for instance, this approach assumes the analysis has captured all relevant effect modifiers and prognostic factors, which is generally unlikely to be achieved, and may introduce bias into the results. 53 In addition, small sample sizes contributed to wide confidence intervals around key efficacy inputs and, by extension, to the uncertainty observed in the sensitivity and scenario analysis results.
Second, the analysis did not include crizotinib, another targeted therapy used in some instances to treat this indication. Although this choice was seen as appropriate given the scope of the analysis (ie, targeted therapies FDA approved for this indication in the United States) and supported by current National Comprehensive Cancer Network guidelines and input from clinical experts, the authors recognize this as a limitation of the study to the extent it implies excluding an intervention that may represent a relevant therapeutic alternative in certain contexts.
Third, as is common in health economic evaluations in oncology, it was necessary to extrapolate beyond the comparatively brief trial follow-up duration, introducing further uncertainty into the results. To mitigate this uncertainty, we applied standard parametric survival analysis techniques to patient-level data from VISION in accordance with current best practice, 37 ultimately concluding that exponential distributions were appropriate for long-term extrapolation of OS, PFS, and TTD to the extent they exemplified both statistical goodness-of-fit and clinical plausibility. This was further investigated in scenario analyses, which determined that adopting alternative distributions for OS and PFS does not materially alter the central findings of this study.
Fourth, plausible assumptions and expert opinion were applied to address outstanding data gaps. For example, there is an absence of evidence specific to mNSCLC harboring METex14 skipping, and therefore, wild-type mNSCLC published studies were used to inform several model inputs, although this is not expected to bias the results of the study in favor of or against, tepotinib. Additionally, although VISION and GEOMETRY mono-1 recruited patients with both advanced (stage IIIb) and metastatic (stage IV) tumors, patients with stage IIIb disease comprised a very small proportion of participants in both trials 9,36 and in VISION, because they could not be treated with radiotherapy, were considered prognostically similar to patients with stage IV disease. Accordingly, clinical inputs derived from both trials were assumed representative of patients with mNSCLC.

Conclusions
Tepotinib may be cost-effective compared with capmatinib in treating patients with mNSCLC harboring METex14 skipping in frontline and line agnostic settings from the US Medicare perspective; in subsequent lines, cost-effectiveness depends upon the value payers attach to additional QALYs. These results are subject to uncertainty and should be revisited as the evidence base for this class of therapies matures.

Supplementary Material
Refer to Web version on PubMed Central for supplementary material.   1L indicates treatment naïve; 2L+, previously treated; WTP, willingness to pay.