Comparative cost-effectiveness of cabozantinib as second-line therapy for patients with advanced hepatocellular carcinoma in Germany and the United States.

Background Cabozantinib was approved by the European Medicines Agency and the Federal Drug Administration as an option for sorafenib-resistant advanced hepatocellular carcinoma, increasing overall survival and progression-free survival compared with placebo. We evaluated the cost-effectiveness of cabozantinib in the second-line setting for patients with an advanced hepatocellular carcinoma from the German statutory health insurance perspective compared with an US scenario using US prices. Methods A Markov model was developed to compare the costs and effectiveness of cabozantinib with best supportive care in the second-line treatment of advanced hepatocellular carcinoma over a lifetime horizon. Health outcomes were measured in discounted life years and discounted quality-adjusted life years. Survival probabilities were estimated using parametric survival distributions based on CELESTIAL trial data. Utilities were derived from the literature. Costs contained drugs, monitoring and adverse events measured in US Dollars. Model robustness was addressed in univariable, scenario and probabilistic sensitivity analyses. Results Cabozantinib generated a gain of 0.18 life years (0.15 quality-adjusted life years) compared with best supportive care. The total mean cost per patient was $56,621 for cabozantinib and $2064 for best supportive care in the German model resulting in incremental cost-effectiveness ratios for cabozantinib of $306,778/life year and $375,470/quality-adjusted life year. Using US prices generated costs of $177,496 for cabozantinib and $4630 for best supportive care and incremental cost-effectiveness ratios of $972,049/life year and $1,189,706/quality-adjusted life year. Conclusions Our analysis established that assuming a willingness-to-pay threshold of $163,371/life year (quality-adjusted life year) for the German model and $188,559/life year (quality-adjusted life year) for the US model, cabozantinib is not cost-effective compared with best supportive care. Sensitivity analyses showed that cabozantinib was not cost-effective in almost all our scenarios.


Background
Hepatocellular carcinoma (HCC) has a constantly increasing incidence in Germany and the United States, as developed countries, with approximately 6628 (8.2/100, 000 inhabitants, Germany) and 24,223 (7.7/100,000 inhabitants, United States) new diagnoses in 2012 [1,2]. Former hepatitis B and C infections, alcoholic cirrhosis and especially the increasing risk of nonalcoholic steatohepatitis are the main drivers of its development [1]. The relative 5-year survival rate remains low at 15%, and curative options such as resection and transplantation are suitable only for locally limited HCC [3,4]. Therefore, advances in systemic therapy for advanced HCC are highly necessary.
Since the breakthrough of sorafenib enhanced the treatment of HCC, patients have suffered a lack of second-line options after progression. Brivanib and everolimus failed to improve overall survival (OS) versus placebo in phase III trials [5,6]. In 2017, regorafenib showed a significant OS prolongation from 7.8 months to 10.6 months versus placebo, leading to approvals by the European Medicines Agency (EMA) and US Food and Drug Administration and its implementation in the European and US market as the first second-line option for systematic HCC therapy [7].
The Federal Joint Committee (GBA) is the German statutory authority that assesses the degree of benefit of newly approved therapies and decides on their reimbursement by the statutory health insurances (SHI). They acknowledged the drug as a nonquantifiable added benefit in the previously approved indication of thirdline treatment for advanced colorectal cancer. As a consequence, the pharmaceutical producer Bayer withdrew regorafenib from German markets because they had identified no opportunity to achieve a sufficient reimbursement amount [8]. Therefore, oncologists could not provide the expensive regorafenib therapy for HCC in Germany, because its costs were no longer reimbursed.
In June 2019, the GBA voted to follow the recommendations of the German Institute for Quality and Efficiency in Health Care (IQWiG) acknowledging a minor added benefit of cabozantinib in the second-line therapy of HCC after progression with sorafenib [9]. Cabozantinib is a tyrosine kinase inhibitor (TKI) that targets tyrosine kinases, such as vascular endothelial growth factor receptor (VEGFR) and hepatocyte growth factor receptor (MET) and it has already been approved as a first-line treatment for advanced medullary thyroid carcinomas and advanced renal cell carcinomas (RCC). This appraisal is based on the findings of Abou-Alfa et al. regarding the CELESTIAL trial results [10,11]. Patients treated with cabozantinib showed an OS of 10.2 months compared with 8.0 months with placebo. Progressionfree survival (PFS) was prolonged from 1.9 to 5.2 months [10].
National healthcare systems are facing rising costs to provide new effective therapies. Our aim is to determine the cost-effectiveness of cabozantinib therapy compared with best supportive care (BSC) in patients whose HCC was inadequately treated by sorafenib within the context of the German and the US healthcare system. We considered only BSC as a comparator in the model, as other second-line systemic HCC drugs were not available or reimbursable in Germany or didn't match the target population. Our BSC definition includes health items of monitoring, consulting and adequate treatment of adverse events. The cabozantinib group received equal BSC and cabozantinib. The primary outcome measure is the incremental cost-effectiveness ratio (ICER) quantified in $ per life year (LY) and $ per quality-adjusted LY (QALY). We assumed cost-effectiveness thresholds estimated as multiples of the gross domestic product (GDP) per capita deduced from the 3 times GDP per capita rule of the WHO [12]. The impact of uncertainties on the ICER was investigated in sensitivity analyses.

General
The selection of data sources and methods follow the German health economic evaluation recommendations of the IQWiG [13]. Our Markov model was implemented in TreeAge Healthcare Pro 2019 software (Tree-Age Pro 2019, R1.1; TreeAge Software, Williamstown, MA). The data were obtained from published material of the CELESTIAL trial and the submitted GBA dossier of IPSEN Pharma and completed by a literature review on cabozantinib, TKIs and HCC [14]. We considered these sources as adequate clinical effectiveness data, because the CELESTIAL trial met its primary endpoint, was sufficiently reported and 48% of the trial subjects are from Europe and 24% from the United States or Canada [10].

Target population, setting and perspective
Our target population in the model was based on the CELESTIAL trial subjects: adult patients with HCC who showed progression under prior sorafenib therapy. The methods of the CELESTIAL trial were described in the published study protocol [15]. In summary, 707 patients were randomized into the cabozantinib group or placebo group and were treated with 60 mg cabozantinib per day or the placebo. The inclusion of patients with Child-Pugh A liver function and the exclusion of patients with an uncontrolled clinically significant illness allowed only relatively healthy patients to be included the trial population [15]. We used published data with reference to other TKIs treating HCC if it was necessary to fill data gaps because of identical drug classes and comparable side effects.
88% of people in Germany were insured by SHIs in 2015 and 12% by private health insurances [16]. The US population is covered by multiple overlapping insurance forms: 55.4% of patients are insured employment-based, 10.8% use direct-purchase and the public plans Medicare and Medicaid cover 17.8 and 17.9% [17]. Furthermore, 8.5% were insured [17]. We chose the perspective of German SHIs for costs induced by cabozantinib therapy and BSC. As a consequence of the diverse US insurance forms and the resulting difficultness to measure exemplary costs, we quantified the chosen health item of the German SHI perspective with adequate US equivalents described in Cost calculation.

Model structure
A Markov model was constructed to estimate the costs and utility gains of the target population illustrated with Fig. 1. It consists of three health states representing the natural process of HCC: stable, progressive and dead. All patients started in stable and either stayed at that stage or transitioned to progressive or dead. The transition to progressive represents the CELESTIAL trial progression definition [15]. Once in the progressive stage, patients were able only to either remain in that stage or to die.

Time horizon, cycle length and discounting
A time lifetime horizon (84 months) was selected to simulate a lifetime horizon providing a practically complete cost and utility estimation of a therapy in a palliative setting. A 30-day cycle length was adopted. The discounting of costs and utilities was performed with a rate of 3%, as recommended by the IQWiG [13].

Transition probabilities
Transition probabilities were deduced from published Kaplan-Meier (KM) curves of the CELESTIAL trial. Parametric distributions were fitted to the KM data for PFS and OS via minimizing the sum of squared residuals to extrapolate from the trial duration to the 84-month horizon. We followed the recommendations of Ishak et al. fitting five parametric distributions to the KM data: exponential, Weibull, Gompertz, log-logistic and lognormal [18]. The best-fitting distribution for all curves was Weibull selected via the sum of squared residuals, Akaike information criterion, Bayesian information criterion and the long-time hazard of the distribution, matching the characteristics of a palliative setting. Goodness of fit values are presented in Table 1 and the detailed transition probabilities and distribution parameters are described in Additional file 1.

Utilities
Evaluating quality of life (QoL) represents an essential step in determining the effectiveness of novel therapies with high rates of adverse events (AEs). Abou-Alfa et al. published the differences in mean total QALYs during cabozantinib treatment with a significant increase of 0.092 for the entire follow-up using the EQ-5D-5 L QoL questionnaire without reporting total QALYs. The study was limited by low questionnaire return numbers (82-100%) [19]. In our base case, we used 0.76 for stable and 0.68 for progressive. These estimations refer to the findings of Thomsen et al. about QALYs in sorafenibtreated RCC and were used in many cost-effectiveness analyses, including the submission about sorafenib for HCC to the British National Institute for Health and Care Excellence (NICE) [20][21][22]. The findings of Bruix et al. examining the QALYs of the RESORCE population under regorafenib therapy support these values (0.76 under regorafenib and 0.77 under placebo) [23]. As described under Target Population, Setting and Perspective, we found it appropriate to use these values. QALY reductions by AEs were not included into the base case QALYs, because the high AE rates would lead to lower QALYs in the cabozantinib group compared with BSC disagreeing with the QoL findings of Abou-Alfa et al. described above.

Cost calculation General
We considered direct medical costs, including drug, monitoring and AE costs using $ for easier comparison, converted by purchasing power parities of 2019 with 0.741 €/$ and 0.689 £/$ as the exchange rates [24]. The frequency and forms of supporting health items were primarily deduced from German clinical practice guidelines and completed by the study protocol recommendations [15].
The German SHI perspective requires health items to be matched with the German diagnosis related groups (DRG) system for hospitalizations and the German Uniform Value Scale catalog for outpatient procedures [25]. DRG values were estimated using the DRG-Research Group Webgrouper. Drug prices and reimbursement amounts were deduced from the pharmacy database Lauer-Taxe of 15th April 2019.
In contrast to Germany, US prescription drug prices have no standardized maximum prices and are affected by multiple rebates and reimbursement programs. We determined the model costs using the US drug price portal GoodRX.com via extracting the average cash prices in April 2019. We estimated physician outpatient fees, other services and hospitalizations using the 2019 physician fee schedule, clinical laboratory fee schedule and Medicare-Severity DRG classifications and software (HCPCS-DRG V1.0 Software) of Centers for Medicare and Medicaid Services and the methods of Tumeh et al. [26].

Costs of cabozantinib medication
Producers of patented drugs and the SHIs negotiate discounts for every newly approved drug in Germany regarding the acknowledged added benefit by the GBA and the costs of appropriate alternative therapies According to the CELESTIAL trial protocol, cabozantinib was applied until radiographic progression or discontinuation induced by high-grade AEs [15]. The median time to cabozantinib discontinuation (3.8 months) and the rate of discontinuation owing to AEs related to the cabozantinib trial regime (16%) were the only available published data on the time to discontinuation [10]. Regarding the lack of explicit data on the discontinuation rate over time and the minority of toxicityrelated discontinuations, we considered it appropriate to include cabozantinib medication costs in all stable months.

Costs of disease monitoring
The German HCC guideline and the CELESTIAL study protocol recommend dynamic contrast-enhanced magnetic resonance imaging or computed tomography scan every 2 months as follow-up imaging [15,27]. The mean costs of both imaging methods were used to match these conditions. The chosen laboratory panel was constructed following the German HCC guidelines on sorafenib for HCC [27].

Costs of treating adverse events
Our model included treatment-related AEs of every grade reported in ≥5% of patients in either treatment arm in the CELESTIAL trial. We performed a chisquared test to exclude AEs that demonstrated no significant (p < 0.05) difference between the two intervention groups to decide whether it was caused by cabozantinib (see Additional file 1). Furthermore, clinically related AEs such as nausea and vomiting were summarized. Required health items included drugs, outpatient services and hospitalizations and were determined along the grade definitions of the Common Terminology Criteria for Adverse Events (CTCAE), Version 4.3. We considered no differences between both comparator groups regarding types, frequencies and costs of the chosen health items. AEs that did not require reimbursable therapy were excluded. The included AE costs were the sum of every grade costs multiplied with its grade incidence.
Adverse events requiring permanent therapy (e. g. hypertension grade 1 requiring antihypertensive drugs) were treated with health items using daily standard dosages with 30 doses per cycle and were matched with an adequate drug compound. The amounted costs were included into all stable cycles until progression. If an adverse events needs a defined treatment with a temporal limit and no need of repetition (e.g. urinary tract infections grade 1 requiring antibiotics for a standardized period), the matching drug compound and its costs were only included into cycle 1 of the state stable. Adverse event grades requiring hospitalizations (e.g. hypertension grade 4, a hypertensive emergency) were matched to an adequate DRG. As we assumed only one hospitalization case per adverse event per patient, we included the costs only in cycle 1 of state stable. All estimated costs per item of both countries are listed in Table 2 below.

Cost-effectiveness thresholds
Because Willingness-to-pay described through ICERs stays difficult to measure and both national healthcare systems have no cost-effectiveness thresholds, we The listed AE costs are already incidence-weighted. AE adverse event, BSC best supportive care. a Deviating costs of second and following months in brackets estimated cost-effectiveness thresholds as multiples of the GDP per capita deduced from the 3 times GDP per capita rule of the WHO [12]. The primary costeffectiveness threshold was the 3 times GDP per capita per gained LY or QALY and was completed by factor 6 and 9 for a broader overview. Table 3 shows the most current GDPs and the deduced thresholds [28].

Sensitivity analyses Cost analyses
The impact of the difference between the median PFS (5.2 months) and median time to discontinuation of cabozantinib (3.8 months) was investigated in two ways. First, cabozantinib therapy was applied for only four standardized cycles (months) even without progression. The second approach was subtracting a monthly cabozantinib price if a patient suffered progression to simulate the case of progression 1 month after discontinuation.

Utility analyses
The high grade of uncertainty regarding QoL under cabozantinib therapy as described in Utilities was investigated through multiple values from findings on QoL regarding HCC and TKI. Approaches and utilities are presented in  [30,31]. A last method was to use a monthly decreasing utility in a progressive state simulating constantly decreasing QoL after progression.

Univariant and probabilistic sensitivity analyses
A univariant sensitivity analysis varied major model inputs at defined intervals to determine their impact. Our probabilistic sensitivity analyses were conducted to explore the likelihood of cabozantinib reaching costeffectiveness by varying major model inputs simultaneously in 10,000 iterations. We used a gamma distribution for costs following the recommendations of the IQWiG to use a distribution with positive skewness and no upper limit [13]. The IQWiG recommended discounting rate borders of 0 and 5% for sensitivity analyses [13]. These rates were implemented via a beta distribution providing the defined borders and a mean of 3%. A normal distribution offered a simple option to include determined confident intervals of utilities into the probabilistic sensitivity analysis. Detailed distribution data are presented in the Additional file 1.

Sensitivity analyses
The results of our univariant and cost-threshold sensitivity analyses are illustrated with tornado diagrams in Figs Table 5.
The mean results of our probabilistic sensitivity analysis were illustrated through an ICER scatter plot (Figs. 4 and 5). The probabilities of achieving the determined cost-effectiveness thresholds are presented in Table 6.

Discussion
Treating HCC after failed sorafenib therapy is effective in our base case with a mean survival gain of 2.2 months versus BSC, supporting the findings of Abou-Alfa et al. and the IQWiG rating of the survival benefit as considerably downgraded to minor owing to the high AE rate [9]. All our scenarios and iterations of our probability analyses showed superior effectiveness and much higher costs of cabozantinib therapy compared with BSC. The base case ICERs were not costeffective using the 3 times GDP per capita/LY (QALY) thresholds of both models. Cabozantinib drug proportion represented the main cost driver generating over 90% of all costs and showed that combined with QoL utilities, it had the largest impact on ICER in the univariant sensitivity analysis. In addition, our cost-threshold analysis indicates that cabozantinib requires high rebates, reducing its list price to be cost-effective. The high rates of AEs during placebo therapy underline the massive disease burden of HCC even in populations with Eastern Cooperative Oncology Group score 0 or 1 performance status. In our analysis, AEs induced an increment of $1232 (Germany) and $4955 (US) between the groups but represented, together with other additional costs, less than 10% of the treatment-related costs.
Our analysis has considerable limitations, which we sought to address through our sensitivity analyses. First, a fundamental limitation of our study was the reliance on published data from the CELESTIAL trial. In 2016, the average age of HCC onset was approximately 70 in Germany [3]. Therefore, the real-world effectiveness of cabozantinib might be worse. Further, the lack of suitable QoL data was solved by testing multiple approaches, but  all examined combinations were not cost-effective. The relevance for therapy decisions, e.g., between cabozantinib or regorafenib, and the massive impact on costeffectiveness indicate the need for more standardized QoL reporting in large intervention trials. Ultimately, the practical value for decision makers of our threshold is very unclear, as the GBA has other approaches for assessing costeffectiveness, and willingness-to-pay remains difficult to measure [32]. Assuming higher cost-effectiveness thresholds led especially in the German model to increased probabilities that cabozantinib is a cost-effective therapy. IPSEN Pharma refused to submit a cost-effectiveness analysis to the NICE about cabozantinib for HCC because no scenario reached the cost-effectiveness thresholds [33]. In July 2016, the National Health Service (NHS) England established the Cancer Drugs Fund to provide early access to new therapies with unclear effectiveness, but cabozantinib for HCC is not in the current list for January 2020, meaning no reimbursement of cabozantinib [34,35]. There are two cost-effectiveness analyses of second-line therapy for HCC from the US payer perspective with non-cost-effective ICERs ranging from $469,374/QALY to $1,040,675/QALY [36][37][38]. These findings and our US pricing scenario match the development of the United States financing the most expensive healthcare system of all countries of the Organisation for Economic Co-operation and Development with expenditures of 16.9% of the gross domestic product (Germany 11.2%) [39]. The high difference ICERs between the US and German scenarios is mostly caused by US drug prices, which are often much higher than German prices (Table 2). Facing various new patented  drugs, German lawmakers set maximum prices and statutory rebates via the AMNOG to limit annual drug spending increases [40]. In consequence, the prescription drug price index decreased 15% from 2004 to 2018, estimated after statistics of the AOK Research Institute [41]. In contrast, the US prescription drug price index increased 56% in the same period but decreased in 2019 as the first time since 1974, estimated after statistics of the US Bureau of Labor Statistics [42].
Regorafenib, with its similar survival outcomes, was investigated by two cost-effectiveness analyses, none of which found cost-effective thresholds with ICERs ranging from $201,797 to $277,463/QALY compared with BSC from US payer perspective [43,44]. The comparability to our US price scenario is limited because these analyses used the much lower drug prices of 2017 and before. As the average cash price of a monthly regorafenib dose in April 2019 was $22,149 according to GoodRx.com (cabozantinib $21,581), current ICERs might be more similar.
Treating advanced medullary thyroid carcinoma with cabozantinib was incorporated in the NHS catalogue after the NICE assessed an ICER versus BSC lower than its cost-effectiveness threshold of £30,000/QALY ($43, 541/QALY) [45]. In contrast, the review and economic model of Tappenden [53,54]. The chimeric antigen receptor T cell therapy for relapsed or refractory pediatric B cell leukemia with its one-time infusion cost of $475,000 was investigated by Lin et al., estimating an ICER of $184,000/QALY versus blinatumomab or chemotherapy from a US payer perspective [55]. The NICE also assessed tisagenlecleucel as not cost-effective but recommended its reimbursement by the Cancer Drugs Fund [56]. Considering all the described findings, our results rank in the upper ICER range of new therapies.
In 2019, there were multiple ongoing phase 3 trials investigating other TKIs or checkpoint inhibitors for HCC [57,58]. After nivolumab demonstrated effectiveness in a phase 1/2 trial as second-line therapy for HCC, a phase 3 trial for first-line therapy will finish in 2020. Additionally, the impact of the sequential or combined use of TKIs and checkpoint inhibitors will be evaluated, e.g., through a phase 3 trial of cabozantinib and atezolizumab as first-line therapy for HCC [59].

Conclusion
In conclusion, cabozantinib is an effective but not costeffective second-line therapy for HCC, as evaluated in two different healthcare systems. Our primary costeffectiveness threshold was not achieved in our base cases or almost all our scenarios. The German model had a higher probability to reach the cost-effectiveness thresholds than the US model caused by the much higher drug prices. The main cost driver is the cabozantinib drug price, and the highest uncertainty arises from QoL inputs.
Additional file 1 Supporting data file. This excel file contains cited supporting information and supplementary material.