Following the modelling guidelines by the International Society for Pharmacoeconomics and Outcomes Research (ISPOR) (21–23), a partitioned survival model (PSM) based on a Markov decision model was constructed to estimate the treatment costs and effects of pembrolizumab+axitinib and sunitinib, respectively. This model included three different health states: PFS, progressed disease (PD), and death ( Figure 1 ). For each treatment arm, the simulated cohort included 1,000 patients who were all assumed to meet the requirements for treatment, and to start treatment in the PFS state. Treatment with the intervention or the comparator would continue until disease progression or death, whichever occurred first. Upon entering the PD state, patients would either receive second-line chemotherapy or best supportive care. Discontinuation of treatment due to adverse events (AEs) was assumed and included in the model based on KEYNOTE-426 trial data (18) as follows: 21% of patients in the pembrolizumab+axitinib arm were assumed to permanently stop receiving pembrolizumab; 20% stopped being treated with axitinib. In the sunitinib arm, 12% of patients were assumed to discontinue treatment due to AEs (18). In both arms, discontinuation was modelled to occur after five cycles (15 weeks) based on corresponding data for the pembrolizumab+axitinib arm in the KEYNOTE-426 trial (16).

The model cycle length was three weeks. A time horizon of 15 years, tantamount to a lifetime horizon, was applied. Regarding the transition of patients to a different health state within a cycle, half-cycle correction was applied (23–25). The median age for each patient group – 62 years for pembrolizumab+axitinib, 61 years for sunitinib – was taken from the KEYNOTE-426 trial data (16, 18).

The primary model outputs were total costs, quality-adjusted life years (QALYs), and the incremental cost-effectiveness ratio (ICER). Life years (LYs) were used as an additional output. The primary data source for the model consisted of the results from the KEYNOTE-426 trial (16, 18). Further data were taken from publicly available databases (e.g., the Dutch Central Statistics Bureau [CBS]) and literature published in peer-reviewed journals. For the construction of the model, Microsoft Excel^® (Microsoft Corporation, Redmond, Washington, USA) was used.

For each treatment arm, parametric extrapolation curves were fitted to the Kaplan-Meier (KM) survival curves presented in the KEYNOTE-426 trial reports (16, 18). WebPlotDigitizer was used to obtain detailed values for time and PFS/OS probability throughout the clinical follow-up period. The obtained data at time points of every 1.5 months were applied to the curve-fitting method by Hoyle & Henley (26). The thus obtained data on progression and/or death and censorships served as input for a curve-fitting code compiled in RStudio. The gathered AIC scores and intercept and log(scale) values were included in the model to compute the extrapolation curves (exponential, Weibull, lognormal, loglogistic). For both treatment arms, the lognormal distribution was used to estimate the PFS, and the Weibull distribution to estimate the OS over the time horizon based on AIC score, face validity, and clinical plausibility (see Supplementary Table 2 for the parameters, Supplementary Figures 1–4 for the extrapolated survival curves) (27). Clinical plausibility of the extrapolation was further tested against the PFS and OS rates in the KEYNOTE-426 trial at the 42-month follow-up point (28). Moreover, assuming better long-term OS for patients treated with pembrolizumab+axitinib, survival rates for sunitinib were modelled not to surpass the rates for the combination treatment at any point within the time horizon.

Background mortality was accounted for to adjust the probability of transitioning to the death state at different ages. For the respective median ages of each treatment arm, the rate of death was calculated based on age-stratified instantaneous mortality data as well as the follow-up life years within the time horizon. Mortality data was taken from life tables for 2020 published by CBS (29). To avoid double-counting, the background mortality coefficient was adjusted for deaths by ccRCC with data taken from IKNL and published literature ( Supplementary Table 4 ) (3–5).

Grade 3/4 AE with an incidence of ≥5% were included in the model (18). In order to avoid double-counting regarding costs and dis-utilities, thrombocytopenia was assumed to be the same as decreased platelet count. The same assumption applied to neutropenia and decreased neutrophil count. The prevalence of AEs to estimate the patient-based dis-utilities and costs was taken from the KEYNOTE-426 trial reports (18).

QALYs were estimated by adjusting the gained LYs by health-related quality of life (HRQoL). Utility-related inputs based on EQ-5D data (30) were derived from literature. The baseline utility as well as utility with the pembrolizumab+axitinib treatment was assumed at 0.76, with the sunitinib treatment at 0.72; the utility in the PD state was assumed at 0.66 for all patients (31, 32).

For AE-related dis-utilities, estimates were taken from published literature as well as from publications by ZIN on comparable treatments for the same indication (14, 33–35). Where dis-utilities for AE were not available for accRCC patients, corresponding values for comparable indications were utilized. Details are shown in Table 1 and Supplementary Table 1 . Differing utility values applied in published cost-effectiveness analyses (CEAs) on pembrolizumab+axitinib versus sunitinib were included as scenarios in the sensitivity analysis (47, 48).

In accordance with the societal perspective chosen for this analysis, healthcare and societal costs were considered. Healthcare-related costs included costs for drug acquisition, administration, AE treatment, healthcare resource use, best supportive care, and end-of-life care. Costs outside of healthcare included travel costs, informal care costs, and productivity loss ( Table 1 ; Supplementary Table 1 ). All costs were inflation-adjusted to 2021 values in euros with the Dutch Consumer Price Index as reference (40, 49). Drug acquisition and administration data as foundation for the corresponding costs were based on the KEYNOTE-426 trial (16, 18). Pembrolizumab was administered intravenously at a 200 mg dose once every three weeks for a maximum of 35 cycles (barring earlier disease progression or death). Axitinib was administered orally at a dose of 5 mg twice daily; treatment routine was assumed to last until disease progression (16, 18); Sunitinib was administered orally at 50 mg once a day for the first four weeks of a six-week cycle (16, 18). As with axitinib, treatment was assumed to continue until disease progression (18). To fit the three-week cycle applied in the model, administration of sunitinib was assumed to occur for 14 out of 21 days. Costs for AE treatment were derived from published literature (50–52).

Regarding treatment upon disease progression, 50% of patients in the pembrolizumab+axitinib arm and 60.7% in the sunitinib arm were assumed to receive second-line treatment (16). The distributions regarding these treatments were based on relevant results of the KEYNOTE-426 trial (16), as well as a Dutch expert estimate, and are shown in Table 1 and Supplementary Table 1 . The treatment durations were taken from the recommendation by ZIN regarding avelumab+axitinib for accRCC (14).

Travel costs incurred for receiving intravenous (IV) chemotherapy and for healthcare resource use at the hospitals, as well as for receiving orally administered drugs (self-administration assumed) at the pharmacy. Costs for informal care were based on the proxy good method (40, 53); published recommendations by ZIN on aRCC treatments served as reference for the hours spent on such care by an informal caregiver ( Table 1 ; Supplementary Table 1 ) (14, 39). Costs regarding absenteeism-related productivity loss were estimated through applying the friction cost method and using relevant data from CBS (30, 43–45). 20% of patients were assumed to still be working in the PFS state, with the remainder already being retired (14, 39). In the PD state, all patients were assumed to have retired. Further, productivity losses for unpaid work on part of the patient and of their informal caregiver during the friction cost period (40) were included. Such losses were assumed to be linked to first- and second-line treatment administrations (46). The costs were based on hourly proxy costs for cleaning labor in the Netherlands (40). On part of the informal caregiver, working hours lost included absenteeism and presenteeism (54). A 40% increase in working hours lost was assumed for informal caregivers of patients in the PD state (14, 39).

In accordance with the ZIN reference case, an annual constant discount rate of 4% for costs and 1.5% for effects was adhered to (30). The willingness-to-pay (WTP) threshold was set at €80,000/QALY as the applicable threshold for aRCC in the Netherlands (14, 55).

A probabilistic sensitivity analysis (PSA) was conducted to illustrate the robustness of our results. 1,000 Monte Carlo simulations were conducted; for parameters constrained between 0 and 1 (utility values and AE incidences), beta distributions were used, and gamma distributions were applied to costs and other parameters constrained to values >0 (56, 57). Multinominal data regarding subgroups of patients in the PD state was subject to a Dirichlet distribution (56). Uncertainty regarding survival data was accounted for by applying variance-covariance matrices using the Cholesky decomposition method in the PSA (56). Where possible, 95% confidence intervals (CI) were determined through applying the input parameter’s standard error (SE) values. Subsidiarily, the SE was assumed as a fraction of the parameter’s mean value. A detailed schedule of the parameters and how the probabilistic values were determined is provided in Supplementary Table 3 . A cost-effectiveness plane was constructed to show the base-case ICER and the uncertainty surrounding costs and effects. Cost-effectiveness acceptability curves (CEACs) were designed to determine the cost-effective intervention (57).

Moreover, several scenario analyses were carried out. These involved alternative distributions for PFS and OS, best-/worst-case scenarios regarding utility values, and utility values based on published CEAs on pembrolizumab+axitinib versus sunitinib (47, 48). Regarding cost parameters, hypothetical lower pricings of pembrolizumab and axitinib (decrease by 10%, 20%, and 50%, respectively) were considered. Further, a shortened first-line treatment duration of 35 and 15 cycles, respectively, was assumed. A different distribution of second-line treatment drugs based on an expert report referenced in a recommendation by ZIN constituted a further scenario. A further scenario applied a second-line treatment patient distribution in accordance with an expert estimate referenced in the ZIN recommendation on avelumab+axitinib ( Table 1 ; Supplementary Table 1 ) (14). An additional scenario exclusively included patients with an intermediate/poor IMDC risk score. Sub-scenarios to this scenario concerned price reductions and shortened first-line treatment durations as mentioned above. The relevant survival probabilities were modelled after the specific KM curves in the KEYNOTE-426 trial (18), with the same selection process for the curve-fitting distribution being applied. For all scenario analyses, all parameters that were not altered in view of the respective scenario were held constant.

Based on the base case results of the CEA, a budget impact analysis (BIA) was conducted from the national-governmental perspective of the Dutch government as the budget holder. Following the ZIN reference case, the analysis included healthcare-related costs over a time horizon of three years (30). Costs were not discounted (58). The patient population was, as a first step, derived from kidney cancer incidence figures for 2019 through 2021 (3). In order to estimate the yearly number of new patients with accRCC, the respective shares of patients with RCC (90%) and ccRCC (70% of RCC patients) were taken from existing studies (4, 5). Further literature was utilized to account for patients whose disease has already advanced at the time of diagnosis (33%) and, respectively, would progress to the advanced stage despite surgical resection (35% as mean value of 20%–50%) (6). The share of accRCC patients eligible for the treatment was derived from the KEYNOTE-426 trial reports (16, 18). No transfer of patients from the standard treatment to the alternative treatment was assumed. To perform the analysis, the BIA calculation tool by The Netherlands Organization for Health Research and Development (ZonMw) was utilized (59). Three scenarios were considered to account for the uncertainty of the cost parameters. The main scenario was built on the base case results of the CEA, the second scenario concerned a 15-cycle treatment for both treatment arms and prices for pembrolizumab and axitinib reduced by 50%, and the third scenario exclusively involved patients with an intermediate/poor IMDC risk score. For this particular scenario, patient data regarding that risk group were obtained from a real-world study on aRCC patients in the Netherlands (60).

A sensitivity analysis was performed to account for the uncertainty regarding patient and cost data. In accordance with the calculation tool’s capacities, ranges of 90%–115% for the patient population and of 80%–130% for costs were assumed. Moreover, given that the base case of the BIA assumed an instantaneous and full replacement of sunitinib by pembrolizumab+axitinib on the market, a scenario analysis was included that assumed a market penetration of the combination treatment of 25% in the first year, 50% in the second year, and 80% in the third year.

The extrapolated survival curves derived from the KM curves for OS and PFS in the clinical trial are shown in Supplementary Figures 1–4 . For pembrolizumab+axitinib, the median PFS rate was estimated at approximately 15.5 months, compared to approximately 9.8 months for sunitinib. According to the extrapolation in the model that was adjusted for background mortality, the median OS rate was estimated at 42.9 months for pembrolizumab+axitinib and 31.9 months for sunitinib. 33.7% of patients treated with pembrolizumab+axitinib were projected to reach 5-year survival compared to 25.6% of patients receiving treatment with sunitinib.

Compared with patients treated with sunitinib, the model projected a gain of 0.55 QALYs (2.79 versus 2.25 QALYs) and 0.65 LYs (4.02 versus 3.37 LYs) for patients receiving pembrolizumab+axitinib over a 15-year time horizon. Detailed cost results are shown in Table 2 .

The total costs for the pembrolizumab+axitinib treatment amounted to €401,071 versus €200,272 for sunitinib, resulting in an increment of €200,799. Acquisition costs in the PFS state were estimated at €194,055 for pembrolizumab+axitinib and at €36,633 for sunitinib. Administration costs in the PFS state amounted to €5,689 for pembrolizumab+axitinib and €80 for sunitinib. Costs for the use of healthcare resources in the PFS state were estimated at €47,861 for pembrolizumab+axitinib and at €18,849 for sunitinib; in the PD state, these costs amounted to €21,583 in the pembrolizumab+axitinib group and to €21,409 in the sunitinib group. Costs outside of the healthcare sector (such as travel costs, costs for informal care, productivity loss) were estimated at €62,930 (pembrolizumab+axitinib) versus €47,496 (sunitinib) in the PFS state and at €32,261 (pembrolizumab+axitinib) versus €27,680 (sunitinib) in the PD state, respectively.

The resulting ICER was €368,396 per QALY gained, which was higher than the set WTP threshold of €80,000/QALY. Costs only concerning the healthcare perspective amounted to €305,881 for pembrolizumab+axitinib and €125,096 for sunitinib, resulting in an increment of €180,785 and an ICER of €331,677.

According to the constructed cost-effectiveness plane ( Figure 2 ), the PSA suggested upon visual inspection that the pembrolizumab+axitinib treatment was clinically more effective, but also more costly than sunitinib since all analysis iterations were in the north-eastern quadrant. The cost-effectiveness probability at the selected WTP threshold was 0%, suggesting that the combination treatment was not cost-effective compared to sunitinib. The point of intersection of the CEACs for each treatment arm, indicating an equal probability of acceptability (i.e., 50%), was estimated at approx. €340,000 ( Figure 3 ), which supported the suggested lack of cost-effectiveness.

Table 3 summarizes the results of the scenario analyses. In none of the scenarios, cost-effectiveness under the WTP threshold of €80,000/QALY was achieved. Reducing the price of pembrolizumab and axitinib by 50% lowered the ICER to; restricting the treatment duration in the PFS state to 15 cycles decreased it further to €132,637/QALY. When only including patients with an intermediate or poor IMDC risk score in the population, the ICER slightly decreased to €357,988/QALY. None of the applied sub-scenarios showed pembrolizumab+axitinib as cost-effective. However, reducing the treatment duration to 15 cycles and the costs for pembrolizumab+axitinib by 50% decreased the ICER by more than two thirds to €118,343.

Based on the relevant patient data for 2019–2021, and accounting for the share of accRCC patients eligible for treatment, the number of patients in the base case scenario as well as in the first alternative scenario (price reduction of 50% for pembrolizumab+axitinib, 15 treatment cycles for both treatment arms) was estimated at 798 in the first year, 729 in the second year, and 781 in the third year. In the second alternative scenario (only patients with an intermediate or poor IMDC risk score), we estimated the population of new patients per year at 91% of the base case population, thus at 726, 663, and 711, respectively. At the base case, the healthcare-related costs per patient amounted to €305,881 (pembrolizumab+axitinib) versus €125,096 (sunitinib), resulting in an increment of €180,785. The expenses per patient in the first alternative scenario were estimated at €153,537 for pembrolizumab+axitinib and €103,823 for sunitinib, respectively, with an increment of €49,713. In the second alternative scenario, costs in the pembrolizumab+axitinib arm were estimated at €284,720, and at €117,610 for the sunitinib arm, resulting in an increment of €167,110. the total budget impact over three years amounted to €417.3 million at the base case. Applying 15 treatment cycles to both treatment arms in the first line and decreasing the prices of pembrolizumab and axitinib by 50% lowered the total impact to €114.8 million. Limiting the patient population to patients with an intermediate or poor IMDC risk score decreased the impact to €350.9 million. Based on the ranges applied in the sensitivity analysis, the total budget impact ranged from €300.4 million to €624.0 million at the base case. The scenario related to treatment duration and costs resulted in a budget impact range of €82.7 million to €171.6 million, and the impact for the scenario regarding patients with an intermediate or poor IMDC risk score ranged from €252.9 million to €524.7 million. The results of the analysis are summarized in Table 4 .

Accounting for increasing market penetration changed the total budget impacts to €214.9 million (base case), €59.1 million (15 treatment cycles, 50% cost reduction for both pembrolizumab and axitinib), and €180.8 million (intermediate/poor IMDC risk score).