The baseline sample for our model was constructed to mirror the CheckMate 9ER trial (Choueiri et al., 2021). The mean age of the individual cohort was 62 years, and all patients had clear cell-type aRCC (Supplementary Table S1). Patients entered the model with untreated aRCC and received either nivolumab (240 mg every 2 weeks) plus cabozantinib (40 mg once daily) or standard sunitinib (50 mg once daily for 4 weeks of each 42-day cycle). After first-line failure, patients who experienced disease progression subsequently received axitinib (5 mg twice per day) and sorafenib (400 mg twice per day) as second-line and third-line treatments, respectively. This predefined treatment sequence was set based on NCCN clinical practice guidelines in oncology: kidney cancer, which listed currently available treatment of aRCC and recommended axitinib and sorafenib as subsequent treatment (National Comprehensive Cancer Network, 2021). All administration and dosage schedules for every line of treatment were obtained from the respective RCTs and are listed in Supplementary Table S2.

We created a microsimulation model to estimate the healthcare cost and clinical benefits associated with nivolumab plus cabozantinib versus sunitinib for patients with treatment-naïve aRCC using TreeAge Pro (TreeAge Software, Williamstown, MA). As illustrated in Figure 1, patients commenced treatment with a nivolumab-plus-cabozantinib or sunitinib followed by axitinib-sorafenib-best supportive care (BSC) treatment sequence until death. A 42-day model cycle was used to match the time interval with the CheckMate 9ERtrial and a lifetime horizon to assess direct healthcare costs and utilities related to each treatment arm (Choueiri et al., 2021). The primary outcomes of the model were used to estimate the additional cost for nivolumab-plus-cabozantinib treatment compared with sunitinib in 2021 US dollars for an incremental quality-adjusted life-year (QALY) yield (incremental cost-effectiveness ratio (ICER)). This study was performed from a US payer perspective with a willingness-to-pay (WTP) threshold of $150,000/QALY (Neumann et al., 2014), and both cost and utilities were discounted by 3% annually (Weinstein et al., 1996).

Patients transitioned between different health states based on transition probabilities calculated from the OS, PFS and discontinuation rate from multiple RCTs associated with the respective strategies (Rini et al., 2011; Motzer et al., 2014; Rini et al., 2020; Choueiri et al., 2021). First, the PFS survival curves from CheckMate 9 ER, AXIS, TIVO-3, and an RCT conducted by Robert J were used to estimate the probability of patients remaining in the PFS state of each treatment line by performing the standard extrapolation method designed by Guyot et al. (Guyot et al., 2012). Getdata Graph digitizer software was used to extract survival data points from survival curves to reconstruct pseudo-individual patient-level data (pseudo-IPD). Next, those pseudo-IPD data were fit to five standard parametric models (Weibull, exponential, lognormal, gamma, and log-logistic distributions), and the most appropriate distribution was selected for all curves based on the goodness of fit (Akaike information criterion) (Li et al., 2021). The PFS data of sorafenib in the two trials (Motzer et al., 2014; Rini et al., 2020) were pooled given the patient baseline characteristics and comparable trial eligibility criteria (Supplementary Table S1) between those two RCTs, similar to the analyses by Wu et al. (Wu and Shi, 2020). We used a log-logistic distribution to model survival, and all the survival parameters are listed in Table 1.

Second, we also took the discontinuation rate associated with AEs into account, with transition probabilities collected from the literature (Rini et al., 2011; Motzer et al., 2014; Rini et al., 2020; Choueiri et al., 2021). Finally, the probability of transitioning to death of each model cycle was defined as the value of combining data concerning treatment-related serious AEs from respective RCTs with an age-specified background mortality rate from the 2019 US Life Table (Arias and Xu, 2019) and observed mortality rate using survival data from each trial (Rini et al., 2011; Motzer et al., 2014; Rini et al., 2020; Choueiri et al., 2021). The probability of death from the BSC phase was calculated on the basis of the OS curve of the RECORD-1 trial using the same approach with the transition probabilities of PFS (Motzer et al., 2008). Baseline evaluations of clinical transition probabilities are displayed in Table 1.

Only direct costs were adopted as follows: drug acquisition costs, administration cost, management of AEs, and BSC. The unit prices of nivolumab in the United States were derived from the Centers for Medicare and Medicaid Services (CMS) on the basis of the 2021 average sale price (ASP Drug Pricing Files, 2021). The costs of oral drugs (sunitinib, cabozantinib, axitinib, and sorafenib) not included in CMS were collected from public literature and databases. (ASP Drug Pricing Files, 2021; Su et al., 2021; Lu et al., 2020; Watson et al., 2020). Medication costs were estimated using a baseline patient with a weight of 70 kg since weight loss effects in disease were considered (Wan et al., 2019; Lu et al., 2020). The overall costs associated with the management of grade 3 or 4 AEs and BSC were derived from the previous literature (Perrin et al., 2015; Wan et al., 2019; Lu et al., 2020; Agency for Healthcare Research and Quality, US Dept of Health and Human Services, 2021). The drug infusion cost was obtained from the 2021 CMS Physician Fee Schedule, with the duration of drug administration based on the CheckMate 9ER trial (Centers for Medicare and Medicaid Services, 2021).

The health utility scores, which range from 0 (death) to 1 (perfect health), reflect the value of QoL in a particular health state. Based on previously published studies, we set the utility values of first-, second-, and third-line treatments and the BSC phases to 0.82, 0.77, 0.66, and 0.494, respectively (Cella et al., 2018; de Groot et al., 2018; Wan et al., 2019; Patel et al., 2021a). We also considered the utility decrement (−0.157) due to AEs (Wu et al., 2018). QALYs were estimated by multiplying the time duration in a specific state by the utility value related to that state.

To test the uncertainty in evaluating input parameters and to assess model robustness, a series of sensitivity analyses, including one-way sensitivity analyses and probabilistic sensitivity analyses (PSAs), were performed. In accordance with established methods, costs were changed by 20% from their baseline values, and the upper and lower bounds were varied over the 95% CI of variables for those parameters with CIs such as utilities (Kohn et al., 2017; Zhang et al., 2012; Goulart and Ramsey, 2011). In the one-way sensitivity analyses, the value of one variable at a time was changed within a predefined range to explore the individual impact of each variable on ICERs for all parameters in Table 1. In PSA, we performed a Monte Carlo simulation of 5,000 iterations of 2000 patients to account for the change in all input parameters at once. All parameters were randomly sampled from the specific distributions. According to recommended distributions in accordance with previous cost-effectiveness analysis, we assumed a gamma distribution for costs, a beta distribution for utility values and incidence of AEs, and a normal distribution for both the weight and starting age of patients (Kohn et al., 2017; Patel et al., 2021b). Based on the PSA, a cost-effectiveness acceptability curve was obtained and used to illustrate the probability that the two treatment strategies could be regarded as the most cost-effective under different WTP thresholds.

We also included three scenario analyses in this study. In the first scenario analysis, we varied the nivolumab to 75, 50, and 25% of its original price. In the second, the time horizon was changed to 5, 10, and 15 years to evaluate the impact of the OS and PFS extrapolations used in the model. In the final scenario analysis, we set patients who would experience a certain proportion (18.9% in the nivolumab-plus-cabozantinib arm and 32.9% in the sunitinib arm) switching to the BSC phase after disease progression from first-line therapy, in accordance with the CheckMate 9ER and AXIS trials.

The nivolumab-plus-cabozantinib treatment strategy was associated with an improvement of 0.56 QALYs and 0.59 LYs compared with sunitinib (2.97 vs. 2.41 QALYs and 3.9 vs. 3.31 LYs, respectively). However, the nivolumab-plus-cabozantinib strategy was associated with dramatically greater healthcare costs ($483,352.70 vs. $198,320.10, respectively), with an additional cost of $285,033. The ICER of nivolumab plus cabozantinib as a first-line treatment was $508,987/QALY compared with standard sunitinib (Table 2).

The one-way sensitivity analyses showed that patients’ age of starting treatment, the utility of first-line therapy, and the drug cost of nivolumab had a considerable influence on model outcomes. Other variables, such as the drug costs of axitinib and sorafenib, the utility of third- and fourth-line therapy, and weight, had a moderate impact on our estimated ICER (Figure 2). The PSA results showed that there was a 100% probability of the nivolumab-plus-cabozantinib strategy being regarded as not cost-effective at the WTP threshold of $150,000/QALY compared with sunitinib (Figure 3).

The first scenario analyses revealed that reductions in the drug price of nivolumab prescribed in the first-line therapy of 75, 50, and 25% would lead to lower ICERs of $410,584/QALY, $295,405/QALY, and $177,747/QALY, respectively. However, it is still unlikely to be cost-effective under the current WTP threshold. The second scenario analyses indicated that the ICERs were $929,570/QALY, $603,897/QALY, and $547,448/QALY when we adjusted the time horizon to 5, 10, and 15 years, respectively. In the final scenario analyses, a certain percentage of patients turned to the BSC phase after progressing from first-line treatment rather than receiving second-line treatment. However, the results found that the model outcome did not vary significantly due to this adjustment, with an ICER of the nivolumab-plus-cabozantinib versus sunitinib strategy of $332,839/QALY. We list all the results of scenario analyses in Supplementary Table S4.