A partitioned-survival model (PSM) was established depending on the clinical data from the phase three study (CAMEL) (Zhou et al., 2021) to estimate the costs and effectiveness of the two treatments, with three health states: PFS, progressive disease (PD), and death (Figure 1). The PSM structure eliminates the need to put forward assumptions for the transition of patients between different health states and enables the direct use of the trials’ Kaplan-Meier (K-M) curves to directly divide patients into different health states. Therefore, the estimation of patients’ proportion in each health state was acquired directly from the cumulative survival probabilities in OS and PFS curves by parametric functions fitting and extrapolation. The model compared the medical cost and health outcomes for two treatments: (C group) treated with chemotherapy (carboplatin and pemetrexed) vs. (CC group) treated with camrelizumab + chemotherapy combination (camrelizumab, carboplatin, and pemetrexed). Patients could receive subsequent treatment if the disease progressed or unacceptable AEs occurred. The population was a cohort with the same characteristics as those in the CAMEL trial (patients with IIIB–IV non-sq NSCLC without EGFRm and ALKm were aged 18–70 years [Median (IQR), 59 (54–64) years and 61 (53–65) years, respectively] and male accounted for 71 and 72%, respectively). The cycle length was 21-days in keeping with the treatment schedule reported by the CAMEL trial (Zhou et al., 2021) and the time horizon was the whole life so as to ensure there were less than 1% survivors. Costs, life years (LYs), quality-adjusted life years (QALYs), and incremental cost-effectiveness ratios (ICERs) were calculated in each treatment group. In line with Chinese pharmacoeconomic guidelines (Liu et al., 2020), both costs and benefits are discounted at 5% (range: 0–8%) per year. The initial state is assumed to be PFS, and death is the absorbing state.

The model transition parameters and proportions were directly obtained from the results of CAMEL (Zhou et al., 2021). We used the GetData Graph Digitizer (version 2.26) to collect the data points from the K–M curves (PFS and OS curves) of the two arms and followed the method of Guyot et al. (2012) to reconstruct estimates of underlying individual patient data (IPD) over the clinical trial time. In terms of the IPD out of the clinical trial time, standard parametric models were used for parametric extrapolation and long-term survival estimates by using the R version 4.1.0 (https://www.r-project.org). Specifically, six parametric functions were considered including exponential, Weibull, Gompertz, log-logistic, log-normal, and generalized gamma distributions. And then, a series of methods were applied to evaluate the goodness-of-fit of each parametric survival model under appropriate circumstances, such as visual inspection and Akaike information criterion (AIC)/Bayesian information criteria (BIC) tests proposed by NICE DSU technical support document 14 (TSD14) (Latimer, 2013) (table in appendix SM1-4). Lower AIC and BIC values indicate better fit of the selected model. Superimposed graphs of the K-M curves from the trial and the predicted curves based on the best fitting parametric survival models are presented in Figure 2, Figure 3, Figure 4, and Figure 5 to intuitively inspect the survival prediction.

The log-logistic model was chosen as the best fit model for the OS curve of C arm and log-normal model for CC arm and PFS curve of C arm. Considerations were as follows: (1) the lowest AIC and BIC values among all survival models; (2) the best fit with the observed curves based on visual inspection.

The health utility score reflects the level of physical, mental, and social functioning associated with a disease correlative health state that varies from 0 to 1, with 0 representing the worst health state/death and 1 representing the best. Average health utilities of the PFS and the PD state were 0.804 (Nafees et al., 2017) and 0.321 (Nafees et al., 2017), respectively, which were obtained from the previously published literature (Nafees et al., 2017). The disutility values owing to 3/4 AEs were considered in our analysis (Nafees et al., 2008; Handorf et al., 2012), but only one-time assessment was carried out during the first cycle for simplification given the trivial influence of AE disutilities. The incidence of AEs was multiplied by the corresponding disutility value to assess the QALYs loss caused by AEs. In addition, ± 20% as the boundaries of the range were used in sensitivity analyses.

In our study, only direct medical costs were considered, including cost of the drug utilization, PD-L1 test, main AEs, treatments for progression (including active treatments and supportive care), monitoring, and terminal care. Drug prices were estimated from the local bid-winning price (Drugdataexpy). Only severe AEs with great clinical impact, including anemia, neutropenia, and thrombocytopenia, were calculated because they had a relatively considerable influence on the economic evaluation by decreasing quality of life and increasing utilization of health resource. In addition, AE costs were calculated only once in the first cycle.

Costs of monitoring, AEs, terminal care, and PD-L1 tests were obtained from previously published studies (Wu et al., 2012; Zheng et al., 2018; Jiang and Wang, 2020; Wan et al., 2020). All patients were assumed to incur one-time PD-L1 test costs in the first cycle and one-time terminal care costs before death. Additionally, costs were discounted at an annual rate of 5% (Sanders et al., 2016). All costs were converted into United States dollars (USD) by exchange rate: 1 USD = 6.47 CYN. All these data are listed in Table 1. If the ICER is below $32,457 threshold (three times GDP per capita of China in 2020, ¥210,000.00), the treatment is generally considered to be cost-effective.

According to the phase three CAMEL trial (Figure 6), the patients in the CC group (N = 205) received camrelizumab 200 mg and pemetrexed 500 mg/m² and carboplatin at an area under the curve (AUC) 5 mg/ml per min, administered as intravenous infusion on Day 1 of each 21-day cycle for 4-6 cycles followed by optional camrelizumab 200 mg and pemetrexed 500 mg/m² every 3 weeks (Q3W) maintenance for the remainder of the study or until recorded PD, intolerable AEs, death, or study completion; the patients in the C group (N = 207) received pemetrexed + platinum on Day 1 per 21-day cycle for 4-6 cycles followed by pemetrexed 500 mg/m² Q3W maintenance. Patients in the C group who were confirmed disease development had an optional crossover to camrelizumab monotherapy according to investigators’ judgment, which was also taken into account in our PSM. The total exposure of camrelizumab should not be more than 2 years. After the initial treatment, 75 (81.5%, PD 92) in the CC group and 120 (94.5%, PD 127) patients in the C group received at least one subsequent anti-tumor therapy, and we assumed that the others (17 (18.5%) and 7 (5.5%), respectively) were treated with supportive care. In the C group, 79 (62.2%) patients transformed to camrelizumab monotherapy and 9 (7.1%) received other anti-PD-1 drugs, alone or combined with other therapies. Subsequent anticancer therapies included immunotherapy and chemotherapy. Referring to the other studies, we assumed that docetaxel was used as the standard second-line chemotherapy, and nivolumab and camrelizumab as the immunotherapy, which is commonly used in China. For dosage calculation, values of body surface area (BSA), weight, and creatinine clearance rate (Ccr) were 1.80 m², 65kg, and 90 ml/min/1.73 m², respectively (Wu et al., 2011).

(Costs and health utilities inputs were summarized in Table 1)

A series of sensitivity analyses were conducted to test the robustness of the model. One-way deterministic sensitivity analyses (DSA) were used to evaluate the impact of uncertainty of a single input variable on the ICER. All parameters were adjusted within the reported 95% confidence intervals (CI) or assuming reasonable ranges of the base case values (±20%) if 95% CIs were unavailable, in accordance with established approach (Zhang et al., 2012). In the probabilistic sensitivity analysis (PSA), a Monte Carlo simulation of 1000 iterations was generated by simultaneously sampling the key model parameters from the pre-specified distributions. The variables about risk of main AEs, proportion of patients and utilities were assigned beta distributions, and the variables about costs used gamma distributions. Based on the data from 1000 iterations, a cost-effectiveness acceptability curve (CEAC) was created to represent the likelihood that the CC regime would be considered cost-effective compared with the C regime on the basis of a willingness to pay (WTP) threshold of $32,457 per QALY in China.

Over a lifetime horizon, the total QALYs for CC and C were estimated to be 1.55 and 1.16, respectively, and the PSM assessed the total costs to be $19,023.42 and $21,922.27, respectively, during that period, resulting in an ICER of $-7382.72/QALY (Table 2).

To our knowledge, this evaluation was the first study to evaluate the cost-effectiveness of camrelizumab and chemotherapy combination therapy for the target patients with metastatic non-sq NSCLC without EGFRm or ALKm in China. Under the circumstance of the decreased price of camrelizumab from $3060.3 to $452.6, our results showed that CC strategy produced an extra 0.39 QALYs while saving $2,898.85, which resulted in an ICER of $-7,382.72/QALY. These findings suggested that the CC strategy was absolutely cost-effective in China, which was also proved by the CEACs (Figure 8). Recently, some scholars (Kazibwe et al., 2021) pointed out that using 1 to 3× GDP per capita as a cost-effectiveness threshold had resulted in a higher proportion of study interventions being found as cost-effective, which might lead to a loss of would-be health gains from deployed resources for health. However, in our study, even if the opportunity cost threshold (or 0.5× GDP per capita) is used, the CC scheme is nearly 90% likely to be cost-effective, which means that our results are robust.

Compared with other PD1/PD-L1-chemotherapy combination therapeutic regimes, including pembrolizumab, atezolizumab, and sindilimab, which were approved in China or abroad as the first line therapy for patients with metastatic non-sq NSCLC without EGFRm or ALKm, only camrelizumab-chemotherapy combination therapeutic regime showed cost-effectiveness in China according to current research. For pembrolizumab, Jiang et al. (2020), Wan et al. (2020) and Wu et al. (2020) had developed studies to assess the cost-effectiveness of combination therapy by partitional survival model, Markov model, and decision tree plus Markov model, respectively. They found that the ICERs of combination treatment vs. chemotherapy alone were $96,644/QALY, $92,533/QALY and higher than $40,000/QALY, respectively, which highly exceed the WTP threshold in China (3 times of GDP per capita). So that, compared with chemotherapy alone, the combination treatments showed no cost-effectiveness. In terms of atezolizumab, Yang et al. (2021) found that the probability of atezolizumab plus chemotherapy (ICER $325,328.71/QALY) was cost-effective was 0% at a WTP value of $30,828/QALY (Table3). Based on these findings, our study may be useful for national medical insurance negotiation and decision-making.

A major strength of the basic analysis was its dependence on a direct comparison of CC and C alone, using data and information from a randomized controlled trial. The second strength was that the population included in the trial and analysis were Chinese, which eliminated the influence of different races. Besides, the PSM structure was unnecessary to produce assumptions for the transition of patients between health states but enabled researchers to partition patients to different health states directly based on the trials’ K-M curves.

Analysis limitations were as follows: First, for the longer-term extrapolation of PFS and OS, there existed an inherent uncertainty. Besides the choice of follow-up treatment would be different according to the individual situation of patients, for which additional data from real world studies could help in validating the model over the long-term. Second, the values of utilities and disutilities were from published studies, some of which were not based on Chinese populations, and the differences caused by different treatment methods were not distinguished. Nafees et al. (2017) figured out that the values of utilities for NSCLC in China were higher than other countries. Third, the published data of the phase three CAMEL trial was the interim results with a median follow-up duration of 11.9 months (IQR 9.0–14.9). Thus, it might be possible that the prediction of cost-effectiveness could be changed with additional follow-up. In addition, this study was a secondary analysis of data from the published literature which might limit the conclusions.