The inclusion and exclusion criteria of the target population were derived from the IMpower132 clinical trial [5]. Inclusion Criteria: adults older than or equal to 18 years of age, histologically or cytologically confirmed stage IV non-squamous non-small cell lung cancer; with the measurable disease according to Response Evaluation Criteria in Solid Tumors (RECIST) version 1.1 [7]; Eastern Cooperative Oncology Group performance status of 0 or 1; no previous treatment for metastatic disease. Exclusion criteria: central nervous system metastasis, autoimmune disease, tumors harboring sensitizing mutations in EGFR gene or ALK genetic alterations or received previous immunotherapy.

According to the IMpower132 clinical trial, permuted-block randomization with a block size of four was used to allocate patients in a one-to-one ratio to the APP and PP before the start of the clinical trial. APP (292 patients): atezolizumab 1200 mg + carboplatin (AUC = 6) or cisplatin 75 mg/m² + pemetrexed 500 mg/m², once every three weeks for four cycles, followed by atezolizumab 1200 mg + pemetrexed 500 mg/m². PP (286 patients): carboplatin (AUC = 6) or cisplatin 75 mg/m² + pemetrexed 500 mg/m², once every three weeks for four cycles, followed by continuous treatment with pemetrexed 500 mg/m².

A partitioned survival model was constructed, and it was divided into three mutually exclusive health states (Figure 1): progression-free survival, progressed disease, and death. At the beginning of the simulation, all patients were in PFS state. After a cycle, patients could remain in PFS state or transition to PD or die, while patients in PD could only maintain PD or die [8]. The cycle is 3 weeks. The simulation until 99% of patients died. According to the “China Pharmacoeconomic Evaluation Guidelines 2020” [9], a discount rate of 5% is used and the willing-to-pay threshold (WTP) is three times Chinese per capita GDP in 2021 (CNY 242,928).

The population distribution of each state during the follow-up period can be obtained directly from the OS curve and PFS curve. After the follow-up period is exceeded, the survival function needs to be calculated by the parametric method. First, use GetData2.20 software to take points from the KM curves of the two treatment regimens, and then use R4.0.4 to reconstruct individual data based on the method of Guyot et al. [10]. After fitting, the median progression-free survival and median overall survival in the APP were 7.8 months and 17.7 months respectively (median PFS and median OS of clinical trials were 7.6 and 17.5 months), and the PP had a median PFS and median OS were 5.3 and 13.8 months respectively (median PFS and median OS of clinical trials were 5.2 and 13.6 months), the gap was within 0.2 months, and the reproducibility was good. Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) combined with visual inspection were used to select the best fitting distribution. The distributions selected for fitting include exponential, gamma, Weibull, log-logistic and lognormal. According to the lowest values of AIC and BIC shown (Table 1), lognormal distribution was selected for the PFS curves of the APP and PP, and Weibull distribution was selected for the OS curves. The parameters are shown in Table 2, and the fitted curves are shown in Figure 2 and Figure 3.

According to the “Oncology Society of Chinese Medical Association guideline for clinical diagnosis and treatment of lung cancer (2021 edition)” [18], patients in the APP were treated with docetaxel, and patients in the PP were treated with tislelizumab in subsequent treatment.

One-way deterministic sensitivity analysis (DSA) was performed to evaluate the impact of parameter changes (Tables 3-5). The analysis results are presented in the form of a tornado diagram. Through 1000 Monte Carlo simulations, the parameters were subjected to probabilistic sensitivity analysis (PSA), with Beta distribution for utility value and Gamma distribution for cost. And the results of PSA analysis are presented in the form of cost-effectiveness acceptability curves (CEAC).

The Base-Case analysis results (Table 6) illustrate that compared with PP, the incremental cost of the APP is CNY 597,040.94, the incremental effect is 0.46 QALY, and the ICER is CNY 1,296,414.83/QALY, which is much higher than 3-times Chinese per capita GDP. Therefore, it can be considered that the atezolizumab plus chemotherapy in first-line non-squamous NSCLC is not cost-effective.

In scenario analysis (Table 7), the ICER is CNY 2,055,935.04/QALY, which is still higher than 3-times Chinese per capita GDP, indicating that the APP is not cost-effective, which is consistent with the base-case results.

The results of deterministic sensitivity analysis illustrated that the highest impact on ICER was the price of atezolizumab, the utility value of PD and PFS status, and the discount rate (Figure 4). Even though the reduction in the price of atezolizumab reduces the ICER value to CNY 974,607.54/QALY, it is still higher than 3-times Chinese per capita GDP, so the APP is not cost-effective, which is consistent with the base-case results.

As the willingness to pay threshold increases, the probability that the APP is more cost-effective also increases (Figure 5). When the willingness to pay threshold is CNY 242,928, which is 3-times Chinese per capita GDP, the probability of the APP being cost-effective is 0. When the willingness to pay threshold is increased to CNY 1,200,000, the probability of the APP being cost-effective is close to 50%. In addition, the results of 1000 Monte Carlo simulations show that the average ICER value is CNY 1286191.31/QALY, indicating that the base-case analysis results are robust.