The target patient population was consistent with that from CheckMate649. The trial enrolled 1581 patients with previously untreated, unresectable, advanced or metastatic GA, GEJA and EA, all aged 18 years or older, from 175 hospitals and cancer centers in 29 countries.

Patients were randomly assigned to the nivolumab plus chemotherapy group (789 patients) and chemotherapy alone group (792 patients). In chemotherapy alone group, 47% of patients received XELOX (oxaliplatin 130 mg/m², day 1, and capecitabine 1000 mg/m² twice a day, days 1 - 14, every 3 weeks) and the other received FOLFOX (oxaliplatin 85 mg/m², day 1, leucovorin 400 mg/m², day 1, and fluorouracil 400 mg/m², day 1 and 1200 mg/m², days 1 - 2, every 2 weeks). In the nivolumab plus chemotherapy group, 46% of patients received nivolumab plus XELOX (nivolumab 360 mg, plus XELOX, every 3 weeks), and the other received nivolumab plus FOLFOX (nivolumab 240 mg, plus FOLFOX, every 2 weeks). Besides, the administration time of nivolumab could not exceed 2 years. Therefore, patients received Subsequent therapy after two years regardless of disease progression.

Microsoft Excel 2016 was used to develop a partitioned survival model from the perspective of the Chinese healthcare system. The model structure consisted of three states, including progression-free survival (PFS) state, progressed disease (PD) state, and death state (as shown in Figure 1). All patients entered the model in the PFS state. According to CheckMate649, the circulation cycle of the model was set to 6 weeks to simulate the patient’s lifetime until 99% of patients died. The main outputs of the model were the cost, quality-adjusted life years (QALY) and incremental cost-utility ratio (ICUR). According to the China Guidelines for Pharmacoeconomics Evaluations (2020), both cost and utility used a discount rate of 5% annually and used 0% - 8% for deterministic sensitivity analysis [7]. What’s more, three times the gross domestic product (GDP) per capita of China in 2020 (CNY 216,000) was used as the willing-to-pay (WTP) threshold [8].

During the trial follow-up period, the proportion of patients in each health state was obtained from Kaplan-Meier (KM) curves in CheckMate649, using GetData Graph Digitizer2.20. Beyond the trial period, extrapolation was constructed to estimate the proportion of patients by parameter method, using R4.1.2. Individual patient data were reconstructed by six distributions, including Exponential, Gamma, Gompertz, Weibull, Log-logistic and Log-normal distribution. The

proportion of patients calculated from PFS curve represented patients in PFS state. The proportion of patients calculated from OS curve represented all patients who are alive, including patients in PFS state and PD state.

Optimal survival distribution for extrapolation could be determined according to Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC), combined with a visual inspection. As shown in Table 1, for OS curves, log-logistic distribution was used for both groups with the best AIC and BIC scores. For PFS curves, log-normal was used for the chemotherapy alone group and Log-logistic distribution was used for nivolumab plus chemotherapy group. The fitting curves could be drawn according to relevant parameters in Table 2. As depicted in Figure 2 and Figure 3, the fitting results were consistent with the original results, exhibiting high reproducibility.

Only direct medical costs were included in this study, including the cost of drugs, disease management, subsequent therapies and the management of adverse events

Table 3). It is assumed that the average body surface area and weight were 1.72 m² [9] and 65 kg [10] respectively. The prices of medicines were all sourced from MENET.

The disease management costs were sourced from HAN Jiaqi [11], including drug administration, hospitalization, imaging and PD-1 testing cost. The PD-1 testing, performed once before starting treatment, was considered a one-time cost for this study. The tumor imaging assessment was done every 6 weeks for the first 48 weeks and every 12 weeks thereafter.

Grade 3 or higher adverse events that occurred in ≥5% of patients in either treatment in CheckMate649 were included in the model and considered a one-time cost. As illustrated in Table 3, adverse events of the nivolumab group versus chemotherapy group were anemia (6.0% vs. 2.7%), neutrophil count decreased (25.7% vs. 20.9%), lipase increased (5.8% vs. 2.1%). The related costs were derived from other cost-utility analysis from the Chinese perspective [12] [13].

The costs of subsequent therapies could be calculated by costs from MENET and the proportion given in CheckMate649.

Note. PFS: progress-free survival; PD: progressive disease.

The utility values of the three health states of PFS, PD and death in the model were 0.797, 0.577 and 0, respectively, and were sourced from published literature [14].

Deterministic sensitivity analysis (DSA) was carried out to evaluate the influence of parameter changes within a certain range. The variation ranges of parameters were 0% - 8% for the discount rate and ±20% of the base value for the other (as shown in Table 3). The analysis result is presented with a tornado diagram.

In addition, probabilistic sensitivity analysis (PSA) was conducted using 1000- times Monte Carlo simulation. The analysis result is presented in the form of a cost-effectiveness acceptability curve (CEAC) to evaluate the robustness of the model.

According to the China Guidelines for Pharmacoeconomics Evaluations (2020), ICUR value between 1 time and 3 times the GDP per capita of China indicates that the regimen is economical, and ICUR lower than 1 times the GDP per capita of China indicates that the regimen is extremely economical.

Therefore, in this study, the WTP thresholds were set to be about 1 time and 3 times the GDP per capita of China, respectively, in order to provide a price reference for medical insurance negotiations.

The results of the base-case analysis shown in Table 4 indicated that compared with the chemotherapy alone group, patients in the nivolumab plus chemotherapy group received more benefits but cost more. It is calculated that ICUR value was CNY 267498.67 per QALY, far more than 3 times the GDP per capita of China in 2020 (CNY 216,000). Thus, under 3 times the GDP per capita of China in 2020 as the WTP threshold, nivolumab plus chemotherapy was not a cost-effective option over chemotherapy alone.

According to tornado diagram depicted in Figure 4, the ICUR value was more sensitive to the utility value of PFS state, the cost of nivolumab, discount rate, the cost of subsequent therapy for nivolumab group and chemotherapy group, and the utility value of PD state, etc. In addition, the cost of testing and adverse event management had a weaker influence on the ICUR value. Generally speaking, the ICUR value ranged between CNY 216919.36/QALY and CNY 325246.55/QALY, all higher than 3 times the GDP per capita of China in 2020 (CNY 216,000).

Note. ICUR: incremental cost-utility ratio.

Therefore, nivolumab plus chemotherapy was not more cost-effective than chemotherapy alone.

The results of the Monte Carlo simulation are shown in Table 5, compared with the results of the base-case analysis. The results showed that both mean and median ICUR value in the Monte Carlo simulation were high than 3 times the GDP per capita of China in 2020 (CNY 216,000), illustrating that nivolumab was not economical.

The results of probabilistic sensitivity analysis in the form of CEAC are depicted in Figure 5. Nivolumab plus chemotherapy group became more economical with an increase in the willingness-to-pay threshold. When the willingness-to-pay threshold reached CNY 216,000/QALY, the probability of cost-effectiveness of camrelizumab plus chemotherapy was approximately 29% was consistent with those of base-case analysis, proving that the results were robust. Under the WTP threshold of CNY 265,580/QALY, two curves intercross, which means that nivolumab plus chemotherapy has a 50% chance of being cost-effective compared to chemotherapy alone. Therefore, the results of probabilistic sensitivity analysis were basically consistent with those of the base-case analysis, proving that the results of base-case analysis were robust.

Note. ICUR: incremental cost-utility ratio.

As mentioned above, the cost of nivolumab had the greatest impact on ICUR value. Thus, further study was carried out to explore the price of nivolumab when it was economical.

According to Table 6, when the price of nivolumab (4 ml: 40 mg) was CNY 3652.71, the ICUR value was calculated to be CNY 215999.58/QALY, lower than 3 times the GDP per capita of China in 2020 (CNY 216,000), indicating that nivolumab was economic.

When the price of nivolumab (4 ml: 40 mg) was CNY 1041.01, the ICUR value was calculated to be CNY 71999.73/QALY, lower than 1 time the GDP per capita of China in 2020 (CNY 72,000), indicating that nivolumab was extremely economical.