This study followed the Consolidated Health Economic Evaluation Reporting Standards (CHEERS) guidelines (17). Eligible patients were between 18 and 75 years of age and had histologically confirmed, locally advanced, unresectable, or metastatic GC/GEJC. None of the patients had previously received systemic therapy for advanced disease. Patient characteristics and inclusion criteria were consistent with those described in the COMPASSION-15 clinical trial.

The participants were randomly assigned to receive either cadonilimab (10 mg/kg, administered intravenously) or a placebo every 21 days for up to 24 months. Both groups received concurrent chemotherapy with capecitabine (1,000 mg/m² orally, twice daily on days 1–14) and oxaliplatin (130 mg/m² intravenously on day 1), repeated in 21-day cycles for up to six cycles (XELOX regimen). Following the combination phase, the patients continued with cadonilimab or placebo as monotherapy.

Subsequent treatments, including PD-1 inhibitors, targeted agents, chemotherapy, or best supportive care, were performed in accordance with the NCCN (16) and Chinese Society of Clinical Oncology (CSCO) guidelines for gastric cancer (18) and were consistent with post-treatment strategies used in the COMPASSION-15 trial (15). Tumor assessments were performed every six weeks during the first 54 weeks after enrollment and every nine weeks thereafter.

Adverse events (AEs) were monitored with a particular focus on severe (grade ≥3) events occurring in more than 3% of patients. These included anemia, neutropenia, thrombocytopenia, and hypokalemia ( Table 1 ).

A three-state Markov model was constructed using TreeAge Pro 2022 (Williamstown, MA, USA) and R version 4.2.4 (Vienna, Austria). Health states included PFS, progressive disease (PD), and death ( Figure 1 ). The model employed a 3-week cycle length over a 10-year time horizon, representing the lifetime of the patient population and capturing over 99% mortality.

The analysis was conducted from the perspective of healthcare payers in both China and the United States. The Chinese model adopted a system-wide healthcare payer perspective, whereas the U.S. analysis focused on direct medical costs relevant to both public and private payers (25).

The model evaluated total life years, quality-adjusted life years (QALYs), incremental cost-effectiveness ratios (ICERs), incremental net health benefits (INHB), and incremental net monetary benefits (INMB). Annual discount rates were applied to both costs and utilities—3% for the U.S. and 5% for China—in accordance with established pharmacoeconomic guidelines (26, 27).

Chinese cost data were converted to 2024 U.S. dollars using an exchange rate of $1 = ¥7.1217 and were adjusted for inflation using the local consumer price index. The willingness-to-pay (WTP) thresholds were set at $40,354.27 per QALY in China (three times the national gross domestic product per capita) and $150,000 per QALY in the U.S., consistent with standards established by the WHO and U.S. healthcare payers (28).

Probabilities for OS and PFS were extracted from Kaplan–Meier (KM) curves in the ASTRUM-005 trial using the GetData Graph Digitizer (http://getdata-graph-digitizer.com), and individual patient data were reconstructed following the method described by Guyot et al. (29).

Due to limited follow-up, extrapolation was required to extend survival estimates across the model’s full time horizon (30). The reconstructed time-to-event data were then fitted with a series of parametric models, including classic models (exponential, Weibull, Gompertz, gamma, log-logistic, log-normal, and generalized gamma).

Model selection was guided by a combination of statistical goodness-of-fit based on the Akaike information criterion (AIC), extrapolation performance based on log likelihood (LogLik), and visual inspection. Using this framework, the most suitable parametric model was chosen to extrapolate KM curves for OS and PFS beyond the follow-up period of the COMPASSION-15 trial (consistent with the trial referenced earlier for treatment protocols) ( Figures 2 – 4 ). Before implementing the Cox proportional hazards (PH) model, the PH assumption—a core prerequisite for valid model inference—was validated using two complementary methods: visual inspection of log-log survival curves and quantitative assessment of Schoenfeld residuals (31, 32). While the PH assumption yielded p-values > 0.05 for both OS and PFS across all patient groups (nominally suggesting the assumption was satisfied), two critical observations indicated potential violation: crossing cumulative hazard curves between the treatment and control arms, and a non-horizontal trend in the smoothed Schoenfeld residuals. The variation in predicted hazards across different parameter distributions is shown in Figures 5 – 7 . The corresponding survival function parameters are detailed in Tables 2 – 4 .

This analysis focused exclusively on the direct medical costs associated with the management of GC/GEJC. These costs include drug acquisition, laboratory testing, enhanced computed tomography (CT), intravenous drug administration, subsequent therapies, best supportive care, end-of-life care, and management of severe adverse events (grade 3 or 4). Medication prices were obtained from public Chinese databases and institutional pricing schedules, whereas other cost components were derived from published economic evaluations and relevant literature.

Owing to the absence of a listed market price for cadonilimab in the United States, its cost was estimated using a comparative approach. Specifically, pricing was approximated based on analogous immunotherapies such as toripalimab and tislelizumab (33). Drug prices for both China and the U.S. were converted to U.S. dollars and adjusted using a price index to ensure cross-national comparability. Tables 1 and 5 summarize the clinical and cost parameters used in this analysis (19–23, 28, 34, 35).

Health outcomes in the model were adjusted using utility values obtained from previously published sources, as EQ-5D-5L (European Quality of Life-5 Dimension-5 Level) data were not directly reported in the COMPASSION-15 trial. Utility values were anchored on a scale ranging from 0 (representing death) to 1 (representing perfect health).

For patients in the PFS state, the utility was set at 0.797 based on data from the TOGA trial and calculated using the Japanese EuroQol (EQ-5D) scoring algorithm (24). The utility for patients in the PD state was 0.577, derived from evaluations conducted by the National Institute for Health and Clinical Excellence (NICE). Quality-of-life decrements (disutilities) associated with severe adverse events, including anemia, thrombocytopenia, neutropenia, and hypokalemia, were also incorporated into the model (24). All AEs were assumed to occur during the initial treatment cycle, with detailed incidence rates provided in Table 1 .

To explore heterogeneity in cost-effectiveness outcomes, subgroup analyses were performed for patients with PD-L1 CPS ≥5 and CPS <5 in both China and the United States. These analyses employed the same modeling structure and assumptions as those used in the base-case scenario. Due to the lack of subgroup-specific data on follow-up treatments, adverse event rates, or healthcare resource utilization in the COMPASSION-15 trial, these parameters were assumed to be consistent with those observed in the overall study population.

Owing to uncertainties in the key input parameters, particularly drug pricing, scenario analyses were conducted to evaluate a range of potential pricing outcomes. In the Chinese context, cost-effectiveness was assessed with and without the inclusion of a patient assistance program. The price of cadonilimab varied between $0 and $2,000 per 625 mg dose, and outcomes were compared against the country-specific WTP threshold of $40,354.27 per QALY.

In the United States, where a formal list price for cadonilimab is currently unavailable, an estimated cost of $8,600 per 750 mg dose was used. This estimate was based on price comparisons with other anti-PD-1 agents, including toripalimab and tislelizumab. Scenario analyses in the U.S. setting varied the price from $0 to $10,000 per dose to identify the maximum price at which cadonilimab would remain cost-effective under a $150,000 WTP threshold.

To address potential inconsistencies between the base-case discount rates (3% for the U.S., 5% for China) and World Health Organization (WHO) recommendations, an additional scenario was run where China’s discount rate was lowered to 3% (matching the U.S. rate). This analysis evaluated how aligning discount rates across regions would impact cost-effectiveness conclusions, particularly for long-term survival outcomes. Given the volatility of the yuan–dollar exchange rate in 2024–2025, a second scenario incorporated Purchasing Power Parity (PPP) adjustments to currency conversion—replacing the base-case market exchange rate with 2024 International Monetary Fund (IMF) PPP values (￥1 = $0.2825, equivalent to ￥3.54 = $1). Concurrently, China’s WTP threshold was adjusted to $20,243.85 per QALY to align with PPP-adjusted economic benchmarks, ensuring cross-country comparability of cost-effectiveness results under a standardized economic metric.

The robustness of the model outcomes was evaluated using one-way sensitivity analysis (OWSA) and probabilistic sensitivity analysis (PSA). In the OWSA, each key parameter was independently varied by ±20% from its base-case value to determine its influence on the ICER. The results were visualized using tornado diagrams to identify the most influential variables.

For the PSA, all model inputs were sampled simultaneously based on appropriate probability distributions, beta for probabilities and utility values, and gamma for cost parameters. A total of 10,000 Monte Carlo simulations were performed to quantify the uncertainty in ICER estimates and to calculate the likelihood that cadonilimab plus chemotherapy would be deemed cost-effective at different WTP thresholds.

Over a 10-year time horizon, the base-case analysis indicated that patients receiving cadonilimab in combination with chemotherapy achieved 1.01 QALYs at a total cost of $28,528.60. In contrast, those treated with chemotherapy alone accrued 0.67 QALYs at a cost of $11,730.98. This resulted in an incremental gain of 0.33 QALYs and an additional cost of $16,797.61, yielding an ICER of $5,0582.10 per QALY for the combination therapy ( Table 6 ).

When compared to China’s WTP threshold of $40,354.27 per QALY, this ICER exceeded the acceptable limit. Consequently, the incremental net health benefit (INHB) was -0.08 QALYs, and the incremental net monetary benefit (INMB) was -$3,396.52, suggesting that cadonilimab plus chemotherapy is not cost-effective in the Chinese healthcare setting ( Table 6 ).

In the U.S. scenario, the ICER for cadonilimab plus chemotherapy was estimated at $347,127.52 per QALY—well above the U.S. WTP threshold of $150,000.00. The corresponding INHB and INMB values were -0.41 QALYs and -$61,121.30, respectively, further supporting the conclusion that the combination regimen is not economically favorable under the current U.S. pricing assumptions ( Table 6 ).

Among patients with a PD-L1 CPS ≥5, the ICER for cadonilimab plus chemotherapy was $37,499.27 per QALY—below China’s WTP threshold of $40,354.27 ( Table 6 ). The corresponding INHB and INMB were 0.04 QALYs and $1,674.96, respectively, indicating that cadonilimab was cost-effective in this clinically responsive subgroup.

In contrast, for patients with a PD-L1 CPS <5, the ICER was $66,013.60 per QALY, exceeding the WTP threshold. The INHB was –0.16 QALYs and the INMB was –$6,355.16, suggesting that cadonilimab was not cost-effective in this lower PD-L1 expression group ( Table 6 ).

In the U.S. setting, the ICER for cadonilimab plus chemotherapy reached $240,877.66 per QALY in the PD-L1 CPS ≥5 group and $398,852.61 per QALY in the CPS <5 group, both of which far exceeded the WTP threshold of $150,000.00 ( Table 6 ). The corresponding INHBs were -0.38 and -0.43 QALYs, while the INMBs were -$56,677.19 and -$64,728.99, respectively. These findings indicate that cadonilimab was not cost-effective in either subgroup within the U.S. healthcare context, despite differential clinical responsiveness.

Figure 8 illustrated the results of the price simulation analysis across a range of cadonilimab pricing scenarios. In China, the ICER increased proportionally as the price varied from $0 to $2,000 per 625 mg dose. A similar trend was observed in the United States, where the price range examined ranged from $0 to $10,000 per 750 mg dose.

According to the respective WTP thresholds, cadonilimab would be considered cost-effective in China if the price was below $209.54 per 125 mg. In the U.S., the threshold for cost-effectiveness was $826.46 per 125 mg.

Scenario analysis evaluating a 3% discount rate for China showed the ICER of cadonilimab plus chemotherapy decreased to $48,678.70 per QALY, though the reduction was minimal ( Table 7 ). For this scenario, PSA results indicated a 29.57% probability that the regimen would be cost-effective at China’s defined WTP threshold. Under PPP-adjusted currency conversion, the ICER of the combination regimen was 25,374.68 per QALY—still exceeding the PPP−aligned WTP of 20,243.85 per QALY ( Table 7 ). PSA findings for this scenario showed a 24.86% probability of cost-effectiveness at the defined WTP threshold.

The OWSA results for the overall population and all subgroups in both China and the U.S. are presented in Figures 9 – 11 . The ICER was most sensitive to variations in the cost of cadonilimab, utility values for the PFS and PD health states, and the proportion of patients receiving targeted therapy during subsequent treatment. Despite these sensitivities, the differences in health outcomes between treatment strategies were sufficiently large that parameter variations did not alter the overall conclusions, except in the CPS ≥5 subgroup in China, where the ICER was influenced by changes in drug cost and health state utilities.

The PSA findings are shown in Figures 12 – 17 . In a Chinese setting, the probability that cadonilimab plus chemotherapy would be cost-effective at the defined WTP threshold was 23.35% for the overall cohort, 64.37% for the CPS ≥5 subgroup, and 3.20% for the CPS <5 subgroup. In the U.S., the corresponding probabilities were 2.30% for the overall cohort, 1.48% for the CPS ≥5 subgroup, and 0.09% for the CPS <5 subgroup. These results further support the conclusion that cadonilimab plus chemotherapy offers limited economic value at the current price.