This study employed a Chinese healthcare system perspective and integrated the literature, clinical treatment characteristics of advanced NSCLC, and insights from clinical experts and recommendations. Microsoft Excel 2016 was used to construct a four-state Markov model, an extension of the traditional three-state Markov model. The model categorized disease progression into two stages: initial progression (second-line treatment) and terminal progression (third-line treatment). This enhancement allowed for accurate simulation of patients’ treatment pathways and improved the precision of cost and health outcome measurements. The model structure is depicted in Fig. 1 below. All simulated patients started from a progression-free state (PFS) and could either remain in this state, progress to a progressed disease state (PDS), or die at the next cycle length. Patients who progressed to the PDS state could remain in PDS or progress to terminal progressed disease (TPD) or death, with death being the absorbing state. In China, the incidence of lung cancer in the 50–59 age group is increasing in both sexes. The median age of patients enrolled in RATIONALE 304 and RATIONALE 303 was 61 years, and the simulations were performed in 3-week cycles starting from the age of 61 years, aligning with the drug schedule. The model's time horizon was set at 21.06 years, corresponding to 361 cycles, to capture the lifetime impacts of tislelizumab combination therapy for first-line nsq-NSCLC patients, as more than 99% of patients died during this time.
The target population for this study consisted of patients with locally advanced or metastatic nsq-NSCLC in the first-line setting. In the tislelizumab group, patients received tislelizumab in combination with platinum-based chemotherapy and pemetrexed in the progression-free state, with a treatment duration of 6 cycles. The discontinuation rate among patients in the tislelizumab group was 17.8%, which was attributed to adverse reactions, patient-initiated decisions, physician-led cessation, and concurrent antitumor therapy. In comparison, the discontinuation rate among patients in the chemotherapy group was 28.7%, with similar factors contributing to discontinuation in this group. These findings were derived from data sourced from RATIONALE 304.
Patients in both groups who progressed received docetaxel as treatment, based on findings from the RATIONALE 303 trial, where the discontinuation rate in the docetaxel group was 27.5%. However, due to the lack of real-world data on discontinuation rates for tislelizumab, discontinuation rates were not considered in this study.
According to clinical trials, eighty percent of patients progressed to receive active treatment in the terminal PD state, while the remaining 20% received optimal supportive care. Given the uncertainty surrounding the protocol for receiving active treatment and the variability in the actual clinical management of advanced nsq-NSCLC patients, this study assumed that all patients received optimal supportive care. The clinical parameters for the Markov model were collected from RATIONALE 303 and RATIONALE 304. The study was conducted according to the Consolidated Health Economic Evaluation Reporting Standards (CHEERS).
Transition probability
Typically, due to the relatively short follow-up duration in clinical trials, survival analysis is used to extrapolate survival curves. This process involves extracting patient survival rate data at various time points from the provided survival curves in clinical trials using GetData Graph Digitizer software. Subsequently, R language (version 4.0.3) with the survHE package was used to perform parameter fitting and extrapolation for progression-free survival (PFS) and overall survival (OS) curves. Detailed descriptions of the PFS and OS curves of the clinical trials can be found in the supplementary methods. The main steps include data reconstruction, parameter model fitting, optimal model selection, and obtaining model parameters. The six common parametric distributions are the Exponential, Weibull, Gompertz, Log-logistic, Log-normal and Generalized Gamma distributions. The best-fitting distribution for the parameterized model was selected based on criteria including lower values of the Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) and visual comparison of the models against Kaplan‒Meier survival curves, with detailed results provided in the supplementary materials (supplement Table S2).
In the RATIONALE 304 trial, the best-fitting distribution for the tislelizumab group's PFS curve was Log-logistic, while the OS curve followed a Log-normal distribution. In the chemotherapy group, both PFS and OS curves were best fitted with Log-normal distributions. In the RATIONALE 303 trial, the PFS and OS curves for docetaxel treatment were best fitted with Log-logistic distributions. The estimated parameters of the goodness-of-fit distributions in Table S2 were applied to fit the 21.06-year time horizon survival times of the two groups (Fig.S1, Fig. S2, Fig.S3). Once the survival functions are obtained, time-dependent state-transition probabilities in cycles are computed based on the extracted data and fitted models.
The transition probability from the PFS to the death state is determined by the natural death rate, which varies over time. The study data were obtained from age-specific mortality rate data sourced from the "China Population and Employment Statistics Yearbook, 2022" (Table S5) [14].
Costs and utility
In this study, costs were assessed from the perspective of the Chinese healthcare system, encompassing direct medical costs such as drug expenses, adverse event costs, optimal supportive care costs, and end-of-life treatment costs.
Tislelizumab, now listed in China’s medical insurance directory, is priced at the national negotiated rate. Prices for pemetrexed, cisplatin, and docetaxel, included in China’s centralized drug procurement list, were obtained from the average bidding price in 2023 from Chinese Drug Bidding Data. Consistency with the clinical trial was maintained by utilizing the same drug dosages administered every three weeks, with the body surface area assumed to be 1.72 m², as reported in previous literature[15].
Adverse event costs were primarily considered Grade ≥ 3 adverse events with significant group differences in incidence rate, as Grade 1/2 adverse events were assumed to be manageable. The unit costs of adverse events were obtained from the literature[16][17], with the assumption that all adverse events occurred during the first treatment cycle, and rates were derived from the RATIONALE 304 and RATIONALE 303 trials. The optimal supportive care costs included outpatient doctor visit fees, laboratory examination fees, and other relevant expenses sourced from the literature (Shanghai, China) [18]. End-of-life treatment costs were also obtained from the literature. All costs were converted to a common currency using an exchange rate of 1 USD = 7.30 CNY.
This study conducted a cost-effectiveness analysis with quality-adjusted life years (QALYs) as the health outcome measure. The QALYs were calculated based on patient survival times and utility values collected from the literature. The utility values used in this study were 0.856 for the progression-free state, 0.768 for the progression state, and 0.703 for the end-of-life state[19]. In addition, negative utility values due to grade ≥ 3 adverse events were collected from the literature: 0.07 for anemia, 0.2 for leukopenia, 0.2 for neutropenia, 0.07 for fatigue, 0.12 for thrombocytopenia and 0.07 for musculoskeletal pain. [20][21].
The utility values for adverse effects were calculated by multiplying their incidences, resulting in negative utility values of 0.165 for first-line treatment with tislelizumab combined therapy, 0.126 for first-line treatment with chemotherapy regimens, and 0.88 for second-line treatment. All adverse reactions are assumed to occur during the first treatment cycle.
To determine the economic viability of the tislelizumab group compared to that of the chemotherapy-alone group, the incremental cost-effectiveness ratio (ICER) was compared to the willingness-to-pay (WTP) threshold, which was set at 1–3 times the per capita GDP according to the World Health Organization (WHO) recommendations. Specifically, when the ICER is less than 1 times the per capita GDP, the added cost is considered entirely worthwhile indicating an absolute advantage for the tislelizumab group. When the per capita GDP < ICER < 3 times the per capita GDP, the added cost is deemed acceptable, indicating cost-effectiveness for the tislelizumab group. However, when the ICER exceeds 3 times the per capita GDP, the added cost is deemed not worthwhile suggesting that tislelizumab is not cost effective. Discounting of both costs and utility values at a 5% discount rate was applied in line with Chinese Pharmacoeconomic Guidelines[22].
Table 1
Model parameters: baseline values, ranges, and distributions for sensitivity analysis
Variable | Basic value | Range | Distribution | Source |
T + Ca:incidence of AEs | | | | |
Anemia | 14.9% | 11.92%-17.88% | Beta | [12] |
Leukopenia | 21.6% | 17.28%-25.92% | Beta | [12] |
Thrombocytopenia | 19.4% | 15.52%-23.28% | Beta | [12] |
Nausea | 0.5% | 0.40%-0.60% | Beta | [12] |
ALT Increase | 3.6% | 2.88%-4.32% | Beta | [12] |
AST Increase | 2.3% | 1.84%-2.76% | Beta | [12] |
Neutropenia | 44.6% | 35.68%-53.52% | Beta | [12] |
Fatigue | 1.4% | 1.12%-1.68% | Beta | [12] |
Appetite Decrease | 1.4% | 1.12%-1.68% | Beta | [12] |
Vomiting | 0.5% | 0.40%-0.60% | Beta | [12] |
Musculoskeletal Pain | 0 | Fixed in DSA | Fixed in PSA | [12] |
Cb:incidence of AEs | | | | |
Anemia | 11.8% | 9.44%-14.16% | Beta | [12] |
Leukopenia | 14.5% | 11.60%-17.40% | Beta | [12] |
Thrombocytopenia | 13.6% | 10.88%-16.32% | Beta | [12] |
Nausea | 0.9% | 0.72%-1.08% | Beta | [12] |
ALT Increase | 2.7% | 2.16%-3.24% | Beta | [12] |
AST Increase | 0 | Fixed in DSA | Fixed in PSA | [12] |
Neutropenia | 35.5% | 28.40%-42.60% | Beta | [12] |
Fatigue | 1.8% | 1.44%-2.16% | Beta | [12] |
Appetite Decrease | 0.9% | 0.72%-1.08% | Beta | [12] |
Vomiting | 0.9% | 0.72%-1.08% | Beta | [12] |
Musculoskeletal Pain | 1.8% | 1.44%-2.16% | Beta | [12] |
Doxc:incidence of AEs | | | | |
Anemia | 5.0% | 4.00%-6.00% | Beta | [13] |
Weakness | 3.9% | 3.12%-4.68% | Beta | [13] |
Decrease in White Blood Cell Count | 17.8% | 14.24%-21.36% | Beta | [13] |
Leukopenia | 15.5% | 12.40%-18.60% | Beta | [13] |
Decrease in Neutrophil Count | 27.1% | 21.68%-32.52% | Beta | [13] |
Neutropenia | 27.1% | 21.68%-32.52% | Beta | [13] |
Febrile Neutropenia | 12.8% | 10.24%-15.36% | Beta | [13] |
Drug cost per mg, US($) | | | | |
Tislelizumab | 1.93 | 1.54–2.32 | Gamma | Database |
Cisplatin | 0.19 | 0.11–0.32 | Gamma | Database |
Pemetrexed | 0.56 | 0.09–0.94 | Gamma | Database |
Docetaxel | 1.86 | 4.93–23.63 | Gamma | Database |
Optimal Supportive Care Costs per cycle, US ($) | 216.00 | 174.00-258.00 | Gamma | [18] |
End-of-Life Treatment Costs per cycle, US($) | 2464.50 | 1848.38-3080.63 | Gamma | [18] |
AEs cost, US($) | | | | |
Anemia | 2150.12 | 32.94–49.42 | Gamma | [16] |
Leukopenia | 1267.73 | 383.04-574.56 | Gamma | [16] |
Thrombocytopenia | 1415.63 | 21.69–32.53 | Gamma | [16] |
Nausea | 16.96 | 0.26–0.38 | Gamma | [16] |
ALTd Increase | 292.59 | 4.49–6.73 | Gamma | [16] |
ASTe Increase | 292.59 | 4.49–6.73 | Gamma | [16] |
Neutropenia | 1094.28 | 383.04-574.56 | Gamma | [16] |
Fatigue | 0 | Fixed in DSA | Fixed in PSA | Expert consultation |
Appetite | 171.00 | 136.80-205.20 | Gamma | Expert consultation |
Vomiting | 496.90 | 54.38–81.57 | Gamma | [16] |
Musculoskeletal Pain | 7.39 | 6.60–9.91 | Gamma | [16] |
Weakness | 549.92 | 439.94-659.91 | Gamma | [17] |
Decreased White Blood Cell Count | 478.80 | 383.03-574.56 | Gamma | Expert consultation |
Decreased Neutrophil Count | 478.80 | 383.03-574.56 | Gamma | Expert consultation |
Febrile Neutropenia | 1025.99 | 820.79-1231.19 | Gamma | Expert consultation |
Utility | | | | |
PFS | 0.856 | 0.718–0.994 | Beta | [19] |
PD | 0.768 | 0.595–0.941 | Beta | [19] |
End stage | 0.703 | 0.632–0.774 | Beta | [19] |
Death | 0 | Fixed in DSA | Fixed in PSA | [19] |
Anemia | 0.07 | 0.058–0.088 | Beta | [20] |
Leukopenia | 0.20 | 0.16–0.24 | Beta | [20] |
Thrombocytopenia | 0.11 | 0.086–0.13 | Beta | [20] |
Neutropenia | 0.20 | 0.112–0.168 | Beta | [20] |
Fatigue | 0.06 | 0.048–0.072 | Beta | [20] |
Musculoskeletal pain | 0.069 | 0.058–0.079 | Beta | [21] |
Other | | | | |
Discount(%) | 5% | 0–8% | Fixed in PSA | Guideline |
Body area(m^2) | 1.72 | Fixed in DSA | Fixed in PSA | [15] |
a.T + C: Tislelizumab plus pemetrexed-platinum chemotherapy; |
b. C: Pemetrexed-platinum chemotherapy; |
c. Dox: Docetaxel; |
d. ALT: Alanine aminotransferase; |
e. AST: Aspartate aminotransferase |
Statistical analysis
In this study, we conducted both one-way sensitivity analysis and probabilistic sensitivity analysis to assess the robustness of the results. One-way sensitivity analysis evaluates the impact of changes in one parameter at a time on the results, varying parameters within predefined ranges, typically determined by confidence intervals. Parameters without confidence intervals were adjusted by a certain percentage, with costs and utilities fluctuating ± 20% and ± 10% from their base values, respectively, following previous studies[23].
The discount rate was varied within the recommended range of 0–8%. The results of the one-way sensitivity analysis are presented in tornado diagrams, illustrating the sensitivity of the model to changes in individual parameters. Probabilistic sensitivity analysis was performed via Monte Carlo simulation to address the uncertainty in multiple parameters simultaneously. Parameters were assumed to follow specific probability distributions, with costs modeled using a gamma distribution and probabilities and utility values modeled with a beta distribution. The analysis was repeated 10,000 times to generate a distribution of cost-effectiveness outcomes.
The results of the probabilistic sensitivity analysis are presented in cost-effectiveness scatterplots and cost-effectiveness acceptability curves (CEAC), offering a visual representation of the uncertainty in the model and the likelihood of the tislelizumab group being cost-effective compared to the chemotherapy-alone group under different WTP thresholds. These sensitivity analyses aid in assessing the robustness of the economic evaluation by considering variations in key parameters and capturing the overall uncertainty in the model's results.