The target population of the model was based on the clinical trial of Kang Ai injection. This trial included 326 patients from multiple hospitals in China, all aged 18–75 years and diagnosed with IIIB–IV stage NSCLC according to pathology or cytology (16). The initial average age of patients was assumed to be 59 years old in our research. To calculate costs, we also assumed that average body surface area was 1.6 m² and creatinine clearance rate was 80 ml/min.

At present, there are a variety of Chinese patent medicines for adjuvant treatment in clinical practice, among which Shenqi Fuzheng injection can be combined with chemotherapy for adjuvant treatment of NSCLC, and it is also in the National Reimbursement Drug List (NRDL). Therefore, Shenqi Fuzheng injection combined with chemotherapy was selected as the control group in this study. The intervention of treatment group is Kang Ai injection combined with gemcitabine or paclitaxel plus platinum-based chemotherapy. All drugs were administered with a cycle of 3 weeks.

The Markov model is applicable for the simulation of disease progression among a finite number of health states. Under appropriate assumptions, it is convenient to simulate patients’ long-term health state changes. Many published studies built a three-state Markov model to evaluate the cost-effectiveness of drugs for patients with NSCLC (17–20). This demonstrated the suitability of the Markov model. Therefore, we chose the Markov model for simulation.

A Markov cohort simulation model was developed with Microsoft Excel (Version 2019) to estimate the costs and outcomes of Kang Ai injection or Shenqi Fuzheng injection combined with platinum-based chemotherapy for the treatment of patients with advanced NSCLC. The Markov model incorporated three states: progression-free (PF), progressive disease (PD) and death, as shown in Figure 1. We assumed that all patients were in the PF state when they entered the model at the very beginning. According to the drug regimens, the cycle length was set as 3 weeks and the time horizon was 10 years based on the poor prognosis of advanced NSCLC and follow-up period, allowing nearly 100% of the patients in all treatment groups to reach the state of death. The main outcomes of the model were total costs, life years, quality-adjusted life years (QALYs), and incremental cost-effectiveness ratio (ICER).

The Kaplan–Meier curves for progression-free survival (PFS) and overall survival (OS) of Shenqi Fuzheng injection combined with chemotherapy in the first-line treatment of advanced NSCLC have not been reported in the existing clinical trials. Therefore, we could not recreate pseudo-individual-level data to fit the survival curves. Meanwhile, although individual patient data from clinical trials of Kang Ai injection can be obtained, the OS curve is not mature and only the PFS curve of Kang Ai injection can be analyzed for survival. Because of the above limitations, this study firstly extrapolated the objective response rate (ORR) value of Shenqi Fuzheng injection group based on the results of reported ORR value of Kang Ai group and existing network meta-analysis (NMA). According to the existing research, the median PFS (mPFS) and median OS (mOS) were calculated by ORR values. Then survival curves were estimated by the method of declining exponential approximation of life expectancy (DEALE), which is based on the assumption that survival follows a declining exponential function (14, 15).

The ORR value of Kang Ai injection combined with platinum-based chemotherapy in first-line treatment of advanced NSCLC was 28.2% (16). Li et al. (21) conducted a NMA on clinical trials of 10 Chinese traditional medicine injections combined with gemcitabine plus platinum-based chemotherapy in the treatment of advanced NSCLC, and found that the odds ratio (OR) value of ORR of Kang Ai injection group compared with Shenqi Fuzheng injection group was 1.23. Then, we calculated that the ORR value of Shenqi Fuzheng injection group was 24.2%. Rui et al. (22) reviewed and analyzed clinical trials of drugs in the treatment of advanced NSCLC in China, and obtained the correlation between clinical short-term surrogate endpoint indicators (ORR, etc.) and endpoint indicators (mPFS and mOS). The calculation equations are as follows:

Ln(mPFS) refer to natural logarithm of mPFS; mPFS refer to median progression-free survival; mOS refer to median overall survival; ORR refer to objective response rate.

According to the above equations, the mPFS, mOS of Shenqi Fuzheng injection group and mOS of Kang Ai injection group were calculated. The results are shown in Table 1. After obtaining the mPFS, mOS of Shenqi Fuzheng injection group and mOS of Kang Ai injection group, the three curves were fitted with exponential distribution. In order to eliminate the influence of different distribution selection, the PFS curve of Kang Ai injection group was also fitted with exponential distribution. The related parameters of survival curves are shown in Table 2. Transition probabilities were calculated from the estimated survival curves. Age- and sex-specified general population mortality which derived from the sixth national census results were extracted to obtained probability of PF to PD state (23).

From the perspective of Chinese healthcare system, we only considered direct medical costs, which included drug acquisition cost, follow-up cost, disease management cost, subsequent treatment cost, and end life cost. Due to the synergistic and attenuating effects of Chinese patent medicines in NSCLC patients, the incidence rate of grade 3 or 4 adverse events (AEs) is lower than 5% (16, 24). Therefore, the costs associated with AEs were not considered in this study. All the costs were updated to 2023 US dollars ($1 = ¥7.10) (25).

Drug acquisition cost included the costs of Kang Ai injection, Shenqi Fuzheng injection, platinum (carboplatin or cisplatin) and gemcitabine or paclitaxel. The costs of Kang Ai injection, Shenqi Fuzheng injection were sourced from current price provided from the original manufacturer. Costs of carboplatin and cisplatin were the median bidding price gathered across provinces in recent 1 year. The costs of gemcitabine and paclitaxel were sourced from median price in the fifth volume-based procurement scheme. In the calculation of drug acquisition cost, the ratio of patients to gemcitabine or paclitaxel was assumed to be 1:1, and the ratio of patients to carboplatin or cisplatin was assumed to be 1:1.

In the first cycle of PF health state, the follow-up cost included the costs of routine blood test, routine urine test, and blood chemistry examination. In the second cycle of PF health state, the follow-up cost included the costs of computed tomography (CT), magnetic resonance imaging (MRI), B-ultrasound, chest X-ray, bone scan, routine blood test, routine urine test, and blood chemistry examination. In the rest cycles of PF health state, the follow-up cost included the costs of CT, MRI, B-ultrasound, chest X-ray, and bone scan. The follow-up cost of each PD health state cycle included CT, MRI, B-ultrasound, chest X-ray, and bone scan.

The cost of disease management was calculated according to PF and PD health state, respectively. The disease management cost of the first two cycles of PF health state included costs of diagnosis, intravenous injection, bed, and nursing. The disease management cost of the remaining cycles of PF health state and PD health state included costs of diagnosis, bed, and nursing.

Follow-up cost and disease management cost were derived from the medical service price lists of 13 provinces including Beijing, Shanghai, Hebei, Shanxi, Jiangsu, Zhejiang, Chongqing, Fujian, Shanxi, Shandong, Guangzhou, Henan, and Jiangxi. Subsequent treatment cost and end life cost were derived from the literature (26, 27). Key inputs related to costs are listed in Table 3.

The health utilities were derived from the research of Nafees et al. (28), which was based on the original Nafees et al. (29) study published in year 2008. Because of the limitations of methodology and geographical applicability in the original study, the study published in 2016 used time trade-off (TTO) method to capture utilities for metastatic NSCLC from local populations in the United Kingdom, Australia, France, China, Taiwan, and Korea. From the research, we applied the PF utility of 0.804 and PD utility of 0.321 based on the results of Chinese people. The details of utility parameters are displayed in Table 3.

According to the China Guidelines for Pharmacoeconomics Evaluations (2020), the annual discount rate applied to costs and health outcomes was 5% (30).

We carried out one-way sensitivity analysis, probabilistic sensitivity analysis, and scenario analysis to evaluate the robustness of base-case results. According to the guidelines, the willingness-to-pay (WTP) threshold value for QALY is recommended to be 1–3 times of the GDP per capita (30). We deemed one time the GDP per capita of China in 2022 ($12,070/QALY) as the threshold.

For one-way sensitivity analysis, we varied a number of parameters and displayed 10 most influential parameters in tornado diagrams. For parameters, of which we could not obtain the upper and lower limits, we assumed a 20% fluctuation from the baseline value. For drugs included in the NRDL, such as Kang Ai injection and Shenqi Fuzheng injection, the prices of these drugs were stable. Hence, we assumed that the upper limits of the drug prices were the prices used in base-case analysis and the lower limits of the prices were taken as 80% of the base-case prices. Besides, we conducted a probabilistic sensitivity analysis (PSA) via Monte Carlo simulation in which every key input was assumed to fit a theoretical distribution. The results of 1,000 iterations were drawn in the form of scatter plot and cost-effectiveness acceptability curves (CEACs). Parametric distributions and ranges used in the analysis are displayed in Table 3.

In order to further explore the uncertainty of the economic results, we carried out a scenario analysis. In order to eliminate the impact of different distributions, the PFS curve of Kang Ai injection group was fitted with an exponential distribution in the base-case analysis. While in the scenario analysis, the PFS curve of Kang Ai group was fitted with a Log-Normal distribution, which was the best-fitting distribution. The time horizon was also reduced to 5 years.

In the 10 years of simulation, per patient in Kang Ai injection group obtained 0.914 QALYs and 1.602 life years and needed to pay $5,997.32. In Shenqi Fuzheng injection group, per patient obtained 0.697 QALYs and 1.459 life years and needed to pay $5,893.94. The ICER derived from Kang Ai injection group compared with Shenqi Fuzheng injection group was $476.41/QALY, which was less than the willingness to pay threshold of one time the GDP per capita of China in 2022. Hence, Kang Ai injection demonstrated a supremely good cost-effectiveness in the base-case analysis. The costs, life years gained, QALYs and ICERs in base-case analysis are given in Table 4.

One-way sensitivity analyses were conducted to test the robustness of the base-case ICER, and the results were shown in the form of tornado diagrams in Figure 2. The base-case ICER was most sensitive to changes in OR value, followed by the cost of Shenqi Fuzheng injection, cost of Kang Ai injection, the PF health state utility. The remaining parameters had less influence on the base-case results. It is worth noting that changes in all parameters cannot reverse the base-case results.

A total of 1,000 Monte Carlo simulations were conducted to obtain a scatter plot, as shown in Figure 3. Under the threshold of one time the GDP per capita, the scatters which represented the ICERs of Kang Ai injection were almost all below the threshold line, meaning the probability for Kang Ai injection to be cost-effective when compared with Shenqi Fuzheng injection was 100%.

When willingness to pay threshold varied from $0 to $70,000 per QALY gained, we drawn a cost-effectiveness acceptability curve (CEAC), as shown in Figure 4. According to the CEAC curve, Kang Ai injection group had a nearly 100% probability to be cost-effective under the threshold of 1–3 times GDP per capita.

We considered seven parametric distributions to fit the PFS curve of Kang Ai group: exponential, gamma, generalized F, generalized gamma, weibull, loglogistic, and log-normal. The evaluation index of assessment of fit was based on Akaike information criterion, Bayesian information criterion, as presented in Table 5. SurvHE package was used to fit various survival models in R (version of 4.2.1). Information about individual patient data of Kang Ai injection are presented in Supplementary Table 1.

Finally, the Log-Normal distribution was used, which was the best-fitting distribution, to fit the PFS curve of Kang Ai group. The value of natural logarithm of the mean and standard deviation parameters were 1.91 and 1.66, respectively. With a time horizon of 5 years, the ICER significantly increased from $476.41/QALY in the base-case analysis to $4,081.83/QALY, but still far below the willingness-to-pay threshold of one time the GDP per capita of China in 2022 ($12,070/QALY), indicating that Kang Ai injection group was to be cost-effective when compared with Shenqi Fuzheng group. Scenario analysis results are displayed in Table 6.