2.1 Model structure
Both the Markov model and partitioned survival model (PSM) are the most frequently used approach in the cancer health economic evaluation. A published systematic review[10]indicated that although the Markov model and PSM yielded similar results under the same model structure and assumptions, the PSM was more recommended and was easier to construct when individual patient data (IPD) were available. Furthermore, the PSM is the most common model employed in evaluating the cost-effectiveness of cancer treatment according to a review of the National Institute for Health and Care Excellence (NICE) appraisals[11].
Therefore, a PSM was developed in Microsoft Excel to evaluate long-term health outcomes and costs for tislelizumab monotherapy versus camrelizumab monotherapy as the second-line treatment of locally advanced or metastatic ESCC. Three mutually exclusive states were included in the model, namely progression-free (PF), progressive disease (PD), and death. All patients were assumed to be in the PF state initially, and either stay in the same health state or move to another health state in the next cycle. The period of patients in each health state over time is determined by the area under the curve (AUC) of a set of mutually exclusive survival curves which commonly are overall survival (OS) curves and progression-free survival (PFS) curves[12]. The trapezoidal method was applied to calculate the AUC, as illustrated in Fig. 1.
The quality-adjusted life-years (QALYs), life-years (LYs), and costs were all measured in this model. The cycle length of the model was 3 weeks, which was aligned with the medication administration cycle in RATIONALE 302, and the time horizon was lifetime (defined by 99% of the patient cohort dead). Only direct medical costs were taken into consideration from China’s healthcare system perspective. Patients in the PF state were assumed to have treatment-related costs, including drug costs, follow-up costs, disease management costs, and costs of adverse events (AE) management. Within the PD state, costs associated with disease management, follow-up visit, and subsequent treatment would occur. If patients died, the cost of end-of-life care was calculated. All costs and health outcomes were calculated based on the 2021–2022 prices and discounted at 5% annually according to the China Guidelines for Pharmacoeconomic Evaluations 2020 [13]. All costs have been converted into US dollars, and the exchange rate was 1 US dollar = 7.23 Chinese RMB Yuan (US Dollar to Chinese Yuan Renminbi conversion – Last updated 20 October 2022, 06:19 UTC). In addition, the threshold of willingness to pay (WTP) for a QALY was assumed to be 1–3 times GDP per capita of China ($11,207–$33,621 per QALY gained in 2021).
2.2 Patient population
Based on the trial eligibility criteria of ESCORT and RATIONALE 302, the target patient population of the model was patients with a histologic or cytologic diagnosis of locally advanced or metastatic ESCC who had progressed on or were intolerant to first-line chemotherapy. As all patients included in the ESCORT clinical trial were Chinese, patients included in RATIONALE 302 were also limited to the Chinese population (including Taiwan). Patients with age greater than 75 years old or brain metastasis in RATIONALE 302 were excluded. Patients in the intervention group received 200 mg tislelizumab on day 1 of each 21-day. Camrelizumab 200 mg was administered intravenously over 30 minutes on day 1 of each 14-day cycle for patients in the control group. All patients received tislelizumab or camrelizumab monotherapy until disease progression, unacceptably toxic adverse reactions occurred or withdrew for other reasons.
2.3 Clinical input
Clinical data on the efficacy and safety of tislelizumab and camrelizumab were derived from RATIONALE 302 trial and the published clinical trial ESCORT, respectively[9]. Since there is no head-to-head clinical trial of tislelizumab monotherapy versus camrelizumab monotherapy, the anchored matching adjusted indirect comparison (MAIC) method was adopted in the model to adjust the baseline of patient characteristics. In the adjustment process, key baseline demographic and disease characteristics factors, namely age, gender, histological grade, disease metastasis, lymphatic metastasis, PD-L1 expression level, and Eastern Cooperative Oncology Group (ECOG) score, prior therapies (PT) surgery, PT. radiotherapy, PT. platinum-based chemotherapy, which were reported in ESCORT, were included. The results of baseline patient characteristics and clinical efficacy data before and after adjustment are shown in Table 1 and Table 2.
Table 1
Baseline characteristics before and after adjustment
| Before adjustment | After adjustment |
Adjustment factor | Camrelizumab group[1](N = 448) | Tislelizumab group (N = 495) | P value unweighted | Tislelizumab group (ESS = 137) | P value weighted |
Age ≤ 60 | 50% | 48% | 0.66 | 50% | 1 |
Male | 89% | 90% | 0.72 | 89% | 1 |
Histological grade 3(Poorly differentiated) | 29% | 18% | 0 | 29% | 1 |
Disease metastasis | 84% | 97% | 0 | 84% | 0.91 |
Lymphatic metastasis | 85% | 76% | 0 | 85% | 1 |
High PD-L1 expression level (vCPS ≥ 10%) | 43% | 29% | 0 | 43% | 1 |
ECOG PS = 1 | 80% | 83% | 0.29 | 80% | 1 |
PT. Surgery | 49% | 45% | 0.25 | 49% | 1 |
PT. Radiotherapy | 66% | 65% | 0.8 | 66% | 1 |
PT. Platinum-based chemotherapy | 95% | 98% | 0.06 | 95% | 1 |
Abbreviations: ECOG = Eastern Cooperative Oncology Group; PS = performance status; ESS = effective sample size; P.T. = Prior therapies. |
Table 2
Results of progression-free survival and overall survival after MAIC.
Adjustment results | HR-PFS | HR-OS |
Tislelizumab monotherapy |
Before MAIC | 0.85 (0.65–1.10) | 0.74 (0.58–0.95) |
After MAIC | 0.78 (0.55–1.11) | 0.68 (0.49–0.94) |
Camrelizumab monotherapy | 0.69 (0.56–0.86) | 0.71 (0.57–0.87) |
HR, hazard ratio; PFS, progression-free survival; OS, overall survival; MAIC, matching adjusted indirect comparison. |
Six parametric distributions, including the exponential, Weibull, Gompertz, log-normal, log-logistic, and gamma distributions, were used to extrapolate the survival curves to capture survival outcomes in lifetime horizon. Since the PFS curves of tislelizumab before and after MAIC adjustment relative to the chemotherapy group and the PFS curves of camrelizumab relative to the chemotherapy group did not meet the PH assumption (Supplementary Figures S1-S2), the MAIC-adjusted HR values were not used to adjust the efficacy. The standard parameters were fitted separately for the camrelizumab monotherapy in ESCORT and the adjusted tislelizumab monotherapy in RATIONALE 302.
The survival curves of the camrelizumab monotherapy were derived from the published literature[9] and the IPD was reconstructed using the method of Guyot et al study[14]. The results of the parameter fitting are shown in the Supplementary material table S1. Akaike information criterion (AIC), Bayesian information criterion (BIC), visual inspection, and logic error checking were used to evaluate best-fitting parametric distributions. As a result, the best-fitting distribution for PFS data of tislelizumab monotherapy was log-normal distribution, while the best-fitting parametric distributions for OS data of tislelizumab and PFS along with OS data of camrelizumab monotherapy were log-logistic distributions. The fitting and extrapolation results of PFS and OS curves are shown in Figs. 2–3.
Only AEs with incidence ≥ 1% and grade ≥ 3 of patients with camrelizumab or tislelizumab were included for this study. AEs to the camrelizumab monotherapy were obtained from ESCORT[9], including somasthenia, diarrhea, hyponatraemia, anemia, lymphopenia, and reactive cutaneous capillary endothelial proliferation (RCCEP). Thereinto, hyponatraemia, anemia and lymphopenia were AEs associated with tislelizumab monotherapy only.
2.4 Cost input
Only direct medical costs were estimated from the perspective of China’s healthcare system including acquisition costs of drugs, subsequent treatment, follow-up visits, disease management, AE management, and end-of-life care.
Drug costs consisted of the costs of tislelizumab and camrelizumab. The prices of tislelizumab and camrelizumab were derived from the cost database available on Menet[15]. Follow-up costs were divided into three parts, including the PF state of the tislelizumab monotherapy, the PF state of the camrelizumab monotherapy, and the PD state, each of which included the cost of contrast computed tomography (CT), routine blood tests, blood chemistry, and routine urine tests. The frequency of contrast CT (chest and upper abdomen) was once every two months for both the PF state of the tislelizumab monotherapy and camrelizumab monotherapy, while the rest of the program was performed before each treatment. For patients in the PD state, the contrast CT (chest and upper abdomen) was performed once every two months, and the rest of the program was once every three weeks. The cost of disease management was also calculated separately for the PF state of the tislelizumab monotherapy, the PF state of the camrelizumab monotherapy, and the PD state, each of which included the costs of diagnosis, intravenous administration, nursing, and hospitalization. The costs of follow-up visits and disease management were obtained from the medical service price in ten randomly selected regions, namely Shanghai, Tianjin, Guangdong, Shandong, Anhui, Liaoning, Ningxia, Yunnan, Hebei, and Heilongjiang, after a stratified sampling of cities nationwide according to their level of economic development. Subsequent treatment is the administration of a different chemotherapeutic agent after patients progressed on tislelizumab or camrelizumab which was used in the first course of treatment. Since both tislelizumab and camrelizumab are PD-1 inhibitors, have similar mechanisms and both are second-line treatment drugs recommended by CSCO guidelines, they are clinically substitutable. This study, therefore, assumed that the subsequent treatment after disease progression for the camrelizumab group is consistent with that of the tislelizumab group, based on expert opinion. The subsequent treatment pattern after progression of tislelizumab was derived from IPD from RATIONALE 302, and only regimens with greater than 5% patient use were included in this study, as shown in Table 3.
Table 3
Subsequent treatment options for tislelizumab group
Subsequent treatment | Proportion of patients |
Anrotinib hydrochloride | 9.6% |
Apatinib mesylate | 8.9% |
Paclitaxel | 8.9% |
Docetaxel | 7.5% |
Gemeracil or Otilacil potassium or Tegafur (Tegeo tablets) | 7.5% |
Irinotecan | 5.5% |
Cisplatin | 8.2% |
Only grade 3 or 4 AEs occurring in ≥ 1% of patients were considered in the costs of AE management, including somasthenia, diarrhea, hyponatraemia, anemia, lymphopenia, and reactive cutaneous capillary endothelial proliferation (RCCEP). Since no therapy was clinically essential for non-severe RCCEP, only the cost of severe RCCEP treatment was considered in the model. AEs treatment drugs, days of treatment and costs were from expert opinions except for the duration of RCCEP, which was derived from the camrelizumab package insert (update May 29, 2019). In addition, the cost of end-of-life care was based on published literature[16]. This study assumed that the mean weight of patients was 60 kg and the mean body surface area was 1.6 m2 to estimate the dosages of drugs, according to the recommendation from the National Healthcare Security Administration (NMPA) in China. All cost parameters are listed in Table 4.
Table 4
Parameters | Deterministic | Distribution | Low | High | Source |
Cost($) | | | | | |
Unit drug costs ($) | | | | | |
Tislelizumab | 200.55(per 100mg) | Constant | 160.44 | 200.55 | MENET* |
Camrelizumab | 404.98(per 200mg) | Constant | 323.98 | 404.98 | MENET* |
Subsequent treatment per cycle($) | | | | | |
| 721.9 | Gamma | 675.96 | 845.46 | RATIONALE 302 IPD |
Unit follow-up costs ($) | | | | | |
PF state of the tislelizumab (per cycle) | 25.45 | Gamma | / | / | Expert opinion |
PF state of the camrelizumab (per cycle) | 29.59 | Gamma | / | / | Expert opinion |
PD state (per cycle) | 25.45 | Gamma | / | / | Expert opinion |
Contrast CT | 25.9 | Gamma | 10.37 | 69.16 | Health care document** |
Blood chemistry | 6.43 | Gamma | 3.96 | 13.14 | Health care document** |
Blood routine | 1.31 | Gamma | 0.69 | 2.35 | Health care document** |
Urine routine | 0.52 | Gamma | 0.28 | 0.83 | Health care document** |
Unit drug administration costs ($) | | | | | |
PF state of the tislelizumab (per cycle) | 9.34 | Gamma | / | / | Expert opinion |
PF state of the camrelizumab (per cycle) | 14 | Gamma | / | / | Expert opinion |
PD state (per cycle) | 10.44 | Gamma | / | / | Expert opinion |
Consulting fee | 0.97 | Gamma | 0.55 | 3.04 | Health care document** |
Intravenous injection | 0.83 | Gamma | 0.55 | 1.66 | Health care document** |
Hospitalization | 3.04 | Gamma | 1.66 | 6.64 | Health care document** |
PF state nursing | 1.45 | Gamma | 0.41 | 3.04 | Health care document** |
PD state nursing | 3.39 | Gamma | 3.39 | 6.22 | Health care document** |
Unit AE management costs ($) | | | | | |
Somasthenia | 106.64 | Gamma | 0 | 127.97 | Expert opinion |
Anemia | 467.5 | Gamma | 86.31 | 730.29 | [2] |
Diarrhea | 406.4 | Gamma | 222.96 | 733.61 | Expert opinion |
Lymphopenia | 82.99 | Gamma | 3.73 | 733.61 | [2] |
Hyponatraemia | 326.42 | Gamma | 14.94 | 829.88 | Expert opinion |
RCCEP | 276.63 | Gamma | 8.3 | 1383.13 | [3] |
End-of-life care costs ($) | | | | | |
| 3949.22 | Gamma | 3159.38 | 4739.07 | [4] |
Probabilities |
Tislelizumab Arm |
Hyponatraemia | 3.10% | Beta | 1.38% | 5.64% | RATIONALE 302[5] |
Anemia | 3.88% | Beta | 1.92% | 6.67% | RATIONALE 302[5] |
Lymphopenia | 3.10% | Beta | 1.38% | 5.64% | RATIONALE 302[5] |
Camrelizumab Arm |
Somasthenia | 1.32% | Beta | 0.27% | 3.15% | ESCORT[1] |
Diarrhea | 1.32% | Beta | 0.27% | 3.15% | ESCORT[1] |
Hyponatraemia | 1.32% | Beta | 0.27% | 3.15% | ESCORT[1] |
Anemia | 2.63% | Beta | 0.98% | 5.07% | ESCORT[1] |
Lymphopenia | 1.32% | Beta | 0.27% | 3.15% | ESCORT[1] |
RCCEP | 79.82% | Beta | 74.39% | 84.77% | ESCORT[1] |
Severe RCCEP | 0.44% | Beta | 0.01% | 1.61% | ESCORT[1] |
Utilities | |
PF state | 0.741 | Beta | 0.593 | 0.889 | [6] |
PD state | 0.581 | Beta | 0.465 | 0.697 | [6] |
Disutilities | | | | | |
Somasthenia | -0.07 | Beta | -0.056 | -0.084 | [7] |
Anemia | -0.07 | Beta | -0.06 | -0.09 | [7] |
Diarrhea | -0.07 | Beta | -0.056 | -0.084 | [7] |
Lymphopenia | -0.2 | Beta | -0.16 | -0.24 | [3] |
Hyponatraemia | 0 | Beta | 0 | 0 | [8] |
RCCEP | -0.1 | Beta | -0.08 | -0.12 | [7] |
Discount rate | | | | | |
| 5% | constant | 0% | 8% | [9] |
PF, progression-free; PD, progressive disease; IPD, individual patient data. *The price of the drug was obtained from MENET, the online price database in China. (https://menet.com.cn). **The price of follow-up and drug administration were obtained from the healthcare document of 10 provinces in China. ***The inclusion criteria of AEs were that the incidence of AEs ≥ 1% and grade ≥ 3. |
2.5 Utility
Due to the lack of studies using the EuroQol Five Dimensions Questionnaire (EQ-5D) to measure health utility values in second-line ESCC patients directly, this study traced the source of health utility values in published studies of second-line treatment for ESCC and found that the majority health utility values of studies[9, 17–22] (PF = 0.741, PD = 0.581) were derived from a quality-of-life study of ramucirumab plus paclitaxel in patients with previously treated gastric or gastroesophageal junction adenocarcinoma[16]. In that study, the baseline health utility value of patients in the intervention group was 0.74 and the health utility value of patients who discontinued treatment was 0.581[16]. A study published by the National Institute for Health and Care Excellence (NICE) noted that the baseline health utility value of patients could be used as their PF state utility value and the health utility value for patients who discontinue treatment could be used as their PD state utility value[23]. Therefore, the utility value for this study was 0.741 for the PF state, 0.581 for the PD state, and 0 for death (Table 4).
AE’s disutilities were also considered in the model, which were obtained from the published literature.[24–26]. Since RCCEP occurred continuously with the use of camrelizumab, the disutility of camrelizumab during treatment was calculated based on the median duration of RCCEP occurrence in the camrelizumab package insert, rather than just considering the disutility of severe RCCEP. The disutility of RCCEP was assumed to be consistent with that of rash in the Chinese population and the disutility of lymphopenia was assumed to be consistent with that of white blood cell count reduced. The costs and disutilities of AEs were only calculated in the first cycle. Details of incidence rate, disutility values, and sources are listed in Table 4.
2.7 Sensitivity analyses
2.7.1 Deterministic sensitivity analyses
To verify the robustness of model results, individual uncertainty parameters were identified, and one-way sensitivity analyses were performed separately within their range of possible variation. The tornado diagram was drawn to identify the factors that have a greater impact on the results. The following key parameters were included in the one-way deterministic sensitivity analyses (DSA): discount rate, costs of drug acquisition, disease management, follow-up, subsequent treatment, and end-of-life care (varied by the standard error, 95% confidence interval), incidence and duration of AEs (varied by 95% confidence interval), utilities and disutilities (varied by 95% confidence interval). Specifically, the price of tislelizumab and camrelizumab was varied from 80–100% of the deterministic value because, in China, there is little space for pharmaceutical manufacturers to increase the price of drugs.
2.7.2 Probabilistic sensitivity analysis
Probabilistic sensitivity analysis (PSA) was performed using a Monte Carlo simulation with 1,000 iterations. The parametric distribution assumptions were based on the recommendations in Decision Modelling for Health Economic Evaluation[27], where the incidence of AEs and utility parameters obeyed the beta distribution, cost parameters and duration of AEs obeyed the gamma distribution. Scatter plots and cost-effectiveness acceptability curves (CEACs) were plotted based on the simulation results to determine the probabilities of being cost-effective for each alternative under different WTP thresholds. In addition, it should be noted that Cholesky decomposition was conducted to take into account the correlation between parameters.
2.7.3 Scenario analysis
Since the control group in ESCORT was docetaxel or irinotecan, while that of RATIONALE 302 was paclitaxel, docetaxel, or irinotecan. A scenario analysis was conducted to consider the influence of different control groups on the outcomes. In this scenario, patients with pre-randomized investigator choice of paclitaxel in RATIONALE 302 were excluded and adjustments for patient baseline were performed. The cost-effectiveness of tislelizumab monotherapy in comparison to camrelizumab monotherapy was then evaluated.