Study Population
We used data from a published phase III RCT of patients with V600-mutated melanoma, in which the patients were randomized to receive as first-line treatment either trametinib plus dabrafenib or vemurafenib alone.18 In this trial, the median follow-up time of the TRAM+DAB combination treatment group and the VEM monotherapy group were 11 months and 10 months, respectively. The median age of the patients was 55 years for the TRAM+DAB treatment and 54 years for the VEM monotherapy.
Markov Model
A Markov model was established using TreeAge Pro 2011 to estimate the clinical and economic outcomes of vemurafenib and of trametinib plus dabrafenib. The Markov model consists of three states: PFS, progressive disease (PD), and death (D; Figure 1). Cycle lengths were 4 weeks, and the time horizon was lifetime. All patients were in the PFS state initially and could experience PD or D. They were randomized to receive either vemurafenib or trametinib plus dabrafenib orally until disease progression. Some patients chose to continue taking the study medication after the disease progression.
The primary model outcomes were the utility and cost of the two treatment modalities, with utility expressed in quality-adjusted life years (QALYs), while the ICER was used to express the cost per QALY gained. Effects on the incremental cost-effectiveness ratio (ICER) were assessed, with discounting for costs and effects at 5 % per year after the first year. Combining state values (e.g., life years, health-related quality of life, and cost) with the average number of cycles a patient stays in each state, it is possible to estimate life expectancy, life expectancy after quality adjustment, and expected costs.19 According to the National Bureau of Statistics of China, China's GDP per capita for 2020 was CNY 72 447; the willingness-to-pay (WTP) threshold for this study was set as CNY 72 447/ QALY.
Model Inputs
Inputs in our Markov model were 1) transition probabilities for PFS, PD, and D in each cycle; 2) probability of adverse effects under each treatment; 3) the utility of PFS and PD status under the two treatment regimens; 4) cost. From the two treatment regimens, PFS and OS data, incidence of adverse events, and transition probabilities were calculated from the clinical trial results of Robert et al.18 All parameter inputs are shown in Table 1.
Cost & Utility
This analysis was from the perspective of China's health system, so only direct medical costs were considered in the model, including the costs of drugs, second-line treatment, general physical examinations, tumor assessment examinations, and the management of adverse reactions. The prices of vemurafenib, trametinib and dabrafenib were derived from the centralized drug procurement platform of Yunnan Province. Values for hematological examinations, CT scans, and other expenses came from public medical institutions. Only adverse reactions above grade 3 were considered; for vemurafenib treatment these were hypertension, arthralgia, skin rash, and increased levels of alanine aminotransferase and aspartate aminotransferase. For trametinib and dabrafenib treatment, the adverse reactions were fever and hypertension; these adverse reactions are generally alleviated after short treatment, and there are fewer cases of drug withdrawal. Therefore, in our Markov model, only the adverse-reaction management costs were included in the first-cycle cost calculation.
In the clinical trials, the researchers did not investigate the health-utility value of the patients in the PFS and PD stages; accordingly, the health-state utilities used in the model were obtained from the literature.20–22
Sensitivity analyses
Due to the uncertainty of parameters and assumptions, we conducted a one-way sensitivity analysis and probabilistic sensitivity analysis (PSA) to evaluate the robustness of the base-case results. The result of one-way sensitivity analysis is shown in Figure 2.
The PSA further explored the uncertainty of input parameters by randomly sampling the parameters (Table 1). Monte Carlo simulations were repeated over 1000 iterations to generate a distribution of ICER outcomes shown as a scatterplot (Figure 3). The cost-effectiveness acceptability curves showed the probabilities of each treatment being cost-effective under a wide range of WTP thresholds (Figure 4).