Analytic Overview
A microsimulation model was constructed to estimate the lifetime cost and health outcomes of 4 treatment strategies for patients with BRAF-mutant advanced melanoma by using TreeAge 2020 Software. (Fig. 1 &2) Each model cycle represented 28 days over a lifetime horizon. Based on the cohort enrolled in the respective randomized control trials (RCTs) published previously, we generated a 50,000 baseline sample of patients.9–11 Patients had a mean age of 54 years (calculated by averaging the mean age of the patients in the 3 RCTs), and all patients had BRAF mutation advanced melanoma.
Patients who were included in the model were treated with 1 of 4 frontline interventions: (1) atezolizumab-vemurafenib-cobimetinib, (2) vemurafenib-plus-cobimetinib, (3) dabrafenib-plus-trametinib, or (4) encorafenib-plus-binimetinib. Patients were prescribed nivolumab as second-line treatment after disease progression, and finally, all patients were switched to the best supportive care (BSC) state until death. (Fig. 1) Dosing and administration schedules for all strategies were based on the respective RCTs and are listed in eTable 2 in the supplement.
The primary outcomes of the model included total cost, quality-adjusted life-years (QALYs) and life-years (LYs), which were used to estimate the incremental cost-effectiveness ratios (ICERs). We conducted this analysis from a US payer perspective, adjusting all the costs to 2021 U.S. dollars (USD) and discounting both cost and utilities by 3% annually.14 Moreover, a willingness-to-pay threshold (WTP) of $150,000/QALY was applied in this cost-effectiveness analysis. 15
Model Structure
The model was classified as four states: PFS, progress-disease (PD), discontinuation due to AEs, or death due to other causes. (Fig. 2) Patients were decided to remain on current treatment or switched to next line therapy on the basis of transition probability which was estimated by using PFS or OS survival data obtained from previous respective studies.9–11, 16,17 The overall mortality rate (except for BSC state) during each line of active treatment corresponded to the transition probability for death state, calculated as the background mortality rate from the 2017 US Life Table 18 and treatment-related serious AEs from RCTs.9–11, 16 (The US Life Table is listed in eTable 3 in the supplement.) As same as previous study, we applied Formula 1 as follows to transform the rates in the Life Table and treatment-related serious AEs to the transition probability.19 The minimum value was taken between the probability for death state and 1.20 Based on the mortality data for metastatic melanoma, the probability of death from the BSC state was calculated by implementing the standard extrapolation method derived by Guyot et al.21 Briefly, we extracted the data points from each OS Kaplan-Meier curve from the respective RCT to generate pseudoindividual patient-level data by using Getdata digital software.17 These reconstructed survival data were then used to fit 4 standard parametric models (log-logistic, Weibull, lognormal, and exponential). On the basis of the goodness-of-fit method (Akaike information criterion), we selected the most appropriate survival distribution to estimate the transition probability.
\(P=1-exp\left\{-r\left.t\right\}\right.\) Formula 1
*The P is representing the probability; r is the rate; and t is the time.
The probabilities of patients remaining on current treatment or progressing to subsequent therapy were estimated on the basis of the PFS curves derived from respective RCTs using the same approach as the transition probabilities of OS.9–11 The fitting results showed that log-logistic distribution was the optimal survival distribution for all PFS and OS curves. The fitting curves of PFS and OS are presented in eFigures 4 to 9 in the supplementary material. The transition probability of the event was calculated by using 1 to minus the ratio of the survivor function at the end of the interval to the survivor function at the beginning of the interval.19 (Formula 2) And the survival function of log-logistic is represented in Formula 3, so Formula 2 could be rewritten to the probability density function of log-logistic as follows.19 (Formula 4) We also considered discontinuation of treatment related to AEs, with transition probabilities obtained from the literature.9–11 All the estimated transition probabilities are displayed in Table 1.
Table 1
Input parameters of model.
Parameters
|
Mean
|
Range
|
distribution
|
Reference
|
Survival model of PFS in the full cohort
|
Atezolizumab + vemurafenib + cobimetinib
|
Shape = 1.375;
Scale = 13.807
|
|
Loglogistic
|
9
|
Vemurafenib + cobimetinib
|
Shape = 1.663;
Scale = 10.926
|
|
Loglogistic
|
9
|
Dabrafenib + trametinib
|
Shape = 1.4075;
Scale = 12.9753
|
|
Loglogistic
|
11
|
Encorafenib + binimetinib
|
Shape = 1.668;
Scale = 14.080
|
|
Loglogistic
|
10
|
Nivolumab
|
Shape = 1.2021;
Scale = 4.6024
|
|
Loglogistic
|
16
|
OS in the best support care
|
Shape = 1.757;
Scale = 7.007
|
|
Loglogistic
|
17
|
Probability of treatment discontinuation as a result of AE (%)
|
Atezolizumab + vemurafenib + cobimetinib
|
13
|
|
Beta
|
9
|
Vemurafenib + cobimetinib
|
16
|
|
Beta
|
9
|
Dabrafenib + trametinib
|
26
|
|
Beta
|
11
|
Encorafenib + Binimetinib
|
13
|
|
Beta
|
10
|
Nivolumab
|
|
|
Beta
|
16
|
Probability of treatment mortality as a result of AE (%)
|
Atezolizumab + vemurafenib + cobimetinib
|
NA
|
|
Beta
|
9
|
Vemurafenib + cobimetinib
|
NA
|
|
Beta
|
9
|
Dabrafenib + trametinib
|
1.5
|
|
Beta
|
11
|
Encorafenib + Binimetinib
|
0
|
|
Beta
|
10
|
Nivolumab
|
0
|
|
Beta
|
16
|
Probability of background death
|
-
|
-
|
-
|
18
|
Drug cost, $ (per mg)
|
Atezolizumab
|
7.82
|
6.26–9.38
|
Gamma
|
22
|
Vemurafenib
|
0.16
|
0.13–0.19
|
Gamma
|
23–25
|
Cobimetinib
|
4.32
|
3.46–5.18
|
Gamma
|
23–25
|
Dabrafenib
|
0.89
|
0.71–1.07
|
Gamma
|
23–25
|
Trametinib
|
144.95
|
115.96-173.94
|
Gamma
|
23–25
|
Encorafenib
|
0.89
|
0.71–1.07
|
Gamma
|
24–26
|
Binimetinib
|
4.63
|
3.70–5.56
|
Gamma
|
24,25,27
|
Nivolumab
|
28.54
|
22.83–32.25
|
Gamma
|
22
|
Cost of best support care
|
4319
|
3455.2-5182.8
|
Gamma
|
29
|
Management of AEs, $
|
Atezolizumab + vemurafenib + cobimetinib
|
972
|
687.6-1166.4
|
Gamma
|
30
|
Vemurafenib + cobimetinib
|
1080
|
864–1296
|
Gamma
|
30
|
Dabrafenib + trametinib
|
1229
|
921-1536a
|
Gamma
|
29
|
Encorafenib + Binimetinib
|
1587
|
1269.6-1904.4
|
Gamma
|
30
|
Nivolumab
|
2688
|
2150.4-3225.6
|
Gamma
|
25
|
Administration cost
|
IV infusion, single or initial drug (≤ 1 hour)
|
148.3
|
118.64-177.93
|
Gamma
|
28
|
Utilities
|
Complete/partial response
|
0.88
|
0.70-1.00
|
Beta
|
4
|
Stable disease
|
0.80
|
0.64–0.96
|
Beta
|
4
|
Progressive disease
|
0.52
|
0.42–0.62
|
Beta
|
4
|
Disutility due to AEs (grade ≥ 3)
|
0.077
|
0.074–0.08a
|
Beta
|
29
|
Average patient weight (kg)
|
70
|
|
Normal
|
4
|
Terminal care, $
|
17346
|
13009–21682a
|
Gamma
|
29
|
a The range is the reported or estimated 95% CI. |
\(tp\left({t}_{u}\right)=1-S\left(t\right)/S(t-u)\) Formula 2
\(S\left(t\right)={[1+(t/\gamma )^\lambda ]}^{-1}\) Formula 3
\(f\left(t\right)=\frac{{ \left(\frac{t}{\gamma }\right)}^{\lambda }-{ \left(\frac{t-u}{\gamma }\right)}^{\lambda }}{1+{ \left(\frac{t}{\gamma }\right)}^{\lambda }}\) Formula 4
*the γ is representing the shape of the distribution and the λ parameter regarded as the scale.
Costs and Utilities
All the model input parameters, including the costs and utilities, are listed in Table 1. In this model, we incorporated only direct costs, including drugs, administration, BSC, management of AEs, and terminal care. The unit costs of atezolizumab and nivolumab were derived from the 2021 average sale price from the Centers for Medicare & Medicaid Services (CMS).22 The prices of oral drugs, including vemurafenib, cobimetinib, dabrafenib, trametinib, encorafenib, and binimetinib, were based on public databases and the literature.23–27 The 2021 average wholesale price from RED BOOK Online was discounted at a rate of 17% to account for contract pricing and to be consistent with evaluates for Medicare reimbursement.24 Although the average patient weight is 74.7 kg in the US, a patient weight of 70 kg was used to calculate the total cost of treatments so that the weight loss effects of the advanced disease could be taken into consideration.4 The administration fee was derived from the 2021 CMS Physician Fee Schedule.28 The overall costs related to BSC, management of grade 3 or 4 AEs and terminal care were obtained from the published literature.25,29,30
The mean utility for each health state was obtained from published analyses. We allocated a utility of 0.88 for patients who had a complete or partial response to therapy, 0.80 for patients in first-line treatment, and 0.52 for progressive disease.31,32 The utility decrement (-0.077) was also adopted to specify the reduction in the valued QALY for an adverse drug reaction.4 QALYs are a way to measure individual- or group-obtained health benefits (length of life), which are adjusted to reflect the quality of life. In this study, QALYs were estimated by calculating the years of life remaining for the patient with BRAF-mutant advanced melanoma following our predefined treatment sequences and weighing over the lifetime horizon with a quality-of-life score. In addition, 1 QALY is equal to 1 year of life in full health. The calculation of the LYs was the same as that of the QALYs; however, LYs was regarded as the health state of a patient in perfect health (the utility was equal to 1).
Sensitivity Analyses
Multiple sensitivity analyses were performed to test the uncertainty of the model and to evaluate the robustness of our outcomes. In the univariable sensitivity analyses, we varied the model parameters based on their upper and lower limitations by using their 95% CIs or changing them by 20% from baseline to examine the impact of variables on outcomes, in accordance with the existing approach.4,33,34 Moreover, a Monte Carlo simulation of 3000 iterations of 6000 patients was generated to perform probability sensitivity analyses (PSAs) by using suitable distributions to sample the key model parameters. Utilities were represented by beta distribution, costs by gamma, and weight and median starting age by normal distribution. Based on PSA, a cost-effectiveness acceptability curve (CEAC) was developed to portray the likelihood that a competing strategy would be considered a cost-effective option at different WTP thresholds for health benefits (QALYs).
We also incorporated 3 scenario analyses in this study. First, to evaluate the influence of survival curve extrapolations simulated in the model, the time horizon was varied in different time spans (10, 20, and 30 years). Second, we assumed that some patients would experience discontinuation and switch to the BSC state in a certain proportion (10–30%). Finally, the price of atezolizumab was decreased to 75%, 50%, or 25% of its original cost.