Development of a Decision Model to Estimate the Outcomes of Treatment Sequences in Advanced Melanoma

Abstract

Background: A decision model for patients with advanced melanoma to estimate outcomes of a wide range of treatment sequences is lacking. Objectives: To develop a decision model for advanced melanoma to estimate outcomes of treatment sequences in clinical practice with the aim of supporting decision making. The article focuses on methodology and long-term health benefits. Methods: A semi-Markov model with a lifetime horizon was developed. Transitions describing disease progression, time to next treatment, and mortality were estimated from real-world data (RWD) as a function of time since starting treatment or disease progression and patient characteristics. Transitions were estimated separately for melanoma with and without a BRAF mutation and for patients with favorable and intermediate prognostic factors. All transitions can be adjusted using relative effectiveness of treatments derived from a network meta-analysis of randomized controlled trials (RCTs). The duration of treatment effect can be adjusted to obtain outcomes under different assumptions. Results: The model distinguishes 3 lines of systemic treatment for melanoma with a BRAF mutation and 2 lines of systemic treatment for melanoma without a BRAF mutation. Life expectancy ranged from 7.8 to 12.0 years in patients with favorable prognostic factors and from 5.1 to 8.7 years in patients with intermediate prognostic factors when treated with sequences consisting of targeted therapies and immunotherapies. Scenario analyses illustrate how estimates of life expectancy depend on the duration of treatment effect. Conclusion: The model is flexible because it can accommodate different treatments and treatment sequences, and the duration of treatment effects and the transitions influenced by treatment can be adjusted. We show how using RWD and data from RCTs can harness advantages of both data sources, guiding the development of future decision models. The model is flexible because it can accommodate different treatments and treatment sequences, and the duration of treatment effects as well as the transitions that are influenced by treatment can be adjusted. The long-term health benefits of treatment sequences depend on the place of different therapies within a treatment sequence. Assumptions about the duration of relative treatment effects influence the estimates of long-term health benefits. We show how the use of real-world data and data from randomized controlled trials harness the advantages of both data sources, guiding the development of future decision models.