Conceptual Framework and Methodological Challenges for Modeling Effectiveness in Oncology Treatment Sequence Models.

In this review, we summarize the challenges faced by existing oncology treatment sequence decision models and introduce a general framework to conceptualize such models. In the proposed framework, patients with cancer receive at least two lines of therapy (LOTs) followed by palliative care throughout their lifetime. Patients cycle through progression-free and progressive disease health states in each LOT before death. Under this framework, four broad aspects of modeling effectiveness of treatment sequences need exploration. First, disease progression, treatment discontinuation, and the relationship between the two events should be considered. Second, the effectiveness of each LOT depends on its placement in a treatment sequence as the effectiveness of later LOTs may be influenced by the earlier LOTs. Third, the treatment-free interval (TFI; time between discontinuation of earlier LOT and initiation of later LOT) may impact a therapy's effectiveness. Fourth, in the absence of head-to-head trials directly comparing LOTs, indirect treatment comparison (ITC) of outcomes for a specific LOT or even for the entire treatment sequence is important to consider. A search of decision models that estimated effectiveness of at least two lines of oncology therapy was conducted in PubMed (N = 20) and technology appraisals by the National Institute for Health and Care Excellence (N = 26) to assess four methodological aspects related to the model framework: (1) selection of outcomes for effectiveness in a treatment sequence, (2) approaches to adjust the efficacy of a treatment in consideration of its place in the sequence, (3) approaches to address TFIs between LOTs, and (4) incorporation of ITCs to estimate comparators' effectiveness in the absence of direct head-to-head evidence. Most models defined health states based on disease progression on different LOTs while estimating treatment duration outside of the main model framework (30/46) and used data from multiple data sources in different LOTs to model efficacy of a treatment sequence (41/46). No models adjusted efficacy for the characteristics of patients who switched from an earlier LOT to a later LOT or adjusted for the impact of prior therapies, and just six models considered TFIs. While 11 models applied ITC results to estimate efficacy in comparator treatment sequences, the majority limited the ITC to one LOT in the sequence. Thus, there is substantial room to improve the estimation of effectiveness for treatment sequences using existing data when comparing effectiveness of alternative treatment sequences.