Session (CTI)

ISSUE: There has been increased recognition of the formulation of research questions through an estimand, and of the importance of appreciating the differences between marginal and conditional summary measures. When effect modification is present, conditional and marginal effect estimates can yield conflicting conclusions, creating challenges for HTA decision-making. These conflicts arise in both effectiveness-only and cost-effectiveness evaluations. For economic evaluation, a critical methodological question emerges: does the structure of the economic model and the nature of the net health benefit function align with the type of baseline risk and treatment effect estimates (i.e. conditional or marginal) being used as inputs? Furthermore, different model types have varying capabilities to handle patient heterogeneity. Understanding when and why these conflicts occur, and how to resolve them through appropriate model-estimand alignment, is essential for valid HTA conclusions. OVERVIEW: Ms. Cope will provide a session overview and outline the scope of decision-making challenges. Dr. Remiro-Azócar will provide an overview of conditional and marginal estimands, when they differ, and their implications for transportability to different target populations. Dr. Phillippo will focus on network meta-analyses and population-adjusted indirect treatment comparison methods for deriving conditional or marginal treatment effects. Dr. Jansen will address the alignment between baseline risk and treatment effect estimand types for use in model-based cost-effectiveness evaluations, how the net health benefit function determines appropriate inputs, and the capabilities and limitations of different model types to obtain the cost-effectiveness estimates of interest. The session concludes with 15 minutes of debate lead by Ms. Cope on these questions: When should treatment decisions be based on conditional versus marginal effects? How should economic models be structured to properly handle the type of baseline risk and treatment effect estimates? Where should any marginalization occur in the modeling process? This session will benefit modelers, statisticians, and decision-makers.