There is debate about whether marginal or conditional treatment effect estimates are of interest in the context of health technology assessment (HTA). Traditionally, the question about the “estimand” of interest has been overlooked, ignored, or implicitly assumed in HTA. In answering this question, it is important to understand the difference between marginal and conditional treatment effects. This becomes pertinent with the rapid development of methods for indirect treatment comparisons and network meta-analysis that make use of individual participant level data and not just published aggregated level data. In this session, relevant terminology will be summarized, methods to obtain marginal and conditional treatment effect estimates presented, and potential implications to the practice of indirect treatment comparisons and network meta-analysis in the context of HTA will be discussed.

DESCRIPTION:

In Part 1, Ms. Cope will introduce the terminology, explain challenges arising with non-collapsible effect measures, and define target estimands using a simple example. In Part 2, Dr. Campbell will illustrate that marginal and conditional treatment effect estimates do not necessarily equate to unadjusted and adjusted analyses and will show how to ‘marginalize’ adjusted estimates in a single study for odds ratios, carrying forward the earlier example. In Part 3, Dr. Remiro-Azócar will provide an overview of methods for population adjustment in the context of indirect comparisons and network meta-analysis with a specific emphasis on the “marginal versus conditional” issue. Dr. Jansen will provide a summary of the discussion and will facilitate a debate among the panelists regarding the implications for indirect treatment comparisons and network meta-analysis in the context of health technology assessment. Attendees will have the opportunity to pose questions after this discussion.