ABSTRACT
Unweighted Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56 4.4 General Considerations in the Multiple-Trial Setting . . . . . . . . . . . 57 4.5 Prediction of Treatment Effect: Surrogate Threshold Effect . . . . 59 4.6 Case Study: The Age-Related Macular Degeneration Trial . . . . . 61
Over the years, it has become clear that the single-trial setting is too restrictive for the evaluation of surrogate endpoints (for details, see Chapter 3). Nowadays, general agreement has grown that there is a need for trial-level as well as individual-level replication. A first formal proposal along these lines used Bayesian methods (Daniels and Hughes, 1997). These ideas were subsequently extended using the theory of linear mixed-effects models (Buyse et al., 2000) and generalized estimating equations (Gail et al., 2000). This chapter focuses on the approach proposed in (Buyse et al., 2000). The methodology assumes that both endpoints are normally distributed random variables that are measured cross-sectionally. In Chapters 5-11, non-normally distributed
with
endpoints (e.g., categorical, time-to-event, etc.) and longitudinally measured continuous endpoints will be considered.
