Quantifying the indirect treatment effect via surrogate markers.

As to the proportion of treatment effect (PTE) explained by surrogate markers, existing research has been focused on how to decompose the treatment effect into two parts: the treatment effect via surrogate markers and the treatment effect not explained by surrogate markers. Most proposed methods quantify the PTE explained by a single surrogate marker, or quantify the PTE explained by multiple surrogate markers without considering the cause-effect relationship among surrogate markers. The change of one marker may sometimes be due to changes in other markers. In this case, quantifying the association between multiple treatment effects via surrogates may also be important. In this paper, a method to quantify the possible causal relationship between surrogate markers is proposed. The new method can also be related to path analysis, a widely used analysis in sociology. Therefore, the proposed method can be viewed as a generalization of path analysis to generalized linear models and Cox regression models. Copyright (c) 2005 John Wiley & Sons, Ltd.