I am involved in planning some studies in which we are interested in investigating a variety of potential biomarkers as possible surrogate endpoints. Our long term plan is to develop these surrogate endpoints to help improve the efficiency of clinical trials in our content area.
There is a long history in the statistical literature regarding the validation of biomarkers as surrogate outcomes for clinical trials. While we had anticipated that this was a place where the use of SEM techniques (e.g. mediation modeling) might be very helpful, we have found very little in the way of methodological work.
Do you know of any applications of SEM to this area?
Hello, The Clinical Trials paper addresses some links between surrogate endpoints and structural equation modeling or structural equation like approaches. There are several articles cited in this paper that are relevant for your question. The paper is on the Mplus web site under papers, mediational modeling. MacKinnon, D. P., Lockwood, C. M., Brown, Wang, W., C.H., & Hoffman, J. M. (2007). The intermediate endpoint effect in logistic and probit regression. Clinical Trials, 4, 499 - 513. [M2] http://ctj.sagepub.com/cgi/content/abstract/4/5/499 Also see Chapter 11 (and Chapter 2) in my book on mediation analysis. http://www.psypress.com/9780805839746
There are several differences between surrogate endpoints and mediators. One primary difference is often one surrogate endpoint is investigated versus several in other areas. There is also typically complete mediation by the surrogate so there is not an effect of a program on the ultimate endpoint that is not through the surrogate endpoint. No direct effect of the intervention on the ultimate endpoint can make the exclusion restriction assumption of instrumental variables valid. The notion of multiple chain mediation is also relevant where X causes M1, M1 causes M2,..leading to the ultimate endpoint.