Casual effects in randomised trials with partial compliance

Specialeforsvar ved Morten Thomas Rytgaard

Titel: Causal effects in randomised trials with partial compliance

 Abstract: This thesis considers causal effect estimation in randomised trials with partial compliance. We are motivated by a study performed by Novo Nordisk A/S in which patients are randomised to follow a treatment over time so that there may be patients that only partially comply to the treatment they were randomised to. In the thesis we give a general introduction to the mathematical formulation of causal effects which allows us to characterise randomised trials and the problems arising with non-adherence. We introduce instrumental variables to bypass these problems concerning lack of adherence and present an estimation procedure. We show that our estimator is consistent and derive an asymptotic distribution for the estimator. In the data analysis of the study conducted by Novo Nordisk A/S, we compare this esti-mator to other methods, and find that using our proposed method the treatment has an average causal effect on the body weight change. We close with a summary and a discussion about the advantages of our estimator compared to the disadvantages concerning the effect of the control treatment.

Vejleder:           Torben Martinussen, Frank Eriksson, SUND
Medvejleder:   Susanne Ditlevsen
Censor:             Søren Andersen, Novo