On October 5, 2018, Aniek Sies defended her thesis Towards precision medicine: Identifying relevant treatment-subgroup interactions and estimating optimal tree-based treatment regimes from randomized clinical trial data
When multiple treatment alternatives are available for a certain problem or disease, one may wish to look for a treatment regime, which is a decision rule that specifies for each patient the preferred treatment given his or her pretreatment characteristics. An important challenge is to find optimal treatment regimes, which are the ones leading to the greatest benefit if the entire population would be subjected to them. An interesting class of treatment regimes is that of the tree-based ones, because they provide a straightforward and most insightful representation of the decision structure underlying the associated regimes.
Recently, several methods for the construction of tree-based regimes have been proposed. Up to now, however, only partial information is available concerning their absolute and relative performance. To address this issue, my first project will be to compare and evaluate four tree-based methods by means of a simulation study. There is some preliminary evidence that skewness and outliers might influence the performance of these methods. I will look into this to get a better understanding of how and why these properties play a role. Subsequently, I will examine to what extent the methods’ performance would be improved by robustifying them in one way or another.
A second project relates to the outcome variables on which the treatment regimes are based. Many existing tree-based methods can only handle continuous outcomes. Extending these methods in a way they can handle categorical outcomes as well, is part of my second project. Another part of it will be dealing with multiple outcome variables.
Prof. Dr. Iven van Mechelen, prof. dr. Eva Ceulemans, prof. dr. Johan Vlaeyen
1 October 2014 – 1 October 2018