Title: Qualitative Treatment Interaction Trees, Version 2.0

Authors: Juan Claramunt Gonzalez, Elise Dusseldorp

Affiliation: Leiden University, Institute of Psychology, Methodology &
Statistics

Abstract:
In randomized control trials, we often find differences across (subgroups of)
patients in treatment efficacy (i.e., treatment A outperforms treatment B for
some subgroups of patients while for other subgroups treatment B outperforms
treatment A). This phenomenon is called qualitative subgroup-treatment
interaction. QUINT (Qualitative INteraction Trees; Dusseldorp & Van Mechelen,
2014) is a statistical method to detect subgroups involved in such
interactions. QUINT generates a binary tree that divides the patients into
three groups (treatment A better than B, treatment B better than A, and no
difference between A and B) by assigning each final node of the tree to one of
these three groups. A recent simulation study (Sies & Van Mechelen, 2017)
showed that an initial implementation of QUINT (R-package quint 1.2;
Dusseldorp, Doove, & Van Mechelen, 2016) was less effective finding
qualitative interactions, particularly, compared to other methods such as
Model-based Recursive Partitioning (MOB; Zeileis, Hothorn & Hornik, 2008). To
improve the power of QUINT, we modified the algorithm (quint 2.0). In quint
1.2, the qualitative interaction condition was tested after the first split,
while in quint 2.0 it is tested at the end of the pruning procedure. This
modification was tested through a simulation study. The results showed an
improvement in the three evaluation criteria used: Type I and Type II error
rates and proportion of correctly assigned patients. As a consequence of this
simulation study, the corresponding update of the R package quint was released
on CRAN in September 2018. During this presentation, we will discuss the new
simulation study and a practical application of QUINT.

References:
Dusseldorp, E., & Van Mechelen, I. (2014). Qualitative interaction trees: a
tool to identify qualitative treatment-subgroup interactions. Statistics in
Medicine, 33(2), 219-237.

Dusseldorp, E., Doove, L., & Van Mechelen, I. (2016). Quint: An R package for
the identification of subgroups of clients who differ in which treatment
alternative is best for them. Behavior Research Methods, 48(2), 650-663.

Sies, A., & Van Mechelen, I. (2017). Comparing four methods for estimating
tree-based treatment regimes. The International Journal of Biostatistics,
13(1).

Zeileis, A., Hothorn, T., & Hornik, K. (2008). Model-based recursive
partitioning. Journal of Computational and Graphical Statistics, 17(2),
492-514.