Title: Detection of treatment-subgroup interactions in clustered datasets:
Combining model-based recursive partitioning and random-effects estimation.

Author: Marjolein Fokkema, Niels Smits, Achim Zeileis, Torsten Hothorn,
Henk Kelderman

Affiliation: VU University Amsterdam

Abstract: Identification of subgroups of patients for which treatment A is
more effective than treatment B, and vice versa, is of key importance to
the development of personalized medicine.  Several tree-based algorithms
have been developed for detection of such treatment-subgroup interactions
in datasets from single randomized clinical trials (RCTs).  In many
instances, however, datasets from several RCTs are pooled, or single RCTs
are carried out in multiple research centers, introducing a clustered data
structure.  In the current paper, we propose an algorithm that allows for
the detection of treatment-subgroup interactions, as well as the estimation
of cluster-specific random effects.  The new algorithm uses model-based
recursive partitioning (Zeileis et al., 2008) to detect treatment-subgroups
interactions, and a linear mixed-effects model (Pinheiro & Bates, 2000) for
estimation of random-effects parameters.  In a simulation study, we
evaluate the performance of the new algorithm, in terms of recovery of
treatment-subgroup interactions and prediction of treatment differences,
and compare it with that of model-based recursive partitioning without
random effects.  The new algorithm outperformed model-based recursive
partitioning without random effect in terms of recovery of true
treatment-subgroup interactions, and prediction of differences in treatment
effects.  We conclude that the new algorithm is an accurate and promising
algorithm for the detection of treatment-subgroup interactions in clustered
datasets, and discuss directions for future research.