Keywords: Qualitative Interaction Trees, Validation
Identifying the optimal treatment for a given patient is the primary goal of data-driven methods for subgroup identification in clinical trials, particularly in palliative care for advanced illness where time and resources are limited. In the recently developed subgroup identification methodology, Qualitative Interaction Trees (Dusseldorp et al, Statistics in Medicine, 2014), the authors propose a bootstrapped internal validation procedure for small sample sizes. This approach has not been thoroughly explored, so we will compare their proposed validation procedure with external validation. Internal validation methods will be illustrated with a randomized trial (n=135) of a reflection and life completion intervention versus relaxation mediation for patients with advanced illness in a University outpatient setting. Preliminary results indicate qualitative treatment interactions for subgroups with higher/lower functional status and preparation for dying. External validation methods will be conducted in a comparable randomized trial (n=149) with identical interventions for Veterans with advanced illness.