Abstract:
|
Treatment response heterogeneity has long been observed in patients affected by chronic diseases, which calls for a shift from a non-personalized approach to a personalized approach. Administering individualized treatment rule (ITR) offers an opportunity to achieve personalized medicine. In clinical practices, an informative and useful ITR should be simple and interpretable. It should also maintain certain flexibility and lead to improved benefit in subgroups of patients. Current statistical methods provide ITRs that lack transparency. We propose a tree-based robust machine learning method to estimate simple ITR and identify subgroups of patients with large benefit. We simultaneously identify qualitative and quantitative interactions and fit piece-wise linear rules. We show the proposed machine learning method has much improved performance comparing to existing ones via simulation studies. Lastly, we apply the method to Sequenced Treatment Alternative to Relieve Depression trial.
|