Optimizing Personalized Treatment Rules in Complex Settings
Online in Zoom
Online in Zoom

Title: Optimizing Personalized Treatment Rules in Complex Settings Abstract: Heterogeneity in treatment effects in public and behavioral health settings drives the need for personalizing treatment assignments - determining which treatments are most effective for which subgroups of a population. In this talk, we discuss approaches for learning optimal treatment assignment rules in experiments with many treatment arms. We propose an 'honest' tree and forest-based approach - the optimal assignment forest - to estimate individualized assignment rules, adapted to a setting with a very large number of discrete treatment arms. We apply this method using data from a ‘mega’ randomized control trial conducted in collaboration with a national gym chain, with multiple behavioral interventions promoting the formation of lasting exercise habits. We also compare our approach to existing regression and classification-based approaches

Optimizing Personalized Treatment Rules in Complex Settings

Rahul Ladhania, PhD Assistant Professor of Health Informatics, Department of Health Management & Policy, University of Michigan School of Public Health

October 22, 2020
3:30 pm - 4:30 pm
Online in Zoom
Online URL: https://umich.zoom.us/j/95897108285
Contact Information: Amanda Larson, larsoama@umich.edu

Title: Optimizing Personalized Treatment Rules in Complex Settings Abstract: Heterogeneity in treatment effects in public and behavioral health settings drives the need for personalizing treatment assignments - determining which treatments are most effective for which subgroups of a population. In this talk, we discuss approaches for learning optimal treatment assignment rules in experiments with many treatment arms. We propose an 'honest' tree and forest-based approach - the optimal assignment forest - to estimate individualized assignment rules, adapted to a setting with a very large number of discrete treatment arms. We apply this method using data from a ‘mega’ randomized control trial conducted in collaboration with a national gym chain, with multiple behavioral interventions promoting the formation of lasting exercise habits. We also compare our approach to existing regression and classification-based approaches