Assessing Moderation from Intensive Longitudinal Data: Application to MHealth Interventions

• Activity Number: 319
• Type: Contributed
• Date/Time: Tuesday, August 11, 2015 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Audrey Boruvka* and Daniel Almirall and Katie Witkiewitz and Predrag Klasnja and Susan A. Murphy
• Companies: University of Michigan and University of Michigan and University of New Mexico and University of Michigan and University of Michigan
• Keywords: longitudinal data ; causal inference ; mHealth

Abstract

In mobile health (mHealth) for behavior change and maintenance, interventions are frequent but momentary. Typically a great deal of information on patient states, environmental factors, and behavioral responses is generated over time. To investigate specific contexts in which intervention delivery is most effective, causal inference methods are needed. The structural nested mean model (SNMM) offers a framework for assessing time-varying treatment effect moderation, but model specification and estimation can prove difficult in practice. In this talk we consider an SNMM where the reference treatment strategy is the assigned or observed treatments---a choice that simplifies estimation and offers robustness to misspecification of main effects. We illustrate our approach with two mHealth interventions, targeting heavy alcohol and tobacco use among college students and sedentary behavior in adults.