A Case Study of the Micro-Randomized Trial of HeartSteps V2

• Activity Number: 104 - Novel Methodology for the Analysis of Physical Activity Data Measured by Accelerometers
• Type: Invited
• Date/Time: Monday, August 9, 2021 : 1:30 PM to 3:20 PM
• Sponsor: Biometrics Section
• Author(s): Xiang Meng* and Peng Liao and Susan Murphy and Predrag Klasnja and Walter Dempsey and Nick Reid
• Companies: Harvard University and Harvard University and Harvard University and University of Michigan and University of Michigan and University of Washington
• Keywords: mobile health; micro-randomized trial; causal inference; HeartSteps

Abstract

HeartSteps is a mobile health physical activity intervention paired with an activity tracker to collect minute level step count and heart rate data. In two micro-randomized trials that used HeartSteps (HeartSteps V2 and V3 MRTs) on individuals with stage 1 hypertension, trial participants were repeatedly randomized every five minutes between receiving an anti-sedentary message and receiving no message based on whether the activity tracker classified them as currently sedentary. The randomization probability was determined by an algorithm that attempts to ensure that an average of 1.5 messages are delivered per day to a participant.

In this presentation, we discuss the use of recently developed statistical methods to assess a causal excursion effect of the anti-sedentary messages on a binary outcome in HeartSteps V2 and V3 MRTs. We define the causal excursion effect and discuss challenges that arise in inference for this effect from data from trials like HeartSteps V2 and V3 MRTs. These challenges include the choice of the proximal binary outcome, construction of both moderators and control variables (features), and missing data in both outcome and features.