Model-Twin Randomization (MoTR) for Estimating the Recurring Individual Treatment Effect

Area of Science:

• Personalized health
• Causal inference in longitudinal data
• Behavioral science

Background:

• Wearable sensors and mobile apps generate dense, single-person time-series data.
• Caregivers and self-trackers aim to use this data for health behavior change.
• Distinguishing correlation from causation in individual data is crucial for effective interventions.

Purpose of the Study:

• To estimate the within-individual average treatment effects of physical activity on sleep duration.
• To introduce a novel method, model-twin randomization (MoTR), for analyzing intensive longitudinal data.
• To demonstrate how causal inference can improve personalized health recommendations.

Main Methods:

• Developed the model-twin randomization (MoTR) method, an application of the g-formula under serial interference.
• Estimated stable, recurring individual treatment effects, akin to n-of-1 trials and single-case experimental designs.
• Analyzed nearly eight years of personal Fitbit step count and sleep data.

Main Results:

• The MoTR method was applied to estimate the causal effect of physical activity on sleep duration.
• Compared MoTR to standard methods, highlighting its ability to handle potential confounding.
• The analysis provided insights into personalized health behavior change using individual time-series data.

Conclusions:

• MoTR offers a robust approach for analyzing intensive longitudinal data to infer causal relationships.
• Causal inference methods, like MoTR, are essential for generating effective personalized health behavior change recommendations.
• Individualized analysis of personal health data can lead to better health outcomes.