A meta-learning method for estimation of causal excursion effects to assess time-varying moderation

Area of Science:

• Health Informatics
• Biostatistics
• Mobile Health Interventions

Background:

• Mobile health (mHealth) interventions, enhanced by wearable tech and smartphones, are increasingly accessible.
• Micro-randomized trials (MRTs) evaluate mHealth effectiveness and introduce causal excursion effects to study time-varying intervention impacts.
• Current MRT analysis methods struggle with real-world mHealth data complexities like uncertain randomization and missing observations.

Purpose of the Study:

• To develop a flexible and robust method for estimating causal excursion effects in mHealth.
• To address limitations of existing methods regarding uncertain randomization probabilities and incomplete data.
• To provide a meta-learner perspective for analyzing complex mHealth intervention data.

Main Methods:

• Proposed a doubly robust inferential procedure, DR-WCLS, for estimating causal excursion effects.
• Investigated bidirectional asymptotic properties of the proposed estimators.
• Compared DR-WCLS with existing methods theoretically and through simulations.

Main Results:

• DR-WCLS provides consistent and more efficient estimates compared to existing methods.
• The proposed method performs well even with missing observations or uncertain treatment randomization probabilities.
• Demonstrated practical utility by analyzing data from a cohort of medical residents.

Conclusions:

• The DR-WCLS method offers a flexible and robust approach for analyzing causal excursion effects in mHealth.
• This method enhances the reliability of mHealth intervention evaluations, particularly in complex, real-world scenarios.
• The findings support the use of DR-WCLS for more accurate assessment of time-varying mHealth intervention effects.