|
Abstract:
|
Treatment discontinuation can significantly limit the long-term effectiveness of mobile-health interventions. However, negative impacts of treatment, e.g., fatigue, habituation, or burden, which can lead to discontinuation can be difficult to observe and incorporate into just-in-time adaptive intervention algorithms. We model these negative effects as latent (unobserved) processes and estimate an optimal intervention strategy using Bayesian Dynamic programming. We establish frequentist performance guarantees and demonstrate its empirical performance through a suite of simulation experiments.
|
