A0631
Title: Zero-inflated latent class mixed models for characterizing longitudinal engagement patterns
Authors: Nicholas Illenberger - NYU Langone Health (United States) [presenting]
Abstract: The digital monitoring of patient and personal health data provides unique opportunities to improve population health outcomes. Digital health applications aimed at tracking food intake and exercise over time, for example, have shown promise in reducing the risk of diabetes when used in consultation with a patient's physician. However, the reach of these applications may be limited by differential uptake across different sectors of the population. Latent class approaches may be useful in identifying different patterns of engagement with these applications, enabling researchers to tailor future developments towards participants that may be insufficiently reached. However, measuring participant engagement with these digital platforms can be complicated in the presence of zero-inflated or missing observations. A zero-inflated extension of the latent class linear mixed effects model is developed, which can be used to identify classes of engagement trajectories based not only on expected levels of engagement but also on expected probabilities of meaningful non-engagement or missingness at each time point. The proposed methodology is applied to identify longitudinal engagement patterns among users of a digital diabetes prevention program application.
