Engagement Dynamics in Unstructured Digital Journaling: Insights from Eight Months of Multimodal Data in Early Outpatient Psychiatric Care

Predicting the natural course and treatment response of individuals seeking psychiatric care remains elusive using traditional clinical instruments alone. The PREDiCTOR study addresses this by leveraging open-ended audio and text journals from a diverse population (ages 15-45) enrolled at their first outpatient psychiatric visit, regardless of specific diagnosis. This paper presents an initial analysis of 8 months of active data collection from 365 subjects, providing insights into subjects engagement patterns aiming to support the development of models designed to predict treatment outcomes, including emergency department visits, hospitalizations, and treatment disengagement. The data describe longitudinal trends in submission behavior across monthly cohorts and the distribution of data density relative to daily push notifications. Furthermore, we report on variations in submission characteristics-such as recording duration and word count-observed between initial and subsequent entries. The insights derived from this initial analysis provide a detailed empirical account of how unstructured digital journals are utilized within a longitudinal clinical framework, serving as a guide for future studies in optimizing data collection protocols and participant engagement strategies.