Piloting the Depression Digital Forecasting Tool (DEDICAT): Feasibility and Acceptability in an At-Risk Population

Objective: This pilot study examined the feasibility, acceptability, and data quality of the firstfunctional iteration of the Depression Digital Forecasting Tool (DEDICAT) in an at‑riskpopulation characterized by moderate to high rumination.Methods: Twenty‑one adults participated in a three-week observational study across twowaves during which smartphone and smartwatch data (heart rate, activity index, social mediause) were collected. Participants installed the DEDICAT app, wore a Samsung Galaxy Watch5, completed daily ecological momentary assessments (EMA), and provided feedback on stresspredictions generated via an unsupervised machine learning algorithm. We focused on(challenges related to) recruitment, adherence, passive data availability, and patterns ofmissingness, analysed using descriptive statistics and generalized linear mixed‑effects models.Outtake interviews provided additional insights into overall user experience and directions forfuture improvements.Results: With a daily average of 3.48 (SD = 1.27) out of 6 answered EMA prompts, adherenceaveraged at 57.6% (SD = 21.2%) completed EMA signals, while showing comparablecompletion rates across waves. Participants completed an average of 2.62 (SD = 1.81)stress‑prediction evaluations, though availability and contextualization of predictionsinfluenced engagement. Passive sensing completeness was moderate: 36.7% (SD = 19.6%) ofcompleted EMAs were accompanied by all three passive data streams. Smartwatch wear timeaveraged 11.4 hours/day. Missingness in both EMA and passive data increased modestly overstudy days, suggesting that fatigue and user burden could have played a role. However,missingness did not differ by wave. Interviews with participants contextualized the quantitativeresults and provided further insight into their experience with the study and the app.Conclusions: The pilot showed that intensive multimodal monitoring with DEDICAT waspartially feasible and generally acceptable in an at‑risk sample, though adherence and passivedata completeness show substantial inter‑individual variability. These findings highlight keyareas for refinements in study procedures, app design, and prediction delivery ahead of largerclinical and at‑risk trials.