Evaluation of a Multi-Layered Digital Phenotyping, Digital Safety Plan, and Pharmacogenomic Intervention for Suicide Prevention in Adults With Depression

This is a multi-center, 6-month, 3-arm, parallel-group randomized controlled trial conducted across 10 outpatient psychiatric units in Spain. The study evaluates the integration of active and passive digital phenotyping, an Ecological Momentary Intervention (EMI) via a Digital Safety Plan (DSP), and prospective pharmacogenomics (PGx) and therapeutic drug monitoring (TDM). Study Design and Interventions: Participants are randomized in a 1:1:1 ratio to one of three parallel arms: * Arm 1 (Control - TAU + Digital Phenotyping): Receives standard outpatient psychiatric care (TAU) along with active Ecological Momentary Assessment (EMA via MeMind delivering 2-4 daily prompts) and passive digital phenotyping (via eB2 Mindcare tracking actigraphy, mobility, and device usage). Pharmacogenetic and TDM profiling are performed silently, with reports masked and delivered retrospectively at Week 24. * Arm 2 (Intervention - TAU + Digital Phenotyping + EMI): Receives clinical TAU, active/passive EMA monitoring, and a personalized, interactive Digital Safety Plan (DSP) integrated into the MeMind platform. The DSP acts as a Just-In-Time Adaptive Intervention (JITAI), providing real-time coping strategies and emergency contact pathways that can be user-activated or triggered automatically by algorithmic crisis detection. PGx/TDM reports remain masked until Week 24. * Arm 3 (Advanced Intervention - TAU + Digital Phenotyping + EMI + Active PGx-Guided Care): Receives clinical TAU, active/passive EMA, and the DSP-based JITAI, combined with active prospective precision clinical decision support. Treating clinicians receive actionable PGx reports at longitudinal visits (Weeks 4, 8, 12, and 24) to guide antidepressant selection and dosing according to national guidelines, supplemented by liquid chromatography-mass spectrometry (LC-MS) therapeutic drug monitoring at Week 8. Data Streams and Biomarker Integration: The trial incorporates a multimodal data architecture combining active/passive EMA, historical digital trace extraction via Google Takeout (capturing spatiotemporal mobility, search queries, and digital usage footprints), and georeferenced exposomic linkage. Biological samples are collected serially at baseline and follow-up visits (Weeks 4, 8, 12, and 24) for local biochemical safety panels and centralized multi-omics assays-including high-throughput genome-wide association studies (GWAS via Axiom™ Spain Biobank Array), epigenomic methylation profiling, transcriptomics, proteomics, metabolomics, and serum drug quantification. Statistical and Analytical Plan: Longitudinal primary and secondary outcomes will be modeled using multilevel hierarchical multivariate regression within an Intention-to-Treat (ITT) framework, incorporating fixed effects for treatment assignment, baseline characteristics, site-level covariates, and time, alongside random effects for individual baseline variability. Genomic associations will be evaluated via mixed logistic regression (SAIGE), gene-set and pathway over-representation analyses (MAGMA/WebGestalt), and Bayesian fine-mapping (susie / coloc) to identify biological targets underlying suicidal behaviour and treatment response.