A FHIR-Based Semantic Data Model for Automated Multicenter Digital Mental Health Studies: Design and Implementation Study (Preprint)

BACKGROUND Digital mental health interventions for children, adolescents, and young people, including gamified cognitive behavioural therapy (CBT) and experience sampling, show promise for scalable, low-threshold care, but their evaluation is constrained by fragmented data infrastructures and limited interoperability support for multicentre studies. Existing common data models address either retrospective research or clinical records, and none fully support automated intervention-study orchestration combined with multimodal digital biomarker collection in youth mental health. OBJECTIVE This study aimed to design, implement, and evaluate the SMILE Common Semantic Data Model (CSDM), a Fast Healthcare Interoperability Resources (FHIR) R5–based semantic data model supporting automated protocol execution and multimodal real-world data collection for a gamified digital mental health intervention delivered to children, adolescents, and young adults in community settings across 7 European countries. METHODS The CSDM was specified using the HL7 FHIR R5 information model and implemented as a HAPI FHIR 6.10.0 repository integrated with a serious game, a companion app, an identity provider (Keycloak), and an attribute-based authorization service (SAPL). Study structures, protocol orchestration, and multimodal data streams were mapped to standard FHIR resources, and automated workflows were driven by CDS Hooks. The framework was deployed across 7 study sites and evaluated with respect to interoperability and workflow automation, system performance, and data quality. RESULTS During 284 days of operation, the platform enrolled 1,713 participants across 7 study sites and 1 testing site (for research and development), generating 1,334 protocol definitions, 7,358 individualized care plans, and 85,426 executable tasks without manual intervention after enrolment. The repository contained 32,814 participant-related data resources, including 26,894 questionnaire responses, 5,617 multimodal compositions, and 303 document references, with an overall completeness score of 96.6% and reference integrity of 99.1%. Stress testing supported up to 5,000 concurrent simulated users, with REST API response times of 40-60 ms under high load and a 1-week endurance test recording 1,108,989 error-free interactions. Prior to deployment, load testing of the integration and questionnaire subsystem sustained up to 5000 concurrent simulated users, with mean questionnaire completion times of approximately 1.3 seconds and REST API response times of 40-60 ms under the highest load. A 1-week endurance test with 500 simulated users processed 9,614,925 queries across 1,108,989 error-free interactions, alongside 4000 HTTP 503 and 11 TLS errors. CONCLUSIONS Standard FHIR resources can orchestrate multicentre digital mental health intervention studies and unify structured assessments with digital biomarker streams in a single unified repository. The SMILE CSDM offers a reusable template for multicentre, real-world digital intervention studies in child and adolescent mental health. CLINICALTRIAL Trial Registration: ISRCTN32523126 (prospectively registered on April 29, 2025).