A Collaborative Approach to Building a Hospital-Centered OMOP Data Layer for Clinically Meaningful ML/AI Applications in Sepsis

Swiss hospitals are adopting the OMOP Common Data Model, building on national semantic foundations established by the Swiss Personalised Health Network (SPHN). Yet clinicians and data teams still lack operational guidance to represent complex conditions in ways that preserve clinical meaning and support reproducible artificial intelligence (AI) applications. We present a collaborative, hospital-driven approach to co-designing an OMOP-aligned data layer for sepsis that connects clinical modeling decisions with AI-readiness and lifecycle requirements. Rather than proposing a new standard, the approach integrates existing semantic frameworks, hospital constraints, and AI reproducibility needs through a modified study-a-thon methodology. The work resulted in shared modeling conventions, ETL patterns, and early specifications for AI metadata and prediction integration, documented in a public OSF project and implemented in multi-hospital test beds. The approach provides a reusable, hospital-centered blueprint for building AI-ready OMOP data layers across clinical domains.