Abstract
The translation of Artificial Intelligence (AI) into clinical practice demands high-quality Real-World Data (RWD), yet unstructured healthcare information poses a significant barrier. To address this, we developed S-RACE, a secure, cloud-based platform designed to systematically transform raw hospital data into high-quality, research-grade evidence. S-RACE features an end-to-end pipeline, starting with on-premises anonymisation for data privacy, followed by Natural Language Processing (NLP) to extract and standardise clinical information into the FHIR format. This curated data foundation is essential for building robust AI models. The platform's integrated "Data Science Lab" supports responsible AI development, incorporating explainability techniques and adhering to governance standards like ISO 42001 and the EU AI Act. Currently, S-RACE is populated with data from 31,276 patients, powering 19 research projects across fields including oncology, cardiology, and diabetes. We demonstrate its utility in kidney cancer and aortic stenosis, where models trained on S-RACE's automatically processed RWE showed performance comparable to those trained on manually curated data. S-RACE provides a scalable, governed environment for RWD curation, offering a trustworthy foundation to accelerate the clinical adoption of responsible AI.