EHR-to-Registry Pipeline: Savana's NLP-Driven Approach to Real-World Evidence Generation

J Marin-Corral, N Iglesias, S Menke, et al.

Summary: This article describes a system called the Savana Next-Generation Registry (SNGR), which is designed to build research-ready datasets by reading and extracting information from the free-text portions of electronic health records — clinical notes, discharge summaries, pathology reports — using a technology called natural language processing (NLP). NLP software reads text the way a human would, identifying medically relevant information and converting it into structured data that can be analyzed. The system covers multiple languages, incorporates review by medical experts, and is designed to meet European data protection and regulatory requirements. The platform has been used in more than 65 real-world research projects and has contributed to over 30 published studies, primarily in cancer care.

Why it matters: A large share of the clinically meaningful information recorded during patient care lives in free-text notes rather than in the standardized fields that most research databases rely on. Systems that can reliably extract that information could make registry and research datasets far richer and more complete — particularly for complex conditions where a patient's full story is rarely captured in a few checkboxes. In cancer care, for instance, how a tumor is staged, what treatments were given, and how the patient responded are often described in narrative notes rather than coded fields.

What to keep in mind: This article describes the platform and its intended capabilities rather than reporting the results of a rigorous independent evaluation. The authors do not provide systematic data on how often the NLP system correctly identifies and extracts key clinical variables — measures such as accuracy and completeness across different languages, institutions, and record types. The number of publications and use cases is cited as evidence of the platform's usefulness, but publication volume reflects adoption rather than verified accuracy. Readers should view this as a credible description of a functioning system with real-world deployment, while recognizing that independent, standardized testing of extraction accuracy across varied clinical settings remains an important gap that the article does not fill.

Relevance: Real-World Data | Real-World Evidence | Clinical Informatics | Data Quality | Interoperability

Source: https://pubmed.ncbi.nlm.nih.gov/42830900/

---