AI news ·
OpenMed Releases Models for De-Identifying Spanish Clinical Reports
OpenMed released two open models that automatically mask patient identifiers in Spanish clinical reports, but they still require human review to catch missed data.

OpenMed has released two open models designed to mask patient identifiers in Spanish clinical reports. The tools are intended to support privacy-preserving workflows by automatically detecting and obfuscating personal health information within unstructured text.
The models address a specific gap in language resources for clinical natural language processing, as most existing de-identification tools have been developed primarily for English. By making the models openly available, the release allows healthcare organizations and researchers working with Spanish-language records to integrate automated de-identification into their data processing pipelines.
OpenMed noted that the tools carry important limitations. False negatives remain a risk, meaning some identifiers may pass through undetected. Local identifiers and institution-specific patterns can also escape masking. The organization stressed that human review continues to be a critical component of any de-identification workflow using these models.