Assessing large language models for data standardization in healthcare data interoperability

Abstract

A large amount of public data is being generated, but the lack of standard formats and field names makes it hard to combine and use data efficiently. This lack of standardization limits the ability to analyze and apply data effectively across platforms. The development of structured formats and standard naming of fields guarantees interoperability for maximum effect. In healthcare, LLMs have been improved in natural language, but struggle with more complex tasks that require specialized medical knowledge. Such limitations make it difficult to rely on LLMs for advanced healthcare applications. This study examines how efficiently LLMs can map clinical terminology to SNOMED Clinical Terms identifiers in healthcare datasets. Although LLMs show some promise, several clear challenges reduce their effectiveness within health interoperability. These findings underscore the need for better model training and domain-specific fine-tuning with thorough testing before deployment in clinical functions.