What question did this study set out to answer?

This research aims to evaluate the effectiveness of large language models as encoders for electronic health records in clinical prediction tasks.

July 9, 2026npj Digital Medicine5 citationsOpen Access

Large Language Models as Effective Encoders for Electronic Health Records

Large language models are powerful electronic health record encoders

View Full Paper

Ask AI

Bookmark

Authors

Randomized trial shows large language models enhance predictive accuracy in electronic health records, hinting at new data processing methods.

This research aims to evaluate the effectiveness of large language models as encoders for electronic health records in clinical prediction tasks.
Converted EHR data to plain text by replacing medical codes with natural-language descriptions.
Compared performance of LLM-based embeddings with a specialized EHR foundation model across 15 clinical tasks from the EHRSHOT benchmark.
Conducted external validation using UK Biobank data.
LLM-based embeddings perform on par with CLMBR-T-Base across 15 clinical tasks.
In external validation, LLM-based model shows statistically significant improvements for some tasks.
Higher vocabulary coverage and better generalization contribute to LLM performance advantages.

Hegselmann et al. (2026) studied this question.

View Full Paper

Ask AI

Bookmark

July 9, 2026npj Digital Medicine5 citationsOpen Access

Large language models are powerful electronic health record encoders

View Full Paper

Ask AI

Bookmark

Authors

Randomized trial shows large language models enhance predictive accuracy in electronic health records, hinting at new data processing methods.

This research aims to evaluate the effectiveness of large language models as encoders for electronic health records in clinical prediction tasks.
Converted EHR data to plain text by replacing medical codes with natural-language descriptions.
Compared performance of LLM-based embeddings with a specialized EHR foundation model across 15 clinical tasks from the EHRSHOT benchmark.
Conducted external validation using UK Biobank data.
LLM-based embeddings perform on par with CLMBR-T-Base across 15 clinical tasks.
In external validation, LLM-based model shows statistically significant improvements for some tasks.
Higher vocabulary coverage and better generalization contribute to LLM performance advantages.

Hegselmann et al. (2026) studied this question.

View Full Paper

Ask AI

Bookmark