AI xPress

AI and computer science research.

Back to Foragentis

Healthcare & Bio AI

Adapted large language models can outperform medical experts in clinical text summarization

Healthcare & Bio AI November 21, 2026

An MRI scanner in a hospital room

Adapted large language models can outperform medical experts at summarizing clinical text, according to a 2024 study in Nature Medicine. Researchers tuned eight large language models, AI systems trained on large amounts of text, for four clinical summarization tasks. In a reader study, ten physicians rated summaries from the best adapted models as preferable to human-written summaries for completeness and correctness.

What did the study test?

The four tasks were radiology reports, patient questions, progress notes, and doctor-patient dialogue. The researchers used domain adaptation, a process that adjusts a general model to a specific type of text, across six datasets. Their quantitative assessment found trade-offs between models and adaptation methods, including cases where recent language model advances may not improve results.

What did the reader study find?

Ten physicians compared summaries from the best adapted models with human-written summaries. The physicians preferred the adapted model summaries for completeness and correctness. The authors describe this as the first evidence of large language models outperforming medical experts in clinical text summarization across multiple tasks.

What could this mean for health records?

The authors suggest that adding these models to clinical workflows could ease documentation burden. That might free clinicians to focus more on personalized patient care and the human parts of medicine.

What is clinical text summarization?

Clinical text summarization condenses long health records, such as radiology reports and progress notes, into shorter summaries for clinicians.

What kind of machine learning is this?

Large language models are a type of machine learning system trained on large amounts of text. The study used domain adaptation to tune general models for medical writing.

Did physicians use the models in a live medical setting?

The abstract describes a reader study with ten physicians evaluating summaries. It does not report testing in a live clinical setting.

Limits and what's next

The reader study included ten physicians. The model evaluation used six datasets and four clinical summarization tasks. This is one study, and the abstract does not describe independent replication or testing in a live clinical setting. The authors describe the possible reduction in documentation burden as a future implication, not as a measured outcome.

Source: Dave Van Veen et al., “Adapted large language models can outperform medical experts in clinical text summarization,” Nature Medicine, 2024.