Researchers use GPT-5 to automatically extract timelines from diabetes case reports, revealing GLP-1 drugs may reduce respiratory complications by 74%

Scientists created a dataset of 136 PubMed case reports on GLP-1 receptor agonists with manually-annotated clinical timelines to train LLMs

GPT-5 achieved 87.1% event coverage and 84.3% accuracy in correctly sequencing symptoms, diagnoses, and treatments from clinical text

Time-to-event analysis found GLP-1 users had significantly lower respiratory risk compared to non-users (hazard ratio of 0.259, p<0.05)

The automated timeline extraction addresses a key challenge in converting narrative case reports into structured data for longitudinal medical research