<ul><li>PMOA-TTS is introduced as the first openly available dataset of 124,699 PubMed Open Access case reports, each converted into structured timelines containing over 5.6 million timestamped clinical events.</li><li>The dataset was created using a combination of heuristic filtering, Llama 3.3, and DeepSeek R1 to extract single-patient case reports and generate prompt-driven extraction.</li><li>Evaluation against a clinician-curated reference set showed high quality in terms of event-level matching, temporal concordance, and timestamp alignment with wide diagnostic and demographic coverage.</li><li>In a survival prediction task, embeddings from extracted timelines demonstrated predictive value with time-dependent concordance indices up to 0.82, showcasing the potential for temporal reasoning and longitudinal modeling in biomedical NLP.</li></ul>

PMOA-TTS: Introducing the PubMed Open Access Textual Times Series Corpus

Discover more