@inproceedings{elgaar-etal-2026-overview,
    title = "Overview of the Medical Decision Extraction, Analysis, and Classification Task ({M}ed{E}x{ACT}) of {B}io{NLP} 2026",
    author = "Elgaar, Mohamed  and
      Cheng, Jiali  and
      Vakil, Nidhi  and
      Sadrolashrafi, Mehrnaz  and
      Mohtarami, Mitra  and
      Wong, Adrian  and
      Amiri, Hadi  and
      Celi, Leo",
    editor = "Demner-Fushman, Dina  and
      Ananiadou, Sophia  and
      Roberts, Kirk  and
      Tsujii, Junichi",
    booktitle = "{B}io{NLP} 2026",
    month = jul,
    year = "2026",
    address = "San Diego, California",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingest-acl-workshops/2026.bionlp-1.63/",
    pages = "760--769",
    ISBN = "979-8-89176-434-7",
    abstract = "This paper presents an overview of the Medical Decision Extraction, Analysis, and Classification task (MedExACT) of BioNLP 2026. The focus of this task is the extraction and labeling of medical decisions in ICU discharge summaries. The task is built on MedDec, a MIMIC-III-based dataset of 451 expert-annotated summaries, and asks systems to extract and classify spans of text that contain medical decisions according to the decision categories defined in the Decision Identification and Classification Taxonomy for Use in Medicine (DICTUM). The official ranking combines span F1 and token F1 with a worst-group robustness metric computed over sex, race, and English-proficiency subgroups. MedExACT attracted broad international interest, with 130 official submissions from 36 teams comprising about 60?100 participants, and has improved information extraction performance by nearly 15{\%} over the previous state of the art. The submitted systems predominantly use long-context encoder models, ensemble decoding, boundary-refinement modules, and robustness-aware training or model selection, with the best submitted run reaching a final fairness-based F1 of 0.596."
}