In this work, we propose our top-ranking (2nd place) pipeline for the generation of discharge summary subsections as a part of the BioNLP 2024 Shared Task 2: “Discharge Me!”. We evaluate both encoder-decoder and state-of-the-art decoder-only language models on the generation of two key sections of the discharge summary. To evaluate the ability of NLP methods to further alleviate the documentation burden on physicians, we also design a novel pipeline to generate the brief hospital course directly from structured information found in the EHR. Finally, we evaluate a constrained beam search approach to inject external knowledge about relevant patient problems into the text generation process. We find that a BioBART model fine-tuned on a larger fraction of the data without constrained beam search outperforms all other models.

Clinical knowledge graphs lack meaningful diagnostic relations (e.g. comorbidities, sign/symptoms), limiting their ability to represent real-world diagnostic processes. Previous methods in biomedical relation extraction have focused on concept relations, such as gene-disease and disease-drug, and largely ignored clinical processes. In this thesis, we leverage a clinical reasoning ontology and propose methods to extract such relations from a physician-facing point-of-care reference wiki and consumer health resource texts. Given the lack of data labeled with diagnostic relations, we also propose new methods of evaluating the correctness of extracted triples in the zero-shot setting. We describe a process for the intrinsic evaluation of new facts by triple confidence filtering and clinician manual review, as well extrinsic evaluation in the form of a differential diagnosis prediction task.