A group of researchers from King’s College London, King’s College Hospital NHS Foundation Trust and University College London have studied methods and models to automatically produce discharge summaries.
The researchers explored different approaches to automatically create and populate Brief Hospital Course (BHC) summaries from inpatient documentation, noting that, if realised in practice, this “would be invaluable in reducing clinician manual burden of summarising documents under high time-pressure to admit and discharge patients”.
The authors note that “producing these summaries from the inpatient course is complex” and highlight that notes are written from various narratives including commentary from multiple care teams, specialisms and perspectives with varying scope, detail, structure and timespan covered.
The authors state that challenges around ‘computational linguistics’ can be framed as a multi-document summarisation task, with the “model required to adapt to varying numbers of documents (simple vs complex cases), large time variances between notes, differences between note types, varying source document authors aims and focus areas.”
During the study the researchers explored a range of methods for BHC summarisation, demonstrating the performance of deep learning summarisation models across extractive and abstractive summarisation scenarios. They also tested a model that incorporates SNOMED medical concept ontology as a clinical guidance signal, noting that it “showed superior performance in 2 real-world clinical data sets”.
However, the authors add that “automatic generation of BHC sections from source notes is still a long way off” and that “in any real-use scenario, a generated summary would likely only be used with explicit supervision and ultimate responsibility for the produced summary, ensuring factual correctness and coherence”.
They highlight that the area requires engineering a solution well beyond a research project, along with the challenge to ensure factual accuracy in BHC generation as they are legal records and likely to be used by follow-up care upon discharge.
The authors conclude that they hope their research motives further work in the area.
Reference: Summarisation of Electronic Health Records with Clinical Concept Guidance Thomas Searle, Zina Ibrahim, James Teo, Richard Dobson, https://doi.org/10.48550/arXiv.2211.07126.
