Large language models could make natural language again the universal interface of healthcare
Jakob Nikolas Kather,Dyke Ferber,Isabella C. Wiest,Stephen Gilbert,Daniel Truhn
DOI: https://doi.org/10.1038/s41591-024-03199-w
IF: 82.9
2024-08-26
Nature Medicine
Abstract:Human healthcare providers communicate in natural language, such as "a 25-year-old male, 2 days after laparoscopic appendectomy for acute appendicitis, with normal bowel movements, no fever, mobilized, and ready for discharge today". Recorded in medical codes, this might read: "a patient with K35, having undergone CPT 44970, set for discharge". During their training, healthcare professionals are taught mnemonics to structure text using terms such as 'airway, breathing, circulation, disability, exposure' (ABCDE), 'subjective, objective, assessment, plan' (SOAP) or 'symptoms, allergies, medications, past medical history, last meal, events leading up to the illness or injury' (SAMPLE). Such mnemonics, which are in natural language, are meant to ensure that essential information is obtained from a patient. Mnemonics are loose structuring systems that are sufficient to transmit information and serve as the de facto standard in healthcare communication.
biochemistry & molecular biology,cell biology,medicine, research & experimental