ai-assisted discharge note

This study out of Korea looked at if using a large language model assistant could improve efficiency in writing discharge notes in the emergency department without sacrificing accuracy or quality. The researchers from this huge tertiary care hospital developed their own AI software for use within their EHR that pulled patient specific data to write what they call a "discharge note." In this virtual training environment, they first had ED docs write their own note based on a patient chart (manual note). They then had the software generate a note (LLM draft), and the physician could edit this note (LLM-assisted note). The time it took the doctor to write each note was measured. Then 3 blinded ED docs evaluated each note (300 in total) for completeness, conciseness, correctness, and clinical utility. 

What did they find? The LLM-assisted note won nearly every category compared to both the draft and the manual note, and it took about half the time (32 seconds vs 69 seconds for the manual). The only categories the LLM-assisted note scored lower on compared to the LLM draft was for completeness and correctness. 
​
Bottom Line: while this wasn't specifically targeted at pediatric specific ED discharge instructions, I think it shows promise that using an LLM/AI could streamline your discharge process and improve efficiency without much concern for losing out on completeness, conciseness, correctness, or clinical utility (as long as you go back and check it!)