Ensemble Hospitalier de la Côte pilots generative AI to assist doctors drafting discharge letters with Smile

Overview

Ensemble Hospitalier de la Côte worked with Smile to reduce the time doctors spend drafting hospital discharge letters in internal medicine. Two priority features were developed: contextual prediction of the investigation text based on the patient's diagnoses and laboratory results, and automated retrieval and formatting of laboratory results. Nearly 3,000 discharge letters were used for training. After a proof-of-concept live demo to 15 doctors, physicians self-estimated time savings of 10 minutes for a simple letter to 30 minutes for a complex one, and it was decided to proceed with phased industrialization.

The challenge

The hospital wanted to optimize medical time, particularly in drafting hospital discharge letters, which are mandatory after each stay and require considerable time for drafting and proofreading, and in the manual transcription of laboratory results, a time-consuming and error-prone task.

The solution

After analysis, two priority features were identified: a contextual prediction of the investigation text based on the patient's diagnoses and laboratory results, and the automation of the retrieval and formatting of laboratory results. The model was trained on nearly 3,000 discharge letters.

Reported business value

At the end of the proof of concept, all physicians present felt the solution could save them time when writing discharge letters; their self-estimates ranged from 10 minutes for a simple letter to 30 minutes for a complex one. These are self-estimates rather than measured results.