Acceptability and Efficiency of Using a Large Language Model to Create Written Patient Summaries

Introduction: Providing families of ICU patients with daily written summaries—highlighting the patient's primary ICU issues and management plan—can improve communication, reduce stress, and enhance satisfaction. The use of a Large Language Model (LLM) may reduce the time required to create written communication, a primary barrier to implementing this approach.


Methods: Since July 2025, in a medical ICU at an urban academic center, we asked ICU clinicians (physicians and advanced practice providers) to identify families of non-decisional patients with whom they were experiencing a communication challenge. We developed an LLM to create written updates of 250-300 words in length and at a 6th-grade reading level based on ICU progress notes. Each ICU day, we used the LLM to generate a written update and asked a member of the patient’s ICU team to rate the AI-summary across four communication domains (clarity, conciseness, clinical utility, comprehensiveness). ICU clinicians reported the time spent editing the AI-summary prior to the final version being shared with families.


Results: The were 57 written summaries among seven patients. On a 5-point Likert scale with higher numbers indicating better quality, AI-summaries were rated as follows: clarity 4.6 (SD 0.8), conciseness 4.4 (SD 0.9), clinical utility 4.6 (SD 1.0), and comprehensiveness 4.5 (SD 1.0). Clinicians spent less than one minute editing the AI-summary in 75% of cases and never more than 5 minutes editing. Clinicians reported the following communication challenges and frequencies: complex medical issues (77%), difficult to provide updates to family (25%), complex family dynamics (21%), and high anxiety/stress (21%). In general, clinicians believed that the written communication helped address these challenges.


Conclusions: LLM-generated written summaries of ICU care were rated highly by ICU clinicians in multiple communication domains and required minimal editing prior to being shared with families. Further study is needed to learn how to integrate this approach into the ICU environment in ways that improve the experience of both ICU care providers and families.