Improving patient-centered inpatient clinical decision-making by applying large language models to identify and summarize serious illness conversation documentation in real-time for patients with cancer being admitted to the hospital

Patients with cancer often receive end-of-life care that may not align with their preferences or provide significant benefit. Serious illness conversations between patients and clinicians improve well-being and reduce intensive care near the end of life by aligning care with patients’ goals and preferences. However, documentation of these conversations is often buried in extensive free-text within electronic health records, making it difficult for inpatient clinical teams to identify and act on patient preferences to provide high-quality, goal-concordant care during hospitalizations.

Application of large language models (LLMs, such as GPT) may overcome this challenge. We will pilot a tool that uses an LLM to summarize documented serious illness conversations in the electronic health record for patients with cancer upon hospital admission. These summaries will be shared with both inpatient and outpatient clinical teams to enhance clinician awareness of patients' goals and preferences, prompting further conversations to ensure patient-centered care. This knowledge will accelerate the use of LLMs to summarize and improve patient-clinician communication, particularly in the hospital setting.