Critical care studies using large language models based on electronic healthcare records: A technical note

Abstract

The integration of large language models (LLMs) in clinical medicine, particularly in critical care, has introduced transformative capabilities for analyzing and managing complex medical information. This technical note explores the application of LLMs, such as generative pretrained transformer 4 (GPT-4) and Qwen-Chat, in interpreting electronic healthcare records to assist with rapid patient condition assessments, predict sepsis, and automate the generation of discharge summaries. The note emphasizes the significance of LLMs in processing unstructured data from electronic health records (EHRs), extracting meaningful insights, and supporting personalized medicine through nuanced understanding of patient histories. Despite the technical complexity of deploying LLMs in clinical settings, this document provides a comprehensive guide to facilitate the effective integration of LLMs into clinical workflows, focusing on the use of DashScope's application programming interface (API) services for judgment on patient prognosis and organ support recommendations based on natural language in EHRs. By illustrating practical steps and best practices, this work aims to lower the technical barriers for clinicians and researchers, enabling broader adoption of LLMs in clinical research and practice to enhance patient care and outcomes.

Keywords: Critical care; Large language model.

Figure 1

Patient outcomes based on adherence to recommended support. This bar plot illustrates the distribution of patient outcomes according to their adherence to organ support recommendations made by a LLM. The data categorize patients into four groups: (1) Recommended and Received: Patients for whom the LLM recommended organ support, and who actually received the support. (2) Recommended but Not Received: Patients for whom the LLM recommended organ support, but who did not receive the support. (3) Not Recommended but Received: Patients for whom the LLM did not recommend organ support, but who received the support regardless. (4) Not Recommended and Not Received: Patients for whom the LLM did not recommend organ support, and who did not receive the support. The plot shows the count of patients with outcomes classified as “improved” or “worsened” within each group. The “x” axis represents the four patient groups, while the “y” axis indicates the number of patients. The outcomes are color-coded: blue for “improved” and red for “worsened.” This visualization helps in understanding the effectiveness of adhering to the LLM's support recommendations. For instance, a higher count of “improved” outcomes in the “Recommended and Received” group compared to the “Recommended but Not Received” group would suggest that following the LLM's recommendations positively impacts patient outcomes. The data used for this analysis includes 100 patients, and the statistical significance of the differences in outcomes was assessed using a chi-squared test, with results indicating a potential association between adherence to recommendations and improved outcomes. LLM: Large language model.