Optimizing Chart Review Efficiency in Pressure Injury Evaluation Using ChatGPT.

Abstract

Introduction: Wound care is an essential discipline in plastic surgery, especially as the prevalence of chronic wounds, such as pressure injuries, is increasing. The escalating volume of patient data and the numerous variables influencing wound outcomes are making traditional manual chart reviews in wound care and research increasingly complex and burdensome. The emergence of Natural Language Processing (NLP) software based on large language models (LLMs) such as ChatGPT presents an opportunity to automate the data extraction process. This study harnesses the capabilities of ChatGPT, hosted by our medical center's secure, private Azure OpenAI service, to automatically extract and process variables from patient charts following sacral wound visits. We assess ChatGPT's potential to revolutionize chart review through improved data retrieval accuracy and efficiency.

Methods: We evaluated the use of the medical center's internal ChatGPT in chart review. ChatGPT and a Python script were integrated into the existing chart review process for patients with sacral wounds from 2 hospital cohorts to extract and format variables related to wound care. Metrics include time taken for review, accuracy of extracted information, and assessment of ChatGPT-generated insights.

Results: ChatGPT reduced the average time per chart review from 7.56 minutes with the manual method to 1.03 minutes using ChatGPT. Furthermore, it achieved a 0.957 overall accuracy rate compared to manual chart review, ranging from 0.747 to 0.986 across extracted data elements. ChatGPT was also able to synthesize accurate narrative descriptions of patient wounds.

Conclusions: We highlight ChatGPT's potential to enhance speed and precision of chart review in the context of both clinical care and wound care research, offering valuable implications for integration of artificial intelligence in healthcare workflows.