Archives of Pathology & Laboratory Medicine
Context: Despite years of efforts, the pathology report lacks standardization for its gross description. This inevitably leads to the occasional omission of key data elements, and difficulty to identify key elements in a lengthy unstructured report. A novel artificial intelligence (AI) tool, ChatGPT by OpenAI (San Francisco, California), was reportedly proficient in extracting keywords from text content. We assess the effectiveness of ChatGPT to quickly find multiple data elements in reports and detect missing data elements.
Design: Gross description of 444 pathology reports of several common organ types was submitted to ChatGPT (https://chat.openai.com/chat) that extracted and listed key data elements from the gross description section according to a predetermined list of data for each organ, and the results were compared with those by human proofreading as a control. The cases were grouped into 3 types based on the results for easy analysis (Table).
Results: ChatGPT correctly extracted the key data elements from the gross description in nearly all cases when the elements were present, and accurately detected missing data elements, with a negligible error rate of 0.45% (Table). It presented both contained and missing elements in a list for easy viewing.
Conclusions: ChatGPT can be utilized to identify key data elements in pathology reports in natural language. This can reduce time and effort to look for key information when signing out the case and help reduce report deficiency. This finding also justifies a future effort to make a ChatGPT-based software tool to facilitate data extraction in pathology reports and other types of medical documents.
Qu Z, Wang M, Santiago C, Qu JW, Zhang P. Enhancing pathology report analysis with artificial intelligence: a validation study of ChatGPT's effectiveness in 444 cases. Arch Pathol Lab Med. 2023 Sep;147(9):e141-e142. doi:10.5858/arpa.2023-0258-AB