Document Type

Conference Proceeding

Publication Date

9-2023

Publication Title

Archives of Pathology & Laboratory Medicine

Abstract

Context: A tumor synoptic report (TSR) audit is an important quality control tool for compliance with reporting requirements; however, no simple audit tool is readily available to most pathologists. ChatGPT (https://chat.openai.com/chat) is an emerging artificial intelligence tool with wide applications. Its ability to extract data from text content has been widely reported but lacks evidence-based validation for its application in medicine. We assess the effectiveness of ChatGPT to identify and extract key data elements of TSRs and detect missing elements.

Design: Forty synoptic reports of malignant tumors of the lung, kidney, colon, and uterus were collected randomly. Key data elements from each of the reports were extracted by ChatGPT and listed as found and missing elements according to standard templates of synoptic reporting by the American Joint Committee on Cancer. The results were compared with results by human proofread as a control.

Results: ChatGPT accurately identified missing data elements from TSRs in 75% of total cases, but in 25% of cases ChatGPT could not accurately identify elements from the TSR chart (Table). ChatGPT presented both found and missing elements in a list for easy viewing. These results indicate that ChatGPT can quickly help find data elements and identify deficient reports with missing elements that will be helpful to enhance the auditing process.

Conclusions: This study demonstrates that ChatGPT can effectively identify key data elements in pathology tumor reports. The use of ChatGPT could be a valuable aid in the analysis and audit of TSRs.

Volume

147

Issue

9

First Page

e148

Comments

College of American Pathologists 2023 Annual Meeting CAP23, October 7-10, 2023, Chicago, IL

DOI

10.5858/arpa.2023-0258-AB

Download

Included in

Pathology Commons

Share

COinS