Using ChatGPT for Extracting Structured Data from Clinical Notes: A Critical Assessment
AI technology continues to revolutionize the field of medicine, with recent advancements showing promise in extracting structured data from unstructured clinical notes. A study published in npj Digital Medicine evaluated the effectiveness of using ChatGPT, a large language-based model, for this purpose.
Traditional natural language processing approaches often require specific annotations and costly model training, making them challenging to implement in healthcare settings lacking human-annotated data. In contrast, models like ChatGPT leverage logical reasoning and knowledge to process language efficiently, offering a viable alternative for extracting structured data from clinical notes.
In the study, researchers utilized the ChatGPT 3.50-turbo model to extract structured data from lung tumor pathology reports and pediatric osteosarcoma reports. By transforming unstructured text into analyzable data, ChatGPT demonstrated its ability to accurately classify pathological classifications, grades, margin status, and tumor stage with high accuracy rates.
However, the model’s performance was influenced by the design of instructional prompts and misclassifications related to pathology terminologies and TNM staging guidelines. Despite some shortcomings, ChatGPT consistently outperformed keyword algorithms and deep learning-based approaches, showcasing its potential in converting healthcare information into organized representations.
Furthermore, the study compared the performance of ChatGPT version 3.50 with GPT-4, revealing improvements in model accuracy and classification abilities. With the ability to handle large volumes of clinical notes and extract structured data without extensive human annotation or training, ChatGPT demonstrates its utility in supporting research and clinical decision-making in the future.
Overall, the findings of this study present exciting possibilities for utilizing large language-based models like ChatGPT to enhance healthcare data analysis and facilitate advancements in medical research. As AI technology continues to evolve, it holds great potential for transforming the way healthcare professionals extract and interpret valuable information from unstructured clinical notes.