基于大语言模型的电子病历文本结构化OA

Objective To conduct text structuring of electronic medical records using a large language model(LLM)and verify the superiority of LLM in text structuring.Methods Baidu's ERNIE Bot was adopted.Based on the optimized prompt,the text of the electronic medical record was structured by calling the LLM application programming interface with Python,thereby obtaining structured features.The ratio of the number of accurately extracted features to all the extracted features was defined as the extraction accuracy,which was used to evaluate the performance of LLM in text structuring.Results Text structuring was performed on 100 electronic medical records of stroke patients.According to the statistics of the medical records,the average extraction accuracy of each electronic medical record text was 98.7%,the precision was 97.5%,and the recall rate was 98.9%.According to the statistics of feature words,the average extraction accuracy,precision,and recall of all feature words were 98.7%,96.4%,and 98.6%,respectively.According to the statistics of major characteristic categories,the extraction accuracies of symptoms,past medical history,medication use and diagnostic results were 99.3%,98.0%,98.8%and 100%respectively.Conclusion It is feasible to use LLM for the structuring of electronic medical record texts.Baidu's ERNIE Bot has superiority in the structuring of electronic medical record texts.