基于人工智能的CDSS诊断符合率研究

A Study of the Compliance of AI-based CDSS Diagnosis Study

目的通过分析临床决策支持系统(CDSS)推出的诊断,包含目标病例最终诊断的覆盖情况及准确度,评价系统诊断疾病的效果,评估现有临床知识库标准程度。方法选定医院心内科1850例出院确诊病历作为测试病历,对比CDSS推出的诊断与患者病案首页最终诊断的覆盖情况。结果通过3轮评测,最终得到CDSS召回率为66.5%,准确率93.6%。结论 CDSS推出诊断绝大多数都是合理的,多数情况下覆盖了临床医生认为重要的确诊诊断,能够帮助临床医生作出完整诊断,并快速聚焦诊断得出鉴别结果。

Objective By analyzing the coverage and accuracy of CDSS recommendations including the final diagnosis of the target case, to evaluate the system effectiveness of diagnosing diseases, and further assess the accuracy of the existing clinical knowledge base. Methods A total of 1850 medical records from Cardiology Department were selected as test cases, and the coincidence between CDSS recommendations and the final diagnosis of the patient＇s medical record homepage were compares. Results With 3 rounds of evaluation, the CDSS recall rate was 66.5% and the accuracy rate was 93.6%. Conclusions The vast majority of CDSS recommendations are reasonable, in most cases they cover the diagnosis considered important by clinicians, which can help the clinician to make a complete diagnosis, and quickly focus on the diagnosis results.