ICD编码人工智能审核质控模式的设计与效果研究

Design and Effect Study of ICD Coded Artificial Intelligence Audit Quality Control Mode

目的设计ICD编码人工智能审核质控模式并进行效果评价,以提高编码准确率。方法抽取经考核合格的质控人员按照《住院病案首页数据填写质量规范(暂行)》(围卫办医发[2016]24号)和《某院住院病案首页填写和质控规则》,对抽取的人工智能审核模块上线前2021年2月1日-2021年10月31日住院病案首页27359份,上线后2022月1日-2022年10月31日住院病案首页60239份编码质量逐份进行质控,对主要诊断及编码漏编、错编、选错等情况进行记录。结果系统上线后医师填写正确率高于系统上线前(χ^(2)=390.596,P<0.05),医师填写各项错误原因错误率明显下降(手术/操作漏填χ^(2)=147.13,P<0.05、漏诊断χ^(2)=6.50,P<0.05、主要诊断选择错误χ^(2)=5.76,P<0.05、诊断填写错误χ^(2)=3.88,P<0.05、手术/操作选择错误χ^(2)=13.08,P<0.05、多填诊断χ^(2)=25.16,P<0.05);编码员编码正确率高于系统上线前(χ^(2)=546.384,P<0.05);编码员各项错误类型错误率明显下降(主诊编码错误χ^(2)=571.922,P<0.05、手术/操作编码错误χ^(2)=36.1,P<0.05、漏编手术/操作码χ^(2)=4.18,P<0.05、漏编码χ^(2)=36.05,P<0.05、疾病与编码不符χ^(2)=74.05,P<0.05、未合并编码χ^(2)=101.78,P<0.05)。结论人工智能ICD编码审核质控模式的实施可实现ICD编码全面实时质控,编码完整性、合理性、正确性获得大大提升,可显著提高编码质控效率和质量。

Objectives To design the ICD coding artificial intelligence audit quality control model and evaluate the effect,so as to improve the coding accuracy.Methods Qualified quality control personnel were selected according to the"Quality Specification for Filling in the Frontt Page Data of Inpatient Medical Records(Interim)"(Health Office Medical[2016]No.24)and"Rules for filling in the front page of Inpatient Medical Records and Quality Control of a Hospital".Before the launch of the extracted artificial intelligence review module,27359 front pages of inpatient medical records from February 1st,2021 to October 31st,2021,and 60239 front pages of inpatient medical records from October 1st,2022 to October 31st,2022 were quality-controlled one by one,and the main diagnosis and the situation of missing coding,wrong coding and wrong selection were recorded.Results After the system was launched,the correct rate of doctors'filling in was higher than that before the system was launched(χ^(2)=390.596,P<0.05),and the error rate of doctors'filling in the causes of errors decreased significantly(surgery/operation omissionχ^(2)=147.13,P<0.05,missed diagnosisχ^(2)=6.50,P<0.05,major diagnosis selection errorχ^(2)=5.76,etc.).P<0.05,diagnosis filling errorχ^(2)=3.88,P<0.05,surgery/operation selection errorχ^(2)=13.08,P<0.05,multiple diagnosisχ^(2)=25.16,P<0.05;The coding accuracy rate of coders was higher than that before the system went online(χ^(2)=546.384,P<0.05).The error rate of various error types of coders decreased significantly(main diagnosis code errorχ^(2)=571.922,P<0.05,surgery/operation code errorχ^(2)=36.1,P<0.05,missed surgery/operation codeχ^(2)=4.18,P<0.05,missed codeχ^(2)=36.05,P<0.05,missed codeχ^(2)=36.05,P<0.05,disease does not match codeχ^(2)=74.05,P<0.05,uncombined codeχ^(2)=101.78,P<0.05).Conclusions The implementation of artificial intelligence ICD code audit quality control mode could realize comprehensive and real-time quality control of ICD code,and the integrity,rationality and correctness of code were greatly improved.It could significantly improve the efficiency and quality of coding quality control.