An Empirical Study Of Automatic ICD Code Assignment To Chinese Diagnosis

International Classification of Diseases(ICD),supported by World Health Organization,is taxonomy of diagnostic codes,which is the international standard for reporting diseases and health conditions globally.In China,diagnoses of disease described in first pages of inpatients' medical records are required to be encoded to ICD-10.The quality of assignment of ICD codes directly affects the accuracy of disease classification and works as the foundation for hospital statistics,single disease quality management and so on.In contrast with its importance,however,the efficiency and accuracy of ICD encoding in medical and health institutions are doubtable,either due to task complexity itself or lack of well-trained and professional coders.Besides,compared with relevant research in English context,both the scope and depth of studies in Chinese context is limited.Based on those considerations,this paper focuses on the task of “automatic ICD code assignment to Chinese diagnosis”,that is,enabling computer systems to encode Chinese diagnoses to ICD-10 diagnostic codes without specific instructions from human,so as to simplify coding work and to enhance coding efficiency in medical practice.Specifically,this research is divided into four parts totally to elaborate above defined task,including algorithms,data,experiments and system development.In the algorithm part,two feasible algorithms: example-based and machine-learningbased methods are introduced(Algorithms).Afterwards,supported by expanded diagnosticcode datasets and relevant corpus(Data),a series of experiments are carried out to verify and to compare effectiveness of each algorithm when applied to automatic code assignment task(Experiments).In addition,this paper offers a preliminary probe into a study on a few universal problems existent in the field of Chinese medical text processing in the context of Chinese automatic ICD-10 assignment,including the choice of text similarity function and corpus as well as the impact of word segmentation.In the last part,automatic code assignment tool is developed to better serve medical practice(System development).The completion of relevant system development marks the completion of study on automatic code assignment on operational level.