Research On Health Portraits Of Multimorbidity Patients Based On Electronic Medical Records

ObjectiveTo build the health portraits of multimorbidity patients,based on patient diagnosis and treatment data in electronic medical records,this paper uses association rules to mine the comorbidity pattern of patient groups.According to the health portrait label system,the named entity recognition technology is used to extract the information of each dimension of the patient to complete the personalized patient health portrait and the health portraits of multimorbidity patients.We hope the health portraits can provide a reference for the diagnosis and treatment of patients with similar comorbidities and reflect patient information comprehensively.MethodBased on the investigation and analysis of domestic and foreign health portrait related studies,considering the comorbidity characteristics of patients and the unstructured characteristics of electronic medical record text,FP-growth association rule algorithm was adopted to mine the comorbidity patterns of patients based on a large number of electronic medical record data.Then,the patient axis and the disease axis information were considered to construct the health portrait label system,which includes the basic information,main manifestations,examinations,other diagnoses and treatment methods of the patients.And the named entity recognition technology was used to extract the label entities of each dimension.Relying on the Neo4 j graph database.We form the individual-centered personalized health portraits and group health portraits under different comorbidity patterns.Results(1)Fp-growth algorithm was used to mine comorbiditis patterns of 25390 electronic medical record data in orthopedics and traumatology ward,respiratory and critical medicine ward,spleen and stomach diseases ward and cardiovascular diseases ward.The results showed that chronic colitis and chronic gastritis,hyperlipidemia and chronic colitis,chronic gastritis and liver cyst were the top 3 comorbidity combinations with the highest support.The top three comorbidity combinations were hypertension,carotid atherosclerosis and lacunar cerebral infarction,hypertension,hyperlipidemia and lacunar cerebral infarction,hyperlipidemia,fatty liver and carotid atherosclerosis.The top 3 association rules with the highest support were chronic colitis and chronic gastritis;Chronic colitis,fatty liver and chronic gastritis;Lacunar cerebral infarction,carotid atherosclerosis,and hypertension.(2)Considering the information of the disease axis and the patient axis,we identify the patient health portrait description frame.The patient personalized health portrait was constructed from the five dimensions of basic information,main manifestations,examination and test,other diagnosis and treatment methods.And the comorbidiosis health portrait was constructed from the three dimensions of main manifestations,examination and test and treatment methods.(3)In order to evaluate the effectiveness of the Ro BERTa-wwm dynamic fusion entity recognition model in this paper,Bi LSTM-CRF,BERT-Bi LSTM-CRF and Ro BERTa-wwm-Bi LSTM-CRF are used as comparative models.Both sets outperform the comparison models,achieving F1 scores of 94.08% and 90.08% respectively.(4)Taking a patient with chronic colitis as the main diagnosis as an example,we construct a personalized health portrait of the patient.And taking "chronic colitis,fatty liver and chronic gastritis" as an example,which is the most supported ternary comorbidity model,we construct a health portrait of this comorbidity group.Conclusion(1)In view of the unstructured text of electronic medical record,the named entity recognition method is used to complete the extraction of image label entities,which can build a more detailed patient health portrait model.(2)Based on the internal association of comorbidity model,the patient’s personalized health portrait is constructed and then the comorbidity health portrait is formed,these portraits can describe the characteristics of patients from the perspective of individual diagnosis and group diagnosis.