Data Mining Methods Predict The Coronary Artery Lesions In Kawasaki Disease

BackgroundThe most serious complication of Kawasaki disease(KD)is coronary artery lesions(CAL),which may cause ischemic heart disease,myocardial infarction and sudden death.To find the risk factors of CAL in KD to study the potential risk of CAL in KD has important significance for the prevention and treatment of KD.ObjectiveThe electronic medical record(EMR)data of KD is the research object of this paper.The risk factors of CAL in KD are found and the model are constructed for the detection and classification of CAL in KD using data mining methods to provide decision support for the treatment of KD.MethodsCollect the clinical and laboratory data,echocardiographic data and diagnosis report data of KD patients,which were pre-processed into the initial data sets.Association rules were used to analyze the 343 KD patients with CAL to find the indicators correlated with CAL in KD.The CAL prediction model are constructed based on RF,BN and NN.The performance of them are evaluated respectively.ResultsMale patients who are younger than 2 years old,higher CRP,NEU,ESR,and PLT are risk factors of CAL in KD.CRP,PLT,ASAL,ESR are co-related with CAL in KD.In particular,higher CRP,EOS and NEU,lower ALB,PA,MCH,and RDW,positive KET were associated with the status of CAL.The performance of RF model is the best comparing with NN model and BN model.The sensitivity is 0.825,which means that the correct identification of KD patients with CAL is 82.5%.ConclusionData mining methods were applied to the EMR data of KD patients.The risk factors of CAL in KD and risk factors complicated with the aggravation status of CAL were discovered to provide advice and decision support for the experimental research,clinical diagnosis and treatment of KD.