Cognitive Map Of Kawasaki Disease With Coronary Artery Lesions In Early Stage

ObjectiveTo construct the prediction model and cognitive map(CM)of coronary artery lesions in acute stage of Kawasaki disease,and to explore the clinical research methods in this field.MethodsThrough visiting the database of PubMed,Web of Science,EBSCCO,China Knowledge Resource Integrated Database,WanFang Database,Vip Database and so on,the biological indicators which concerned as the risk factors of KD complicated with CAL in the past studies were extracted.The biological indicators were be extracted from the database which main diagnosis was KD.The indicators would be deleted if the missing rate exceeding 5% and the data would be processed by different methods.The prediction model established by Back Propagation Artificial Neural Network(BPANN)and the generalize weight(GW)of each indicator was determined.Bayesian network(BN)method was used to excavate the relationship between each biological indicators and CAL.Combined GW of each indicators with the result of BN model,determined the risk of KD complicated with CAL.And then build the CM of KD complicated with CAL.ResultsThe results showed that the classification rate of the training set,the sensitivity and the specificity was 94.64%,63.76% and 99.85%,respectively.The coincidence rate of the test set was 94.63%,the sensitivity was 53.45% and the specificity was 99.59%.The GW scatter plot showed that red blood cells and white blood cells were close to 0 in the whole model.BN model results also showed the red blood cells,white blood cell and CAL were independent.There are seven relationships among the remaining nodes,age and hematocrit,age and neutrophilic granulocyte,hemoglobin And CAL,hematocrit and hemoglobin,CAL and neutrophilic granulocyte,platelet and neutrophilic granulocyte,neutrophilic granulocyte and C-reactive protein,respectively.ConclusionsAccording to the data of the first biochemical examination of children after admission,the BPANN method was used to establish the predictive model of CAL in children with KD,which improved the accuracy of prediction and has reference significance for early detection of high-risk children with CAL.Combined with BN results,construct CM between age,platelet count,hematocrit,hemoglobin,C-reactive protein,and neutrophilic granulocyte,and red blood cell.CM intuitively showed the correlation between the indicators and the correlation strength.Because of the sensitivity of the prediction model was low,we considered the risk factors of KD complicated with CAL can not be limited to the study of biochemical indicators,but also need to incorporate more non-biochemical indicators,and compared with inflammatory and immune related disease,combined with data mining methods to study.