Analysis Of Influencing Factors Of Children's Kawasaki Disease Based On Mixed Effects Model

Kawasaki disease is a common vascular inflammatory disease in pediatrics.Currently,the standards set by the American Heart Association are widely used worldwide,that is,laboratory test indicators such as C-reactive protein(CRP),total white blood cells,hemoglobin,and total number of platelets are important references.In order to study the intrinsic links between indicators and their role in predicting Kawasaki disease,this paper analyzed the clinical data of 391 Kawasaki children from a hospital in Shanghai.Because of the large differences between different children,a mixed-effect model with random intercept is used to model and statistically infer.Specifically,this paper considers two mixed effects models with random intercepts: linear mixed effect model(LMM)and semiparametric mixed effect model to model Kawasaki disease data.Three methods of constrained maximum likelihood method(REML),RE-EM tree method and unbiased RE-EM tree method are considered to fit the model,and then model interpretation and statistical inference are carried out.Because the Kawasaki disease data studied in this paper are unbalanced longitudinal data,the RE-EM tree method and the unbiased RE-EM tree method mentioned in the literature only consider the fitting problem of the semi-parametric mixed effect model under balanced longitudinal data..Therefore,this paper first generalizes these two methods to unbalanced longitudinal data,and compares REML,RE-EM tree and unbiased RE-EM tree method by numerical simulation.Specifically,the REML,RE-EM tree method and unbiased RE-EM tree method are compared from the perspective of the real model as the LMM with random intercept and the semiparametric mixed effect model.The results show that when the real model is a semi-parametric mixed effect model,the RE-EM tree method and the unbiased RE-EM tree method are significantly better than REML regardless of whether it is iterated multiple times;when the real model is LMM,the sample size is relatively small.The prediction accuracy of REML is relatively high,and its advantages are significantly reduced as the sample size increases.Further,the above three methods were used to analyze the clinical data of 391 children with Kawasaki disease,and the prediction accuracy of the five methods was evaluated by a cross-validation test.The results show that the mean square error of the unbiased RE-EM tree method is the smallest and has the highest prediction accuracy.Therefore,this method is further used to model the semi-parametric mixed effect model.The total number of white blood cells and hemoglobin are found to be important factors affecting CRP changes through tree structure.At the same time,according to the splitting conditions of the intermediate nodes of the tree,it is found to be consistent with the current diagnostic criteria,and there are close internal correlations among the eight criteria in the diagnostic guideline.The two indicators of anemia and CRP?30mg/L can be synthesized.One standard,hemoglobin HGB<95 g/L,simplifies the criteria for determining whether a child has Kawasaki disease.