Clinical Validation And Optimization Study Of Risk Prediction Model Of Children’s Henoch-Sch(?)nlein Purpura Nephritis In Traditional Chinese And Western Medicine

Objective: To verify the clinical application value of the risk prediction model for children with Henoch-Sch(?)nlein Purpura Nephritis(HSPN),and apply machine learning to optimize the model,in order to improve the early prediction level of HSPN in children.Methods: This study is divided into two parts.In the prospective study,255 children with Henoch-Sch(?)nlein Purpura(HSP)who visited the Affiliated Hospital of Shandong University of Traditional Chinese Medicine from January 1,2020 to December 31,2021 were included.Predict the risk of renal damage in children with HSP,and verify the accuracy of the model according to the follow-up results.In a retrospective study,the risk factors related to HSPN in children were further screened,and Logistic Regression,Random Forest,XGBoost,GBDT and Stacking model fusion were used to optimize the model.Results:1.In the prospective study,the application of the model for graded prediction showed that the specificity of the first-diagnosis prediction(75.64%)was slightly higher than that of the second-diagnosis prediction(69.87%),and the sensitivity of the second-diagnosis prediction(79.80%)was slightly higher than that of the first-diagnosis prediction(72.72%).The difference between the two was not statistically significant(P>0.05).2.In the retrospective study,after univariate feature selection and feature importance ranking,the variables whose repetition ranks in the top 10 are: age≥8 years old,clinic time,damp-heat with stasis syndrome,persistent purpura≥4 weeks,anti-streptolysin “O”,Epstein-Barr virus infection,Mycoplasma pneumoniae infection,serum albumin,cystatin C,and triglyceride were important risk factors for HSPN in children.The accuracy,precision,recall,F1 value and AUC of the combined Stacking model are0.94,0.89,0.79,0.80,and 0.896,respectively,which are significantly higher than the four single models.Conclusion:1.The children’s HSPN risk prediction model constructed in the previous study has good performance and is suitable for clinical Chinese and Western medicine,but the model does not consider the time attribute of the predictor.2.Age≥8 years old,clinic time,damp-heat with stasis syndrome,persistent purpura≥4 weeks,relapse,anti-streptolysin “O”,Epstein-Barr virus infection,Mycoplasma pneumoniae infection,serum albumin,cystatin C,and triglyceride were important risk factors for HSPN in children.3.The combined Stacking model performs better in child HSPN prediction.