摘要
目的探索机器学习算法对儿童过敏性紫癜发生肾损害的风险预测价值,并从中分析儿童紫癜性肾炎的相关危险因素。方法选取2018年11月—2021年6月山东中医药大学附属医院收治的过敏性紫癜住院患儿268例资料,包括患儿的一般资料、临床症状、实验室检查结果,共42个特征,使用递归特征消除法进行特征筛选,通过交叉验证的方式执行,得到最佳特征子集后分别使用logistic回归、朴素贝叶斯、决策树、LightGBM算法进行分类预测,使用精确率、召回率、AUC等比较模型的性能。结果建立的LightGBM模型的精确率、准确率、召回率、F1及AUC均最高,分别为0.87、0.80、0.81、0.79、0.884。根据LightGBM模型提供的特征重要性排序,居于前十的特征是血清白蛋白、血小板、抗链球菌溶血素“O”、C反应蛋白、肌酐、皮疹持续≥4周、淋巴细胞绝对值、D2聚体、皮疹反复、单核细胞绝对值是儿童紫癜性肾炎的危险因素。结论基于LightGBM算法构建儿童紫癜性肾炎的预测模型具有更好的预测效果,可为临床早期诊断提供决策依据。
Objective To explore the value of machine learning algorithm in predicting the risk of renal damage in children with henoch-schonlein purpurain,and to analyze the related risk factors of henoch-schonlein purpurain nephritis in children.Methods Clinical data of 268 hospitalized children with henoch-schonlein purpura from November 2018 to June 2021 were collected in the Affiliated Hospital of Shandong University of Traditional Chinese Medicine.which included 42 features as patient′s demographic information,clinical symptoms and laboratory test results.The recursive feature elimination method was used for feature screening.After obtaining the best feature subset,logistic regression,Naive Bayes,Decision Tree and LightGBM algorithm were used for classification prediction,and the performances of models were compared based on accuracy,recall and AUC.Results The constructed LightGBM model had the highest scores of accuracy,recall,F1 and AUC,which were 0.87,0.80,0.81,0.79 and 0.884,respectively.According to the ranking of feature importance generated by lightgbm model,the top ten features were serum albumin,platelet,antistreptolysin"O",C-reactive protein,creatinine,rash lasting≥4 weeks,lymphocyte absolute value,D2 polymer,rash recurrence,monocyte absolute value,which were potential relevant risk factors for henoch-schonlein purpura nephritis in children.Conclusions LightGBM algorithm based prediction model is able to effectively predict henoch-schonlein purpura nephritis in children,which may provide valuable evidence for early clinical diagnosis.
作者
褚会敏
王金娟
潘月丽
CHU Hui-min;WANG Jin-juan;PAN Yue-li(College of Traditional Chinese Medicine,Shandong University of Traditional Chinese Medicine,Jinan,Shandong 250355,China;不详)
出处
《中国预防医学杂志》
CAS
CSCD
北大核心
2022年第1期62-67,共6页
Chinese Preventive Medicine