基于随机森林的川崎病诊断关键实验室指标预测研究

Prediction of key laboratory indicators for diagnosis of Kawasaki disease based on random forest

目的通过随机森林算法构建诊断模型,分析不完全川崎病区别于完全川崎病的关键实验室指标,以提高不完全川崎病的诊断率及降低其误诊率。方法选取2018年1月至2023年6月就诊于鄂尔多斯市中心医院儿科确诊为完全川崎病或不完全川崎病的患儿145例,其中不完全川崎病34例(不完全组),完全川崎病111例(完全组)。比较两组红细胞计数、血红蛋白、红细胞沉降率、血钠、白细胞计数、血小板计数、平均血小板体积、中性粒细胞计数、中性粒细胞百分比、淋巴细胞计数、淋巴细胞百分比、C反应蛋白、丙氨酸转氨酶、总蛋白、清蛋白、总胆红素、天冬氨酸转氨酶、降钙素原(PCT)、肌酸激酶、肌酸激酶M多肽(CK-M)、N末端B型钠尿肽前体(NT-proBNP)、凝血酶原时间、凝血酶时间、D-二聚体、尿白细胞计数的差异,利用随机森林算法构建诊断模型,分析不完全川崎病区别于完全川崎病的实验室指标。结果模型结果显示,贡献度前5位的实验室指标依次为血小板计数(0.381)、尿白细胞计数(0.372)、淋巴细胞百分比(0.260)、凝血酶时间(0.255)及中性粒细胞百分比(0.163),在区分川崎病类型方面具有重要的指示意义。构建的诊断模型ROC曲线下面积为0.864,灵敏度为1.00,特异度为0.74。结论通过随机森林算法得出,血小板计数、尿白细胞计数、淋巴细胞百分比、凝血酶时间以及中性粒细胞百分比可作为诊断不完全川崎病的关键实验室指标,这为患儿的诊断与及时治疗提供理论依据。

Objective To construct the diagnostic model by the random forest algorithm,and to analyze the key laboratory indicators distinguishing incomplete Kawasaki disease from complete Kawasaki disease in order to increase the diagnostic rate of incomplete Kawasaki disease and reduce its misdiagnosis rate.Methods A total of 145 children patients with definitely diagnosed complete or incomplete Kawasaki disease in the pediatric department of Ordos Central Hospital from January 2018 to June 2023 were selected,including 34 cases of incomplete Kawasaki disease(incomplete group)and 111 cases of complete Kawasaki disease(complete group).The differences in red blood cell count,hemoglobin,erythrocyte sedimentation rate,blood sodium,white blood cell count,platelet count,mean platelet volume,neutrophil count,neutrophil percentage,lymphocyte count,lymphocyte percentage,C-reactive protein,alanine transaminase,total protein,albumin,total bilirubin,aspartate transaminase,procalcitonin(PCT),creatine kinase,creatine kinase M polypeptide(CK-M),N-terminal B-type natriuretic peptide precursor(N-proBNP),prothrombin time,thrombin time,D-dimer and urine white blood cell count were compared between two groups.The random forest algorithm was used to construct the diagnostic model and analyze the laboratory indicators for distinguishing incomplete Kawasaki disease from complete Kawasaki disease.Results The model results showed that the top 5 laboratory indicators of contribution degree in order were platelet count(0.381),urine white blood cell count(0.372),lymphocyte percentage(0.260),thrombin time(0.255)and neutrophil percentage(0.163),which have important indicative significance in distinguishing Kawasaki disease types.The area under the receiver operating characteristic(ROC)curve of the constructed diagnostic model was 0.864,with a sensitivity of 1.00 and a specificity of 0.74.Conclusion Platelet count,urine white blood cell count,lymphocyte percentage,thrombin time and neutrophil percentage through the random forest algorithm could serve as the key laboratory indicators for the diagnosis of incomplete Kawasaki disease,which provides the theoretical basis for the diagnosis and timely treatment of the children patients.