A novel decision tree-based algorithm for differentiation of incompleted Kawasaki disease from infectious diseases
10.3760/cma.j.issn.1673-4912.2020.10.001
- VernacularTitle:决策树模型在不完全川崎病辅助诊断中的应用
- Author:
Yuanjie ZHOU
1
;
Nan SHEN
;
Lijuan LUO
;
Tingliang LIU
;
Lanping WU
;
Qing CAO
Author Information
1. 上海交通大学医学院附属上海儿童医学中心感染科 200127
- From:
Chinese Pediatric Emergency Medicine
2020;27(10):721-725
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To establish a novel decision tree-based algorithm in complete Kawasaki disease(cKD)and evaluate its diagnostic value in incomplete Kawasaki disease(iKD)and pediatric infectious disease(IF)with common clinical characteristics, which facilitates early and accurate diagnosis of iKD.Methods:Based on inclusion criteria of KD and IF, clinical and laboratory data of patients with cKD, iKD and IF from Shanghai Children′s Medical Center between December 2018 and December 2019 were collected.The training data set included cKD and random half number of IF patients, and validation data was constituted with iKD and the rest of IF patients.The decision tree algorithm analysis was performed in training data set to generate a clinical diagnostic panel for cKD.Finally, the decision tree-based algorithm was verified and evaluated among the iKD patients.Results:A single statistical analysis was performed on 26 examination indexes of constructing decision tree-based algorithm.It was found that 16 examination indexes were obviously different between cKD and IF patients, and 17 examination indexes were significantly different between iKD and IF patients.According to date set of cKD and IF patients, the decision tree-based algorithm was established.The erythrocyte sedimentation rate>35mm/h, N-terminal atrial brain natriuretic peptide precursor≥315 pg/ml, CD3 -/CD19 + %≥21%, and the amount of neutrophil≥8.5×10 9/L were constructed as key elements.The algorithm had a sensitivity of 0.947 and a specificity of 0.963, and correctly classified subjects with iKD who were difficult to be distinguished from patients with IF. Conclusion:A decision tree-based algorithm based on the examination indexes of cKD is one of the effective methods to identify iKD and IF, which provides strong support for the early clinical diagnosis of iKD.
