Construct a machine learning model for differential diagnosis of bloodstream infection based on laboratory indicators
10.3969/j.issn.1673-9701.2024.31.013
- VernacularTitle:基于实验室指标构建血流感染鉴别诊断的机器学习模型
- Author:
Mei ZHANG
1
;
Miaoling JIN
;
Cui LI
Author Information
1. 浙江大学医学院附属第一医院浦江分院(浦江县人民医院)检验科,浙江浦江 322200
- Keywords:
Bloodstream infection;
Nomogram model;
Decision tree model;
Random forest model
- From:
China Modern Doctor
2024;62(31):55-59
- CountryChina
- Language:Chinese
-
Abstract:
Objective To construct a diagnosis models for differential diagnosis of bloodstream infection(BSI)using nomogram model,random forest model,and decision tree model,respectively.Methods A retrospective analysis was performed on 225 BSI patients diagnosed and treated in Pujiang County People's Hospital from January 2022 to January 2024,and the patients were divided into a training set and a validation set according to a ratio of 7:3.The differential diagnostic models for Gram negative BSI(GN-BSI)and Gram positive BSI(GP-BSI)were established by nomogram model,random forest model,and decision tree model,and the differential diagnostic efficacy of different models were analyzed.Results Binary Logistic regression analysis showed that neutrophil to lymphocyte ratio(NLR),C-reactive protein(CRP),interleukin-6(IL-6),red cell volume distribution width to platelet ratio(RPR),procalcitonin(PCT)were diagnostic variable between GN-BSI and GP-BSI(P＜0.05).In training set,area under the curve(AUC)of nomogram model,random forest model and decision tree model to identify GN-BSI and GP-BSI were 0.900,0.911 and 0.884,respectively,and AUC of random forest model was significantly higher than that of decision tree model(Z=3.521,P=0.038).In verification set,AUC of nomogram model,random forest model and decision tree model for identifying GN-BSI and GP-BSI were 0.908,0.916 and 0.893,respectively,and AUC of random forest model was significantly higher than that of decision tree model(Z=3.412,P=0.042).Conclusion The three models have good identification value for GN-BSI and GP-BSI,among which the random forest model and nomogram model have better identification performance.
