Development of a predictive machine learning model for pathogen profiles in patients with secondary immunodeficiency

被引：3

作者：

Liu, Qianning ^{[1
]}

Chen, Yifan ^{[1
]}

Xie, Peng ^{[2
]}

Luo, Ying ^{[2
,4
]}

Wang, Buxuan ^{[1
]}

Meng, Yuanxi ^{[3
]}

Zhong, Jiaqian ^{[3
]}

Mei, Jiaqi ^{[3
]}

Zou, Wei ^{[2
]}

机构：

[1] Jiangxi Univ Finance & Econ, Sch Stat, Nanchang 330013, Jiangxi, Peoples R China

[2] Nanchang Univ, Affiliated Hosp 1, Jiangxi Med Coll, Dept Infect Dis, Nanchang 330006, Jiangxi, Peoples R China

[3] Nanchang Univ, Clin Med Coll 1, Jiangxi Med Coll, Nanchang 330006, Jiangxi, Peoples R China

[4] Third Peoples Hosp Jiujiang, Dept Infect Dis, Jiujiang 332000, Jiangxi, Peoples R China

来源：

BMC MEDICAL INFORMATICS AND DECISION MAKING | 2024年 / 24卷 / 01期

基金：

中国国家自然科学基金;

关键词：

Secondary immunodeficiency; Pathogens; Imbalanced data; K nearest neighbour; Boosted logistic regression; Random forest; Gradient boosting machine; INFECTIONS; GUIDELINES; MANAGEMENT; PNEUMONIA; DISEASES;

D O I：

10.1186/s12911-024-02447-w

中图分类号：

R-058 [];

学科分类号：

摘要：

BackgroundSecondary immunodeficiency can arise from various clinical conditions that include HIV infection, chronic diseases, malignancy and long-term use of immunosuppressives, which makes the suffering patients susceptible to all types of pathogenic infections. Other than HIV infection, the possible pathogen profiles in other aetiology-induced secondary immunodeficiency are largely unknown.MethodsMedical records of the patients with secondary immunodeficiency caused by various aetiologies were collected from the First Affiliated Hospital of Nanchang University, China. Based on these records, models were developed with the machine learning method to predict the potential infectious pathogens that may inflict the patients with secondary immunodeficiency caused by various disease conditions other than HIV infection.ResultsSeveral metrics were used to evaluate the models' performance. A consistent conclusion can be drawn from all the metrics that Gradient Boosting Machine had the best performance with the highest accuracy at 91.01%, exceeding other models by 13.48, 7.14, and 4.49% respectively.ConclusionsThe models developed in our study enable the prediction of potential infectious pathogens that may affect the patients with secondary immunodeficiency caused by various aetiologies except for HIV infection, which will help clinicians make a timely decision on antibiotic use before microorganism culture results return.

引用

页数：11