Related Risk Factors That Predict Moderate to Severe Asthma Attack in Children: Analysis Based on Logistic Regression and Decision Tree

Qianqian Li, Yinghong Fan, Ronghua Luo, Jie Hu, Li Wang, Tao AiInternational Journal of General Medicine 2025 · DOI

Abstract

Allergic history, medical history, MP infection, and increased NEU% were related risk factors that predict moderate to severe asthma attack in children. Multivariate logistic regression and decision tree both had a good predictive effect for analyzing the risk factors of moderate to severe asthma attack in children.

Introduction

Asthma is the most common chronic inflammatory airway disease in children, and its global prevalence continues to rise. While many pediatric cases are manageable, moderate-to-severe attacks can be life-threatening and place a significant burden on healthcare systems and families. Early identification of high-risk patients is critical for effective intervention.

Traditional statistical methods often struggle to capture the complex, non-linear interactions between various risk factors like age, genetics, and environmental triggers. This study addresses this challenge by combining traditional multivariate logistic regression with machine learning decision tree analysis to identify clinical and biological markers that predict attack severity.

Methods and Datasets

The researchers conducted a retrospective analysis of clinical data from pediatric patients at the Chengdu Women’s and Children’s Central Hospital, covering the period from January 2020 to August 2023.

Flowchart of the patient selection process and group distribution.

As shown in the patient selection flowchart, 805 children were eligible for analysis after exclusions. These patients were divided into "Mild" (n=459) and "Moderate to Severe" (n=346) groups based on the 2016 Chinese Guidelines for Children’s Bronchial Asthma. The study tracked 15 independent variables, including demographic data, medical history, and laboratory markers such as C-reactive protein (CRP) and neutrophil percentage (NEU%).

Analytical techniques included:

  • Multivariate Logistic Regression (LR): Used to identify independent risk factors and calculate Odds Ratios (OR).
  • Decision Tree (DT) Analysis: Specifically using the CHAID (Chi-squared Automatic Interaction Detector) method to visualize hierarchical decision paths.
  • Validation: The decision tree was validated using a stratified ten-fold cross-validation method.

Key Results and Insights

Pathogen and Inflammatory Triggers

The study highlighted the significant impact of infections on asthma severity. Rhinovirus (RV) infection was the strongest independent predictor in the regression model (OR 4.546). Meanwhile, the decision tree identified Mycoplasma pneumoniae (MP) as the most important root node for classification.

While asthma is traditionally associated with eosinophilic inflammation, this study identified Neutrophil percentage (NEU%) as a significant risk factor. This suggests that severe attacks may involve distinct inflammatory pathways that require specific clinical attention.

Predictive Modeling and Clinical Rules

Children aged 6 and older were found to be at significantly higher risk (OR 1.636), possibly due to increased exposure to allergens in school environments.

Decision tree model for predicting the severity of asthma attacks in children.

The decision tree model (Figure 6) generated eight specific clinical rules for risk stratification. For example, a child with an allergy history and an MP infection has a 76.19% probability of a moderate-to-severe attack if their CRP is > 6.9 mg/L.

In terms of overall performance:

  • Logistic Regression: AUC = 0.733 (95% CI: 0.698–0.767)
  • Decision Tree: AUC = 0.694 (95% CI: 0.658–0.731)

The two models provided complementary insights; for instance, the Decision Tree highlighted CRP as a key predictor, whereas the Logistic Regression did not find it significant in a multivariate context, providing a more robust toolkit for clinical decision-making.

Limitations and Future Work

The study is limited by its single-center design and reliance on Chinese diagnostic criteria, which may affect generalizability. Additionally, environmental pollutants (like PM2.5) and specific genetic markers were not included in the current analysis. Future work aims to incorporate multi-center data and more granular genetic information to improve predictive accuracy beyond the current 66.2% achieved by the decision tree model.

Summary updated

Abstract

Allergic history, medical history, MP infection, and increased NEU% were related risk factors that predict moderate to severe asthma attack in children. Multivariate logistic regression and decision tree both had a good predictive effect for analyzing the risk factors of moderate to severe asthma attack in children.
BibTeX
@article{li2025related,
  author = {Qianqian Li and Yinghong Fan and Ronghua Luo and Jie Hu and Li Wang and Tao Ai},
  title = {{Related Risk Factors That Predict Moderate to Severe Asthma Attack in Children: Analysis Based on Logistic Regression and Decision Tree}},
  journal = {International Journal of General Medicine},
  year = {2025},
  doi = {10.2147/ijgm.s530736},
}
Figure