Comparison of Predictive Models for Severe Dengue: Logistic Regression, Classification Tree, and the Structural...

Area of Science:

• Epidemiology
• Biostatistics
• Infectious Diseases

Background:

• Predicting severe dengue illness is crucial for timely intervention.
• Existing models often rely on logistic regression or classification trees.
• Comparing the performance of different statistical models is essential for improving diagnostic accuracy.

Purpose of the Study:

• To compare the predictive performance of logistic regression, classification tree, and Structural Equation Models (SEM) for severe dengue illness.
• To evaluate the effectiveness of these models using demographic and laboratory data.
• To identify the most accurate model for predicting severe dengue.

Main Methods:

• Utilized a modified dengue severity classification based on WHO 1997 guidelines.
• Developed predictive models using demographic and laboratory indicators from Thai pediatric cohorts.
• Employed logistic regression, classification tree, and SEM for model development.
• Performed external validation using independent patient datasets.

Main Results:

• Structural Equation Models (SEM) demonstrated strong predictive performance (AUC 0.73-0.85).
• Logistic regression models also showed good discrimination (AUC 0.65-0.84).
• Classification trees exhibited high sensitivity (0.95-0.99) but low specificity (0.10-0.44).

Conclusions:

• Structural Equation Models (SEM) are a viable and comparable alternative to traditional methods for predicting severe dengue.
• The findings suggest SEM can be effectively used alongside logistic regression and classification trees.
• This research contributes to better predictive tools for severe dengue illness management.