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

Abstract

Abstract Background This study aimed to compare the predictive performance of 3 statistical models—logistic regression, classification tree, and structural equation model (SEM)—in predicting severe dengue illness. Methods We adopted a modified classification of dengue illness severity based on the World Health Organization’s 1997 guideline. We constructed predictive models using demographic factors and laboratory indicators on the day of fever occurrence, with data from 2 hospital cohorts in Thailand (257 Thai children). Different predictive models for each category of severe dengue illness were developed employing logistic regression, classification tree, and SEM. The model’s discrimination abilties were analyzed with external validation data sets from 55 and 700 patients not used in model development. Results From external validation based on predictors on the day of presentation to the hospital, the area under the receiver operating characteristic curve was from 0.65 to 0.84 for the regression models from 0.73 to 0.85 for SEMs. Classification tree models showed good results of sensitivity (0.95 to 0.99) but poor specificity (0.10 to 0.44). Conclusions Our study showed that SEM is comparable to logistic regression or classification tree, which was widely used for predicting severe forms of dengue.