Development of prediction models for the incidence of pediatric acute otitis media using Poisson regression analysis and XGBoost.

Abstract

Otitis media has profound health and economic impact, and its occurrence is known to be influenced by air pollution and climate. The purpose of this study was to develop prediction models using climate and air pollution indicators for the occurrence of acute otitis media (AOM). The study was conducted from January 1, 2014, to December 31, 2019, and included pediatric patients (age < 12years) diagnosed on their emergency room visit in our tertiary medical institution. We obtained data on the weekly number of AOM patients and the weekly average values of air pollution and climate indicators. Poisson regression analysis and eXtreme Gradient Boosting (XGBoost) were used to develop prediction models for the overall pediatric patients and for the patients classified according to sex (male and female) and age (< 2years and ≥ 2years). For the overall population, the correlation coefficients between the original and estimated data in the testing set were 0.441 (p < 0.001) and 0.844 (p < 0.001) for the models developed using Poisson regression analysis and XGBoost, respectively. The root-mean-square errors in the testing set were 3.094 and 1.856, respectively. For patients classified according to sex and age, the prediction models developed using XGBoost showed better performance than the models developed using Poisson regression analysis. In conclusion, this study successfully developed prediction models with air pollution and climate indicators for the incidence of pediatric AOM, using XGBoost. This model can be further developed to prevent pediatric AOM.