Identifying the contribution of physical and chemical stressors to the daily number of hospital admissions implementing an artificial neural network model

Abstract

The relative contribution of chemical (air pollution) and physical (temperature and humidity) health stressors to urban hospitalization rates is the objective of the current study. The data used in the study included the daily number of hospital admissions due to cardiorespiratory diseases, hourly mean concentrations of CO, NO2, SO2, O3, and black smoke in several monitoring stations, as well as meteorological data (temperature, relative humidity, wind speed/direction) in Athens, Greece. The relations among the data above were studied using Generalized Linear Models (GLMs) and Artificial Neural Networks (ANNs). Elevated particulate concentrations are the dominant parameter related to hospital admissions (an increase of 10 μg/m3 leads to an increase of 8.6% in hospital admissions), followed by O3 and the other atmospheric pollutants (CO, NO2, and SO2). Meteorological parameters also play a decisive role in the formation of air-pollutant levels affecting public health. Both models performed adequately, however the ANN adaptation in complicate environmental issues results in improved modeling outcomes compared to the GLMs. The major finding of the study lies on the flexibility and the adaptation of the methodological approach for assessing non-linear problems and specifically the effect of non-linear parameters, such as the temperature. Moreover, the importance of temperature is established even when the whole dataset is modeled, reflecting the dual mode effect of temperature on cardiorespiratory admissions. Considering the urgent challenge to predict climate change effects on public health, a mathematical tool that successfully captures the direct impact of the affecting meteorological parameters (temperature and humidity) to health outcomes is of high added value.