Air pollution epidemiology: A simplified Generalized Linear Model approach optimized by bio-inspired metaheuristics

• Published in: Environmental research Vol. 191; p. 110106
• Main Authors: Belotti, Jônatas T., Castanho, Diego S., Araujo, Lilian N., da Silva, Lucas V., Alves, Thiago Antonini, Tadano, Yara S., Stevan, Sergio L., Corrêa, Fernanda C., Siqueira, Hugo V.
• Format: Journal Article
• Language: English
• Published: Netherlands Elsevier Inc 01.12.2020
• Subjects: aerodynamics; Air Pollutants - analysis; air pollution; Air Pollution - analysis; algorithms; ambient temperature; Brazil; Brazil - epidemiology; computer software; epidemiology; Hospital admissions for respiratory diseases; hospitals; Humans; Linear Models; Particle swarm optimization; PM10; quality of life; relative humidity; Respiration Disorders; Splines; statistical analysis; Brazil; PM10; Splines; Hospital admissions for respiratory diseases; Particle swarm optimization; PM
• ISSN: 0013-9351; 1096-0953; 1096-0953
• DOI: 10.1016/j.envres.2020.110106

Studies in air pollution epidemiology are of paramount importance in diagnosing and improve life quality. To explore new methods or modify existing ones is critical to obtain better results. Most air pollution epidemiology studies use the Generalized Linear Model, especially the default version of R, Splus, SAS, and Stata softwares, which use maximum likelihood estimators in parameter optimization. Also, a smooth time function (usually spline) is generally used as a pre-processing step to consider seasonal and long-term tendencies. This investigation introduces a new approach to GLM, proposing the estimation of the free coefficients through bio-inspired metaheuristics - Particle Swarm Optimization (PSO), Genetic Algorithms, and Differential Evolution, as well as the replacement of the spline function by a simple normalization procedure. The considered case studies comprise three important cities of São Paulo state, Brazil with distinct characteristics: São Paulo, Campinas, and Cubatão. We considered the impact of particles with an aerodynamic diameter less than 10 μm (PM10), ambient temperature, and relative humidity in the number of hospital admissions for respiratory diseases (ICD-10, J00 to J99). The results showed that the new approach (especially PSO) brings performance gains compared to the default version of statistical software like R. •A new approach of Generalized Linear model to epidemiology studies is proposed.•The use of Artificial Intelligence Approaches to coefficients' estimation and smoothing splines were replaced.•Using PSO brings performance gains to air pollution risk studies.•The best result was achieved for one day after exposure to PM 10.