Premium
Characterization of exposure to extremely low frequency magnetic fields using multidimensional analysis techniques
Author(s) -
Verrier A.,
Souques M.,
Wallet F.
Publication year - 2005
Publication title -
bioelectromagnetics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.435
H-Index - 81
eISSN - 1521-186X
pISSN - 0197-8462
DOI - 10.1002/bem.20075
Subject(s) - principal component analysis , linear discriminant analysis , statistics , bioelectromagnetics , pattern recognition (psychology) , sample (material) , computer science , discriminant , artificial intelligence , mathematics , magnetic field , physics , quantum mechanics , thermodynamics
Abstract Our lack of knowledge about the biological mechanisms of 50 Hz magnetic fields makes it hard to improve exposure assessment. To provide better information about these exposure measures, we use multidimensional analysis techniques to examine the relations between different exposure metrics for a group of subjects. We used a combination of a two stage Principal Component Analysis (PCA) followed by an ascending hierarchical classification (AHC) to identify a set of measures that would capture the characteristics of the total exposure. This analysis gives an indication of the aspects of the exposure that are important to capture to get a complete picture of the magnetic field environment. We calculated 44 metrics of exposure measures from 16 exposed EDF employees and 15 control subjects, containing approximately 20000 recordings of magnetic field measurements, taken every 30 s for 7 days with an EMDEX II dosimeter. These metrics included parameters used routinely or occasionally and some that were new. To eliminate those that expressed the least variability and that were most highly correlated to one another, we began with an initial Principal Component Analysis (PCA). A second PCA of the remaining 12 metrics enabled us to identify from the foreground 82.7% of the variance: the first component (62.0%) was characterized by central tendency metrics, and the second (20.7%) by dispersion characteristics. We were able to use AHC to divide the entire sample (of individuals) into four groups according to the axes that emerged from the PCA. Finally, discriminant analysis tested the discriminant power of the variables in the exposed/control classification as well as those from the AHC classification. The first showed that two subjects had been incorrectly classified, while no classification error was observed in the second. This exploratory study underscores the need to improve exposure measures by using at least two dimensions: intensity and dispersion. It also indicates the usefulness of constructing a typology of magnetic field exposures. Bioelectromagnetics 26:266–274, 2005. © 2005 Wiley‐Liss, Inc.