Portal de Periódicos da CAPES

Whole‐fish versus filet polychlorinated‐biphenyl concentrations: An analysis using classification and regression tree models

1999; Wiley; Volume: 18; Issue: 8 Linguagem: Inglês

10.1002/etc.5620180831

1552-8618

James F. Amrhein, Craig A. Stow, Clay Wible,

Data Mining Algorithms and Applications

Abstract Fish polychlorinated‐biphenyl (PCB) measurements usually represent one of two different sample types: filets or homogenized whole fish. Filet measurements are more appropriate for use if the goal of analysis is estimating human PCB consumption, while whole‐fish analysis may be more useful for quantifying and understanding processes of contaminant flow and bioaccumulation. While it is generally assumed that whole‐fish PCB concentrations exceed filet concentrations because of the presence of fatty internal organs in whole‐fish samples, the literature contains no reported comparisons of filet versus whole‐fish PCB concentrations. We measured total PCB concentrations in filets and whole‐fish samples from the same individuals in Lake Michigan coho salmon (Oncorhynchus kisutch) and rainbow trout (Oncorhynchus mykiss). The average whole‐fish to filet PCB concentration ratio was 1.70 for coho salmon and 1.47 for rainbow trout, but it varied considerably among individuals, with a few fish exhibiting a higher concentration in the filet than in the whole‐fish sample. Classification and regression tree (CART) models indicated that filet PCB concentration and fish length were the best predictors of whole‐fish PCB concentration, whereas filet and whole‐fish lipid concentrations where less important predictors. Lipid normalization of the PCB data decreased within‐individual variability, was equivocal with respect to variability among individuals, and accentuated the between‐species difference. Both species exhibit a pronounced 1:1 relationship between the whole‐fish to filet PCB concentration ratio and the whole‐fish to filet lipid concentration ratio; however, we point out that there is a strong spurious component to this relationship, which indicates that the relationship may be more algebraic rather than an indication of underlying mechanisms.