The analysis of asbestos count data with “nondetects”: The example of asbestos fiber concentrations in the lungs of brake workers

Abstract

In the analysis of tissue for asbestos fibers, some measurements may be below the analytical detection limit (nondetects). The use of maximum likelihood and survival analysis methods have been recommended to perform comparisons between subjects in the presence of nondetects. When the data consist of "counts" another method is useful. This method is discussed, and illustrated with an analysis of asbestos lung burden data among brake mechanics previously analyzed by other methods. Statistical models for count data, namely Poisson and negative binomial regression, were used to compare the asbestos fiber concentrations in the lungs of brake mechanics with those of control subjects. The fit of the models was assessed with an analysis of residuals. The negative binomial regression models fit the data well. The concentrations of Quebec asbestos fibers in the lungs of the brake mechanics were significantly higher than in the control population. Helsel recommended the use of maximum likelihood and survival analysis methods to perform comparisons in the presence of nondetects. When analyzing asbestos fiber count data, or other count data arising in occupational or environmental health, the use of models such as the Poisson and negative binomial may be added to the analyst's toolbox. Benefits are that neither of these methods requires the substitution of arbitrary values for the nondetects and that programs for the computation of count data models are contained in popular statistical software packages.