Estimating values for nondetectable samples.
Sanderson-W; Echt-A; Ewers-L; Hornung-R
American Industrial Hygiene Conference and Exposition, May 19-23, 1997, Dallas, Texas. Fairfax, VA: American Industrial Hygiene Association, 1997 May; :78
When estimating worker exposures to a particular toxic agent, sample measurements are sometimes below the limit of detection (LOD) of the analytical method. The industrial hygienist is then challenged to select valid estimates for those samples in order to calculate an unbiased average exposure concentration. Several techniques have been proposed for estimating average concentrations from data containing nondetectable samples. These methods include a maximum likelihood statistical method and two methods involving the limit of detection. The maximum likelihood method is complex and requires laborious calculations, while the other methods involve simply substituting the LOD/2 or the LOD/square-root-of 2 for each nondetectable value. Using Monte Carlo simulation, the LOD/square-root-of-2 has been shown to provide more accurate estimates of the mean and standard deviation when the measurement data are not highly skewed, but the LOD/2 is recommended when the data are highly skewed (GSD > 3.0). Samples for airborne lead concentrations from bridge painters were used to evaluate these recommended values for the LOD. A total of 105 samples were collected on 37-mm polyvinyl chloride filters at a flow rate of 1 liter per minute (Lpm) and analyzed using flame atomic absorption spectroscopy (flameAA); 25 (23.8%) of the samples were below the analytical LOD. To estimate the true concentrations on these nondetectable samples, they were reanalyzed using graphite furnace atomic absorption spectroscopy, which is a more sensitive analytical method. All 25 samples were above the LOD for the graphite furnace method, and these values were substituted for the flame-AA LOD. The geometric mean (GM) and geometric standard deviation (GSD) of all 105 samples were then compared to the GM and GSD calculated by substituting the nondetectable values with the LOD/2 and the LOD/square-root-of-2. The GM (GSD) for the measurements with the graphite furnace measurements substituted for the nondetectable values were 28.1 (7.0) micrograms/cubic meter, while the GM (CSD) with the LOD/2 and the LOD/square-root-of- 2 substituted for the nondetectable samples were 29.1 (6.7) and 31 .6 (6.2), respectively. This analysis supports the recommendation that substitution of nondetectable values by the LOD/2 should be used when data are highly skewed.
Workers; Work-environment; Exposure-levels; Toxins; Sampling; Analytical-methods; Industrial-hygienists; Industrial-hygiene; Statistical-analysis
American Industrial Hygiene Conference and Exposition, May 19-23, 1997, Dallas, Texas