inadequately-fitting respirators. These percentages consider only face-seal leakage. Any additional leakage through filters or sorbent elements will increase the percentage of wearers not achieving APF-level protection. It should be recognized that those wearers with inadequate respiratory protection will not be identifiable except possibly for those contaminants with adequate warning Properties or those very few contaminants for which an overexposure can be biologically detected (e.g., urine- or blood-monitoring techniques).
Additionally, it should be also recognized that the Hyatt/LASL APF's do not consider uncertainty present in the underlying performance data that is due to sampling errors. However, this is not necessarily a critical weakness in the approach. With
the Hyatt/LASL approach it is not particularly relevant nor critical to know with a
high degree of certainty the precise percentage of all workers expected to achieve a
given APF before proper fit testing is performed. That is, it is basically irrelevant
whether 5, 9, 10, 12, or 15% are incapable of achieving a satisfactory fit. This is be-
cause the purpose of subsequent fit testing is to screen out 100% of these individuals.
NIOSH concluded that APFs computed according -to the definition used by Myers et al. and other researchers in the 1980s can result in less protection to at least 1 in 20 respirator users when compared to APF's computed according to the criterion of Hyatt/LASL in the 1970s even if the fit testing in 100% effective. This occurs because Myers et al. APF values are predicated on the less strict requirement that only an estimated 95%, not an assured 100%, of respirator users in the workplace must attain a class APF after proper fitting (i.e., fit testing) has been performed by the employer.
NIOSH has concluded that the Myers et al. approach implicitly considers it accept-
ates an equivalent situation to one that would result if OSHA were to permit 5% of American workers to exceed permissible exposure limits (PELs).
The results from any WPF must be evaluated both in terms of internal validity and external validity. With regard to internal validity, suppose a researcher were to conclude that the WPF-performance of respirator A is better than some performance criterion. If the conclusion was based on random errors occurring during WPF mea- surements, rather than truly superior pert , the conclusion would have no internal validity. Variability in WPF results exist and this variability casts doubt on the internal validity of any conclusions drawn from WPF results.
However, there is a widely accepted solution to the problem of questionable inter- nal validity. The answer is to perform a statistical analysis that takes into account not only the differences among WPF percentile point estimates, but also considers the variability of the WPF results.
