Page:NIOSH DM DFM respirator evaluation draft.pdf/60

This page has been proofread, but needs to be validated.

52

Performance Evaluation of DM and DFM Filter Respirators—WORKING DRAFT 9.15.92

methods^[1]). As with birth control methods, the ideal public health goal for respirator wearers is a zero control failure rate.
For respirator performance, control failure rate will be defined as the number of users per 100 users that fail to achieve individual working protection factors equal to or exceeding the assigned protection factor for their respirator (i.e., WPF's 2 APF). Where respirators are used, the reasonable expectation of both purchasers and users is that none of the users will receive less protection than the class APF (when the masks are properly selected, fit tested by the employer, and properly worn by the users).
One type of statistical analysis suitable for WPF failure-rate analysis computes the following two values:

A point estimate for the number of users that fail to achieve a given WPF per 100 users (failure rate) and
A 1-sided, 95% upper confidence limit (UCL_1::95) for the actual number of user WPFs less than a given WPF per 100 users (actual failure rate under the conditions of the study).^[2]

If one were to able to conduct multiple WPF respirator-performance studies under conditions identical to any given study reported by a research team, the resulting study-to-study failure-rate point estimates would vary considerably due to sampling error. Generally, the smaller the sample size in a study, the larger the potential sampling error. Thus computation of confidence limits is essential so that one can create a confidence interval (interval estimate), This is a range of values around the point estimate within which we are confident (at a specified confidence level) that the actual failure rate lies. With a confidence interval one can then assess the amount of uncertainty or margin of error associated with the point estimate of the actual failure rate in each study. Regarding the 95% confidence level associated with each particular UCL_1,95, statistical theory predicts for any given sample of WPF's that in 19 of 20 similarly conducted studies the similarly computed UCLs will exceed the actual

↑ Trussell, J. and K. Kost: Contraceptive Failure in the United States: A Critical Review of the Literature, Studies in Family Planning 18(5):237-288 (1987).
↑ Leidel, N. A. and K. A. Busch: Statistical Design and Data Analysis Requirements. Chapter 8 of Patty's Industrial Hygiene and Toxicology, Volume III, Theory and Rationale of Industrial Hygiene Practice, Second Edition, Volume 3A, The Work Environment, Cralley, L. J. and L. V. Cralley, Editors, John Wiley & Sons, Inc., New York, (1985), Section 6.8, p. 493–497.

[1] Trussell, J. and K. Kost: Contraceptive Failure in the United States: A Critical Review of the Literature, Studies in Family Planning 18(5):237-288 (1987).

[2] Leidel, N. A. and K. A. Busch: Statistical Design and Data Analysis Requirements. Chapter 8 of Patty's Industrial Hygiene and Toxicology, Volume III, Theory and Rationale of Industrial Hygiene Practice, Second Edition, Volume 3A, The Work Environment, Cralley, L. J. and L. V. Cralley, Editors, John Wiley & Sons, Inc., New York, (1985), Section 6.8, p. 493–497.

[1]

[2]