Inward leakage variability between respirator fit test panels - Part II. Probabilistic approach.

This study aimed to quantify the variability between different anthropometric panels in determining the inward leakage (IL) of N95 filtering facepiece respirators (FFRs) and elastomeric half-mask respirators (EHRs). We enrolled 144 experienced and non-experienced users as subjects in this study. Each subject was assigned five randomly selected FFRs and five EHRs, and performed quantitative fit tests to measure IL. Based on the NIOSH bivariate fit test panel, we randomly sampled 10,000 pairs of anthropometric 35 and 25 member panels without replacement from the 144 study subjects. For each pair of the sampled panels, a Chi-Square test was used to test the hypothesis that the passing rates for the two panels were not different. The probability of passing the IL test for each respirator was also determined from the 20,000 panels and by using binomial calculation. We also randomly sampled 500,000 panels with replacement to estimate the coefficient of variation (CV) for inter-panel variability. For both 35 and 25 member panels, the probability that passing rates were not significantly different between two randomly sampled pairs of panels was higher than 95% for all respirators. All efficient (passing rate ≥80%) and inefficient (passing rate ≤60%) respirators yielded consistent results (probability >90%) for two randomly sampled panels. Somewhat efficient respirators (passing rate between 60% and 80%) yielded inconsistent results. The passing probabilities and error rates were found to be significantly different between the simulation and binomial calculation. The CV for the 35-member panel was 16.7%, which was slightly lower than that for the 25-member panel (19.8%). Our results suggested that IL inter-panel variability exists, but is relatively small. The variability may be affected by passing level and passing rate. Facial dimension-based fit test panel stratification was also found to have significant impact on inter-panel variability, i.e., it can reduce alpha and beta errors, and inter-panel variability.