The study protocol was approved by the NIOSH Institutional Review Board. To choose the sample size for this study, we chose 5% as an estimate of the failure rate and 3% maximum error of the estimate; these resulted in a sample size of 200 and 95% confidence interval of 2–8% for the estimated failure rate. The reasons the expected failure rate of 5% was chosen are that (1) Lord Corporation conducted fit testing annually and found less than 1–3% of employees switching to different sizes and/or models; and (2) Hoffmann-La Roche conducted fit testing every two years and found that 7% of the employees switching to different sizes and/or models. Since some subjects may drop out of the study, 229 subjects were recruited at the beginning of the study.

Subjects were recruited according to the frequency of the U.S. workforce in each face size category of the new NIOSH respirator fit test panel based on face width and face length.^[8] Subjects were recruited from the NIOSH subject pool for certification testing and physiology study where they maintain a medically cleared status for testing by undergoing an annual physical. Additional subjects were recruited from the general public of the southwestern Pennsylvania region; these subjects were evaluated only using the OSHA Respirator Medical Evaluation Questionnaire (Appendix C to OSHA 29 CFR 1910.134).^[2] A NIOSH medical doctor reviewed the answers and determined if the subjects were cleared to participate. In the rare instance that an electrocardiogram was needed during the medical screening, it was conducted by the doctor in an onsite physiology lab. Subjects were excluded if they had a history of uncontrolled chronic asthma, pneumonia, or high blood pressure (i.e., systolic > 160 mm Hg, diastolic > 95 mm Hg).

Pregnant women were permitted to participate. Efforts were made to recruit some subjects who were overweight or on diets or on weight loss programs. No one was excluded from this study because of race, gender, or facial characteristics unless the sampling goal in each size category was exceeded by more than 25%. We accepted any subject meeting our sampling criteria and who met the criteria specified by OSHA 29 CFR 1910.134 for respirator wearers, such as exclusion of beards, sideburns, etc. We also attempted to recruit a similar number of male and female subjects. Subjects aged 18–60 were allowed to participate in this study. Both experienced and inexperienced respirator users were accepted because all participants underwent training on how to don the respirator and were also assisted during donning.

Participants signed a consent form on the first visit of the study. All volunteers received monetary reimbursement for their participation. Height, weight, blood pressure, and anthropometric facial dimensions were measured on each visit; shoes were removed prior to height and weight measurements. Height was measured using a Seca 242 Digital Stadiometer and weight using a Seca 882 Digital Scale (Hanover, MD). Blood pressure was measured using an Omron HEM-870 blood pressure monitor (Bannockburn, IL). Thirteen facial dimensions (head breadth, minimal frontal breadth, nasal root breadth, interpupillary breadth, face width, nose breadth, bigonial breadth, lip length, head circumference, nose length, nose protrusion, face length, and menton subnasale) were measured using spreading calipers and sliding calipers (GPM Instruments, Zurich, Switzerland) and a Lufkin steel measuring tape (Cooper Tools, Apex, NC). Face length and face width measured on Visit 1 were used to determine cell number according to the NIOSH bivariate panel (Figure 1).^[8]

Seven different sizes and styles of N95 filtering-facepiece respirators (FFR), none having exhalation valves, were chosen from those included in the pre-pandemic CDC Strategic National Stockpile (SNS) (Table 1),^[9] representing a variety of sizes. We focused on FFRs because they are the most commonly-worn form of negative pressure half-mask respirator^[1] and the predominant user—healthcare industry—has frequently questioned the need for annual fit testing.^[5]

Fit factor (FF) is a quantitative estimate of the fit of a particular respirator to a specific individual, and typically estimates the ratio of the concentration of a test agent (for example, ambient particles) in ambient air to the concentration inside the respirator when worn. Inward leakage (IL) consists of particle leakage through two pathways: respirator filter (filter penetration) and the interface between the respirator and a wearer’s face (faceseal leakage). A method for evaluating respirator fit by first determining IL by means of a test subject fit test was validated in a previous pilot study^[7] and those methods were used for the current study. A PortaCount® Plus (Model: 8020A, TSI, Inc., Shoreview, MN) condensation particle counter (without the N95-Companion accessory) was used to measure IL of ambient particles ranging from 0.02 to > 1 μm.^[10] The PortaCount^® Plus without N95-Companion counts a larger size range of particles (as compared to using the N95-Companion) and increases the calculated maximum fit factor value to > 10,000. Using the PortaCount Plus alone was also shown in a previous study to have a smaller combined error (alpha, beta, and assignment error) than the PortaCount^® Plus with N95 Companion.^[11] Based on results of a pilot study assessing changes in respirator fit over a four-week period, it was determined that this fit test method can adequately measure various variance components to allow accurate comparison of fit between visits.^[7]

While seated to accommodate subject comfort, subjects completed a fit test comprised of five individual exercises: normal breathing, deep breathing, breathing while moving their head from side to side, breathing while moving their head up and down, and a return to normal breathing. Each individual exercise had a total duration of 50 s comprised of (in order) a 4 s ambient sample line purge, 5 s ambient sample, 11 s mask sample line purge, and a 30 s mask sample; thus, each complete fit test comprised of 5 exercises lasted for approximately 4.2 min. Each subject visit required nine fit tests (3 samples × 3 donnings each), totaling approximately 40 min of fit testing. Subjects were prompted to conduct each exercise of a fit test by visual cues from a laptop computer running FitPlus for Windows (software developed by TSI, Inc. for fit test data collection). Three exercises—bending over, talking, and grimace—performed in the OSHA ambient aerosol condensation nuclei counter (CNC) quantitative fit test protocol (CFR 1910.134) were omitted to economize on time to encourage on-going subject participation. Moving the head up and down has been found to be as challenging as bending over and talking.^[12] The grimace exercise was excluded from our fit test protocol because this study aimed to evaluate changes in respirator fit due to weight gain/loss and was not concerned with how well a respirator is able to reseat on the user’s face following the grimace.

The respirator model selection process only occurred on Visit 1. On Visit 1, each subject randomly selected a respirator model that they felt was most comfortable for them from the seven models and then watched the appropriate training video for donning and doffing that model (Figure 2). Comfort was evaluated by assessing the accommodation for eyeglasses (if worn) and being able to comfortably position the respirator on the face. We chose FF ≥ 100 as the criterion for a passing FF value for this study. If the subject obtained a FF < 100 on three consecutive fit tests using the same respirator sample, he/she was asked to select another respirator model, watch a training video for that model, and then perform three fit tests using a sample of the new model. This process was repeated until a respirator model with adequate fit was tested. If the subject obtained a FF ≥100 on one of the three consecutive fit tests using the first respirator sample, he/she tested two more samples of that model, performing three fit tests on each sample. During subsequent visits (Visits 2–7) a subject received three new respirator samples of the model assigned on Visit 1 with all samples taken from the same box for all subsequent visits (Figure 3).

Subjects were fit tested in a laboratory room using the ambient air in the room as the test agent. The ambient particle concentration of the laboratory was typically between 5,000 and 15,000 particles per cm³. Periodically when the ambient aerosol concentration fell below the minimum concentration of 1,000 particles per cm³ needed to operate the PortaCount^®, the ambient aerosol concentration was supplemented with sodium chloride aerosol using a generator (Model: 8026, TSI, Inc.). For each donning, subjects first performed a user seal check, and while seated waited for three minutes with the respirator donned to purge the ambient particles inside the respirator. This 3-min acclimation period is shorter than the 5 min specified by the OSHA ambient aerosol CNC quantitative fit test protocol (CFR 1910.134) and was selected to shorten the overall visit time to encourage subject retention. Next, the subject completed the five exercise fit test while seated. At the completion of the fit test, the PortaCount^® Plus outputs an overall FF by calculating the harmonic mean of the FFs from the five individual exercises. The inverse of the overall FF is the IL.

After subjects completed fit testing, filter penetration was measured for each respirator by sealing the respirator’s faceseal edge (the periphery of the respirator which contacts the wearer’s face) to a Plexiglas plate using melted beeswax. The plate, containing a centrally-located 3.5 cm diameter circular opening, was placed into a holder with an airtight connection to a vacuum line drawing air continuously at approximately 10.3 L/min to simulate the breathing minute volume of a person while seated.^[13,14] A PortaCount^® Plus was used to measure the ambient aerosol concentration in the room and the concentration at approximately 2 in downstream of the filter. The same test time duration used for the test subject for the five exercise IL test (approximately 4.2 min) was employed for filter testing. A FF was obtained for the test; the inverse being filter penetration because this test measured only particle penetration through filter. The test aerosol concentration for the filter penetration test was typically between 5,000 and 15,000 particles per cm³. Ambient aerosol was used for this testing, except for occasionally being supplemented with sodium chloride aerosol as described above when the ambient concentration fell below 1,000 particles per cm³.

Each respirator was tested three times and a mean filter penetration was calculated. Mean filter penetration for each respirator sample was subtracted from the IL measurement of each donning measured on the subject’s fit test for the corresponding sample to calculate face seal leakage (FSL) of each of the nine donnings. The reciprocal of FSL is the calculated FF used in this study as part of the criteria for determining whether a subject achieved acceptable fit on a particular visit.

The 90th percentile FSL value of each set of nine donnings was calculated for each subject at each visit utilizing the geometric mean and geometric standard deviation of the nine FSL values: (1)X0.90=GM×GSD1.282 where: X_0.90 = 90th percentile FSL

GM = geometric mean of FSL

GSD = geometric standard deviation of FSL

On each visit, a subject’s overall respirator fit was classified as “acceptable” if the 90th percentile FSL was ≤ 5% for the set of 9 fit tests and at least 1 of the 9 tests resulted in a FF ≥100. If either criterion was not met, the fit for that visit was classified as “unacceptable.” We chose the criterion of 90th percentile FSL ≤ 5% (corresponding to FF ≥ 20) because it is the OSHA Assigned Protection Factor for FFRs (10) multiplied by a safety factor of 2.

Faceseal leakage data were log-transformed and Shapiro-Wilk normality tests were performed for data grouped by subject and visit (1197 data sets); because 83% of sets were log-normally distributed, log-transformed FLS data was used for analysis. Analysis of Variance (ANOVA) tests were performed on log-transformed FSL values to test if group means were significantly different between fit acceptability categories of each visit. All analyses were performed using SAS for Windows 9.3 (SAS Institute Inc., Cary, NC).

Based on OSHA’s consideration that a 7% change in respirator model after 2 years was unacceptable and thus adopted the annual fit test requirement, we considered a 7% or greater change in fit acceptability over a 1-year period as the a priori criteria for our study.

Table 2 shows the percent of subjects with unacceptable fit for each of the seven visits, where 195 subjects were tested for Visit 1 and fewer subjects were tested on subsequent visits due to subject attrition. An unacceptable fit was observed for 13.8%, 10.4%, 6.9%, 12.3%, 14.9%, and 15.7% of subjects for Visits 2–7, respectively (average = 12.3%). The majority of these changes in fit acceptability at 6-month intervals are higher than the a priori criterion of 7%.

The projected risk of obtaining an unacceptable fit if fit testing is only performed at the one-, two-, or three-year intervals was calculated as the sum of subjects with unacceptable fit on the indicated visit plus subjects with unacceptable fit on any two consecutive previous visits plus subjects with unacceptable fit on the previous yearly visit(s) (Table 3). The risk of an unacceptable fit increased from 10% at Year 1 to 20% at Year 2 to 25% at Year 3.

A large percentage of subjects (63.4%) experienced acceptable fit on all 7 visits (Table 4). A total of 88% of subjects experienced unacceptable fit once, twice, or not at all during the seven visits and 12% of subjects experienced unacceptable fit 3 or more times over the 3-year period. The percentage of subjects that changed from acceptable to unacceptable fits between consecutive visits ranged from 2.9–13.8% (Table 5).

Changes in fit acceptability higher than the a priori criterion of 7% were observed in 4 of the six 6-month visit intervals (Table 5). These data raise a concern for the possibility of workers in actual workplaces to have a change in respirator fit resulting in decreased protection even after a six-month interval between fit tests. However, performing fit testing every six months would pose a significant additional burden on employers in terms of expense and testing time and would need to be weighed against the benefits from the marginal improved protection that it would achieve.

For subjects with acceptable fit, mean weight change ranged from −2 to 3 lb. For those with unacceptable fit, mean weight change ranged from −2 to −13 lb (Table 6). For each visit, ANOVA results indicate that GM FSL for subjects in the acceptable fit category was significantly less (P < 0.05) than GM FSL for subjects in the unacceptable fit category for each visit.

Across all visits, 24% of subjects experiencing the greatest weight loss (≥ 20 lb) had an unacceptable fit (Table 7). The percentage of test subjects in each group is obtained by dividing the number of test subjects with unacceptable fit on the indicated visit in that group by the total number of subjects for that visit (both those classified with acceptable or unacceptable fit) experiencing the indicated weight change. Compared to subjects who lost > = 20 lb, the percentages of subjects with unacceptable fit decreased for subjects who experience lower weight loss or higher weight gains. Only 7% of subjects experiencing the greatest weight gain (≥ 20 lb) had an unacceptable fit; however, the total number of subjects in this category was small (n = 43).

Odds ratio analyses were performed using the total number of subjects falling into each weight change category of Table 7. The predictor variable is weight change category and the response variable is fit acceptability status. Weight gain < 10 lb was used as the positive status for “weight change category” and “unacceptable fit” is the positive status for “fit acceptability status.” An odds ratio > 1 indicates a weight gain < 10 lb is more likely to result in unacceptable fit than the weight change category being compared to. An odds ratio < 1 indicates a weight gain < 10 lb is less likely to result in unacceptable fit than the weight change category being compared to. Subjects in two weight change categories (i.e., subjects loosing > = 20 lb (P-value < 0.01) and subjects loosing > = 10 lb but <20 lb (P-value < 0.02)) had significantly higher percentages of subjects with unacceptable fit than subjects gaining < 10 lb. One limitation of this analysis was that the number of subjects whose weight gain was ≥ 20 lb was smaller for each visit compared with the other weight change groups. Chi-Square analyses showed that there was an association between weight change category and fit acceptability status using the data from subjects gaining <10 lb compared to subjects loosing > = 20 lb (P-value < 0.01) and loosing > = 10 lb but <20 lb (P-value < 0.02).