We employed a retrospective cohort study design to investigate the effect of self-reported use of hearing protectors on the 5-year incidence of hearing shifts among workers exposed to hazardous noise. We analyzed audiometric records from 2005–2009 for 19,911 male and female workers ages 18–65 years. These data were collected by the NIOSH OHL Surveillance Project from audiometric service providers (i.e., providers) who perform audiometric testing for U.S. workers in regulated industries. Audiometric service providers shared de-identified audiograms with NIOSH as described in Masterson et al. [2012]. Because the audiograms were de-identified before being sent to NIOSH, a project exemption was obtained from the NIOSH Institutional Review Board.

Each audiogram includes: an arbitrary employee identification number unique to the individual worker (employee ID); date of test; hearing threshold values at frequencies 0.5, 1, 2, 3, 4, 6, and 8 kHz; industry code classified using the 2007 North American Industry Classification System (NAICS); and limited demographic data including the worker's age and sex. Some audiograms also include responses to questions posed to workers during testing, such as whether they wear hearing protection, have non-occupational noise exposures (hunting, military, or other recreational), are taking any medications, have an infection at the time of the test, etc. However, the extent to which these fields are populated varies, such that some providers never recorded these responses, some did so intermittently, and others did so fairly consistently.

This analysis utilized audiometric records from one provider that consistently recorded data on workers’ self-reported use of hearing protection at work. Our analytic sample dataset included the most recent (last) audiogram from each of the years 2005–2009 for all workers ages 18–65 years who had at least one valid audiogram in both 2005 and 2009 and who had at least one additional valid audiogram during the three intermediate years of the study period (i.e., had no more than 2 years without an audiogram during the study period). These criteria were adopted to provide sufficient evidence to infer continuous employment with exposure to hazardous noise. Substantial reductions in sample size would have resulted if audiograms were required for every year of the study period.

While there are exceptions, such as when companies choose to test all of their employees regardless of noise exposure, the vast majority of the audiograms included in the database were conducted because the workers were exposed to hazardous noise and were included in government-mandated hearing conservation programs (HCPs). For example, the OSHA Occupational Noise Exposure regulation for general industry requires that employers administer a continuous, effective HCP—including baseline and subsequent annual audiograms—for all employees whose noise exposures equal or exceed an 8-hr time weighted average of 85 dBA [Occupational Safety and Health Administration, 1983]. In this study, audiometric testing is assumed to imply that the worker was exposed to hazardous noise.

Because the audiograms included in the NIOSH repository were originally collected for administrative rather than research purposes, records may be incomplete or inaccurate [Laurikkala et al., 2000]. Consequently, audiograms that did not meet quality standards were removed from the analysis. Audiograms indicative of hearing impairment due to non-occupational factors or pathology were also removed. Specifically, audiograms were excluded if they contained unlikely threshold values suggesting testing errors, negative slope indicative of excessive background noise during testing, or large inter-aural differences suggesting non-occupational pathology and possible “crossover” from the bad ear to the good. Procedures for identifying these audiograms are described elsewhere [Masterson et al., 2012]. Audiograms with missing values for age, sex, NAICS (industry) code, or threshold values at frequencies 2, 3, 4, or 6 kHz were also dropped from the analysis. Finally, audiograms from workers in the mining industry sector were excluded due to insufficient numbers.

The dependent variable in this study was change in hearing between each worker's 2005 audiogram (designated baseline) and 2009 audiogram (current). We analyzed two measures of hearing shift. The first measure was an OSHA standard threshold shift (OSTS), defined as a 10 dB or greater increase in the average pure-tone hearing threshold across the frequencies 2, 3, and 4 kHz in either ear from the baseline audiogram to the current audiogram. OSTS is commonly used among hearing conservationists and incorporates frequencies within the speech range. The second measure was a high frequency threshold shift (HFTS), defined as a 10 dB or greater increase in the average pure-tone hearing threshold across the frequencies 3, 4, and 6 kHz in either ear—again comparing the 2009 audiogram with the 2005 baseline. HFTS is a measure of hearing change at the frequencies which are usually the first to be impaired by exposure to hazardous noise. Both outcome variables were dichotomous and coded as 1 if a shift was present and 0 if not. Because OSTS and HFTS were defined in relation to a 2005 baseline—corresponding to the beginning of our study period—all shifts were considered incident cases.

Our main independent variable, self-reported use of hearing protection at work, was constructed from 2005–2009 questionnaire responses collected from workers prior to audiometric testing and included in their audiometric record. Workers completed a survey which included the question “Do you normally work in noise WITHOUT using your hearing protection in required areas?” Workers’ responses were used to create an ordered categorical exposure variable with three levels: 1-never, 2-sometimes, and 3-always. If a worker reported not wearing hearing protection (i.e., answered “yes”) for every available audiogram during 2005–2009, the variable was coded 1-never. Conversely, if a worker reported on every audiogram that he or she did wear hearing protection (i.e., answered “no”), it was coded 3-always. If, however, a worker answered “yes” on at least one audiogram and “no” on at least one audiogram during 2005–2009, the variable was coded 2-sometimes.

Worker age (in 2009) and sex were included in the analysis as covariates. We also controlled for industry sector, comprised of groupings of two-digit NAICS codes. Additionally, dichotomous variables reflecting workers’ self-reported exposure to recreational target shooting or hunting and other non-occupational hazardous noise were included. These questions were also part of the questionnaire that workers completed prior to audiometric testing. Workers who reported these exposures on any audiogram from 2005 to 2009 were considered positive for that exposure. Finally, we included a continuous variable reflecting the number of years each worker had an audiogram prior to 2005 as a proxy measure to control for variability in the duration of workers’ exposure accumulated before the 2005–2009 observation period. Data on other potentially important covariates such as smoking, race, income, education, and occupation were unavailable.

The 5-year cumulative incidence of OSTS and HFTS was calculated as the proportion of workers with a given shift at their last audiogram in 2009. Point estimates and 95% confidence intervals for the incidence proportion of both shift measures were produced for the full sample and by sex, industry sector, exposure to recreational shooting or hunting, exposure to other recreational noise, and hearing protector use.

The bivariate association between self-reported use of hearing protection at work (an ordinal variable) and the incidences of OSTS and HFTS was assessed using the Cochran–Armitage test for trend. Associations between nominal independent variables (sex, industry sector, exposure to recreational shooting or hunting, and exposure to other recreational noise) and the outcome variables were assessed using the Chi-square test for independence. For continuous variables (age and duration of exposure), we used the t-test for equal means or, where necessary, the Cochran and Cox approximation of the P-value for the t-test to account for unequal variances. Statistical significance for all tests was indicated based on an Alpha level of 0.05. All analyses were conducted using SAS 9.3 (SAS Institute Inc., Cary, NC).

To evaluate the relationship between self-reported use of hearing protection at work and the 5-year cumulative incidences of OSTS and HFTS, while controlling for other factors, we modeled adjusted odds ratios and their 95% confidence intervals using logistic regression. However, because our data consisted of workers clustered within companies, and company may strongly influence the safety culture, including the use of hearing protection, we also wished to quantify the inter-company variability in the risk of hearing shifts. Therefore, we employed a multi-level analysis which takes into account the hierarchical structure of the data [Searle, 1987; Goldstein, 1995; Hox, 2002]. Using the logit link function for binomial data, we fit a generalized multilevel, multivariable model using the SAS GLIMMIX procedure to estimate adjusted odds ratios for OSTS and HFTS for all independent variables (fixed effects), while specifying company as a random effect. Also using the SAS GLIMMIX procedure, we performed linear trend tests of the levels of self-reported use of hearing protection at work for the OSTS and HFTS outcomes.

We described the degree to which the company effect influenced the risk of OSTS and HFTS in terms of the median odds ratio (MOR) [Larsen et al., 2000]. The MOR is a simple function of the cluster (company-level) variance and converts it to the familiar odds ratio scale, where 1 is the null value. That is, when the MOR = 1, there is no company effect. The MOR is always ≥1. Here, the MOR is defined as the median value of the odds ratio between the higher and lower risk of two randomly selected companies, that is, the median increase in risk that would result from a worker moving to another company with a higher risk [Merlo et al., 2006]. As the company influence on the risk of workers experiencing an OSTS or HFTS increases, so does the MOR.

For categorical variables, we used the following reference groups: “Female” for sex, “Manufacturing” for industry sector, “No” for both recreational target shooting or hunting and other recreational or non-occupational noise exposures, and “Always” for hearing protection use at work. With the exception of industry sector, we chose these categories to make comparisons with putative lower-risk groups. We chose “Manufacturing,” which is not considered a lower-risk group, as the referent due to very small numbers in the other industry sectors which could result in unstable comparisons [Hardy, 1993]. For continuous variables, the mean value was used as the reference, with risk ratios expressed as the effect of a 1-unit offset from the mean.

The findings of this study are subject to several limitations which arise from both the composition of our study sample and the nature of the data available on sample subjects. First, selection bias may have influenced our results. We utilized a convenience sample from data contributed by a single audiometric service provider who operated primarily in one geographic region of the U.S. and served companies from a limited range of industries. Such a sample does not allow the generalizability which could have been drawn from a representative, probability sample of U.S. workers enrolled in hearing conservation programs. Furthermore, we excluded workers with less than three annual audiograms from the sample. Some of these may reflect non-compliance with audiometric testing programs, and such workers may also be less compliant with the requirement to wear hearing protection in hazardous noise. Removing these workers from the sample may have under-represented the population of noise-exposed workers who do not wear hearing protectors, skewing the results toward the null. In addition, we excluded a large number of audiograms due to quality issues. Audiometric testing for hearing conservation purposes often takes place in a background noise environment compliant with OSHA regulations but not with current practice guidelines, resulting in artificially elevated thresholds [Frank and Williams, 1994; Lankford et al., 1999]. The audiometric technicians who collect the data usually are not audiologists and may not have the experience to notice and retest unlikely thresholds. Excluding a substantial number of audiograms from our dataset may have introduced bias if those exclusions differentially represented subsets of workers with higher or lower risk for shifts in hearing.

Second, like all studies that make use of self-reported data from questionnaires, our study is subject to both recall and social desirability biases. Respondents were asked to recall behaviors occurring over the course of the previous year. Furthermore, workers were required to classify their hearing protector use as either “yes” or “no,” and some workers with mildly intermittent protector use may have had difficulty deciding whether they “normally” wore hearing protectors when exposed to loud noise. Workers also may have been inclined to respond in a way that would reflect compliance with their employers’ policies, leading to misclassification of workers along the independent variable.

Furthermore, even workers who accurately reported hearing protector use may not have been properly protected from their noise exposure due to improper or inconsistent use of the device. Most workers do not receive the level of noise reduction predicted by its labeled Noise Reduction Rating (NRR), and removing a hearing protector for even short amounts of time during a workday substantially reduces the effective level of protection [Occupational Safety and Health Administration, 1983; National Institute for Occupational Safety and Health, 1998].

Misclassification also may have resulted due to lack of actual noise exposure data on the workers in the study sample. Although audiometric testing as part of a mandated hearing conservation program is a strong indicator that a worker is noise-exposed, some proactive employers may offer hearing testing to all of their workers regardless of exposure status. Therefore, our study sample may have included some unexposed individuals who would not be expected to show noise-related shifts regardless of hearing protector use. We could not assess whether or to what extent this was actually the case because those data were not available from the audiometric service provider, but we believe it would have occurred rarely if at all because companies have an economic incentive to test only those workers required by regulation. Additionally, because of the way the survey question assessing HPD use was worded—workers were asked if they worked without hearing protection in noisy areas wear it was required—any unexposed workers included in the sample would likely have answered “no” and been classified as “always” wearers of hearing protection. So any bias introduced would have been in favor of a protective effect of HPD use. Even among workers exposed to hazardous noise, there is likely to have been considerable variation in the intensity and duration of individual workers’ exposure.

Additionally, confirmation of threshold shifts in a subsequent audiogram is recommended to rule out temporary threshold shifts which recover over time or result from short-term medical conditions, subject inattention, or tester error [National Institute for Occupational Safety and Health, 1998]. This was not possible using our dataset, so some of the OSTS or HFTS recorded in this study may have been spurious.

Finally, our study was observational and our findings regarding the minimal association between hearing protector use and hearing shifts do not imply lack of a causal relationship. While we attempted to control for as many confounding variables as possible, our results may have been influenced by other key factors. For example, although we included an adjustment for the effect of non-occupational exposure in the analysis, only crude non-occupational exposure measures were available and identified hearing losses could have resulted from unprotected noise exposure outside of the work place. We were unable to control at all for the effect of other potentially important variables such as smoking status, race and education. And, as no job or medical histories were available, the work-relatedness of identified shifts in hearing must be inferred.

Despite these limitations, this study also had a number of strengths. First, the dependent variables—hearing shifts—were based on audiometric data rather than self-report. We also used two different hearing outcome measures—the traditional OSTS utilized in OSHA-mandated HCPs and the HFTS based on audiometric test frequencies initially affected by noise exposure and therefore more sensitive to early evidence of threshold shift. The repository of available audiograms was large enough to allow excluding records with quality issues or evidence of non-occupational pathology, resulting in a clean dataset from which to evaluate the effectiveness of hearing protectors in preventing noise-induced hearing loss. In addition, sufficient employees had serial audiometric records available across a 5-year period to allow examination of a time interval long enough to observe hearing changes and monitor patterns of hearing protective behavior.

The survey question used to define the independent variable—hearing protector use—was asked consistently across all years included in the study. The response was completed by each worker in writing prior to arriving at the mobile audiometric test unit. Consistency of protector use across the study period was categorized along a continuum of always, sometimes, and never.

Finally, we were able to control for the effect of a number of important covariates in the analysis including, age, sex, industry sector, duration of exposure, prior occupational exposure, and exposure to non-occupational noise.