Examining physiological and perceptual consequences of noise exposure

Excessive exposure to noise has been known to damage the ear, hence regulation of the amount of occupational noise allowed in certain industries (OSHA, 1983; NIOSH, 1998). The primary consequence of noise exposure was traditionally thought to be damage to hair cells and their stereocilia, resulting in either temporary (Dobie and Humes, 2017) or permanent shifts in hearing sensitivity (Neuberger et al., 1992; Clark and Pickles, 1996) depending on level and duration of exposure. Recent research suggests that clinically undetectable damage from noise exposure, termed “hidden hearing loss,” may manifest as an impaired ability to understand speech in the presence of background noise (Schaette and McAlpine, 2011). This hypothesis stems from animal studies which show exposure to noise damages the synapses coupling inner hair cells and auditory nerve fibers (“synaptopathy”), mostly fibers that respond to high sound levels (Kujawa and Liberman, 2009; Furman et al., 2013; Liberman and Liberman, 2015). The extent of synaptopathy in animals is negatively correlated with auditory brainstem response (ABR) wave I amplitudes, in that the number of fibers affected will systematically reduce the ABR wave I amplitude. ABR threshold, however, may not increase until a large portion of fibers is affected (Lin et al., 2011). These findings in animals have spurred research into relating noise exposure history with ABR measures in humans (i.e., noise-induced synaptopathy), and the development of tests to detect hidden hearing loss in the clinic. This has proven to be a difficult task for a number of reasons: quantification of noise exposure must be done through subjective historical accounts, and dose of exposure can only be estimated; a lack of normative data for ABR amplitudes; a potentially lower than suspected prevalence of hidden hearing loss in a normal-hearing population (Hind et al., 2011; Spehar and Lichtenhan, 2018); and issues with clinical classification of “normal” hearing thresholds. The average pure-tone thresholds of healthy normal-hearing adults are 0 dB hearing level (HL) with 95% CI of a maximum of ±10 dB HL up to 8 kHz (Pienkowski, 2017), yet standard practice classifies thresholds up to 20 dB HL as normal. Additionally, synaptopathy can be comorbid (and even interact) with age-related hearing loss and other cochlear pathologies, such as outer hair cell loss, which affect hearing sensitivity. Results of studies in humans have been mixed—some finding relationships between noise exposure and ABR amplitudes suggesting synaptopathy (e.g., Bramhall et al., 2015; Stamper and Johnson, 2015; Grose et al., 2017; Valderrama et al., 2018), others not (e.g., Prendergast et al., 2017; Fulbright et al., 2017; Grinn et al., 2017; Spankovich et al., 2017). Some studies have found relationships between noise exposure history and speech perception (e.g., Liberman et al., 2016) or ABR amplitude and speech perception, dependent on audiometric thresholds (e.g., Bramhall et al., 2015). Yet others have not found relationships between noise history and speech perception (e.g., Yeend et al., 2017; Guest et al., 2018a, 2018b; Valderrama et al., 2018; Bramhall et al., 2018).

Relationships observed between reported noise exposure and ABR wave I amplitudes can be misleading, as there are a number of factors confounding such relationships. For example, both ABR amplitude and history of noise exposure are related to sex. Males tend to partake in noisier occupations and recreational activities and have smaller ABR wave amplitudes (Jerger and Hall, 1980). While one explanation may be that higher levels of noise exposure cause the lower ABR amplitude (i.e., synaptopathy), the two could be entirely unrelated. Some propose that males' larger head size and cochleae could result in smaller ABR amplitudes (Dehan and Jerger, 1990; Don et al., 1993). Age and cumulative measures of noise exposure history have been shown to be positivity correlated (Bramhall et al., 2017; Prendergast et al., 2017), while age and ABR amplitudes are negatively correlated (Jerger and Hall, 1980; Konrad-Martin et al., 2012; Bramhall et al., 2015). Audiometric and ABR thresholds are correlated (Stapells, 2000) and it is well-documented that audiometric thresholds and noise exposure are positively correlated, especially around 4 kHz (NIH, 1990). Such interdependencies confound attempts to observe a direct relationship between noise exposure and ABR wave I amplitude.

The present study aims to explain the variability in ABR wave I amplitudes using noise exposure history and several other factors that may confound a direct relationship between ABR wave I amplitude and noise exposure, including age, sex, and hearing thresholds. We examine ABR wave I amplitude in both a population of adult participants with a large range of ages, noise exposure history, and audiometric thresholds, as well as in a subset of younger adult participants with normal audiometric thresholds. Two measures of noise exposure were used and compared to one another in terms of their ability to explain ABR wave I amplitude. Consequences of noise exposure on speech perception were also examined using word recognition in several difficult listening conditions: background noise, time compression, and reverberation. Finally, the relationship between the ABR wave I amplitude and word recognition was examined. The results of this study will inform ongoing efforts by the research community to relate noise exposure history with ABR measures in humans, and the development of diagnostic tools or test batteries for synaptopathy. The ability to diagnose noise-induced synaptopathy will guide pharmaceutical interventions that are currently under development. For example, a diagnosis that suggests primary neural degeneration, as opposed to hair cell loss, might be treated by application of neurotrophic factors, which have been shown to regenerate synapses following noise-induced synaptopathy in transgenic mice (Wan et al., 2014; Suzuki et al., 2016).

Ninety-seven adults age 20–86 (mean age = 45) years participated in the study (56 female). Tympanometry and standard audiometry determined inclusion into the study. Pure-tone air conduction thresholds (HD 200 over-the-ear headphones; Sennheiser, Wedemark, Germany) at octave frequencies (0.25 to 8 kHz), three inter-octave frequencies (0.75, 3, and 6 kHz), and two extended high frequencies (EHFs) (11.2 and 16 kHz), were measured using an audiometer (GSI AudioStar Pro, Grason-Stadler) in 5-dB steps following the Hughson-Westlake procedure (ASHA, 1978). Thresholds at two frequencies (1.5 and 4 kHz) were measured in 2-dB steps following a modified Hughson-Westlake procedure. Participants were required to have thresholds ≤65 dB HL at all frequencies between 0.25–8 kHz and no air-bone gap >10 dB at any octave frequency from 500 to 4000 Hz. The mean and range of audiometric air conduction thresholds are displayed in Fig. 1(a). Equipment limitations only allowed for testing of levels up to 90 dB HL for 11.2 kHz and 60 dB HL for 16 kHz, therefore participants with hearing thresholds greater than these levels were indicated as having no response (“NR” in Fig. 1). In total there was one participant with NR to 90 dB HL at 11.2 kHz and 35 participants with NR to 60 dB HL at 16 kHz. Inclusion criteria also included middle-ear pressure within the range of −100 to +50 daPa and static compliance between 0.3 and 2.5 cm³ measured via 226-Hz tympanometry (Otoflex 100, Madsen). All additional measures were made monaurally; if both ears met the inclusion criteria, the better ear was chosen for testing. If both ears had similar audiometric thresholds, the test ear was selected randomly. In total, there were 49 right ears and 48 left ears included in the study.

Participants completed all measures within two months over two visits. Average data collection time for each participant was approximately four hours. All procedures were approved by the Boys Town National Research Hospital Institutional Review Board, and informed consent was obtained from all participants. Participants were paid for their participation.

Two measures were used to assess noise-exposure history. First, participants were asked whether they had ever been exposed to impulse noise (e.g., explosion or gunfire) without hearing protection in their lifetime. The response to this question determined assignment of the participant into a high or low noise group. Additionally, participants completed the noise portion of the Lifetime Exposure of Noise and Solvents Questionnaire (LENS-Q) (described by Bramhall et al., 2017). This questionnaire covers a variety of possible sources of noise exposure across military and non-military occupations and recreational activities. While all participants completed the non-military occupational and recreational sections, twenty-two active duty and veteran service members also completed the military-occupational noise section. The questionnaire asks about the frequency and duration of exposure and frequency of any hearing protection used. The LENS-Q was scored by assigning a dBA value to each activity or noise source based on a database of noise level measurements (Berger et al., 2015). For activities where the database included multiple measurements, the mean of available measurements was used. The dBA value was adjusted for use of hearing protection, and then weighted by assigning a value of 1 for exposure equal to 80 dBA and doubling the value with every 3 dB increase over 80 dBA. The weighted value was multiplied by frequency and duration of exposure, and then summed across all activities and sections for a raw score. The logarithm (base 10) of the raw score determined the participant's final score. This procedure is described in Bramhall et al. (2017).

In total, the High-Noise group consisted of 49 participants (16 female) and the low-noise group included 48 participants (40 female). The LENS-Q scores had a range of 3.50 to 12.41. Due to the number of military participants and the nature of activities included in the LENS-Q, the high-noise group and higher LENS-Q scores were skewed toward males. Participant characteristics by noise group are shown in Table I.

Tone-burst-elicited ABR waveforms were recorded at 1.5 and 4 kHz using custom-designed software (Cochlear Response [CResp] version 1.0; Boys Town National Research Hospital, Omaha, NE) on a computer equipped with a 24-bit soundcard (Babyface; RME, Germany). Electroencephalographic (EEG) responses were acquired using surface electrodes placed at the forehead (Fpz, ground), vertex (Cz, noninverting active), and an inverting reference electrode placed in the ear canal (ER3-26A gold foil tiptrodes). Pure-tones at 1.5 and 4 kHz were gated via Blackman window with duration of 1 ms. Stimuli were presented in alternating polarity monaurally at a rate of 11/s to an ER-3A insert earphone (Etymotic Research, Elk Grove, IL) connected to the soundcard. The stimulus sound-pressure level (SPL) was 110 dB peak-equivalent (pe) SPL. Calibration of the stimulus levels was done using a sound level meter (System 824 and SoundTrack LxT1; Larson Davis, Provo, UT) with the ER3 connected to the sound level meter via a 2 cc coupler (G.R.A.S. 60126, Denmark). High levels were chosen to maximize the number of ABR waves observed in participants, especially those with hearing loss (Ridley et al., 2018), and because larger effects of noise exposure history have been seen at such levels (Bramhall et al., 2017). Electrode impedances were ≤5 kΩ in all cases. The EEG signal was amplified (gain = 100 000), filtered (0.01 to 1.5 kHz; Opti-Amp 8001; Intelligent Hearing Systems, Miami, FL), filtered for line interference using a 60 Hz notch filter and directed to the computer via the soundcard for averaging. Responses were separated by even and odd recordings and stored in two buffers which were averaged for the final waveform (total averages = 1500 artifact-free responses). Artifact rejection was based on the peak absolute differences between the buffers and was set at a maximum of ±20 μV.

Two examiners independently identified peaks and troughs of ABR wave I. The software allowed for a resolution of 0.02 μV for amplitude and 0.02 ms for latency. The amplitude of wave I was calculated as the difference between the positive peak and the following trough. Latencies were used to clarify disagreements between examiners, but were not used for any other analyses. The processing delay of the soundcard was taken into account when analyzing the data for latency. Differences >0.02 ms, which occurred in 21 of the total 194 waveforms (∼11%), were resolved by a third examiner.

Stimuli for all behavioral measures were presented monaurally via ER-3A insert earphones (Etymotic Research, Elk Grove, IL) in a sound-treated room.

In addition to audiometric thresholds, a three-alternative forced choice (3AFC) adaptive procedure was used to determined pure-tone thresholds in quiet (TIQ) at 1.5 and 4 kHz. This method mitigates some biases known to occur in standard audiometry such as the interval bias: entrainment to the stimulus interval; and effects of age: older people are more likely to wait to respond until positive they heard the tone (Yost, 1978; Gelfand, 1982). For a 3AFC, three intervals were presented with only one interval containing the target stimulus; in this case two intervals were silent and one interval contained a pure-tone (AudioLab matlab; developed by Lopez-Poveda). The participant was required to indicate which interval contained the pure-tone and feedback was provided for each response. A 2-up, 1-down adaptive procedure was used to track the 71% point on the psychometric function, or threshold.

The initial stimulus level was 20 dB above the participant's audiometric threshold (in dB SPL) at the stimulus frequency, rounded up to the nearest 10 dB. The procedure had an initial step size of 5 dB which lasted for three reversals. The step size was then reduced to 2 dB for six reversals, for a total of nine reversals. The final six reversals were used to determine threshold (dB SPL). Participants completed one training run to familiarize them with the procedure and then two trials which were averaged to determine TIQ. Trials were included when the within-trial standard deviation was ≤5 dB; and two trials were only averaged if the thresholds were ≤6 dB apart. If the two trials did not meet both criteria, additional trials were completed until two trials met qualifications.

Word recognition scores for each participant were assessed in four listening conditions: (1) speech in quiet, (2) speech in the presence of noise, (3) speech that had been time-compressed by 45%, and (4) speech that had been time-compressed by 45% and a reverberation time of 0.3 s (Noffsinger et al., 1994). The stimuli were four 50-word lists spoken by a male talker (NU-6; Auditec, Inc., St. Louis, MO). The words were presented at 65 dB SPL for participants with a pure-tone average (PTA) at 1, 2, and 4 kHz of ≤35 dB SPL. For six participants with a PTA >35 dB SPL, words were presented at 30 dB SL rounded up to the nearest 5 dB for all conditions to ensure that stimuli were audible. Performance in each condition was measured as the percent words correct of the final 45 words in each list. Given the difficulty of some of the listening situations, the first five words were considered as training in order to familiarize participants with the condition.

Regression models were created to explain the variability in ABR wave I amplitude across the participant population. The explanatory variables chosen for the model were age (years), sex (M/F), TIQ at 4 kHz (dB SPL), EHF threshold category, and noise-exposure measure: LENS-Q score or high/low categorization based on impulse noise exposure. TIQ at 1.5 and 4 kHz were correlated (r = 0.7); therefore, to avoid overfitting and collinearity, TIQ at 4 kHz was included in the models. EHF was a binary variable determined for each frequency (11.2 and 16 kHz) based on a threshold cutoff of 20 dB HL. Participants were grouped into those with EHF thresholds <20 dB HL and those with thresholds ≥20 dB HL. This variable served as a proxy for high-frequency outer hair cell function. The numeric data (ABR wave I amplitude, age, TIQ, and LENS-Q score) were centered before inclusion into the model and assumptions of homoscedasticity and normality were met according to the Breusch-Pagan and Shapiro-Wilk statistics, respectively. Statistical analysis was completed in r (R Core Team, 2018) using ordinary least squares regression (olsrr) (Hebbali, 2018). One participant had outlying values of ABR wave I amplitude at both 1.5 and 4 kHz of greater than two standard deviations above the mean. Ordinary least squares analysis is biased by outliers; therefore, this participant was removed from the data before statistical analysis was performed leaving a total n = 96 participants. All possible models using the variables age, sex, TIQ, EHF, and LENS-Q score were compared to find the best-fit model. Best fit was determined by a combination of R², adjusted R², Akaike Information Criteria (AIC), and Sawa's Bayesian Information Criteria (SBIC).

Collinearity of predictor variables was reduced by removing redundant variables; however, some correlated variables such as age and thresholds or sex and noise exposure were purposefully left in the model to account for possible confounds to a direct relationship between noise exposure and ABR wave I amplitude. To accurately break down the variance explained by each predictor variable, an assessment of relative importance was implemented on the predictors included in the best-fitting model (relaimpo) (Grömping, 2015). Relative importance can be defined as the proportionate contribution each predictor makes to R², considering both a direct effect and its effect when combined with other variables in the regression equation (Johnson and LeBreton, 2004). This approach is based on sequential sums of squares but accounts for the dependence on ordering (which is biased by correlated predictors) by averaging over orderings.

Pearson partial correlations were also performed on the best fitting model to quantify the relationship between LENS-Q score and ABR wave I amplitude while controlling for other variables (ppcor) (Kim, 2015).

Separate regression models and Pearson partial correlations analyses (where applicable) were performed for 1.5 and 4 kHz.

In attempt to isolate a population that could be affected by noise-induced synaptopathy as opposed to, for example, presbycusis or outer hair cell loss, the same analysis was completed on a subset of the data that included only participants age 40 years and under and with clinically normal hearing (i.e., audiometric thresholds <20 dB HL). Characteristics of this subset are shown in Table II and mean and range of audiometric air conduction thresholds are displayed in Fig. 1(b).

A comparison of the two measures of noise exposure was made by performing analysis of variance (ANOVA) on the LENS-Q scores of the participants in the high and low noise categories.

Regression models were also created to explain variability in performance on speech recognition tasks in difficult listening conditions: background noise, time compression, and time compression with reverberation. A model was not generated for word recognition in quiet, due to performance clustering at ceiling regardless of noise exposure, age, and TIQ. The explanatory variables were age, sex, TIQ, EHF, LENS-Q score, and ABR wave I amplitude. Normality was achieved by centering the data. Relative importance and Pearson partial correlations were also performed for the variables LENS-Q and ABR wave I amplitude while controlling for other variables.

An example ABR waveform at 4 kHz for an individual participant is displayed in Fig. 2. Locations of the peak and trough of wave I are indicated in the waveform. The peak of wave V is also indicated. Wave I amplitude was calculated as the difference between the peak and the trough. For this example, wave I amplitude was 0.57 μV.

The distributions of ABR wave I amplitudes (n = 97) for the two frequencies can be found in Fig. 3. The participant whose data were considered outliers (i.e., amplitude greater than two standard deviations from the mean) is indicated by plus signs. The total number of ABR wave I amplitudes that were modeled for each frequency was n = 96, after removing the outlier. TIQ at 1.5 and 4 kHz were highly correlated (r = 0.61) and contributed redundant information to the models of ABR wave I amplitude and speech performance; therefore, to avoid overfitting, only TIQ at 4 kHz was used. ABR wave I amplitude at 1.5 and 4 kHz were also correlated (r = 0.62); therefore, to avoid overfitting and collinearity, only ABR wave I amplitude at 4 kHz was included in the models of speech performance. Finally, only EHF at 11.2 kHz was included in the model, as the 16 kHz binary variable did not add information to the model.

Relationships between noise exposure history and ABR wave I amplitude were modeled including confounding variables: age, sex, TIQ, and EHF. The best-fitting model for ABR wave I amplitude was dependent on stimulus frequency and on the number of parameters. The dependence of model performance on the addition of each predictor variable is shown in Fig. 4(a) and Table III. Only those predictors which added to the explained variance and did not result in overfitting were included in the best-fit model.

For the model of ABR wave I amplitude at 1.5 kHz, the best-fit model included four of the five explanatory variables: age, sex, TIQ at 4 kHz, and EHF at 11.2 kHz. These four variables combined explained 43% of the variance in ABR wave I amplitude (Table III). The variables which accounted for the majority of the variance were age and EHF at 11.2 kHz. LENS-Q score did not explain any additional variance and inclusion in the model resulted in overfitting as evidenced by a reduction in adjusted R² and increase in AIC and SBIC. Coefficients of the predictors are displayed in Table IV. EHF at 11.2 kHz was the only significant predictor that showed a negative relationship with ABR wave I amplitude. Age, sex, and TIQ at 4 kHz also trended negatively with ABR wave I amplitude though these predictors were not significant.

The ABR wave I amplitude at 4 kHz was best modeled using four of the variables: age, sex, TIQ at 4 kHz, and EHF at 11.2 kHz. These variables combined accounted for 44% of the variance of ABR wave I amplitude at 4 kHz [Fig. 4(a); Table III]. Age and EHF at 11.2 kHz accounted for the majority of the variance. The addition of LENS-Q score did not significantly improve the model and resulted in overfitting as evidenced by a reduction in adjusted R² and increase in AIC and SBIC. Coefficients of the predictors are displayed in Table IV. Age and EHF at 11.2 kHz showed a significant negative relationship with ABR wave I amplitude. TIQ at 4 kHz and males showed a negative (but insignificant) trend with ABR wave I amplitude.

The best-fit model for ABR wave I amplitude in a younger subset of the participants with normal hearing was different than that of the larger population. Age and TIQ were still included in the model to account for the age range (20–40 years) and the range of TIQ at 4 kHz which spanned almost 30 dB; however, these variables were expected to contribute less to a model of ABR wave I amplitude in this population. Statistically significant models for 1.5 and 4 kHz were not achieved, and the best-fit models accounted for less of the variance in ABR wave I amplitude than the large cohort.

The best-fit model at 1.5 kHz included four of the five variables: age, sex, EHF at 11.2 kHz, and LENS-Q score. These variables combined accounted for 25% of the variance in ABR wave I amplitude but was not statistically significant [Fig. 4(b); Table III]. The addition of TIQ at 4 kHz did not improve the model. Coefficients of the model are displayed in Table IV. Because sex was a significant factor, partial correlations were performed for LENS-Q in males (n = 9) and females (n = 27) separately (Table V). For females, a positive correlation of LENS-Q score with ABR wave I amplitude approached significance. The correlation for males was not significant.

At 4 kHz, the best-fit model included age, EHF at 11.2 kHz, and LENS-Q score. These variables accounted for 15% of the variance in ABR wave I amplitude but the model was not statistically significant [Fig. 4(b); Table III]. Coefficients of the model parameters can be found in Table IV. Since sex was not a significant factor in ABR wave I amplitude at 4 kHz, partial correlation analysis was performed on male and female groups together (Table V). LENS-Q score was positively correlated with ABR wave I amplitude and approached significance (p = 0.07).

Figure 5 plots the observed ABR wave I amplitude against predicted wave I amplitude for the best fit model. Separate model predictions are shown for the model based on all participants, and the model based on the younger participants with normal hearing, for 1.5 and 4 kHz. A hypothetical perfect linear relationship (unity slope) is indicted by the dashed line, while the actual model fit to the data is plotted as a solid line. Overall, the slope of the predicted amplitudes was shallow.

The two noise exposure measures were LENS-Q score and a categorical (yes/no) assignment based on asking the participant whether they had ever been exposed to an impulse noise without hearing protection. The distributions of LENS-Q score for the high and low noise groups are shown in Fig. 6 using a stacked bar chart. While there is some overlap in LENS-Q score between the groups, there are two distinct distributions of scores that can be explained by exposure to impulse noise alone. An ANOVA resulted in a significant difference between the two groups (F^1,96 = 114.7, p < 0.0001).

Each of the regression models above was repeated, replacing LENS-Q score with the categorical variable of noise exposure. In the large cohort, the high/low noise category did not improve model performance and in fact was nearly identical to LENS-Q score in its ability to explain ABR wave I amplitude. This was true for both frequencies. In the young, normal hearing subset, the substitution of a categorical High/Low Noise variable for LENS-Q score resulted in poorer models with a 5% decrease in the variance explained for both frequencies.

A model of performance for word recognition in background noise showed no relationship with noise exposure or ABR wave I amplitude at either frequency (Table VII). The only significant predictor was TIQ at 4 kHz which explained 26% of the variance in word recognition (Table VI).

The best-fitting model of performance on time compression explained 38.7% of the variance and included sex, TIQ at 4 kHz, and ABR wave I amplitude at 4 kHz [Fig. 7(b); Table VI]. Coefficients of the model are presented in Table VII. Partial correlations showed ABR wave I amplitude was positively correlated with performance, but this only approached significance in females (Table VIII).

The best fit model for time-compression with reverberation included only age and TIQ at 4 kHz, and explained 45% of the variance in performance [Fig. 7(c); Table VI]. Increased age and TIQ at 4 kHz negatively influenced the model of performance (Table VII).

Figure 7 shows predictions for models describing speech in background noise [Fig. 7(a)], time compression speech [Fig. 7(b)], and time compression with reverberation [Fig. 7(c)] based on all participants. No model was generated for word recognition in quiet, due to performance clustered at ceiling regardless of noise exposure, age, and thresholds. In each plot, the dashed line indicates a perfect linear relationship and the solid line indicates the actual relationship between predicted and observed word recognition (% correct). The effect of ABR wave I amplitude for both 1.5 and 4 kHz stimuli was interchangeable in the models, therefore only ABR wave I at 4 kHz was included.

Similar to models of ABR wave I amplitude, the models for speech recognition in the younger participants with normal thresholds were relatively poor, with the exception of a significant model of time-compressed word recognition. Neither sex nor EHF were included in the best-fit model of performance on any speech task.

The model for speech recognition in noise for the young participants with normal hearing fit poorly (Table VII). The only predictors which could explain variance in performance were age and ABR wave I amplitude but were not significant (Table VI). The predicted performances are plotted against observed performance in Fig. 7(d).

The best model for performance on time compression included LENS-Q score, TIQ at 4 kHz, and ABR wave I amplitude (Table VII) and combined, predicted 24% of the variance in performance. The predicted performance on time compression is plotted against observed performance in Fig. 7(e). Partial correlation of LENS-Q score showed a significant positive relationship with time-compressed speech (Table VIII).

Performance on word recognition of time compression with reverberation was best modeled by TIQ and ABR wave I amplitude, but was a poor fit for this subpopulation (Table VII) and explained only 14% of the variance. Figure 7(f) compares the predicted performance and observed performance. Partial correlation of ABR wave I amplitude showed a positive relationship with performance which approached significance (Table VIII).

The primary goal of this study was to assess the relationship between self-reported noise exposure and ABR wave I amplitude. A model of ABR wave I amplitude was created using age, sex, pure-tone thresholds, and noise exposure history as predictor variables. Relative importance and partial correlations were used to control for collinearity of the predictor variables and quantify the variance explained and strength of their relationship with ABR wave I amplitude. Two ABR stimulus frequencies (1.5 and 4 kHz) in two sample populations were modeled: one which included a large cohort of a broad range of ages and audiometric thresholds, and a younger subset of participants with clinically normal audiometric thresholds. In the large cohort, age, sex, TIQ at 4 kHz and EHF at 11.2 kHz were important factors in predicting ABR wave I amplitude, but noise exposure was not. In the younger participants with clinically normal hearing, no combination of these variables was able to explain much of the variance in ABR wave I amplitude; however, a positive relationship between noise exposure and ABR wave I amplitude in females approached significance. The extended high-frequency threshold variable served as a proxy for outer hair cell loss. While EHF are not typically tested in the clinic, we found that these thresholds explained a significant amount of the variance in ABR wave I amplitude. Relationship between EHF thresholds, ABR amplitudes, and speech perception have been studied more recently and confirm the importance of high-frequency outer hair cell loss on our current measures of synaptopathy (Liberman et al., 2016; Grose et al., 2017; Prendergast et al., 2017; Yeend et al., 2017). These findings oppose the idea that noise-induced synaptopathy has a negative effect on ABR wave I amplitude when controlling for age and hearing sensitivity, but could indicate that outer hair cell loss at high frequencies may be driving the reduction in ABR wave I amplitude. New reports encourage measuring EHF in all studies of synaptopathy and hidden hearing loss (Bramhall et al., 2019; Guest et al., 2018a; Le Prell, 2019).

Age was one of two most important variables explaining the variability in ABR wave I amplitude. Age has been shown to substantially reduce ABR wave amplitudes (Konrad-Martin et al., 2012), and our findings agree. In humans, the aging process independent of noise exposure results in synaptopathy (Makary et al., 2011; Viana et al., 2015); and in mice, noise exposure and aging interact to exacerbate synaptopathy (Fernandez et al., 2015; Kujawa and Liberman, 2015) and audiometric hearing loss in old age (Miller et al., 1998). Most studies attempting to link noise exposure and synaptopathy in humans try to avoid contamination of age-related synaptopathy by recruiting young subjects. In the present study, a subset of younger adults participants (40 years and under) was examined separately with the intention of reducing the effect of age, but age still explained variance in ABR wave I amplitude. A study by Guest et al. (2018b) age-matched participants 40 years and under between a group of people with impaired speech perception and normal hearing thresholds and a control group with non-impaired speech perception and normal thresholds, with a similar sample size (n = 32) to the young, normal hearing group in the present study. They found no relationships between impaired speech perception and ABR wave I amplitude or noise exposure. Conversely, we found trends between noise exposure, ABR wave I amplitude, and complex speech perception, but which only approached statistical significance. Guest et al. (2018b) put forth the possibility that synaptopathy unaccompanied by audiometric hearing loss is rare in young people. It is highly possible that noise-induced synaptopathy is more prevalent in older adults, however, so is age-related synaptopathy and audiometric hearing loss—factors which confound attempts to determine a direct relationship between noise and synaptopathy. The future of preventative intervention for hearing loss requires distinction between age-related hearing loss and hearing loss due to environmental factors such as noise exposure (Dobie, 1992; Dobie and Humes, 2017). Diagnosing noise-induced synaptopathy in patients middle-aged or older, or patients with hearing loss may prove difficult, as these both drive reduced ABR amplitudes, which could mask physiologic effects of noise-related synaptopathy.

Another goal of the study was to compare two measures of noise exposure history. Both involved questions and subjective answers, because gathering complete information on an individual's lifetime exposure to noise is impossible. There is no gold standard for quantifying this information. In the search to link noise exposure with synaptopathy in humans, a number of surveys/questionnaires have been created to estimate exposure. We have intuitively judged these surveys based on how comprehensively they capture the variety of potential noise sources to which a given individual may be exposed throughout the lifespan, which consequently requires these surveys to be time-consuming. For a review of various self-reported noise exposure measures and results of those studies, see Guest et al. (2018b).

In the present study, the LENS-Q took up to an hour to complete for those participants with extensive noise history. Is this breadth of noise sources necessary to estimate exposure? Compare the impact of exposure to riding in a motor boat with shooting a revolver, assuming no hearing protection was used in either activity. The Berger et al. (2015) database holds recorded values of 90 dBA for the motor boat and 161 dB SPL for the revolver. The weighting procedure assigns corresponding weights of 10 for the motorboat and 133 159 726 for the revolver. These weighted values are then multiplied by frequency and duration of exposure. It is easy to see how shooting a revolver one time has a much greater impact on LENS-Q score than regular weekend motor boating. This begs the question of how well we can estimate noise exposure from just a couple of noise sources. In this study, we compared a comprehensive survey to a single question: whether the participant had ever been exposed to an impulse noise, such as an explosion or firearm discharge, without hearing protection. The results were interesting. In a direct comparison via ANOVA, LENS-Q scores of two groups: those who answered “yes” (high noise) and those who answered “no” (low noise), were significantly different. In the distribution (Fig. 6) we see two distinct groups. Substitution of the categorical high/low noise groups made no difference to the models of ABR wave I amplitude in the large cohort where participants of all ages and hearing thresholds were included. In the younger participant group, where LENS-Q score was a predictor of ABR wave I amplitude, substitution of score for the high/low noise group resulted in a poorer model. This suggests that the comprehensive nature of the survey does indeed capture information that is useful and is perhaps a more accurate estimation of noise exposure; however, depending on the application, the additional benefit might not be worth the additional time required to administer the survey.

Interestingly, all partial correlations including LENS-Q score showed a positive correlation with ABR wave I amplitude, meaning more noise exposure predicted larger wave I amplitudes. We are unsure what to make of this finding, except to interpret with caution due to the sample size of the younger clinically normal hearing group and insignificance of the positive trend. This contradicts the negative correlation found by Bramhall et al. (2017) which fits with the theory that noise exposure causes synaptopathy, reducing the ABR wave I amplitude. The participants in the Bramhall et al. (2017) study had much greater exposure (LENS-Q scores >16) to noise than the participants in the present study. This could explain the lack of (or weak positive) relationship in the younger participants with normal thresholds.

Another consideration is the potential prevalence of hidden hearing loss, i.e., clinically undetectable damage from noise exposure. In theory, hidden hearing loss may manifest as a suprathreshold hearing deficit like difficulty understanding speech in noise. Studies that have explored the prevalence of these complaints suggest the prevalence may only be 5%–8% (Stephens et al., 2003; Hind et al., 2011; Spehar and Lichtenhan, 2018) and that the etiology of this complaint may not necessarily be due to noise exposure in every case (see Pienkowski, 2017). Therefore, it is not surprising that we found no relationship between self-reported noise exposure history and speech perception in the large cohort. Even in the young normal-hearing group, the sample may be too small to find evidence of noise-induced synaptopathy.

Our measure of speech perception included word recognition scores for three difficult listening conditions: background noise, time compression, and time compression with reverberation. These were chosen to assess the hypothesis that the primary clinical complaint of patients with hidden hearing loss is difficulty understanding speech in noise. Interestingly, TIQ at 4 kHz was the only significant predictor of word recognition performance in noise in the large cohort. In addition to background noise, we included time compression and reverberation as behavioral measures of temporal processing have been hypothesized to be affected by synaptopathy (e.g., Bharadwaj et al., 2015). In females, a positive correlation between ABR wave I amplitude and recognition of time compressed words approached significance, suggesting a relationship between synaptopathy and perception of speech involving intense temporal processing. This was also a trend in the younger participants with normal hearing, indicative of perceptual deficits accompanying synaptopathy but not necessarily noise-induced synaptopathy. In the young group, noise exposure was a significant predictor of performance, but with higher noise exposure relating to better performance on time compression (Table VIII). Again, we are unsure of the mechanisms underlying this finding, other than the small number of younger participants depreciate any conclusions drawn from this sample. There is mounting evidence of additional pathologies that could underlie symptoms of hidden hearing loss [see Kohrman et al. (2019)]. Overall, the results lacked strong evidence for a relationship between subjective noise exposure history and performance on measures of speech in difficult listening conditions. This concurs with other studies in similar populations (Yeend et al., 2017; Guest et al., 2018b; Valderrama et al., 2018).

One reason for the insignificant findings in the younger adults with clinically normal hearing is the small number of participants under 40 years. Other studies which had similar sample sizes for this group recruit the extremes of the noise-exposure spectrum. Our sample was recruited randomly from the general population of Omaha, NE. Because we did not specifically recruit based on participant demographics, there was also a sex imbalance across the noise exposure groups. Although sex was included in the models, sex is so highly correlated with noise exposure in this cohort (almost all low noise participants are female, almost all high noise participants are male) that adjusting for sex may skew the noise exposure effect. More than the sex imbalance, the variance of ABR wave I amplitudes in this group are small, requiring a larger sample to tease apart differences based on factors such as noise exposure. Future studies should aim for larger samples in order to include confounding variables such as age, sex, and EHF thresholds into a model of ABR wave I amplitude. Another limitation of the model is the lack of interaction effects. We did not include interactions in order to produce independent variances explained by each predictor. Bramhall et al. (2015) found an interaction effect of ABR wave I amplitude and pure tone thresholds on speech-in-noise performance.

Limitations were also present in the tests chosen for this study. TIQ was only collected at two frequencies. Although audiometric thresholds through 8 kHz were obtained in each participant, a 3AFC procedure was deliberately chosen to reduce age-related bias and improve reliability. Due to the longer test time, we sacrificed additional test frequencies. 1.5 kHz was chosen to represent low-frequency hearing which codes for many speech sounds. 4 kHz was chosen as the common locus of increased thresholds due to noise-exposure. Thresholds between 4 and 11.2 kHz were not included in this study, but could have affected the ABR wave I amplitudes at 4 kHz (Don and Eggermont, 1978). Using a single high frequency threshold (4 kHz) in the model is over-simplistic and may fail to account for outer hair cell dysfunction in many participants. Distortion-product otoacoustic emissions (DPOAE) were also obtained in these participants, but only at 1.5 and 4 kHz (supplementary document¹). Otoacoustic emissions would be an alternative method to account for differences in outer hair cell function between participants, as these have been shown to be more sensitive to noise-exposure and ototoxicity-related changes than pure tone thresholds (Engdahl and Kemp, 1996; Marshall et al., 2009). Replacement of TIQ at 4 kHz with DPOAE level (dB SPL) at 4 kHz did not add to the explained variance in any model and virtually took the place of TIQ and EHF in the order of importance and amount of variance explained (supplementary table¹).

The choice of a questionnaire to measure noise exposure was an unavoidable limitation. No one noise exposure questionnaire has proven to be more valid and sensitive than all other questionnaires. One issue with the LENS-Q, which is currently being addressed, is the mismatch in intensity values for steady-state (dBA) and impulse (dB SPL) noise sources.

The lack of evidence for a substantial relationship between noise exposure and ABR wave I amplitude or speech perception may have a number of explanations. One could be that subclinical damage from exposure to noise may only be one of many factors contributing to small ABR amplitudes and poor performance on tasks of speech recognition. Another could be the measures of noise exposure and hearing used in this study are not sensitive to true noise exposure or the consequential damage. Future studies are needed to account for confounding variables to these measures and to improve on the measures themselves.

1. Age and outer hair cell function (especially at high frequencies) are confounding variables that need to be addressed in the search to link noise exposure with synaptopathy via ABR wave I amplitude.

2. Self-reported noise-exposure history is not related to ABR wave I amplitude in a cohort with a broad range of ages and hearing thresholds.

3. Self-reported noise-exposure history is not related to performance on word recognition in difficult listening situations in a cohort with a broad range of ages and hearing thresholds.

4. A single question based on exposure to impulse noise (yes/no) can predict scores on a comprehensive noise-exposure survey.

This research was funded by NIH Grant Nos. 5R01DC016348-02 and T32DC000013.