The participant pool for the current study was drawn from a larger multi-visit study designed to examine MEMC under a variety of experimental conditions. During the initial visit of the larger study, participants completed questionnaires related to hearing health and noise exposure, underwent standard clinical audiological procedures including otoscopy, audiometric testing, and middle ear assessment, participated in a brief screening of cranial nerves V and VII and completed a short pupillary conditioning experiment. The main goal of this initial visit was to determine if participants met the eligibility criteria for the remaining part of the larger study.

The data used in the current study included results from all participants who completed this first visit and for whom middle ear assessments were available. No effort was made to exclude participants based on audiometric, otoscopic, cranial nerve screening or questionnaire results. A total of 285 participants ranging in age from 18–68 years were used for this study. The median age of the participants in the study was 21 years old (interquartile range: 20 to 23 years). The sample included 199 (70%) females and 86 males. Both the young age of participants and the large number of female participants likely resulted from the location participants were recruited. The fields of study at the College of Health and Human Services (CHHS) at Western Michigan University (WMU) are primarily dominated by young adult females.

Participants were recruited using informational flyers distributed throughout the CHHS building at WMU and surrounding areas in Kalamazoo, MI and through word-of-mouth from previous participants. This study was reviewed and approved by the Human Subjects Institutional Review Board of WMU and the Human Research Protection Office of the U.S. Army Medical Research and Materiel Command.

During a participant’s first visit for the larger study, the data for this retrospective study was obtained. Participants completed a questionnaire which included questions on history of tinnitus, noise exposure (occupational noise, leisure noise, and firearm use), facial nerve disorders, such as Bell’s palsy, and concussions. The questionnaire was followed by a video-otoscopic examination of the external ear and pure tone threshold audiometry. Participants were given an opportunity for a short break followed by a cranial nerves screening. The final two sets of measures taken during this visit were middle ear assessment and a test of the participant’s pupillary response to light and sound.

The data collection procedure was controlled using a custom MATLAB script that controlled the data acquisition process (i.e. otoscopy, cranial nerve testing and pupil conditioning) and automatically launched proprietary software (i.e. audiometry and middle ear assessment) during protocol execution. The visit that included the procedures used in this paper took approximately one hour after the participant provided informed consent to participate.

The audiometric data collected for each participant were classified into audiometric configurations described in Ciletti & Flamme (2008), see Table 1. Each configuration is labelled with a letter followed by a number. The letter represents the thresholds in the low frequencies (0.5 to 2 kHz), and the number represents the threshold of the worst frequency above 2 kHz (Ciletti & Flamme, 2008). Of the 285 participants included in this study, 261 (92%) participants had audiometric thresholds that were best described by the A11 configuration. The thresholds in the A11 configuration represented the group with the best hearing sensitivity in the NHANES dataset used by Ciletti & Flamme (2008). This configuration was characterized by bilateral symmetry and mean pure tone thresholds of 10 dB HL or less between the stimulus frequencies of 0.5 to 8 kHz, with the exception of a mean threshold of 11 dB HL at 6 kHz. Following A11, A19 and A22 were the next most frequent configurations and each accounted for approximately 2 % of the sample (i.e., A11, A19, and A22 accounted for approximately 96 % of the total sample). These configurations are very similar, but were identified separately for men and women in the Ciletti & Flamme study.

Analyses were performed on all reflex traces collected. Two methods of reflex detection were developed and implemented using custom written MATLAB scripts. These methods are similar, but not identical to the Frequentist and Bayesian/Kalman methods described in Flamme et al (2017). The first method, referred to as the conventional method, identifies a reflex as present if the maximum change in the reflex trace is greater than 0.02 ml, based on the underlying assumption that this much change would be infrequent in the absence of an MEMC. One disadvantage of the conventional method is that the morphology of the reflex is not taken into consideration. The second method, referred to as the correlational method, identifies a reflex as present if the correlation between that trace and one of a set of reflex prototypes exceeds a correlation coefficient “cut-point”. This procedure required a number of analysis steps.

The prototypical shapes were selected based on the presence of a stimulus-linked change in the trace. All traces collected for each frequency/laterality combination were viewed on the same figure, showing reflex growth and morphology patterns. Similar stimulus-linked changes across sampled traces were considered candidates for prototypes, and the 11 original prototypes were selected from emblematic examples in the sample and to maximize the differences among the prototypes. Further analyses of the prototypes determined that four prototypes were highly correlated (e.g., > 0.93) to at least one of the remaining prototypes, and therefore were considered redundant. The seven remaining prototypes (Figure 1) were used for the remaining analyses.

In order to determine correlation cut-point, 406 reflex traces were randomly selected from the dataset. Four authors (KM, GF, ST, KD) independently viewed each trace and made binary judgments on the presence/absence of a stimulus-linked change in impedance (i.e., an AR). Presence of a reflex was judged only on the morphology of the reflex. The time axis scaling remained so that the judge could confirm that the reflex was time-linked to the elicitor.

Traces in which all four judges agreed that a reflex was present were classified as true reflexes. All other traces, including those where only 2 or 3 of the judges agreed, were labeled as no reflex present. Interrater agreement, as represented by Cohen’s kappa statistic was (κ=.826), which is considered excellent by conventional interpretation guidelines (Lindis & Koch, 1977). The binary indicators for all 406 traces were then used to complete a Receiver Operating Characteristic (ROC) curve, which produced the sensitivity and specificity of this identification method at various correlation cut-points. The area under the ROC curve for this analysis was 0.96, and sensitivity and specificity was maximized at a cut-point of 0.8617.

Proportions of AR observations and 90 % confidence intervals and analyses of NHANES data were calculated using the Stata software package (StataCorp, College Station, TX). The 90 % confidence interval was used because the lower bound of the 90 % confidence interval represents the point above which 95 % of observations would be expected to fall under infinite repeated random sampling of a population. Thus, a lower bound of the 90 % confidence interval that is greater than a proportion of 0.95 would indicate greater than 95 % certainty that 95 % of the sample would exhibit an AR. Analyses of NHANES were conducted using complex sample variables (e.g., sample weighting, clusters, and population sampling units) in accordance with guidance from the CDC/National Center for Health Statistics (CDC/NCHS, 2017b).

The results for the correlational detection method are shown in Table 2. Each ear was analyzed separately, for each elicitor frequency and each laterality. The bilateral category represents the proportion of participants with present AR at the elicitor frequency in both ears. An individual is less likely to have an AR in both ears than in only one, and therefore, the proportions of AR are lower for the bilateral category than for left or right ear separately. Each category also includes “1 or 2 kHz” in the ipsilateral condition for comparison with results from the NHANES study (Flamme et al, 2017), analyzed using the same criteria. We have calculated the proportion of participants with a reflex present at either 1 or 2 kHz for the left ear, right ear, and in both ears for the ipsilateral condition only. This proportion will be higher than either 1 or 2 kHz on its own. Similar results for the conventional detection method are shown in Table 3. As discussed previously, an AR is considered pervasive when the lower bounds of the 90% confidence interval exceed 0.95. We were unable to find a category, using either detection method, that met this criterion. It is, however, important to notice two trends in these results. First, the proportion of AR was lower for the contralateral AR than for the ipsilateral AR at every elicitor frequency. This trend was found with both detection methods. The difference (as a percentage) between ipsilateral and contralateral AR ranges from 6.7% to 20.4% (mean: 11.9%, SD: 3.87%).

Another important trend in these results is that the proportions of AR at 0.5 and 1 kHz were higher than those at 2 and 4 kHz. To quantify this difference, we found the average of the proportions at 0.5 and 1 kHz (referred to as 5–1 average) and at 2 and 4 kHz (2–4 average) for each condition. The 2–4 average was subtracted from the 5–1 average to calculate the difference (presented as a percentage). The mean differences were as follows: 4.3% (SD:0.43%) for ipsilateral, correlational method; 5.55% (SD:1.90%) for contralateral, correlational method; 8.28% (SD:1.81%) for ipsilateral, conventional method; 2.55% (SD:0.17%) for contralateral, conventional method.

To compare the results from the current study to the results presented in Flamme et al (2017), four participant groups were further analyzed to determine the proportion of individuals in each group with an AR at 1 or 2 kHz. The four participant groups, based on age and audiometric results, included: any age and any audiogram configuration (all participants), any age and an audiogram best described by the A11 configuration (Ciletti & Flamme, 2008), individuals 18–30 years old with any audiometric configuration, and individuals 18–30 years old with an audiogram best described by the A11 configuration. The numbers of participants in each group, as well as the proportions of individuals in each group with an AR at 1 or 2 kHz in both ears and the 90% confidence intervals for both detection methods are shown in Table 4. The results represented in Table 4 show that the AR is very common in these participant subgroups. The proportion of AR was above 90% in all participant groups for both detection methods. However, the lower bounds of the 90% confidence intervals did not exceed 0.95 for any of the groups using either detection method. Even in the fourth group, which includes individuals who are young (18–30 years) and have very good hearing (hearing levels best described by the A11 configuration), the AR rates did not meet the criterion for pervasiveness. We confirmed the results of Flamme et al (2017) which concluded that the AR was very common, but was not pervasive.

The mean thresholds of individuals with audiometric configurations best described by A11 (Ciletti & Flamme, 2008) for the current study were typically better than the thresholds used originally to define the A11 audiometric configuration. For example, the threshold at 1 kHz as described by A11 is 4 dB HL, but the mean threshold at 1 kHz for the current study was 0 dB HL for both the left and right ears. The mean thresholds for the current study and the A11 configuration are shown in Figure 2. The mean thresholds for individuals with thresholds best described by the A11 audiometric configuration were better than the A11 audiometric configuration at all frequencies (0.5 kHz to 8 kHz). In order to provide a more precise comparison, the NHANES data used in Flamme et al (2017) were re-analyzed using only participants whose thresholds fell within the interquartile range (IQR) of individuals best described by A11 in the current study. The same analysis and reflex detection methods described in Flamme et al (2017) were used.

The proportion of bilateral AR reported in Flamme et al (2017) for the youngest population (aged 18–30 years) with the best hearing (audiograms best described by A11) using either detection method from that study was 0.91 with a 90% confidence interval from 0.89 to 0.92. Using only NHANES participants with audiograms within the IQR of the current study, the proportion increased to 0.94 with a 90% confidence interval from 0.91 to 0.98. These highly screened results from NHANES are most comparable to participant group 4 in Table 4, where the proportions in the current study matched the highly screened NHANES group within one percent, and continue to demonstrate lack of AR pervasiveness.

The current study was undertaken to determine whether the proportions of acoustic reflexes observed in a laboratory study using a diagnostic middle ear analyzer with a friction-fit probe were similar to the nationally-representative (NHANES) results obtained using a screening device and handheld probe. The results obtained in the current study confirm the nationally-representative results of Flamme et al (2017) and reaffirm the conclusion that acoustic reflexes are common, but not pervasive. In spite of the large sample of young participants with excellent hearing, we remain unable to find a subpopulation in which AR are pervasive.

It is possible that cases of absent acoustic reflexes could be associated with reductions to the numbers of synapses the inner hair cell and auditory nerve fibers, which can be described as cochlear neuropathy. Recent work with mouse models (Valero et al. 2016) suggests that neuropathy induced by noise exposure might reduce or degenerate synapses to auditory nerve fibers and thus disturbing acoustic reflexes. It is not clear whether this phenomenon was a large contributor to the results of this study or if cochlear neuropathy might explain differences in the probabilities of responses across frequency and stimulus-probe configurations. However, the results of Valero et al. (2016) and other related work are relevant to the issue of reliance on AR by DRC for impulsive noise.

The proportions of responses found in the current study were considerably higher than would be observed in the general population. Prior studies (Flamme et al, 2017; Hall, 1982; Hodges & Ruth, 1987) have shown that the probability of reflex detection declines with a decrease in hearing sensitivity, male gender, and increasing age. It is also possible that the proportion of individuals exhibiting AR would have been higher if broadband elicitors (e.g., white or pink noise) had been used. We elected not to evaluate broadband noise stimuli for this component of the study because broadband elicitors were included in other aspects of the overall project and because the main purpose of these analyses was to make a comparison to the tonal elicitor results described in Flamme et al. (2017).

The median age of participants was 21 years old (interquartile range: 20 to 23 years) and 70 % of the individuals were female, which both lead to what is likely an overestimate of the proportion of AR seen in the general population. If the gender distribution of the participants had been more equal and/or the average age of participants had been higher, it is likely that the proportions of AR observed would decrease (Flamme et al, 2017; Hall, 1982).

It is important to note that the results of the current study apply to stimulus durations that are considerably longer and at lower levels than the impulsive sounds addressed by a DRC for impulsive noise. The observation of an AR for a long-duration signal does not necessarily imply that short-duration signals like gunshots will also elicit an AR among people with robust AR for long-duration signals. In this study, presentations were not made at higher levels because the levels tested here were similar to those used in the prior study and to ensure the safety of the volunteer research participants (Hunter et al., 1999; Schairer, et al., 2007), who were exposed to high-level signals in other study procedures. It is also important to note that responses at higher levels could be more likely to elicit middle ear muscle contractions through alternate neural pathways (e.g., eye blinks), which are very likely to produce middle ear muscle contractions (Tasko, et al., 2017).

The current study and Flamme et al (2017) examined only the AR at 1 and 2 kHz in relation to the pervasiveness of the AR. This was a limitation of the NHANES data set, as those were the only two frequencies obtained. However, as our results showed, AR at 0.5 and 1 kHz were more common than AR at 2 and 4 kHz. Therefore, a more comprehensive study including all four test frequencies could lead to a different decision about inclusion in DRC and may be worthwhile to explore if the investigators in that study manage the capitalization on chance across multiple observations. One complication to such a study, however, is that the inclusion of AR in a DRC would involve the generalization of results obtained with non-impulsive sounds to impulsive sounds. In the current study, we judged a person as having an AR if reflexes were observed in at least one of the two elicitor frequencies under consideration, which represents as little as 50 % of the elicitors presented. In such a case, the response observed to one elicitor stimulus was not shown to generalize to a similar stimulus of equal duration. The finding of AR for only one elicitor stimulus confirms the integrity of some portions of the reflex arc, but it is not sufficient evidence to ensure that this arc will be activated by a substantially different stimulus such as an impulse.

The finding that the proportions of AR were lower for the contralateral condition than for the ipsilateral condition suggests that bilateral contractions are not sufficiently dependable to justify their inclusion in DRC for impulsive noise, even among individuals with very good hearing sensitivity. This finding is perhaps an outcropping of the impressions of Mukerji et al (2010), who expressed the position that the neural pathways triggering the MEMC are more complex than suggested by conventional wisdom. Future research surveying the conditions leading to MEMC could augment neurophysiological studies of the anatomy and physiology of the nuclei involved in the MEMC.