In univariate analyses, WAI in one frequency band was compared with the magnitude of the ABG at the corresponding frequency or at one other frequency (Piskorski et al. 1999; Keefe et al. 2012), or the general determination of CHL in an ear based on the presence of ABGs (Beers et al. 2010; Keefe et al. 2012; Prieve et al., 2013). In all studies, tympanometry was measured using a 226 Hz probe tone for children and probe tones of 226, 678, and 1000 Hz in infants. In infants, tympanometry using probe tones of 678 and 1000 Hz and |R|2 in the 1/3 octave bands from 800 to 2500 Hz and 6300 Hz identified CHL well, with positive likelihood ratios (LR+s) ranging from 8.1 to 38 and LR− ranging from 0.068 to 38. No one measure identified CHL significantly better than others, based on overlapping 95% confidence intervals of Cohen’s d.
In the studies examining CHL in children (Piskorski et al. 1999; Beers et al. 2010; Keefe et al. 2012), receiver operating characteristic (ROC) analysis was performed. The two measures of interest from ROC curves are the area underneath the curve (AUC), and point of symmetry on the ROC curve, which signifies the greatest equivalence of sensitivity and specificity. The standard for the ROC curves was determination of CHL for the ear or ABG at specific frequencies. In addition, SA and TW measured with tympanometry using a 226 Hz probe tone were compared with those constructed using WAI. Measures of |R|2 detected CHL well, with AUC ROC curves that ranged from 0.84 to 0.97 (Beers et al. 2010). Figure 3A, from Beers et al. (2010), illustrates ROC curves for three 1/3 octave frequencies (top panel), and Figure 3B illustrates distributions of |R|2 in the 1250 Hz band, along with the calculated sensitivity and specificity using a criterion of R 2 >71.7 for identification of CHL (bottom panel). In two other studies, ABG was the measure used to determine CHL at each audiometric frequency. In these studies, whether WAI or tympanometry had higher ROC curve areas depended on the frequency bands used for the comparison. Keefe et al. (2012) reported that an ABG of at least 10 dB at any frequency was identified similarly by tympanometry and A. For identification of ABGs of 20 and 30 dB, ROC areas for SA and TW at 226 Hz were higher than A in the 250 Hz frequency band (Keefe et al. 2012). However, ROC areas for identification of ABGs at 1000, 2000, and 4000 Hz were higher for the univariate A centered at the same frequency than for SA and TW using a 226 Hz probe tone. When A was compared with the ABG at any frequency as the standard, the area underneath the ROC curve was approximately 0.95. In a different study, ROC curve areas were approximately 0.80 (Piskorski et al. 1999) when |R|2 was used to identify ABGs of 20 to 30 dB. ROC areas for TW were approximately the same as those for |R|2.
Looking across studies, WAI using mid-frequency bands (1200–2000 Hz) identified CHL in infants equally well by tympanometry using probe tones of 678 and 1000 Hz, and |R|2. In children, tympanometry using probe tones of 226 Hz sometimes performed equally well as A or |R|2, but there was variability depending on the frequency of the WAI measure.
Two studies used multivariable measures that combined WAI values across frequency (Keefe et al. 2012) and different measures of WAI as well as SA and TW (Piskorski et al. 1999) into one single-output variable. The single-output variable was then used to evaluate the effectiveness of the identification of CHL. Both studies included children and found that combining WAI measures across frequencies was more effective in identifying CHL than a using WAI in one frequency band or tympanometry using a 226 Hz probe tone. Piskorski et al. (1999) performed discriminant function analysis, which uses a set of input variables and combines them to maximize the mean values between normal and impaired populations and minimize within-group variances. A second statistical analysis used was the logistic regression model. In this method, each variable is transformed into a dichotomous variable of 0 or 1 and the probability of impairment can be estimated. The variables used in the multivariate analysis were G, |R|2 analyzed into 1 octave bands centered at 500 and 2000 Hz, and equivalent volume, which represents all sources of compliance. Combining the WAI measures improved identification of ABG of 15 to 30 dB, with ROC areas of 0.85 or greater. In addition, the authors performed a multivariate analysis on the combined WAI and tympanometry measures and found that for ABGs of 20 to 30 dB, ROC areas were slightly higher than for combined WAI frequencies alone. In a different study, Keefe et al. (2012) combined A across frequency and used a multivariate technique called log LR. This technique compares the WAI from the individual ear in each frequency band with the mean data ±1 standard error from ears with NH and CHL, and calculates an LR that the response came from one of the populations. The LRs are then combined across frequency, resulting in a single variable from which the ROC curves and variance among different ROC curves were computed. The areas under the ROCs using LR were close to 1.0 and superior to those for tympanometric measures of SA and TW. Symmetry values were close to 0.9. On further analysis, ears were classified into the categories of high risk, low risk, or moderate risk of CHL. The majority of ears classified as normal or with CHL were correctly classified as low risk and high risk, respectively. In summary, these two studies indicate that combining WAI measures across frequency and/or across WAI variables provides improved identification of ABG over univariate WAI and tympanometry measures of SA and TW.
Two studies have combined the techniques of WAI and tympanometry, meaning WAI is measured while pressures are introduced into the ear canal ranging from −300 to +200 daPa. This technique allows for analysis of WAI across frequency at particular pressures, looking across pressure for a particular frequency, or both, looking across both pressure and frequency. Both studies measured A and At, and evaluated the accuracy of identifying CHL through magnitude of ABG. Keefe and Simmons (2003) studied ambient and WAI tympanometry in a group of older children and adults. They found that ROC curve areas were greater for multivariate A as compared with univariate A, SA, and TW, similar to results from other studies (Piskorski et al. 1999; Keefe et al. 2012). The ROC curve areas were greatest for multivariate At, resulting in areas of 0.95 compared with 0.89 for A. However, in a more recent study by Keefe et. al. (2012), no differences were found in ROC curve areas for identification of ABG between A and At. There are several reasons why the results from these two studies may conflict. First, the statistical procedures used to combine multivariate inputs were different for the two studies. Keefe et al. used a log-likelihood method, resulting in ROC areas of approximately 0.95, which was sufficiently high that it would be difficult to obtain statistically significant improvements. Piskorski etal. (1999) used two statistical techniques, discriminant analysis and logistic regression, to analyze their data. Another difference between the studies by Keefe and Simmons and Keefe et al. was the age of participants. Keefe et al. included children aged 2.6 to 8.2 years and Keefe and Simmons included participants over a wide range of ages, 10 to 55 years. It is likely the etiology of CHL for the participants in Keefe and Simmons was more variable than that for Keefe et al. It is also possible that participant age is a contributing factor. The use of applied ear-canal pressure, which may change the stiffness of the ear canal in children but not in adults, could influence the results.
Wideband Middle Ear Measures in Adults
Three studies collected WAI in groups of adults with CHL whose etiologies were surgically confirmed. The ears reported in the work by Nakajima et al. (2012) included 14 with otosclerosis, 6 with ossicular discontinuity, and 11 with SCD. For otosclerotic ears, |R|2 was increased relative to normal ears from 400 to 1000 Hz, indicating a stiffer system. Conversely, AdB was reduced in these frequency ranges. Ears with ossicular discontinuity exhibited increased AdB from 500 to 800 Hz, and those with SCD produced an increased AdB around 1000 Hz. The mean |R|2 and AdB for these participant groups are illustrated in Figure 4A, B, respectively. It is encouraging that although these various etiologies produced changes in WAI in different frequency bands, WAI and ABGs identified CHL with similar sensitivity and specificity to measurements of umbo velocity and ABG. Shahnaz and colleagues (2009a) also studied otosclerotic adult ears with measures of tympanometry using sweeping frequency tones and |R|2. They also found that |R|2 was elevated in frequency bands from 300 to 1000 Hz, for which ROC curves are illustrated in Figure 5A. In Figure 5B, the area under the ROC curve area under the curve for |R|2 in a 1/3 octave band centered at 500 Hz and the frequency at which admittance phase was 45° based on sweep-frequency tympanograms were similar, with those for SA being significantly lower. When using the 90th percentile from NH ears as a criterion, |R|2 in the 500 Hz band had a sensitivity and specificity of 82 and 83%, respectively. In a follow-up study, Shahnaz and colleagues (2009b) examined correlations between changes in|R|2 and AC thresholds pre- and poststapedectomy. Data from 10 of the patients had been reported in Shahnaz et al. (2009a). Although correlations between the reduction of AC thresholds and changes in |R|2 were not significant, there was a trend for similar changes in the lower-frequency bands of |R|2 and AC threshold across the frequency range.
WAI and Prediction of ABG
A topic that may be worth exploring is whether WAI can predict the magnitude of ABGs. Although the data from Nakajima et al. (2012) suggest that magnitude of ABG and AdB are not related for adults with otosclerosis, ossicular discontinuity, and SCD, the data from children, in which OME is likely to be the cause of CHL, could be considered. Data from Piskorski et al. (1999) qualitatively illustrate (their Figure 10) that there may be a trend between |R|2 at 2000 Hz and the coefficient for the discriminant function (DF). Data from Keefe et al. (2012) suggest that participants having ABGs ≥ 30 dB have lower A than those having ABGs ≥ 20 dB. Because the data for participants having ABGs ≥ 30 dB were included in the data for those having ABGs ≥ 20 dB, the question of whether there is a relationship between A and ABG could not be further tested.
The WAI measures of A, AdB, or|R|2 identify CHL with excellent accuracy. Univariate ambient WAI and admittance tympanometry using a 1000 Hz probe tone identify CHL with similar accuracy in infants (10 weeks of age) and in children; tympanometry at 226 Hz identifies CHL at 226 Hz better than ambient WAI in the 250 Hz frequency band. Multivariate statistical measures that combine WAI across-frequency are superior to using WAI in one frequency band (univariate) and tympanometry at a single frequency for detecting ABGs of 15 to 30 dB in children. In adults, identification of CHL due to otosclerosis, SCD, and ossicular discontinuity has high sensitivity and specificity. There are conflicting results as to whether WAI tympanometry increases the effectiveness of identifying a CHL. Finally, the ability of WAI to accurately predict magnitude of CHL is doubtful, but has not been adequately tested.
As a noninvasive tool to measure outer/middle ear properties, WAI is attractive because many frequencies can be tested simultaneously, and in little time. Although WAI is better at detecting CHL than single-frequency admittance tympanometry, tympanometry is a good tool for detecting CHL, especially in infants (Prieve et al. 2013).
The first and most obvious research need is to test large numbers of participants having a wide span of ages with CHL due to a variety of etiologies. The conclusions of this article, while positive, are based on WAI measures in 544 ears with NH and 219 ears with CHL. Of those with confirmed CHL, 130 were from children and 17 were from infants. There were 54 adult participants with confirmed CHL, counting participants once for those whose data were reported in multiple studies. Moreover, the various studies analyzed different WAI measures and analyzed the data through several statistical procedures. There are few published studies that compare the effectiveness of different measures of WAI, such as |R|2, A, AdB, although it is stated in Keefe et al. (2012) that the measure Aa has less variability than other WAI measures. An important issue is how various statistical methods for combining WAI across frequency affects identification of CHL. Another important topic of discussion is whether WAI tympanometry improves identification of CHL. A final, interesting, question that could improve audiometric testing procedures in infants and children is whether WAI can predict the magnitude of ABG, and, with what accuracy. A challenge to answer this question is that to predict magnitude of an ABG, it is necessary to have accurate measurement of AC and BC thresholds, which relies on calibration of signals. For infants, especially for BC stimuli, there are no such standards. With current bone-vibrator standards, there is an apparent “ABG” at 500 Hz in ears with NH (Vander Werff et al. 2009). This is caused by different load impedances and vibration transfer functions of the head for an infant and an adult. It is concluded that exploration of WAI for identification of CHL is a worthy pursuit and could positively impact clinical diagnosis of hearing loss.
Beers A. N., Shahnaz N., Westerberg B. D.. Wideband reflectance in normal Caucasian and Chinese school-aged children and in children with otitis media with effusion. Ear Hear. (2010);31:221–233
Govaerts P., Casselman J., Daemers K.. Audiolgoical findings in large vestibular aqueduct syndrome. Int J Pedeatr Otorhinolaryngol. (1999);51:157–164
Keefe D. H., Sanford C. A., Ellison J. C.. Wideband aural acoustic absorbance predicts conductive hearing loss in children. Int J Audiol. (2012);51:880–891
Keefe D. H., Simmons J. L.. Energy transmittance predicts conductive hearing loss in older children and adults. J Acoust Soc Am. (2003);114(6 Pt 1):3a217–3238
Merchant S. N., Rosowski J. J.. Conductive hearing loss caused by third-window lesions of the inner ear. Otol Neurotol. (2008);29:282–289
Minor L. B.. Superior canal dehiscence syndrome. Am J Otol. (2000);21:9–19
Nakajima H. H., Pisano D. V., Roosli C.. Comparison of ear-canal reflectance and umbo velocity in patients with conductive hearing loss: A preliminary study. Ear Hear. (2012);33:35–43
Piskorski P., Keefe D. H., Simmons J. L.. Prediction of conductive hearing loss based on acoustic ear-canal response using a multivariate clinical decision theory. J Acoust Soc Am. (1999);105:1749–1764
Prieve B. A., Werff K. R., Preston J. L.. Identification of conductive hearing loss in young infants using tympanometry and wideband reflectance. Ear Hear. (2013);34:168–178
Rosowski J. J., Nakajima H. H., Hamade M. A.. Ear-canal reflectance, umbo velocity, and tympanometry in normal-hearing adults. Ear Hear. (2012);33:19–34
Shahnaz N., Bork K.. Wideband reflectance norms for Caucasian and Chinese young adults. Ear Hear. (2006);27:774–788
Shahnaz N., Bork K., Longridge N.. Wideband reflectance (WBR) in normal and otoschlerotic ears. Ear Hear. (2009a);30:219–233
Shahnaz N., Longridge N., Bell D.. Wideband energy reflectance patterns in preoperative and post-operative otosclerotic ears. Ear Hear. (2009b);48:240–247
© 2013 by Lippincott Williams & Wilkins
Vander Werff K. R., Prieve B. A., Georgantas L. M.. Infant air and bone conduction tone burst auditory brain stem responses for classification of hearing loss and the relationship to behavioral thresholds. Ear Hear. (2009);30:350–368