Pulsed DPOAEs in serial measurements

For the combined analysis paradigm presented here, DPOAE level maps and hearing thresholds from a study conducted by the authors [3] were used, in which the test–retest reliability of the level map-based L_EDPT was compared with that of pure-tone thresholds. Measurements were recorded seven times over three months at 14 frequencies between 1 and 14 kHz in 20 ears of ten normal-hearing subjects (PTA_{4 (0.5–4 kHz)} < 20 dB HL; age 32.1 ± 9.7 years). Subjective hearing thresholds, L_TA, were recorded three times at each frequency and at two adjacent frequencies using modified Békésy tracking audiometry, creating a frequency group to smooth out fine-structure effects in the behavioral audiogram. The study was approved by the Ethics Committee of the University of Tübingen (265/2018B01) and conducted in accordance with the Declaration of Helsinki for experiments with humans.

All measurements were conducted with two ER-10C probes (Etymotic Research, Elk Grove Village, IL, USA) using a standard PC with NI measurement cards (National Instruments, Austin, TX, USA). Stimulation and data acquisition were carried out using measurement software implemented in LabVIEW (National Instruments). Specially developed software in MATLAB (The MathWorks, Natwick, MA, USA) enabled automated analysis of the DPOAEs and hearing thresholds. In order to achieve consistent placement of the sound probes across all sessions, the frequency response of the ear-canal sound pressure from 0.3 to 20 kHz was determined for each ear and visually compared with that from previous sessions. The stimulus sound pressure was calibrated before each session by an in-ear measurement and the transmission to the eardrum was corrected using an artificial ear (B&K type 4157, Brüel & Kjær, Nærum, Denmark).

For the bilateral acquisition of DPOAE level maps, measured simultaneously from the two ears, 21 short-pulse DPOAEs of different stimulus levels (L_1,L₂ pairs) were acquired at each frequency pair (f₂ = 1–14 kHz, f₂/f₁ = 1.2). The short-pulse stimulation enables the separation of the nonlinear distortion and coherent reflection components in the time domain by utilizing their different latencies (Fig. 1).

The artifact-free DPOAE amplitude, P_DP (red dot in Fig. 1a), extracted from the time signal, estimates the amplitude of the nonlinear distortion component; it was accepted for statistical analysis for SNR ≥ 10 dB and converted to the DPOAE level, L_DP. The total measurement time for all DPOAE level maps with 21 levels at 14 frequencies in both ears was 12.6 min.

Plotting of the measured DPOAE amplitudes as a function of the stimulus levels L₁ and L₂ enables construction of a DPOAE level map (Fig. 2, three-dimensional surface). The map allows an individually optimal growth function to be determined (Fig. 2, black line); that is, a set of DPOAE amplitudes, specific for a given frequency pair and subject, which are maximal for a given L₂. This growth function was obtained by numerically fitting a mathematical function to the measured DPOAE amplitudes. An individual, frequency-specific stimulus level L_1,opt for a given L₂ is determined from projection of the optimal growth function in the L₁,L₂ plane. L_EDPT is given as the value of L₂ at which the optimal growth function intersects the L₁,L₂ plane (Fig. 2, red arrow). With this method, L_EDPT can be determined without having to define optimal stimulus levels in advance of the experiment [28].

In addition, the model level maps can be used to reconstruct those L_DP that would have been generated with the frequency-independent stimulus paradigm proposed by Kummer et al. [18]; namely, L_1,kum = 0.4L₂ + 39 dB. The maps can also be employed to reconstruct those L_DP that would have been generated with the constant level-spacing paradigm conventionally used as a clinical standard; namely, L_1,std = L₂ + 10 dB.

In this work, reconstructed L_DP are presented exemplarily for L₂ = 45 and 65 dB sound pressure level (SPL). It is important to realize that the reconstructed L_DP only provide meaningful estimates of actual measured values to the extent that the model function faithfully reproduces the actual dependence on the stimulus-level combinations. For this reason, as well as due to the usual limiting conditions of practical measurability (usually residual noise level < −20 dB SPL [4, 12]), reconstructed L_DP < −15 dB SPL were not evaluated and therefore excluded from further analysis.

For the purpose of developing an analysis paradigm that combines simultaneously occurring changes in L_DP, L_EDPT, and L_TA from examination to examination, dependencies between these parameters must be assigned. It is known that L_EDPT and L_TA are correlated approximately with a ratio of 1:1 [28, 30]; i.e., an increase in hearing threshold by 10 dB is accompanied by an increase in L_EDPT of also about 10 dB. For the dependency of L_DP and L_TA, we use the observation made by Kummer et al. (see their Fig. 7b and their Tables II, III; [18]) that an increase in hearing threshold by 10 dB was associated with a decrease in L_DP of only about 5 dB; i.e., correlated with a ratio of approximately 2:1. Overall, this observation means that in order to estimate an increase of a cochlear-induced hearing threshold from a decrease of L_DP, the L_DP decrease would have to be doubled. Therefore, in a combined analysis, a change in L_DP was weighted by a factor of (negative) 2. The analysis investigated the following four combinations of the changes ∆L_TA, ∆L_DP, and ∆L_EDPT: namely, (∆L_TA − 2∆L_DP)/2, (∆L_TA + ∆L_EDPT)/2, (∆L_EDPT − 2∆L_DP)/2 and (∆L_TA + ∆L_EDPT − 2∆L_DP)/3.

The software SPSS Statistics (version 26, IBM Corp., Armonk, NY, USA) was used for the statistical tests. To quantify the test–retest reliability of the L_EDPT, L_TA, and L_DP, the absolute differences (AD) between two visits (1 vs. 2, 1 vs. 3, 1 vs. 4, …, 2 vs. 3, 2 vs. 4, …; N = 21) were used as a metric [20, 21]. Test–retest reliability determines the ability of a method to produce similar results when repeated for the same individual under the same experimental conditions. The statistical significance of the absolute differences between the samples was tested using the Friedman test.

The correlation between L_DP and L_TA depended significantly on the stimulus levels (Fig. 3). For f₂ = 8–14 kHz, in which there was already a high-frequency hearing loss in some subjects and thus a larger L_TA range, there was a strong correlation between L_DP and L_TA (Spearman’s ρ = −0.737; p < 0.001) when L_DP was reconstructed using individually optimal, frequency-specific stimulus levels L_1,opt (blue) at L₂ = 45 dB SPL (Table 1). By contrast, when L_DP was reconstructed with frequency-independent, standard clinically used stimulus levels L_1,std (green, L₁ = L₂ + 10 dB) or L_1,kum (yellow, L₁ = 0.4L₂ + 39 dB), there was only a small correlation between L_DP and L_TA (L_1,std Spearman’s ρ = −0.202, p = 0.003; L_1,kum Spearman’s ρ = −0.282, p < 0.001). For f₂ = 1–6 kHz, there was slight hearing loss and therefore insufficient L_TA range to establish a potential correlation between L_DP and L_TA. The stimulus level L₂ = 45 dB SPL is shown as an example in Fig. 3 and Table 1 because the cochlear amplifier is not yet in compression at this level and a sufficient number of detectable DPOAE signals were generated.

The test–retest reliability of L_DP increased significantly with individually optimized stimulus levels (L_1,opt; Fig. 4). The median AD of L_DP decreased significantly from 2.3 dB (L_1,std) or 2.2 dB (L_1,kum) to 1.8 dB using L_1,opt at L₂ = 45 dB SPL (Table 2, Friedman test, F = 734.65; p < 0.0001). Consequently, the derived reference interval decreased, respectively, from 10 or 9 dB to 6 dB, defined here as the 90th percentile of AD. At L₂ = 65 dB SPL, the median AD using L_1,opt was reduced from 2.4 dB (L_1,kum) or 1.9 dB (L_1,std) to 1.4 dB and the reference interval decreased, respectively, from 9 or 7 dB to 4 dB (Table 2).

L_EDPT and L_TA are correlated linearly (L_TA = 0.86L_EDPT − 6.7 dB, r² = 0.45, SD = 7.7 dB, p < 0.001, Fig. 5). In addition to the high correlation with L_TA, L_EDPT also showed a high test–retest reliability, with a median AD of 3.3 dB for f₂ = 1–14 kHz (Table 3). Table 3 quantifies the test–retest reliability using the median, the interquartile range (IQR), and the 90th percentile of AD for L_TA and L_EDPT as well as for different combined analysis paradigms. The combination of L_EDPT, L_TA, and L_DP,opt,65 (L_1,opt at L₂ = 65 dB SPL), denoted by L_EDPT′, presented the smallest median AD (2.0 dB). An almost equally low median AD (2.1 dB) was observed when both DPOAE measures (L_DP and L_EDPT) were combined. Figure 6a shows the distributions of differences between two sessions for each of L_EDPT and L_EDPT′. The combined measure, L_EDPT′, presents a significant reduction in the number of outliers compared with L_EDPT and a reduction in the standard deviation from 5.6 to 3.9 dB (vertical lines). Figure 6b shows the distributions of absolute differences (AD) for L_EDPT and L_EDPT′. The vertical lines mark the reference range, defined by the 90th percentile. The reference range of L_EDPT′ (6.2 dB) is significantly smaller than that of L_EDPT (9.3 dB). Taken together, these observations imply that the combined measure L_EDPT′ presents significantly higher test–retest reliability than L_EDPT.

In general, DPOAE levels show a relatively limited correlation with cochlear-induced hearing loss, and the complex relationship between DPOAE level and the associated hearing loss is nonlinearly dependent on stimulus level and frequency [4, 12]. Interestingly, after using the individually optimal stimulus level L_1,opt at a moderate stimulus level L₂ = 45 dB SPL, a significantly higher correlation of L_DP and L_TA and a lower scatter was found, especially in the high-frequency range f₂ = 8–14 kHz (Fig. 3). Since the inclusion criterion for the study was defined as PTA_{4 (0.5–4 kHz)} < 20 dB HL, there was hardly any hearing loss for some frequencies in the range 1–6 kHz and, therefore, the L_TA range was insufficient to detect potential correlation between L_DP and L_TA and consequently any associated middle-ear or noise influences.

The idea behind the use of optimal stimulus levels is to achieve an ideal overlap between the travelling-wave envelopes of the two stimulus tones f₁ and f₂ near the characteristic place of the f₂ tone by taking into account the different compression states of the two travelling waves near the f₂ characteristic place [24], so that the distortion produced by nonlinear mechanoelectrical transduction is maximal near that place [2]. Although these two aims were originally introduced by Kummer et al. [18], the algorithms for attaining an optimal L₁ for a given L₂ differ significantly: In our case, the optimization parameters are derived individually for each session, subject, and f₂ and, as such, our algorithm is a major development of the earlier optimizing algorithm (the “scissor’s” algorithm) where the parameters were independent of the subject and f₂ [18]. Individually optimizing L₁ led to a reduction in the inter-subject variability of L_DP, especially for f₂ = 8–14 kHz (Fig. 3b).

Not only the inter-subject variability of L_DP, but also the intra-subject variability, the so-called test–retest reliability, improved significantly by selecting frequency-specific, individually optimal stimulus levels. Consequently, the frequently quoted reference range for an intra-subject DPOAE change from examination to examination at the stimulus level of L₂ = 65 dB SPL was reduced from approximately 6–8 dB to 4–5 dB. In future clinical examinations, pulsed DPOAE signals excited using frequency-specific, individually optimal stimulus levels could therefore be a valuable method for detecting early signs of changes in the functional state of the cochlear amplifier, before they become visible with conventional DPOAE level measurements.

DPOAE level maps capture—with high precision—the intensity behavior of the cochlear amplifier near the f₂ place for different stimulus level pairs L₁,L₂. L_EDPT derived from such maps incorporate information from multiple DPOAE amplitudes and thus allow for a more precise and extended diagnosis of the functional state of the cochlear amplifier, as has already been shown by us and other authors for DPOAE growth functions and their properties [26]. With numerical extrapolation of the DPOAE amplitudes, the growth behavior of the DPOAE amplitudes can also be derived at low stimulus levels. It is precisely at such levels that the test significance of the functional state of the cochlear amplifier is highest due to nonlinear amplification being highest at low stimulus levels [18].

In addition, provided that there is no damage to the inner hair cells and neural pathways, L_EDPT enables an objective quantification of the hearing threshold [8, 28, 30]. Such quantification is not possible using conventional protocols in the clinic, where suprathreshold DPOAEs are measured at one or two stimulus levels. Moreover, changes in the individual transmission characteristics of the middle ear can be captured by DPOAE level maps, whereby losses in anterograde middle-ear transmission shift and distort the DPOAE level maps and losses of retrograde middle-ear transmission reduce DPOAE amplitude [19]. The DPOAE growth function is extrapolated along the ridge of the three-dimensional surface of the model DPOAE level map (Fig. 2) and is thus based on maximum DPOAE amplitudes generated using individual, near-ideal stimulus levels. L_EDPT based on DPOAE level maps estimate hearing thresholds more accurately than conventional DPOAE growth functions that are excited with predetermined stimulus levels [28]. In this study, L_TA for f₂ = 1–14 kHz correlated with the L_EDPT derived from DPOAE level maps with a standard deviation of 7.7 dB (Fig. 5). This error estimate is slightly higher than the standard deviation of 6.5 dB reported by Zelle et al. [30], and is attributable to a factor four reduction in the averaging time per DPOAE signal in the present study together with the extension of the frequency measurement range from 1–8 kHz to 1–14 kHz. In addition, for reasons of the still limited quantity of hearing-loss data, the regression analysis has not been performed at single frequencies, in which case the standard deviation of the estimate of L_TA predicted from L_EDPT for a given frequency can be significantly reduced [30]. Moreover, it is expected that the implementation of a modern calibration procedure such as IPL (integrated pressure level) or FPL (forward pressure level) would further reduce the standard deviation, particularly at high frequencies [20].

L_EDPT not only estimate individual hearing thresholds accurately, but are also stable for follow-up measurements in a given ear [3]. The test–retest reliability of L_EDPT for the entire frequency range f₂ = 1–14 kHz with a median AD of 3.3 dB is comparable to that of L_TA (median AD = 3.2 dB), whereas for the high-frequency range, f₂ = 11–14 kHz, L_EDPT are superior to L_TA [3]. The reference range corresponding to the 90th percentile, above which an ear must be considered in need of control in follow-up examinations, is approximately 10 dB for both L_EDPT and L_TA for f₂ = 1–14 kHz. When L_DP differences are doubled to correct for L_TA being proportional to L_DP with a slope of 2, L_DP show a comparable test–retest reliability; namely, with a median AD of 2.8–3.6 dB and a 90th percentile of 8–12 dB when using L_1,opt (Table 2). Since L_DP and L_EDPT are partly subject to different confounding factors (e.g., middle-ear pathology, noise sources) and physiological mechanisms, the strategy introduced in this study was to combine the two DPOAE parameters and the auditory threshold parameter into a single parameter that is as sensitive and reliable as possible.

To date, changes in the pure-tone hearing threshold and DPOAE level (typically, measured at L₂ = 65 dB SPL, L_1,std = 75 dB SPL) have generally been considered separately in everyday clinical practice for the monitoring of ototoxicity. To the best of our knowledge, the test–retest reliability of concurrent changes in hearing thresholds and DPOAE levels has not been reported in the literature. Only multivariate statistical DPOAE analyses that consider DPOAE level and SNR simultaneously have been presented for predicting hearing threshold [11] and ototoxic hearing loss [16]. As predictors of hearing status, multivariate DPOAE analyses achieve better test quality compared with univariate approaches using either DPOAE level or SNR. However, even with multivariate analyses, there is still considerable overlap between the distributions for normal-hearing and hearing-impaired people, which was found to be more pronounced for the frequency range 0.75–3 kHz than for 4–8 kHz [11]. Multivariate DPOAE analyses also lead to improved test performance for predicting an increase of ototoxic-induced hearing threshold, but only when the cumulative cisplatin dose is included in the analysis [16]. Using a 6-dB change in the DPOAE level as a metric allows for little to no improvement over an analysis based on cumulative cisplatin dose and pre-exposure hearing threshold [16]. The analysis paradigm presented here, which combines changes in L_EDPT, suprathreshold L_DP, and fine-structure-reduced L_TA, significantly improved the test–retest reliability (Fig. 6 and Table 3). It is expected that this approach will lead to higher sensitivity and specificity in future studies for detecting pathological or regenerative changes in the outer hair cells.

Since this study focused on the validation of the methodology of pulsed DPOAEs in follow-up measurements in normal-hearing subjects, there are few data for mild-to-moderate hearing loss for f₂ = 1–6 kHz. Therefore, for the purpose of specifying a metric that combines the various parameters, we assumed that L_DP is correlated with L_TA with a ratio of 1:2, the assumption being mainly based on the findings of Kummer et al. [18]. Given the nonlinear dependence of L_DP on stimulus frequency and level, future applications of the combined analysis paradigm should quantify the relationship between L_DP and L_TA as a function of frequency and level using individually optimal stimulus levels L₁.

Although it was shown here that the combined analysis paradigm together with the pulsed DPOAE protocol yields a higher test–retest reliability than reported to date, it still has to be established that, for example, ototoxic hearing impairment in follow-up examinations of patients receiving chemotherapy with cisplatin can be detected earlier and more sensitively by using DPOAE level maps and combined analysis paradigms compared with other audiological test procedures.

In addition, the procedure could be optimized through further technical adjustments. A modern calibration procedure for sound pressure could be implemented that avoids erroneous stimulus levels due to standing waves within the auditory canal and thus facilitates the detection of DPOAEs in still higher numbers and quality. It would also be advantageous to develop an adaptive algorithm that enables the detection of DPOAE level maps within an L₁,L₂ space depending on the SNR, in order to reliably construct DPOAE level maps in as many patients with residual cochlear hearing as possible in a time-efficient manner.