The relationships between cochlear nerve health and AzBio sentence scores in quiet and noise in postlingually deafened adult cochlear implant users

Abstract

Objectives:

This study investigated the relationships between the cochlear nerve (CN) health and sentence-level speech perception outcomes measured in quiet and noise in postlingually deafened adult cochlear implant (CI) users.

Design:

Study participants included 28 postlingually deafened adult CI users with a Cochlear^® Nucleus^™ device. For each participant, only one ear was tested. Neural health of the CN was assessed at three or four electrode locations across the electrode array using two parameters derived from results of the electrically evoked compound action potential (eCAP). One parameter was the phase locking value (PLV) which estimated neural synchrony in the CN. The other parameter was the sensitivity of the eCAP amplitude growth function (AGF) slope to changes in the interphase gap (IPG) of biphasic electrical pulses (i.e., the IPGE_slope). Speech perception was tested using AzBio sentences in both quiet and a ten-talker babble background noise with +5 dB and +10 dB signal-to-noise ratios (SNR). IPGE_slope and PLV values were averaged across electrodes for each subject, both with and without weighting by the frequency importance function (FIF) of the AzBio sentences. Pearson and Spearman correlations were used to assess the pairwise relationships between the IPGE_slope, the PLV, and age. Multiple linear regression models with AzBio score as the outcome and the PLV and the IPGE_slope as predictors were used to evaluate the associations between the three variables while controlling for age.

Results:

The IPGE_slope and the PLV demonstrated different patterns with regards to their relationships with electrode location, age, and speech perception. The PLV, but not the IPGE_slope, differed significantly across electrodes, where the apical electrodes had larger PLVs (better neural synchrony) than the basal electrodes. The IPGE_slope, but not the PLV, was significantly correlated with participant’s age, where smaller IPGE_slope values (poorer spiral ganglion neuron density) were associated with more advanced age. After controlling for the effect of age, the PLV, but not the IPGE_slope, was significantly associated with AzBio scores in the +5 dB SNR condition, where larger PLVs predicted better speech perception. Neither the PLV nor the IPGE_slope was significantly associated with AzBio score in quiet or in the +10 dB SNR condition. The result patterns remained the same regardless of whether the mean values of the IPGE_slope and the PLV were weighted by the AzBio FIF. The result patterns generally did not change with fitting methods or input/output scales of the AGF slopes.

Conclusions:

The IPGE_slope and the PLV quantify different aspects of CN health. The positive association between the PLV and AzBio scores in the +5 dB SNR condition suggests that neural synchrony is important for speech perception in adult CI users in challenging listening conditions with a relatively high noise level. The lack of association between age and the PLV indicates that reduced neural synchrony in the CN might not be the primary factor accounting for the greater deficits in understanding speech in noise observed in older CI users, as compared to middle-aged CI users.

INTRODUCTION

The cochlear implant (CI), a prosthesis that partially restores hearing through stimulating the cochlear nerve (CN) via electrodes surgically implanted into the inner ear, is a standard treatment option for listeners with sensorineural hearing loss (for a review, see Zeng 2004). Since electrical stimulation takes place at the auditory periphery, subsequent transmission of the signals by the CN is a prerequisite for the central auditory system’s ability to access and process the sound information. Therefore, it is believed that the neural health of the CN is crucial for the success of CI treatment (e.g., He et al. 2017; Zamaninezhad et al. 2023). The association between the CN health and hearing performance in CI users has been supported by post-mortem observations, where within-subject between-ear comparisons showed that the ear with a larger amount of spiral ganglion neurons (SGNs) consistently yielded a better word recognition performance (Seyyedi et al. 2014). However, due to the invasiveness of the histological procedures, direct examination of the CN is not feasible in living human CI users. Rather, non-invasive electrophysiological measures, such as the electrically evoked compound action potential (eCAP), have been used as a proxy of the CN health status of human CI users in research and clinical settings.

The eCAP is a near-field recorded, synchronized response of a population of CN fibers elicited by electrically stimulating a CI electrode, which has been used to evaluate neural encoding of electrical stimulation at the CN, such as spectral resolution (Won et al. 2014), neural adaptation (Hughes et al. 2012; He et al. 2023b), and amplitude modulation encoding (Tejani et al. 2017) in CI users. Morphologically, a typical eCAP waveform consists of a negative peak (N1) at around 0.2–0.4 ms after stimulus onset, followed by a positive peak (P2) at around 0.6–0.8 ms after stimulus onset (e.g., Brown et al. 1990). The amplitude of the eCAP waveform is defined as the difference in voltages between P2 and N1, and it increases with the stimulation level. The relationship between stimulation level and eCAP amplitude can be depicted using an amplitude growth function (AGF), also known as the input/output (I/O) function.

The slope of eCAP AGF has been shown to be associated with the density of SGNs in animal studies, where steeper slopes indicate higher SGN density in pharmaceutically deafened, implanted animals (Pfingst et al. 2015). Aligned with the animal results, shallower eCAP slopes were observed in pediatric CI users with cochlear nerve deficiency (CND) compared to those with normal-sized CNs (He et al. 2018). However, since the raw eCAP responses are susceptible to inter-patient and inter-electrode differences in non-neural factors (Brochier et al. 2021), such as electrode impedance and electrode-modiolus distance (EMD; Imsiecke et al. 2021), researchers have been seeking to overcome this drawback by using the differences between eCAP-derived measurements under various stimulation conditions to assess CN health. One particularly promising measurement is the sensitivity of the eCAP amplitude to changes in the interphase gap (IPG) of biphasic, electrical pulses (i.e., the IPG effects). Theoretically, non-neural factors should affect eCAP amplitude growth functions (AGFs) measured using different IPGs in a similar way. As a result, the effects of non-neural factors should be reduced or even eliminated when the differences between AGFs measured at different IPGs are calculated. Aligning with this theoretical possibility, research studies in both animal models and human CI users have demonstrated that the IPG effects are not affected by non-neural factors, such as electrode impedance (Schvartz-Leyzac & Pfingst 2016, 2018; Imsiecke et al. 2021) and EMD (Schvartz-Leyzac et al. 2020b; Imsiecke et al. 2021).

In animal models, the IPG effects have been shown to be correlated with SGN survival. Specifically, larger effects of IPG on eCAP amplitude (Prado-Guitierrez et al. 2006) and the AGF slope (Ramekers et al. 2014; Schvartz-Leyzac et al. 2019; Schvartz-Leyzac et al. 2020a) are associated with higher SGN density in guinea pigs. Consistent with these results measured in animal models, the IPG effect on the AGF slope (IPGE_slope) has been shown to be positively correlated with sentence and consonant recognition (Schvartz-Leyzac & Pfingst 2018) and speech reception threshold (SRT; Zamaninezhad et al. 2023) in postlingually deafened adult CI users. More importantly, IPGE_slope was larger in children with normal-sized CNs than in children with CND (He et al. 2020; Yuan et al. 2022), a patient population with substantially fewer but otherwise healthy SGNs than other patient populations. Specifically, histological results of human temporal bone studies have shown that the mean SGN count in children with CND is 5,739 (range: 0–15,714), representing about 16.4% of the SGN count expected for the corresponding age, which is roughly 35,500 (Wright et al. 1986; Glueckert et al. 2010; Nelson & Hinojosa 2001; Haginomori et al. 2002; Chen et al. 2020; Da Costa Monsanto et al. 2022). In comparison, the mean SGN count in hearing impaired children with normal-sized CNs ranges from 14,301 to 20,523 (Miura et al. 2002). It should be noted that the axons of existing SGNs in children with CND are myelinated, with no apparent sign of degeneration (Glueckert et al. 2010; Wright et al. 1986). Therefore, the results reported in He et al. (2020) and Yuan et al. (2022) provide a strong piece of evidence supporting the indicative value of IPGE_slope for SGN count. Nevertheless, the sensitivity of IPGE_slope as an index of SGN count in CI users cannot be determined due to the lack of direct and precise measurements of neural structure in living human listeners.

The IPG effect on stimulation level offset (IPGE_offset) has been proposed as another candidate index of CN health. Based on computational modeling results, Brochier et al. (2021) proposed that the IPGE_offset outperformed the IPGE_slope in controlling for non-neural factors. Consistent with this proposed relationship between IPGE_offset and CN health, Jahn and Arenberg (2020b) observed greater IPGE_offset in younger compared to older listeners. However, several empirical studies failed to demonstrate associations between the IPGE_offset and speech perception measurements in CI users with either full-length electrode array (Zamaninezhad et al. 2023) or short array to preserve residual low frequency acoustic hearing (Kim et al. 2010). Recent computational modeling studies on the IPG effects suggested that IPGE_slope and IPGE_offset might reflect different aspects of CN health. As noted in the discussion of Brochier et al. (2021), IPGE_offset likely reflects the properties and health status, rather than the number, of surviving SGNs. Aligned with this distinction, a three-dimensional model of the cochlea by Kipping et al. (2024) and Zhang et al. (2024) demonstrated correlations between IPGE_offset and CN degeneration, defined as reduced diameter in or entire loss of the peripheral axons. In contrast, a two-dimension model by Takanen et al. (2024) demonstrated that the IPGE_slope calculated as the numerical difference between the AGF slopes on a linear I/O scale is dependent on neural survival, and that non-neural factors such as electrode-neuron distance had little interference on the IPGE_slope. Taken together, the existing evidence from animal, human and computational research are converging in suggesting that IPGE_slope is a promising indicator of CN density among the currently available eCAP measurements. As a result, IPGE_slope was used to assess neural survival of the CN in this study.

While having sufficient CN fibers responding to auditory input is a prerequisite for auditory perception, CN density alone does not guarantee good hearing functions in challenging listening environments. In theory, effective and accurate representation of sound signals that allows the listener to separate target signals from noise requires synchronous firing across neurons, which in turn depends on the health status of the CN, as demonstrated in animal and computational modeling studies (Kim et al. 2013; Heshmat et al. 2020). Neural desynchronization leads to a smeared representation of temporal cues, so that even though the ability to detect sound in quiet may be minimally affected, hearing performance in noise would degrade drastically. This scenario is exemplified by some listeners with auditory neuropathy spectrum disorder (ANSD), who have normal or relatively good behavioral audiometric thresholds and speech perception in quiet, but disproportionally impaired signal detection and speech perception in noise (Kraus et al. 2000; Zeng et al. 2005). The electrophysiological measures of patients with ANSD are characterized by a relatively normal cochlear microphonic and/or otoacoustic emission (OAE) response, and an abnormal or absent auditory brainstem response (ABR), which has been interpreted as a lack of synchrony across CN fibers despite a relatively normal hair cell function (e.g., Starr et al. 2008). The crucial role of neural synchrony in acoustic hearing has been further supported by the compound action potential (CAP) recorded in normal hearing (NH) listeners, where the level of neural synchrony, quantified with the phase locking value (PLV) of trial-by-trial CAP measurements, was found to be a strong predictor of recognition scores for speech in noise and time-compressed speech in quiet (Harris et al. 2021).

Due to the difference between acoustic and electrical hearing, the observations in listeners with NH or ANSD may not be readily generalizable to CI users. For many CI users, speech perception in noise is a challenging task despite excellent hearing performance in quiet (Zaltz et al. 2020). Histological observations of SGN dystrophy and demyelination in listeners with various hearing profiles (Nadol 1997; Wu et al. 2019) suggest that reduced neural synchrony could be an underlying cause of poor speech perception in noise in CI users. However, this proposed relationship has rarely been evaluated, largely due to the lack of electrophysiological measures of neural synchrony in the CN. We recently developed a new method to quantify peripheral neural synchrony in CI users, where the PLV of trial-by-trial eCAP responses was used as an index to quantify the degree of neural synchrony in the responses generated by CN fibers across multiple electrical stimulations (He et al. 2024). We demonstrated that higher PLVs are associated with better temporal resolution as indicated by lower behavioral gap detection thresholds (GDTs) and smaller effects of noise on word recognition in post-lingually deafened adult CI users, consistent with the hypothesized effect of neural synchrony on hearing performance in electrical hearing.

In summary, previous research has established both CN survival and neural synchrony as crucial factors contributing to hearing performance in CI users. However, considering that nerve damage can result in both lower neural density and poorer synchronization, as has been shown in computational models (Heshmat et al. 2020), little is known about whether the contributions of neural survival and synchrony to hearing performance are independent or overlapping. Observations in NH listeners by Harris et al. (2021) suggest that neural engagement and synchrony are two separate dimensions that vary differently with changes in stimulus level. In human CI users, the number and synchrony of excited CN fibers have been modeled using retrospective deconvolution performed on intraoperative eCAP recordings, both significantly associated with postoperative speech recognition scores (Dong et al. 2023). However, the model was built upon assumptions about the shape of unitary response from CN fibers, which have not been directly validated in humans (Dong et al. 2020; Dong et al. 2023). While the IPGE_slope and the PLV in CI users have been measured postoperatively in separate experimental studies to assess their association with speech perception (Schvartz-Leyzac & Pfingst 2018; Zamaninezhad et al. 2023; He et al. 2024), it is unclear whether the biological underpinnings of these two indices are orthogonal and impact speech perception differently depending on listening conditions.

To address this critical knowledge gap, we measured the IPGE_slope, the PLV and speech perception in the same group of postlingually deafened adult CI users and evaluated their relationships. Speech perception was measured using AzBio sentences (Spahr et al. 2012), a speech corpus consisting of multiple lists of everyday sentences with similar levels of difficulty. Based on previous studies on the effects of neural survival and neural synchrony on speech perception in listeners with various hearing profiles, we hypothesized that (1) the IPGE_slope and the PLV are two independent measures of CN health (Harris et al. 2021; Kraus et al. 2000); (2) the IPGE_slope is positively associated with speech perception in quiet (Zamaninezhad et al. 2023); (3) the PLV is positively associated with speech perception in noise (Zeng et al. 2005; Harris et al. 2021; He et al. 2024).

MATERIALS AND METHODS

Participants

Study participants included 28 postlingually deafened CI users (age range: 20.39–84.04 years, mean = 61.86 yrs, standard deviation SD = 14.78 yrs; 15 female, 13 male; nine bilateral CI users). Twenty-one of them also participated in our previous study on the development and validation of the eCAP PLV measurement (He et al. 2024), and their PLV data were reused in the current study. All participants were native speakers of American English and used a Cochlear^® Nucleus^™ device (Cochlear Ltd., New South Wales, Australia). All participants had a full 22-electrode insertion with their devices, as confirmed by postoperative computerized tomography scans. Only one ear was tested in each participant. For bilateral users, the tested ear was decided through randomization. With the exception of two participants (A075 and A085), none of the participants had any functional acoustic hearing in either ear. The lack of functional acoustic hearing was self-reported by the patients and was verified by pure tone audiometric thresholds of 70 dB HL or higher, averaged across octave frequencies between 250 Hz and 8,000 Hz. For the participants with functional hearing on the opposite side, an ear plug was used to minimize its potential contribution during the AzBio speech testing. Demographic information and hearing loss etiology of the participants are listed in Table 1. All participants provided written informed consent at their initial visit to the lab prior to data collection and were compensated for their time. The study was approved by the Biomedical Institutional Review Board at the corresponding author’s institution (No. 2017H0131).

Table 1.

Demographic information of all participants.

Participant ID	Sex	Ear tested	Age at testing (yrs)	Internal device and electrode array	Hearing loss etiology	Electrodes tested
A003	M	L	61.79	CI 512	Sudden SNHL	3, 9, 14, 20
A006	M	L	69.02	CI 512	Meniere’s disease	3, 9, 15, 18
A008	F	R	67.57	CI 24RE(CA)	Hereditary	3, 9, 15, 21
A009	F	L	38.01	CI 24RE(CA)	Head Trauma	3, 9, 15, 21
A017	M	R	62.53	CI 24RE(CA)	Unknown	8, 12, 15, 18
A019	F	L	55.36	CI 532	Unknown	4, 9, 15, 21
A020	M	R	62.75	CI 522	Head Trauma	6, 9, 18, 20
A024	M	R	36.79	CI 24RE(CA)	Unknown	3, 9, 15, 21
A029	F	R	59.60	CI 24RE(CA)	Unknown	3, 9, 15, 21
A030	F	R	65.57	CI 532	Unknown	3, 9, 15, 21
A034	M	R	70.43	CI 532	Unknown	3, 9, 15, 21
A035	F	L	76.62	CI 422	Unknown	4, 9, 15, 21
A040	M	L	62.55	CI 632	Unknown	3, 9, 15, 21
A047	F	R	66.90	CI 24RE(CA)	Unknown	3, 15, 21
A052	M	L	67.63	CI 532	Vestibular Schwannoma	3, 9, 15, 21
A061	F	R	84.04	CI 532	Unknown	3, 7, 10, 17
A065	F	L	58.65	CI 532	Unknown	6, 9, 15, 21
A066	F	L	75.65	CI 622	Unknown	6, 9, 15, 21
A070	M	R	69.77	CI 632	Head Trauma	3, 9, 15, 21
A074	M	R	83.70	CI 632	Unknown	3, 9, 15, 21
A075	M	R	51.73	CI 632	Unknown	3, 9, 15, 21
A077	F	L	57.89	CI 632	Unknown	3, 15, 18
A079	M	L	57.46	CI 532	Usher	3, 9, 15, 21
A080	M	L	79.86	CI 632	Unknown	3, 9, 15, 21
A085	F	L	40.47	CI 612	Vestibular Schwannoma	6, 15, 21
A087	F	L	20.39	CI 522	DFNB8/10	3, 9, 15, 21
A091	F	L	52.15	CI 612	Unknown	5, 9, 15, 21
A093	F	R	77.29	CI 24RE(CA)	Unknown	9, 12, 16

M, male; F, female; L, left; R, right; SNHL, sensorineural hearing loss; CA, contour advance; DFNB8, deafness, autosomal recessive 8; DFNB10, deafness, autosomal recessive 10.

Stimuli and Apparatus

For eCAP recordings, the stimulus was a charge-balanced, cathodic leading biphasic pulse with a pulse-phase duration of 25 μs. The IPG between the cathodic and anodic phases, as well as the presentation levels, varied across measurements, as detailed in the “Procedures” section. All eCAP recordings were performed using the neural response telemetry function implemented in the Custom Sound EP software v6.0 (Cochlear Ltd., New South Wales, Australia).

For speech perception tests, the stimuli were meaningful sentences (e.g., “The vacation was cancelled on account of weather.”) from the AzBio sentence corpus (Spahr et al. 2012), recorded by two female and two male native American English speakers. The sentences presented to each participant and for each condition were evenly distributed across the four speakers. The background noise was a ten-talker babble presented at two signal-to-noise ratios (SNRs): +10 dB and +5 dB. All stimuli were delivered via a loudspeaker (RadioEar Corporation, PA) placed 1 m in front of the participant at 0° azimuth in a sound-attenuated booth.

Procedures

Testing Electrodes

The default testing sites for eCAP measurements were electrodes 3, 9, 15, and 21 (i.e., e3, e9, e15 and e21). These electrodes were selected to cover a wide range along the array with relatively equal numerical separations in between, while keeping the testing time reasonable. In the case of an open- or short-circuit or lacking a measurable eCAP at the maximum comfortable level due to high impedance at a default electrode, a nearby alternative electrode was tested. In rare cases when multiple neighboring electrodes are untestable, the sites were selected to cover a majority of the testable regions while keeping relatively equal separations in between. Four participants (A047, A077, A085, and A093) were tested at only three electrodes due to time constraints. The electrodes tested for each participant can be found in Table 1.

Behavioral C Level Measures

The maximum comfortable level (C level) for eCAP stimuli at each IPG level (7 μs and 42 μs) was determined via subjective rating using an ascending procedure. Prior to the measurement, participants were shown a visual scale of 1 (“barely audible”) to 10 (“very uncomfortable”) and were instructed to give a loudness rating using verbal responses or hand gestures following each stimulus presentation. Each presentation consisted of five pulses delivered at a probe rate of 15 Hz. The stimuli were first presented at a relatively low level and gradually increased in steps of 3–5 clinical levels (CLs) until a rating of “7” was reached, then in steps of 1–2 CLs until a rating of “8” was reached. The lowest level that corresponds to a rating of “8” (“maximal comfort”) was recorded as the behavioral C level.

eCAP Measures

The eCAP was measured using a two-pulse, forward-masking paradigm (Brown et al., 1990), where the masker pulse was always presented at 10 CLs higher than the probe pulse. The masker pulses were delivered at the testing electrode, and the eCAP responses were recorded two electrodes away from the testing electrode in the apical direction. There was an exception for electrode 21, which was recorded two electrodes away in the basal direction (i.e., electrode 19). The probe pulses were presented with a masker-probe interval of 400 μs. The total number of trials in each stimulation sequence differed across measurements, as detailed below. Responses were recorded at a sampling rate of 20,492 Hz with a sampling delay of 122 μs, an amplifier gain of 50 dB, and a monopolar-coupled stimulation mode.

Measure of Neural Survival: The IPGE_slope

In this study, the IPGE_slope was operationally defined as the numerical difference (in μV/dB, where dB was calculated in reference to 1 nC) between the AGF slopes with IPGs of 42 μs and 7 μs, and therefore, its measurement involved acquiring an AGF and calculating its slope at each IPG level for each participant. The IPG of 7 μs was chosen because it was the shortest IPG value available for eCAP recording in the Custom Sound EP software, and the IPG of 42 μs was selected to allow for comparisons with previous studies on IPG effects in CI users with a Cochlear^® device (He et al. 2020; Yuan et al. 2022). For both IPG levels, the maximum presentation level of the stimuli was the behavioral C level measured with an IPG of 42 μs. To acquire AGFs, the eCAP measurement started at the maximum presentation level and decreased in steps of 1 CL for five steps, then in steps of 5 CLs until no peaks could be visually identified in the eCAP waveform, i.e., when the threshold is reached. Additional presentation levels in steps of 1 CL for five steps near and above the eCAP threshold were tested. At each presentation level, the eCAP waveform was acquired by averaging the raw responses to 50 pulses with a probe rate of 15 Hz. The visual identification of eCAP peaks, or lack thereof, was performed by the experimenter at the time of testing. To minimize the potential differences in criteria for identifying peaks across experimenters, one expert researcher (the corresponding author) inspected and finalized the peak selection for all data after the experiment sessions.

The AGF slope at each IPG level was calculated using the window method developed by Skidmore et al. (2022), where linear regressions were performed on sliding windows along a resampled AGF, and the largest slope among all windows was regarded as the AGF slope. The window method was suitable for calculating the slope of AGFs with various shapes, both in animals and human listeners. For the results in human listeners, the correlation between AGF window slope and speech perception outperformed traditional fitting methods such as linear or sigmoid fitting. In the present study, we focused on the IPGE_slope results calculated using the window method in the linear-logarithmic I/O scale of μV/dB, following the recommendations by Skidmore et al. (2022). However, since linear fitting and the linear-linear I/O scale of μV/nC were commonly used in previous research on IPG effects (Schvartz-Leyzac & Pfingst 2018; Zamaninezhad et al. 2023), and that fitting method and I/O scale may affect how well eCAP measurements correlate with neural health and speech perception (Brochier et al. 2021; Skidmore et al. 2022), we also calculated the IPGE_slope on a linear-linear scale (μV/nC) with both linear and window fitting methods to evaluate the robustness of the results across fitting methods and I/O scales, briefly reported at the end of the Results section. All calculations were performed using MATLAB 2021b (MathWorks, MA).

Measure of Neural Synchrony: The PLV

For each participant and electrode, the PLV was derived from 400 eCAP trials measured using biphasic pulses with an IPG of 7 μs presented at the behavioral C level, using the method developed by He et al. (2024). The IPG of 7 μs was selected for PLV measurement because it is the default and most used IPG in the programming of Cochlear^® CI devices. The recording at each electrode took about 18 minutes to complete. The PLV is a unitless value between 0 and 1, where 0 means that the phases of the time-frequency decomposed eCAP waveforms (detailed below) were randomly distributed across trials, and 1 means that the phases were perfectly correlated. To calculate the PLV, the eCAP responses were re-referenced to the first sample of each trial and time-frequency decomposed using the Fast Fourier Transform (FFT) at six linearly spaced frequencies (788.2, 1576.3, 2364.4, 3152.6, 3941.0, and 4729.2 Hz) and divided into six partially overlapped time frames with an onset-to-onset interval of 48.8 μs and a length of 1561.6 μs. At each frequency and within each time frame, the unit vectors representing the phases of the 400 individual trials were averaged, and the length of the averaged vector was taken as the time-frequency-specific PLV. The formula for calculating the PLV at the time $t$ and the frequency $f$ based on $N$ individual trials is:

$$PLV\left(f,t\right)=\left|\frac{1}{N}\sum _{k=1}^N\frac{F_k(f,t)}{\left|F_k(f,t)\right|}\right|$$

The PLV of the electrode was then obtained by averaging all time-frequency-specific PLVs calculated from the eCAP responses at that electrode. The PLV results measured in one example participant are shown in Figure 4. The time-frequency decomposition and the calculation of the PLV were performed using MATLAB R2021b and the newtimef.m function from EEGLAB v2022.1 (Delorme & Makeig 2004).

Figure 4.

PLV values measured in one participant (A019). The time-frequency specific PLV values are shown in the heatmaps in the lower panels. The overall PLV values of individual electrodes are listed in the upper panels. Bolded black lines represent eCAP responses averaged across 400 trials. Responses in individual trials are plotted with gray lines.

Measure of Speech Perception: AzBio Scores

Each participant was tested with AzBio sentences (Spahr et al. 2012) under three conditions: in quiet and in a ten-talker babble background noise with SNRs of +10 and +5 dB, respectively. The sentences were presented at 60 dB SPL in all conditions. The participants were tested unilaterally on the side with CI. Specifically, bilateral users were required to take off the CI on the opposite side. For the two participants with functional acoustic hearing in the opposite ear, an ear plug was used to minimize the contribution of acoustic hearing. For each participant and condition, a sentence list was randomly selected from Lists 1–8 of the AzBio corpus, each consisting of 20 sentences. For each participant, different word lists were used for different conditions. Participants were instructed to repeat back after each sentence and were encouraged to guess if they were unsure about what they heard. An experimenter recorded the number of words they correctly repeated in each sentence. The AzBio score was calculated as the number of words in the list correctly repeated by the participant, divided by the total number of words in the list. Following the standard clinical practice for scoring the AzBio sentence test, all words in the list, including prepositions, counted towards the score.

Averaging IPGE_slope and PLV Values across Electrodes

As demonstrated in He et al. (2023a), using eCAP results measured at single CI electrode locations to correlate with auditory perception outcomes in CI users can lead to inaccurate conclusions. Therefore, the values of the IPGE_slope and the PLV were averaged across all tested electrodes for each participant as an overall representation of CN health across the cochlea (He et al. 2023a). Both weighted and unweighted averages were calculated for both parameters. To calculate the weighted average, the results were weighted based on the frequency importance function (FIF) derived from AzBio scores under various spectral filtering conditions and SNRs in NH listeners (Lee & Mendel 2017). The FIF was fitted to a four-parameter Weibull function in SigmaPlot v15 (Grafiti LLC, CA). For each participant, the importance weight of each test electrode was calculated using the fitted Weibull function based on the electrode’s central frequency derived from the frequency-to-electrode table of the participant’s everyday programming map. The individual values of the empirically measured AzBio FIF and the fitted curve are shown in Figure 1. The unweighted average was calculated as the arithmetic means of the IPGE_slope and the PLV values across all tested electrodes for each participant.

Figure 1.

Individual AzBio FIF values (black circles) measured by Lee and Mendel (2017) and the fitted Weibull function (gray line)

Data Analysis

The IPGE_slope analyses reported in the Results section are based on the AGF slopes obtained from the window fitting method in the unit of μV/dB, unless otherwise stated. The electrode-specific IPGE_slope and PLV values were compared across electrode locations using linear mixed-effect models. Pairwise comparisons between the electrodes were performed using the Tukey method for multiple comparisons adjustment.

The IPGE_slope and PLV values were then averaged across the tested electrodes, weighted by the AzBio FIF, as representations of the overall CN neural health. The pairwise relationships between the IPGE_slope, the PLV, and age at testing were assessed using either Pearson or Spearman correlation tests for variable pairs depending on the results of the Shapiro-Wilk normality test. A multiple linear regression model with AzBio score as the outcome and the IPGE_slope and the PLV as predictors was performed to evaluate the associations among the three variables under all testing conditions (quiet, +10 dB, and +5 dB SNR). Participant’s age was included in the regression model as a covariate to control for the potential effects of advanced age on speech perception and/or eCAP measurements, as have been demonstrated in CI users (Roberts et al. 2013; Sladen & Zappler 2015; Xie et al. 2019; Jahn & Arenberg 2020a). Interactions were included between testing condition with age, IPGE_slope, and PLV. The analysis included an unstructured correlation matrix on the residuals to account for unequal variances by condition and the repeated measures with each individual having responses in all three testing conditions. The tests of interest were contrasts for the relationships between AzBio with IPGE_slope and with PLV for each testing condition. Model assumptions were evaluated by visual inspection of the residuals with no violation of assumptions being detected. Effect sizes were computed as $\frac{\beta }{\sqrt{{\sigma }_b^2+{\sigma }_e^2}}$ where ${\sigma }_b^2$ is the random intercept variance and ${\sigma }_e^2$ is the residual variance. This effect size (d) can be interpreted similarly to Cohen’s d where 0.2 is a small effect, 0.5 is a medium effect and 0.8 is a large effect.

Correlation tests were performed in JASP v0.18.3 (JASP Team 2024), and the regressions were performed in R v4.4.1 (R Core Team 2024), with the lme4 (Bates et al. 2015), nlme (Pinheiro et al. 2024), emmeans (Lenth 2024), and lmerTest (Kuznetsova et al. 2017) packages used for the linear mixed-effect models.

RESULTS

The individual IPGE_slope and PLV values measured at each electrode are shown in panels a and b of Figure 2, respectively, with the range, mean, and standard deviation (SD) listed in Table 2. To assess the potential differences in the two measurements across electrodes, two linear mixed-effect regressions were performed with the IPGE_slope and the PLV as the outcome variables, the electrode location as the within-subject fixed effect, and participant as the random effect. The four categories of electrode locations corresponded to and were named after the four default testing locations: e3, e9, e15, and e21. If other electrodes were measured in lieu of the default electrodes, they were assigned to one of the categories based on their locations relative to the other electrodes tested in the same participant. For example, among the four electrodes tested in participant A017, e8 was the most basal electrode and therefore categorized as “e3” in the regression model. The category “e3” was used as the reference level in the regressions. The degrees of freedom were estimated using the Satterthwaite’s method. The results reveal an overall effect for PLV across electrodes (F_{(3, 77.4)} = 5.44, p = .002). Focusing on the pairwise comparisons, only two comparisons showed a statistically significant difference in PLVs which were e3 compared with the PLVs measured at e15 (mean difference = −0.092, t_(77.2) = 3.47, p = .005, d = −0.59) and e21 (mean difference = −0.087, t_(77.2) = 3.27, p = .009, d = −0.56). The other comparisons were not significantly different which were e3 with e9 (mean difference = −0.035, t_(77.7) = 1.28, p = .578, d = −0.23), e9 with e15 (mean difference = −0.057, t_(77.5) = 2.09, p =.167, d = −0.36), e9 with e21 (mean difference = −0.051, t_(77.5) = 1.89, p =.240, d = −0.33), and e15 with e21 (mean difference = 0.005, t_(77.0) = 0.20, p =.997, d = 0.03). For the analysis of electrode location effect on the IPGE_slope, two data points (A024 e9 and A087 e21) were excluded from the analysis because their IPGE_slope values were more than three SDs away from the mean of the corresponding electrode location. The IPGE_slope did not significantly differ across electrodes (F_{(3, 75.2)} = 2.22, p = .092). In the subsequent sections, the IPGE_slope and the PLV refer to the weighted averages of the corresponding values across electrodes for individual participants unless otherwise stated. It is worth noting that no outliers were removed in the subsequent analysis, as all PLV and IPGE_slope values fell within three SDs from the mean when averaged across all tested electrodes of each participant.

Figure 2.

Individual values of (a) the IPGE_slope and (b) the PLV by electrode location. The four categories from basal to apical locations corresponded to, and were named after, the default electrodes e3, e9, e15, and e21, respectively. Values measured at non-default electrodes were categorized based on their locations relative to the other electrodes tested in the same participant. Boxes show the range between the first and the third quartile of the data values. The horizontal bars inside the boxes represent the median. The vertical whiskers show the range of values that are within 1.5 interquartile range (IQR) from the boxes. Asterisks denote pairwise comparisons with significant differences (p < .05). N.S., no significant differences.

Table 2.

Descriptive statistics of the IPGE_slope and the PLV values measured at different electrode locations.

Electrode location	e3	e9	e15	e21
PLV	Min = 0.117	Min = 0.092	Min = 0.109	Min = 0.160
	Max = 0.590	Max = 0.734	Max = 0.782	Max = 0.808
	0.318 (0.137)	0.360 (0.161)	0.409 (0.160)	0.404 (0.166)
IPGEslope	Min = −20.003	Min = −4.982	Min = −8.306	Min = −10.068
	Max = 52.438	Max = 41.041	Max = 45.231	Max = 41.102
	7.394 (18.073)	12.136(16.650)	7.668 (14.652)	3.701 (2.227)

The IPGE_slope values were calculated from AGF slopes in μV/dB fitted using the window method. Two electrodes were identified as outliers for IPGE_slope (more than three SDs away from the mean) and therefore excluded from the descriptive statistics. Values in each cell are listed in the format of “min, max, mean (SD)”.

Correlations between the IPGE_slope, the PLV, and Age

The eCAP AGFs acquired at different IPG durations from one example participant (A019) are shown in Figure 3. Time-frequency specific PLV values of the same participant, measured at the C level of each electrode, are illustrated in Figure 4. Figure 5 shows the standardized values of IPGE_slope and PLV of individual participants averaged across electrodes (panel a), along with their age (panels b-c). All values were within three SDs from the mean. Spearman correlation tests showed that that the association between the IPGE_slope and the PLV did not reach statistical significance (rho₍₂₆₎ = 0.340, p = .077). Although not a focus of this study, we assessed the relationships between age and the two CN health measures in the current dataset, where most of the participants fall within the age range of 50–85 years. Spearman correlation test results revealed a significant negative correlation between the IPGE_slope and age (rho₍₂₆₎ = −0.583, p = .001). Pearson correlation test results showed that PLV was not significantly correlated with age (r₍₂₆₎ = −0.068, p = .730).

Figure 3.

The AGFs measured with IPGs of 7 μs and 42 μs in one participant (A019), where the maximum stimulation level was set to the C level measured with a 42-μs IPG at each electrode. The amplitudes were normalized by dividing the amplitude of each trial by the maximum amplitude among the trials with a 7-μs IPG. Please note that the ranges of the axes are different across panels due to large variabilities in eCAP thresholds, C-levels, and amplitudes across electrodes.

Figure 5.

Correlations between (a) the IPGE_slope and the PLV, (b) the IPGE_slope and age, and (c) the PLV and age. The axes were standardized by converting into z-scores to improve comparability across panels. Each symbol indicates the result measured in one participant. The results of correlation tests are shown in each panel.

The Associations among the IPGE_slope, the PLV and Speech Perception Scores

The relationships between AzBio results in quiet and noise and eCAP measurements are illustrated in Figure 6. The results revealed significant associations between the PLV and AzBio scores in the noise +5 dB SNR condition (slope = 69.2, t₍₂₄₎ = 2.36, p = .027) after adjusting for IPGE_slope and age, where larger PLVs (better neural synchrony) are associated with higher AzBio scores, but not in the noise +10 dB SNR (slope = 68.7, t₍₂₄₎ = 1.61, p = .121) or the quiet condition (slope = 27.6, t₍₂₄₎ = 0.88, p = .390). No significant association between the IPGE_slope and AzBio scores was observed in any testing conditions (p > .10 in all cases) after adjusting for PLV and age. Detailed results of the linear regression models are available in Table 3.

Figure 6.

The residuals of AzBio results measured in quiet and in noise with +10 and +5 dB SNRs calculated using reduced linear regression models, plotted against eCAP measurements. In the top panels, vertical axes represent the residuals of “AzBio score ~ PLV + age” to highlight the relationships between IPGE_slope and AzBio results. In the bottom panels, vertical axes represent the residuals of “AzBio score ~ IPGE_slope + age” to highlight the relationships between PLV and AzBio results. The unstandardized beta and p-values were obtained from the linear regression (Table 3) for the corresponding eCAP measurements shown in the horizontal axis of each panel. Each symbol represents the AzBio result measured in one participant.

Table 3.

Results of linear regression model examining the relationships between the IPGE_slope, the PLV, and AzBio scores measured in quiet and in two noise conditions.

Listening condition	Predictor	Unstandardized β (SE)	t	p	Partial correlation with outcome
Quiet	IPGEslope	0.492 (0.375)	1.311	.2025	0.258
	PLV	27.6 (31.5)	0.876	.3898	0.176
	age	0.447 (0.378)	1.184	.2486	0.235
Noise, +10 dB SNR	IPGEslope	−0.053 (0.507)	−0.104	.9184	−0.021
	PLV	68.7 (42.7)	1.610	.1205	0.312
	age	−0.438 (0.511)	−0.856	.4004	−0.172
Noise, +5 dB SNR	IPGEslope	−0.278 (0.349)	−0.796	.4342	−0.160
	PLV	69.2 (29.3)	2.359	.0270*	0.434
	age	−0.740 (0.352)	−2.105	.0461*	−0.395

The IPGE_slope values were calculated from AGF slopes in μV/dB fitted using the window method. “Partial correlation with outcome” is listed as a measure of effect size and refers to the partial correlation coefficient (Pearson’s r) between one predictor and the outcome, after adjusting for other predictors.

Electrode Weighting by AzBio FIF

To qualitatively evaluate whether weighting the IPGE_slope and the PLV by the AzBio FIF modifies their relationships with AzBio scores, in a separate set of linear regressions, we used unweighted averages of the IPGE_slope and the PLV in lieu of their weighted counterparts. Overall, the result patterns and statistical significance remained the same in the unweighted version of the linear regressions, where the relationships between the PLV and AzBio scores were statistically significant in the noise +5 dB SNR condition (slope = 75.2, t₍₂₄₎ = 2.47, p = .021) but not in the noise +10 dB SNR (slope = 77.7, t₍₂₄₎ = 1.74, p = .094) or the quiet condition (slope = 31.7, t₍₂₄₎ = 0.94, p = .355), and no significant relationship between the IPGE_slope and AzBio scores was observed in any of the tested conditions (p > .10 in all cases).

Comparisons across AGF Fitting Methods and I/O Scales

To evaluate the potential effects of AGF fitting methods and I/O scales on the results, we recalculated the IPGE_slope based on AGF slope used linear fitting and the window method, both in the linear-linear I/O scale of μV/nC. The statistical tests described above were then reperformed using each version of the IPGE_slope values. The result patterns and statistical significance stayed largely the same across fitting methods and I/O scale, with a couple exceptions with regards to statistical significance, using an alpha level of 0.05: When the IPGE_slope values were calculated using window fitting in μV/dB, the correlation between the IPGE_slope and the PLV was not significant (rho₍₂₆₎ = 0.340, p = .077), and the association between the PLV and the AzBio scores in noise with +10 dB SNR was not significant (t₍₂₄₎ = 1.61, p = .120). When the IPGE_slope values were calculated using linear fitting in μV/nC, the correlation between the IPGE_slope and the PLV was not significant (rho₍₂₆₎ = 0.352, p = .067), and the association between the PLV and the AzBio scores in noise with +10 dB SNR was significant (t₍₂₄₎ = 2.35, p = .028). When the IPGE_slope values were calculated using window fitting in μV/nC, the correlation between the IPGE_slope and the PLV was significant (rho₍₂₆₎ = 0.490, p = .009), and the association between the PLV and the AzBio scores in noise with +10 dB SNR was not significant (t₍₂₄₎ = 1.62, p = .119). It is worth noting that the terms “significant” and “non-significant” are used to describe whether the p values are smaller than the alpha level of 0.05, and do not suggest the results labeled with “significant” and “non-significant” are fundamentally different from each other. Overall, the analysis of the current dataset shows that the result patterns are generally consistent across the three combinations of AGF slope calculation methods and I/O scales. Further research is needed to systematically compare various combinations of AGF fitting methods and I/O scales to identify whether one or some of them might be better at representing CN survival status compared to the others.

DISCUSSION

This study assessed the relationships between two CN health measures, the IPGE_slope and the PLV, and evaluated their contributions to speech perception in quiet and noise in postlingually deafened adult CI users. We hypothesized that the IPGE_slope and the PLV are two independent measures predictive of speech recognition scores in quiet and in noise, respectively. The hypotheses were partially supported by the results showing that the IPGE_slope and the PLV had different relationships with electrode location and age, and that the speech perception scores measured in noise with +5 dB SNR were positively associated with the PLV after controlling for the effect of age. However, contrary to our hypothesis, we did not observe significant associations between the IPGE_slope and speech perception measured either in quiet or in noise after controlling for the effect of age. The result patterns stayed largely the same regardless of fitting methods (window vs. linear) or I/O scales (linear-linear vs. linear logarithmic) used in AGF slope calculation.

Peripheral Neural Survival and Synchrony

The IPGE_slope and the PLV demonstrated different variabilities across electrode locations and age, suggesting that they were measuring different aspects of CN health at least to some extent. These results are aligned with the physiological process of neural degeneration of the CN that occurs in patients with sensorineural hearing loss and are consistent with histological results of human temporal bone studies. Detailed explanations are provided below.

The deterioration of the bipolar SGNs starts at the peripheral axon, which connects the SGN soma to the organ of Corti in the cochlea (Xing et al. 2012). The degeneration of the peripheral axon starts with deterioration of the myelin sheath (Xing et al. 2012) and is followed by axonal dystrophy, segmental swelling (ballooning), and ultimately, loss of the peripheral axon (e.g., Chen et al. 2006; Heshmat et al. 2020; Kumar et al. 2022; Liu et al. 2015; Nadol 1990; Wu et al. 2019; Wu et al. 2023). Even after the complete loss of the peripheral axon, the soma and the central axon of the SGN (i.e., unipolar SGN) can survive for decades (Rask-Andersen et al. 2010), and the proportion of unipolar SGNs increases steadily with age (Wu et al. 2023). This process gives rise to a key difference between acoustic and electric hearing. While the loss of peripheral SGN axons likely contributes to impairment in acoustic hearing (Wu et al. 2019; Wu et al. 2020; Wu et al. 2021), electric hearing can be achieved even without the peripheral axons, as the stimulation can directly reach the somata and/or the central axons of SGNs (Javel & Shepherd 2000). Both axonal dystrophy and demyelination alter many neural properties (e.g., Tasaki 1955; Waxman & Ritchie 1993), which reduce the synchronized discharge of affected CN fibers (e.g., Gonzalez 2019; Heshmat et al. 2020; Kim et al. 2013; Tasaki 1955) as well as across the population of CN fibers (Kandel 2002). Additionally, loss of the peripheral axon and altered membrane properties can shift the action potential initiation site distally to the SGN soma or central axon in electrical hearing (e.g., Hartmann et al. 1984; Javel & Shepherd 2000; Van Den Honert & Stypulkowski 1984), further reducing discharge synchronization across CN fibers (Javel & Shepherd 2000). As explained in He et al. (2024), the PLV is influenced by temporal jitter in spike firing of individual CN fibers as well as discharge synchronization among a population of activated CN fibers across multiple stimulations, which is an ideal tool to assess neural synchrony in the CN in CI patients.

The results of human temporal bone studies demonstrated an age-related loss in both SGN soma and the peripheral axon in adult listeners: Compared with the apical regions, the basal area of the cochlea shows the greatest loss in both SGN soma and peripheral, with peripheral axon showing a substantial greater loss than SGN soma axon (Liu et al., 2015; Makary et al., 2011; Wu et al., 2019; Wu et al., 2023). In one CI case who was 70 years of age at the time of death, a total loss of peripheral axons for all surviving SGNs at the basal half of the cochlea was observed. In contrast, peripheral axon counts in the apical cochlea are similar to those of age-matched, nonimplanted listeners (O’Malley et al. 2024).

Based on the literature reported above, smaller PLVs are expected at more basal electrode locations where substantial degeneration in the peripheral axon occurs in middle-aged and older CI patients, which make up a majority of the patient group tested in this study. Consistent with this expectation, our data showed an electrode location effect on the PLV, with significantly smaller PLVs measured at the basal electrode location than apical locations. Our results did not show the effect of age on the PLV. However, this result does not necessarily conflict with those reported in Wu et al. (2019) and Wu et al. (2023) due to the difference in cases/participants included in these two studies. Specifically, the cases reported in Wu et al. (2019) and Wu et al. (2023) were non-CI listeners with acoustic hearing, while the current study focused on CI users. Further studies using animal models, histological studies in human temporal bones and computational modeling techniques are warranted to determine the crucial factors (e.g., age, surgical trauma, etiology) for neural health of the CN in implanted cases. Finally, the lack of association between the PLV and age calls for further investigations on CI users with a wide age range to determine the contribution of poor neural synchrony in the CN to speech perception deficits in noise experienced by many older CI users.

Unlike the PLV, the IPGE_slope is negatively correlated with participant age but not affected by electrode location. The significant correlation between the IPGE_slope and age is consistent with the age-related decline in SGN counts in humans reported in different studies (e.g., Liu et al. 2015; Kusunoki et al. 2004; Makary et al. 2011; Wu et al. 2019; Wu et al. 2023) . The lack of difference in the IPGE_slope across electrodes is consistent with histological results of some human temporal bone studies reporting similar amounts of SGN loss across cochlear locations (Makary et al., 2011; Wu et al., 2023). However, substantial variations in SGN loss at different cochlear locations have also been reported in the literature (e.g., Hinojosa & Marion 1983; O’Malley et al. 2024). For example, a temporal bone study by Seyyedi et al. (2013) reported the SGN counts at four segments along each of 26 ears implanted with a Cochlear^® device. Contrary to the lack of significant electrode location effect on IPGE_slope in the current study, the SGN counts were substantially decreased at the basal location compared to middle and apical locations in Seyyedi et al. (2013). Possible factors accounting for this difference include the older age of the cases included in Seyyedi et al. (2013), which ranged from 42 to 92 years (mean: 73.48 yrs; SD: 12.22 yrs), and the greater degrees of sensorineural hearing loss in these cases due to the stricter CI candidacy criteria used decades ago compared to our study. Both factors result in greater SNG loss at the base of the cochlea than more apical areas. Unfortunately, the numeric relationship between IPGE_slope and SGN count cannot be determined by comparing across studies due to different participants included in our study vs those included in Seyyedi et al. (2013).

It should be noted that both the PLV and the IPGE_slope are eCAP-derived measurements. The eCAP is a near-field recording of the aggregated activity of a group of electrically stimulated CN fibers. As a result, it is affected by firing synchrony both within and across CN fibers as well as the number of activated CN fibers. In addition, each eCAP response used in AGF slope calculation in this study was an average of 50 single sweeps. Its amplitude can be affected by the variability in the peak latencies of the single sweeps, resulting in covariation between IPGE_slope, an amplitude-based eCAP measurement, and the PLV. Therefore, the trend of positive correlation between the PLV and IPGE_slope values observed in this study is not surprising. However, the differences in their associations with electrode location and age indicate that the PLV and the IPGE_slope reflect different aspects of CN health. Further studies using computational modeling techniques are warranted to determine the biological underpinnings of these measures and how different types and degrees of neural degeneration of the CN affect these measures.

Peripheral Neural Synchrony Contributes to Speech Perception in Noise

We observed significant positive associations between the PLV and AzBio scores measured in noise with +5 dB SNR, suggesting that neural synchrony is important for speech perception in high levels of background noise. These results are consistent with and complement the data from our previous study on CI users (He et al. 2024; Skidmore et al. 2023). This observation also agrees with and expands the findings by Harris et al. (2021), where the CAP-derived PLV was shown to be a strong predictor of the perception of time-compressed speech and speech in noise in NH listeners. These results highlight the importance of neural synchrony in the CN in speech perception in listeners with various hearing profiles. The difference between the PLV effects on speech perception across listening environments is likely due to the heightened importance of temporal cues for speech perception in noise (Nie et al. 2006), and the lack of neural synchrony is associated with poor performance in psychophysical tasks requiring fine temporal perception (Zeng et al. 2005). Consistent with the previous findings, He et al. (2024) observed that the PLV is correlated with behavioral GDTs, measured using stimuli with clinically relevant stimulation rate of 900 Hz: Within the same subject, electrodes with higher PLV are associated with lower (better) GDTs. The current results provide further validation for using the eCAP-derived PLV as a measure of neural synchrony in adult CI users. Future studies could measure neural synchrony in the CN in implanted deafened animals using both the eCAP-derived PLV and traditional single-neuron recording methods (e.g., Seki & Eggermont 2003) to further evaluate the validity of using the PLV as an index for neural synchrony in the CN.

Lack of Significant Associations between the IPGE_slope and Speech Perception

The lack of significant associations between the IPGE_slope and speech perception across listening conditions, after controlling for age, does not support our hypothesis on the contribution of neural survival to speech perception. . This result is consistent with those reported by Imsiecke et al. (2021) but not consistent with the observations in a recent study by Zamaninezhad et al. (2023). This discrepancy could be due to some critical differences between the testing materials and methods used in these studies. Specifically, the German matrix sentences in Zamaninezhad et al. (2023) consisted of syntactically correct but semantically unpredictable sentences, while the AzBio corpus used in the current study and the Hochmair-Schulz-Moser (HSM) sentence test used in Imsiecke et al. (2021) consisted of meaningful sentences on everyday topics; the Freiburg monosyllable test in Zamaninezhad et al. (2023) required the listeners to repeat a single word at a time, while the AzBio and HSM tests required them to repeat a full sentence in each trial. These differences allow the AzBio and HSM sentence tests to better simulate real-life listening situations, but also leave room for the effect of cognitive factors such as working memory (Ingvalson et al. 2015) to modulate the speech perception performance on top of CN health condition. Therefore, it is possible that the association between the IPGE_slope and speech perception, if any, has been eclipsed by the individual differences in cognitive factors in the present study. Future research could test both cognitive abilities and the IPGE_slope in CI users from various age groups, to evaluate their relative contributions to speech perception at each stage of the aging process. The involvement of cognitive factors could also be assessed through non-invasive neurophysiological measures, such as electroencephalogram (EEG) and functional near-infrared spectroscopy (fNIRS).

In addition, the IPGE_slope values in Zamaninezhad et al. (2023) were calculated as the difference between the AGF slopes measured with IPGs of 30 μs and 2.1 μs, but the present study used the AGF slopes with IPGs of 42 μs and 7 μs for the same calculation. It is possible that the sensitivity of the IPGE_slope as an index of CN survival varies with the IPG levels used for eCAP recordings, and further investigation is warranted for optimizing the parameters in the IPGE_slope measurement for the purpose of representing CN health condition.

It is important to point out that the lack of a significant association between the IPGE slope and speech perception in quiet observed in this study should not be interpreted as evidence for the absence of a relationship between neural survival and speech perception. Further studies involving a larger cohort of CI participants are needed to confirm this finding.

Frequency Importance Function in Speech Perception Measures

While not a central focus of the present study, in the models evaluating the contributing factors to AzBio scores, we calculated the values of the IPGE_slope and the PLV as both unweighted averages across the tested CI electrodes and weighted averages based on the AzBio FIF. The results were similar regardless of whether the AzBio FIF weights were applied, which seemingly contradicts the definition of the FIF (Lee & Mendel 2017). A possible explanation is that the FIF used in the current study was measured in NH listeners (Lee & Mendel 2017), and thus may not be fully generalizable to CI users due to the differences in the weighting of frequency bands in speech perception between CI users and NH listeners (Sladen & Ricketts 2015) and potentially larger individual differences in CI users than NH listeners (Mehr et al. 2001; Bosen & Chatterjee 2016). Furthermore, the weights of adjacent frequencies could differ considerably in some FIFs (Healy et al. 2013), so that estimating the weights of CI electrodes based on a smooth curve fitted to discrete values of the empirically measured AzBio FIF may have limited validity, even if the overall FIF shape of CI users is similar to that of NH listeners. Future research could develop and validate methods for measuring the FIF in CI users and test whether weighting eCAP-derived indices by the CI-based FIF can improve their capability to predict speech perception. In addition, the variation pattern of eCAP-derived indices across electrode locations could be compared with phoneme confusion matrices in low-context speech perception tasks (e.g., single syllables) to further investigate the potential association between peripheral neural health and speech perception.

Methodological Considerations

Fitting Method and I/O Scale

One methodological factor to consider when calculating the IPGE_slope is the fitting method and I/O scale of the AGF slopes. Skidmore et al. (2022) demonstrated that the window method is broadly applicable to various AGF shapes, and that the window fitting results calculated using linear-linear scale (μV/nC) and linear-logarithmic scale (μV/dB) were highly correlated in human listeners. To further evaluate how the I/O scale affects the calculation of AGF slope at the level of individual electrodes, we compiled the indices of the selected windows (1 to 8, where 1 and 8 correspond to the lower and higher ends of the dynamic range, respectively), and the range of stimulation levels (in CL) corresponding to the selected window of each tested electrode, when calculated both in μV/nC and in μV/dB. When averaged across all electrodes and participants, the selected ranges in CL were similar when using μV/nC (176.1–186.8 CL) and μV/dB (178.1–188.0 CL), while the selected window index differed by 1 window on average (μV/nC: 5.2, μV/dB: 6.2).

For most of the electrodes, similar window indices and CL ranges were selected when using μV/dB and μV/nC. In particular, for those with a concaved AGF shape similar to the lower half of the S-shaped sigmoid function, a larger window index (typically 7 or 8) was selected when using both μV/dB and μV/nC. However, the inherent differences in the units of dB (re 1 nC) and nC would result in different slope values. When the same or similar windows were selected, the direction of the numerical difference between AGF slopes in μV/dB and μV/nC were dependent on the stimulation level of the selected window. A qualitative observation was that when the stimulation level of the selected window was relatively high, with an upper limit around 175 CL or higher, the slope value calculated in μV/dB was generally larger than μV/nC. Alternatively, when the stimulation of the selected window was relatively low, with an upper limit around 165 CL or lower, the slope value calculated in μV/dB was generally smaller than μV/nC.

In rare cases, the selected window indices were very different between μV/dB and μV/nC (e.g., for A008R e15 with an IPG of 42 us, windows 8 and 1 were selected when using μV/dB and μV/nC, respectively). Visual inspection of their AGF curves revealed that their AGF was more linearly shaped (rather than concaved) when the amplitude was plotted against CL (a logarithm unit). The linear unit of nC changes at a smaller step when the stimulation changes by 1 CL at lower levels compared to higher levels, while the logarithmic unit of dB (re 1 nC) changes at roughly equal steps across stimulation levels. Such difference would result in larger slopes in μV/nC at lower stimulation levels compared to higher stimulation levels for AGFs with a linear shape when plotted against CL, so that a small window index (corresponding to low stimulation levels) would be selected when using μV/nC.

In short, the numerical difference in AGF slopes between window fitting using μV/dB and μV/nC is influenced by both the shape of the AGF curve and the stimulation range of the electrode. Such subtle numerical differences at the level of individual electrodes may result in different statistical significance in data analysis. However, given the overall similarity in the AGF slopes calculated using the window method in μV/nC and μV/dB, it is unlikely that the general result trends are drastically different across I/O scales when using the window fitting method, as demonstrated by the IPGE_slope results in the current study. Future research could systematically compare the IPGE_slope measures quantified in various I/O scales (i.e., linear/linear, linear/log, log/linear, and log/log) in CI users with CND and normal-sized CN to evaluate how well they reflect CN survival in human listeners.

PLV Analysis Applied to eCAP Recordings

The eCAP PLV analysis methodology used in this study follows that of He et al. (2024), which was adapted from a PLV analysis approach for click-evoked acoustic compound action potentials (CAPs) developed by Harris et al. (2021). The parameters of the analysis need to be adapted because of the difference between the morphologies of eCAPs and click-evoked acoustic CAPs. Schvartz-Leyzac et al. (2025) have recently expressed concerns about the suitability of this adapted PLV method regarding the sampling rate, time window duration, and lack of measurement of a baseline PLV before the stimulation period. However, in human CI users, the shortest possible N1-P2 interpeak latency of the eCAP response is around 0.2 ms (for a review, see He et al. 2017). In an FFT analysis, this corresponds to the shortest possible fundamental period of 0.4 ms or the highest possible fundamental frequency of 2.5 kHz. The sampling rate of Cochlear^™ Nucleus^® device is 20,492 Hz, and therefore the fundamental frequency and first few harmonics of these shortest possible eCAP waveforms fall below the Nyquist frequency of 10,246 Hz. Thus, this sampling rate, despite being the lowest among all three major CI devices, is sufficient for eCAP recording and is not a factor that could limit the sensitivity of PLV measures. Additionally, in human CI users, the eCAP occurs within the first 1200 μs after stimulus onset (Botros et al. 2007). Therefore, a recording window of 1600 μs is sufficient for capturing the entire eCAP response. It is worth noting that the duration of the time frame used in the FFT analysis, rather than the recording window of the eCAP, affects the frequency range used in the PLV calculation. As demonstrated in He et al. (2024), the number of time frames used to calculate the PLV, when the recording window is fixed at 1600 μs, is not a determining factor for the PLV method. More detailed explanations are in the “Methodological Considerations” section of He et al. (2024).

It is true that the recording window for measuring the eCAP using the neural response telemetry function implemented in the Custom Sound EP software v6.0 (Cochlear Ltd., New South Wales, Australia) does not allow for measurement of a baseline period before the stimulus presentation. However, the duration of the time window can be extended to include 64 samples (around 3,200 μs), which provides an opportunity to compare PLVs calculated for the signal (i.e., eCAP responses) with those calculated for the baseline noise recorded following the eCAP waveform. To fully understand the limitations of the PLV method, we used this extended time window to record the eCAP results in nine participants tested in this study. For the traces measured at each electrode location in these participants, PLVs were calculated for two nonoverlapping 32-sample time windows, consisting of the first and the second half of recorded samples. As has been pointed out previously, the eCAP in human CI users occurs within the first 1200 μs after stimulus onset (Botros et al. 2007) , which is fully captured by the first time window. Therefore, the PLV caculated for the first time window was defined as the “signal” PLV, and the PLV calculated for the second time window represent the PLV of the noise floor. The segments of the recorded traces corresponding to the second time window at four electrode locations in the left ear of A019 and the heat maps of time-frequency-specific PLVs are shown in Supplemental Digital Content 1. The segments of these traces corresponding to the first time window and the PLV heat maps are shown in Figure 4. The PLVs measured for the eCAP signal (i.e., the first time window) and the noise floor (i.e., the second time window) at each electrode location measured in these nine participants are shown in Supplemental Digital Content 2. The results of a LMM showed that PLVs calcaulted for the signal were significantly larger than those measured for the noise floor (F_{(1, 52.14)} = 106.60, p < .001), after controlling for electrode location and individual differences.

Stimulation Rate and the PLV

Due to device limitations, the stimulation rates that can be tested using the device of the current study were lower than those typically employed in clinical CI programming maps (≥500 pulses per second, pps). Nevertheless, as mentioned earlier in the discussion, using the same device setup and PLV calculation methods as the current study, He et al. (2024) found that behavioral gap detection thresholds measured using stimuli with 900 pps are correlated with the PLV. Specifically, within the same subject, electrodes with higher PLVs are associated with lower (better) gap detection thresholds. The result supports that neural synchrony measured by the PLV at lower stimulation rates are associated with temporal acuity at clinically relevant stimulation rates (900 pps). This result is also consistent with previous findings that discharge synchronization of CN fibers is important for encoding temporal envelope cues (Young & Sachs 1979; Joris & Yin 1992) and plays a critical role in neural representation of speech sounds presented in noise (Sachs et al. 1983; Delgutte & Kiang 1984; Heeringa & Köppl 2022).

It is worth noting that even if the PLV can be measured at 900 pps, it may not be an ideal stimulation rate to quantify neural synchrony. One possible approach for evaluating neural synchrony at clinically relevant stimulation rates is to use the modified forward masking paradigm to derive eCAP sweeps from responses to individual pulses within a pulse train, enabling calculation of the PLV (Carlyon & Deeks 2015; Hughes & Laurello 2017; Mussoi & Brown 2019). However, as explained in Chratian et al. (2025), these derived eCAP sweeps are affected by multiple factors, including discharge synchronization, discharge rate, spike amplitude of each discharge, as well as various temporal response properties, such as facilitation, adaptation, recovery from adaptation, refractoriness, and stochasticity. These factors are also affected by stimulation rate, level, and duration (Hughes et al. 2012; Hughes & Laurello 2017; Rubinstein et al. 1999). As a result, interpretation of the PLV calculated based on these derived eCAP sweeps is complicated and cannot be attributed solely to neural synchrony.

Potential Study Limitations

This study has six potential limitations. First, identifying the best parameter and method to quantify the IPG effects is an ongoing research topic. In this study, the IPGE_slope was selected to quantify the IPG effect based on the current literature. Future studies may suggest other parameters or quantification methods as preferred. If this is the case, we will reanalyze these data using the preferred parameter and/or method. Second, the PLV and the IPGE_slope are eCAP-derived measurements. As a result, the possibility that these measures are affected by non-neural factors cannot be completely excluded. This possibility could be further assessed by estimating EMD from CT scans and evaluating the associations between EMD and other variables of interest in future research. Third, only the association between the PLV and IPGE_slope was assessed and reported in this study. The potential associations between the PLV and other parameters that have been used to quantify the IPG effects in published studies are not reported because they are beyond the scope of this study. Fourth, the present study is that only 28 post-lingually deafened adult participants were included in the study. Most of them were within the age range of 50–85 yrs and had generally good speech perception outcomes. Therefore, the variance in speech perception scores explained by the PLV or the IPGE_slope may not represent the variance explained in the entire CI patient population. Further studies in CI users with varied speech perception outcomes are warranted to assess the generalizability of the current observations. Fifth, speech perception was measured only using AzBio sentences, which have high ecological validity but prone to the effects of central auditory processing and cognitive factors. In this study, only one ear was tested for each participant, including those who are bilateral CI users. Therefore, in the current dataset, it is not possible to control for potential individual differences in central auditory processing and cognitive abilities using within-participant between-ear comparisons. Future research can test both ears of bilateral CI users and/or add measurements for central auditory processing and cognitive abilities to pinpoint the crucial factors contributing to speech perception in adult CI users. Finally, the 10-talker babble was used as the competing background noise to assess speech perception performance. It does not fully capture the challenge of understanding speech in more complex environments.

CONCLUSIONS

The IPGE_slope and the PLV are two eCAP-derived indices for different aspects of CN health. The significant positive associations between the PLV and AzBio scores measured in noise with +5 dB SNR suggest that neural synchrony is important for speech perception in challenging listening environments. The lack of association between age and the PLV among the current patient population, which primarily consists of middle-aged and elderly CI users, calls for further investigation to determine whether reduced neural synchrony in the CN is unlikely the primary factor accounting for excessive speech perception deficits in older CI users. Future studies can also investigate the contribution of cognitive factors to speech perception and how they interact with the effects of CN health status, as well as use animal models or computational modeling techniques to better understand the biological underpinnings of the IPGE_slope and the PLV.