Aging Effects on Cortical Responses to Tones and Speech in Adult Cochlear-Implant Users

Abstract

Age-related declines in auditory temporal processing contribute to speech understanding difficulties of older adults. These temporal processing deficits have been established primarily among acoustic-hearing listeners, but the peripheral and central contributions are difficult to separate. This study recorded cortical auditory evoked potentials from younger to middle-aged (< 65 years) and older (≥ 65 years) cochlear-implant (CI) listeners to assess age-related changes in temporal processing, where cochlear processing is bypassed in this population. Aging effects were compared to age-matched normal-hearing (NH) listeners. Advancing age was associated with prolonged P2 latencies in both CI and NH listeners in response to a 1000-Hz tone or a syllable /da/, and with prolonged N1 latencies in CI listeners in response to the syllable. Advancing age was associated with larger N1 amplitudes in NH listeners. These age-related changes in latency and amplitude were independent of stimulus presentation rate. Further, CI listeners exhibited prolonged N1 and P2 latencies and smaller P2 amplitudes than NH listeners. Thus, aging appears to degrade some aspects of auditory temporal processing when peripheral-cochlear contributions are largely removed, suggesting that changes beyond the cochlea may contribute to age-related temporal processing deficits.

INTRODUCTION

Older adults often experience difficulties in speech understanding, particularly in adverse listening conditions including noise (e.g., Dubno et al. 1984; Frisina and Frisina 1997), and when the speech signals are temporally distorted (e.g., Gordon-Salant and Fitzgibbons 1993). Temporal features in speech, such as the temporal envelope, are thought to convey cues critical to speech understanding (Rosen 1992). Auditory temporal processing abilities decline with advancing age (Fitzgibbons and Gordon-Salant 1996; Harris and Dubno 2017), and it has been hypothesized that age-related changes in auditory temporal processing contribute to speech understanding difficulties among older listeners (Gordon-Salant et al. 2011; Schneider and Pichora-Fuller 2001). Nevertheless, most acoustic aging studies are confounded by some degree of peripheral hearing loss in older listeners. Cochlear implants (CIs), which severely degrade spectral information and largely retain temporal envelope information, can provide good speech understanding, particularly in quiet (Blamey et al. 2013; Friesen et al. 2001; Fu 2002). Because CIs bypass the transduction of sound in the cochlea, evaluation of temporal processing in older CI listeners may help to clarify the nature of age-related temporal processing deficits.

Scalp-recorded cortical auditory-evoked potentials (CAEPs) are used to index auditory temporal processing in human listeners (Han et al. 2016; Roque et al. 2019a; Tremblay et al. 2003b). The CAEP P1-N1-P2 complex, an obligatory response in the auditory cortex reflecting the detection of time-varying acoustic cues, is used to quantify auditory timing (latency) and salience (amplitude) (Martin et al. 2008). This complex consists of three consecutive positive and negative waveform features with sound onset latencies of approximately 50 ms (positive peak P1), 100 ms (negative peak N1), and 200 ms (positive peak P2). These peaks are thought to reflect different processes (Crowley and Colrain 2004). N1 may reflect the encoding of acoustic features, and P2 may reflect processing after establishing an auditory object (Näätänen and Winkler 1999; Ross et al. 2013).

Acoustic-hearing adult listeners demonstrate auditory aging effects, such as delayed P2 latencies (e.g., Roque et al. 2019a; Tremblay et al. 2004, 2003b) and sometimes N1 latencies (e.g., Bidelman et al. 2014; but see Roque et al. 2019a; Tremblay et al. 2004, 2003b). Age-related latency delays appear to be influenced by stimulus presentation rate and stimulus complexity. For example, Tremblay et al. (2004) compared the P1-N1-P2 complex elicited with a 1000-Hz tone or a syllable /pa/ at three inter-stimulus intervals (ISIs: 0.51, 0.91, and 1.51 s, corresponding to fast, medium, and slow rates) between younger (21–33 years) and older (63–79 years) normal-hearing listeners. The age-related P2 latency delays were evident across all rates, and N1 latency delays occurred at fast and medium rates. The age-related latency delay occurred for both N1 and P2 in response to the speech stimulus but only occurred for N1 in response to the tone stimulus. Further, aging may affect CAEP amplitudes; older listeners may show larger N1 amplitudes (Bidelman et al. 2014; Herrmann et al. 2016) and P1 amplitudes (McNair et al. 2019; Roque et al. 2019b) compared with younger listeners.

Age-related auditory temporal processing deficits are likely to originate from changes in the peripheral and central auditory systems. Animal and human studies suggest a significant age-related degradation of peripheral auditory structures—outer hair cells, the synapses between the hair cells, auditory nerve fibers (ANFs), and ANF peripheral axons (Sergeyenko et al. 2013; Wu et al. 2019). Significant ANF loss in aged auditory systems (Otte et al. 1978; Sergeyenko et al. 2013; Wu et al. 2019) may degrade encoding of temporal features (Lopez-Poveda 2014; Lopez-Poveda and Barrios 2013). Furthermore, there appear to be age-related declines in auditory temporal encoding at subcortical (e.g., Anderson et al. 2012; Parthasarathy et al. 2014; Walton et al. 1998) and cortical levels (e.g., Hughes et al. 2010; Ng and Recanzone 2018; Tremblay et al. 2003b).

The interconnected nature of the peripheral and central auditory systems, however, may pose challenges to address an important question in the field: What are the relative contributions of peripheral and central origins to age-related temporal processing deficits? Limitations in past studies designed to answer this question about these separate origins may preclude the ability to draw strong conclusions. Existing work in acoustic-hearing listeners attempted to partially address this question by examining older acoustic-hearing listeners with normal or near-normal hearing and with hearing loss. Age-related temporal processing deficits (e.g., prolonged CAEP latencies in older listeners) appear to occur independently of hearing loss (Roque et al. 2019a; Tremblay et al. 2003b), which has been interpreted as supporting central effects of aging. However, it can still be argued that these age-related differences are manifestations of peripheral-cochlear changes with advancing age because standard hearing screening protocols may be inadequate to identify peripheral pathologies (e.g., synaptopathy, Kujawa and Liberman 2015) and extended high-frequency (> 8 kHz) hearing loss (Li et al. 2015).

Unlike acoustic hearing, a CI bypasses the cochlea and directly stimulates the auditory nerve. Therefore, assessment of temporal processing in CI listeners may provide a novel human approach to better separate peripheral-cochlear versus retrocochlear contributions. Hence, the current study aims to examine age-related changes in auditory temporal processing in adult CI listeners using CAEPs. Recently, researchers have begun to explore age-related deficits in speech understanding and temporal processing of CI listeners (Shader et al. 2020a, b, c; Sladen and Zappler 2015; Xie et al. 2019). In light of previous studies demonstrating age-related temporal processing deficits, we hypothesized the observation of similar aging effects in CI listeners, manifested as prolonged cortical peak latencies. However, compared to acoustic-hearing listening, the aging effects on CAEPs in CI listeners may be less pronounced because peripheral-cochlear encoding is largely bypassed. For example, aging may lead to hair cell loss and/or reduced endocochlear potential in the cochlea (Schmiedt et al. 2002; Wu et al. 2020, 2019), which may affect temporal precision (Anderson et al. 2021). Hence, we also included age-matched listeners with normal hearing to compare aging effects with those of CI listeners.

METHODS

Listeners

Four groups participated in the current study: ten younger to middle-aged adult CI users (YCI, 7 females, 28.9–60.8 years, mean age = 45.1 years, SD = 10.8), ten older adult CI users (OCI, 6 females, 65.5–85.6 years, mean age = 72.9 years, SD = 6.7), ten younger to middle-aged adults with normal hearing (YNH, 8 females, 29.1–59.7 years, mean age = 45.4 years, SD = 11.1), and ten older adults with normal hearing (ONH, 10 females, 65.5–80.9 years, mean age = 69.9 years, SD = 4.7). For simplicity, we refer to the YCI and YNH groups as “younger” even though they include middle-aged individuals. The two relatively younger groups (YCI and YNH) were matched in age (± 2.5 years) on a case-by-case basis. Although we were unable to match the oldest CI listeners to ONH listeners because of recruiting difficulty, the mean age was not significantly different between the two older groups (Mann-Whitney-Wilcoxon test, p = 0.364). The sex distribution was not significantly different across the four groups (Fisher’s exact test of independence, p = 0.233).

All listeners were native speakers of American English. They were screened with the Montreal Cognitive Assessment (MoCA; Nasreddine et al. 2005) to ensure normal or near-normal cognitive function (≥ 22 out of 30 possible points) (Cecato et al. 2016; Dupuis et al. 2015). The MoCA score was missing for one ONH listener. All other listeners passed the MoCA screening except for one OCI listener (MoCA = 20). We chose a lower cutoff score (22) considering that sensory impairments such as hearing loss may negatively influence one’s MoCA score (Dupuis et al. 2015). Note that the MoCA scores were 26 or above (the typical cutoff) in most listeners: 8/10 YCI, 9/10 OCI, 10/10 YNH, and 8/9 ONH. Details of demographic information for the two CI groups are provided in Table 1. The two CI groups did not significantly differ on the duration of deafness (Mann-Whitney-Wilcoxon test, p = 0.849) or duration of CI use (Mann-Whitney-Wilcoxon test, p = 0.631). Here, duration of deafness was defined as the interval between the age at CI activation and the age at which they self-reported having a severe-to-profound hearing loss or at which they no longer benefited from (e.g., no longer able to understand speech with) hearing aids. Further, according to Table 1, there were no systematic device-related differences between the age groups including electrode, speech processing strategy, internal device, and clinical rate. Normal hearing was defined as ≤ 25 dB HL at octave frequencies from 250 to 4000 Hz. The average (and individual) thresholds of the test ear in the YNH and ONH groups are displayed in Fig. 1.

Table 1.

Demographic information for CI listeners

Group	Age	Sex	Ear	DoD (year)	CI use duration (year)	Processor	Etiology	Internal device	Processing strategy	Clinical rate (pps)	Electrodes deactivated
YCI	28.9	F	Right	20.0	8.9	CP910	Hereditary	CI512	ACE	900	None
YCI	31.6	F	Right	1.3	29.0	CP910	Bacterial meningitis	Nucleus 22 Series	SPEAK	250	19,20,21,22
YCI	37.5	F	Left	< 1	4.5	Med-EL	Ototoxicity	Med-El Concert	FS4	1263	None
YCI	39.6	F	Left	3.0	19.6	CP910	Unknown	CI24M	ACE	1200	1,2
YCI	43.0	M	Left	37.0	5.0	CP810	Bacterial meningitis	CI422	ACE	900	1,2
YCI	46.3	M	Right	5.0	15.3	CP810	Cogan’s syndrome	CI24R	ACE	900	21,22
YCI	51.6	F	Right	0.0	12.6	CP810	Unknown	CI24RE	ACE	1200	None
YCI	54.9	F	Right	7.0	7.9	CP810	Ototoxicity/trauma	CI512	ACE	900	None
YCI	56.2	M	Left	2.0	9.8	CP910	Unknown	CI24RE	ACE	900	1
YCI	60.8	F	Left	17.0	3.8	Freedom	Meniere’s disease	CI24RE	ACE	900	None
OCI	65.5	F	Right	13.0	7.5	CP910	Nerve damage	CI24RE	ACE	900	None
OCI	65.9	F	Right	7.0	4.9	CP910	Acute and central nervous system Lyme disease	CI24RE	ACE	900	1,2
OCI	66.3	F	Right	1.0	8.3	CP910	Premature birth	CI24RE	ACE	1200	None
OCI	67.6	F	Right	8.0	5.6	CP910	Possibly otosclerosis plus a series of childhood illnesses	CI24RE	ACE	900	1,11,13,17,21
OCI	72.6	M	Right	< 1	14.6	CP910	Unknown	CI24R	ACE	900	1,2,3
OCI	73.0	F	Right	7.0	14.0	CP810	Possibly genetic	CI24RE	ACE	900	1,2,12,14,16,18,22
OCI	74.6	F	Right	5.0	9.6	CP810	Measles	CI512	ACE	500	1
OCI	78.7	M	Right	7.0	10.5	CP810	Unknown	CI24RE	ACE	900	None
OCI	78.9	M	Right	1.0	7.8	CP910	Measles, high-dose antibiotics, aging	CI512	ACE	900	None
OCI	85.6	M	Right	2.0	6.6	CP810	Sudden sensorineural hearing loss	CI24RE	ACE	900	None

DoD duration of deafness, YCI younger to middle-aged adult cochlear-implant users, OCI older adult cochlear-implant users

Fig. 1.

Mean pure-tone thresholds (dB HL) for YNH (gray/thick lines/ triangles) and ONH (red/thin lines/circles) groups. The horizontal dashed line indicates 25 dB HL, the point of audiometrically normal hearing. Error bars denote ± 1 standard deviation. Transparent lines (gray/thick lines for YNH and red/thin lines for ONH) denote thresholds for individual listeners. The threshold data for 8000 Hz were missing for one ONH listener. YNH: younger to middle-aged adults with normal hearing, ONH: older adults with normal hearing

Written informed consent was obtained from all listeners. All materials and procedures were approved by the Institutional Review Board at the University of Maryland. All listeners received monetary compensation for their participation.

Design and Procedure

Stimuli

A 1000-Hz tone and a speech syllable /da/ were used for the current study. Both stimuli were sampled at 20 kHz and were equated in duration (510 ms) and intensity. The tone stimulus was created using custom MATLAB (The MathWorks, Natick, MA) scripts, with 10-ms rise/fall (Hanning window) times. The speech stimulus was created as follows: A 170-ms /da/ with a constant fundamental frequency (F0) of 100 Hz was synthesized with a Klatt-based synthesizer (Klatt 1980) at a sampling rate of 20 kHz (Anderson et al. 2012). Its duration was then time expanded to be 510 ms in Praat (Boersma and Weenink 2009). The F0 remained unchanged for the expanded stimulus. The choice of a relatively long stimulus duration was to minimize the corruption of CAEP peaks, particularly P2, by artifacts generated by the procedures to remove CI artifacts (Presacco et al. 2017).

Design

Each listener participated in two repeated sessions of testing to determine if the cortical responses can be reliably obtained in individuals, particularly in CI listeners. The time interval between the two sessions did not significantly differ across the four groups [Kruskal-Wallis χ²(3) = 6.56, p = 0.087]. During each session, electrophysiological responses were elicited with the two stimuli (1000-Hz tone and /da/) across five inter-stimulus intervals (ISIs: 0.5, 1, 2, 3, and 4 s). Short ISIs represented faster stimulus presentation rates. We varied ISIs and stimulus types because prior work suggests that age effects on CAEPs may be influenced by stimulus presentation rate and/or stimulus complexity (Tremblay et al. 2004). The ten conditions (2 stimulus types × 5 ISIs) were blocked and presented in randomized orders across individuals.

Procedure

The testing was conducted in a sound-treated electrically shielded booth. During testing, the lights were turned off to reduce electrical interference. The stimulus presentation was controlled via Presentation software (Neurobehavioral Systems, Inc.). For CI listeners, the stimuli were presented monaurally via direct audio input to the sound processor with their everyday settings of the self-reported better ear at the most comfortable level. A research processor was used whenever possible; otherwise, their clinical processor was used (4 for YCI and 4 for OCI). For NH listeners, the stimuli were presented monaurally via a single-insert earphone (ER-2, Etymotic, Elk Grove Village, IL) at 75 dB SPL to the right ear except for one YNH listener, because the 500-Hz threshold in their right ear was at 30 dB HL (> 25 dB HL criterion). At the beginning of the study, listeners were instructed to ignore the sounds they heard and to watch muted movies of their choice with subtitles. Each testing session lasted about 2 to 3 hours.

EEG Data Recording

The electrophysiological data were recorded from 32 electrodes (online referenced to the average of left and right earlobe electrodes) in the BioSemi ActiveTwo system (Biosemi B.V., Amsterdam, Netherlands) and sampled at 2048 Hz. Two additional electrodes were placed around the right eye to track vertical and horizontal ocular artifacts. A minimum of 250 sweeps was recorded for each condition in individual listeners.

EEG Data Preprocessing

Data processing was performed with custom MATLAB scripts (The MathWorks, Natick, MA) adapted from the EEGLAB toolbox (Delorme and Makeig 2004). Data analyses were limited to eight frontocentral electrodes (F3, Fz, F4, FC1, FC2, C3, Cz, and C4) where auditory cortical responses are typically most prominent (Tremblay et al. 2003a). The raw electrophysiological data were re-referenced offline to the earlobe electrode contralateral to the stimulation ear. The re-referenced data were bandpass-filtered from 0.03 to 30 Hz (forward-backward 2nd-order Butterworth filter), down-sampled to 256 Hz, corrected for ocular artifacts (Schlögl et al. 2007), and then segmented into epochs of 900 ms (100 ms before and 800 ms after stimulus onset). Each segment was baseline corrected by subtracting the mean voltage of the pre-stimulus region from each point in the segment.

Trials with amplitudes exceeding a predefined range (± 1000 μV for CI listeners and ± 200 μV for NH listeners) were rejected. Conditions with fewer than 150 sweeps after artifact rejection or with noisy electrodes during the recording were removed. Less than 1.5% of data from CI and NH listeners were excluded from further analysis. For the remaining data, the number of trials in individual conditions ranged from 193 to 250 (median = 250) for CI listeners and from 155 to 250 (median = 250) for NH listeners. The artifact-free trials were averaged for each condition in individual listeners.

To further minimize artifacts generated by CI sound processing, we adopted a regression-based approach from previous studies (Mc Laughlin et al. 2013; Presacco et al. 2017). Briefly, the CI-related artifacts were estimated as the averaged response from the electrode that best represents the stimulus features via manual inspection (the selected electrode is usually most adjacent to the CI). The estimated CI-related artifacts were then subtracted from each averaged response across the eight electrodes of interest in the same condition, leading to the final averaged responses for later analysis. Details of the procedures to remove CI-related artifacts can be found in Presacco et al. (2017). Visual inspection of response waveforms across all 32 electrodes before and after CI-artifact removal suggests that the procedures were effective in minimizing CI-related artifacts, particularly for the frontal and central electrodes (see Appendix 2 for an example). After CI-artifact removal, responses from electrodes outside the frontal and central regions generally included more artifacts. Hence, it was reasonable and preferable to restrict analysis to the aforementioned eight frontocentral electrodes.

Analysis of Event-Related Potentials: Latency and Amplitude

An automated peak-picking algorithm was implemented in MATLAB to identify the latencies for prominent cortical peaks in their expected time regions: P1 (35–75 ms), N1 (80–150 ms), and P2 (160–250 ms) (Karawani et al. 2018). The identified peaks were manually checked and adjusted by the first author if necessary and verified by another researcher. The same MATLAB algorithm also calculated the area amplitudes corresponding to a 40-ms time window around each peak (Karawani et al. 2018).

Statistical Analysis

Separate linear mixed-effects models implemented via the lme4 package (Bates et al. 2014) in R version 3.6.2 (Team RC 2013) were used to fit the data for latency and amplitude. The models were applied separately to N1 and P2 peaks. The P1 data were not included in the analyses because those data are likely affected significantly by the CI artifacts based on the inspection of response waveforms in individual listeners. All models included the same fixed effects: age at testing, hearing status (NH and CI), stimulus type (tone or speech), and ISI (0.5, 1, 2, 3, or 4 s). Age was centered using the mean age of all CI and NH listeners and was treated as a continuous variable. Hearing status, stimulus type, and ISI were treated as categorical variables. In the model, the reference levels were CI, an ISI of 0.5 s, and tone. For all models, the initial random effects were set as (1 | stimulus type × ISI | subject/electrode).

To reduce the risk of data overfitting, we removing random and fixed effects that did not contribute significantly to the model (p > 0.05) using the step function in the lmerTest package (Kuznetsova et al. 2017). Results from the simplest, best-fitting models were reported in the “RESULTS” section. We computed the significance values for fixed effects in those optimal models using the anova function in the lmerTest package (Kuznetsova et al. 2017). We conducted post hoc analysis for significant fixed effects, if necessary, with the emmeans (for categorical variables) and emtrends (for continuous variables) functions in the emmeans package (Lenth et al. 2018b). Multiple comparisons were corrected by the Bonferroni method. Descriptive statistics, if reported, represent mean ± 1 standard deviation (SD).

RESULTS

Figures 2 and 3 display grand-average waveforms collapsed across the two test sessions at each of the eight electrodes for the CI and NH groups, respectively. Figure 4 displays grand-average waveforms collapsed across the two test sessions and the eight electrodes for the CI and NH groups, respectively. These three figures show that the peaks were delayed for the two older adult groups (OCI and ONH) compared to their younger counterparts (YCI and YNH), particularly for the P2 peak. Such patterns are consistent across ISIs, stimulus types, and electrodes. The response amplitudes were larger in the ONH group compared to the YNH group, particularly for the P2 peak in response to the speech stimulus. Figure 5 displays the mean (± 1 SD) latencies of N1 (Fig. 5a, b) and P2 peaks (Fig. 5c, d) to the two stimuli separately. Figure 6 displays the mean (± 1 SD) amplitudes of N1 (Fig. 6a, b) and P2 peaks (Fig. 6c, d) to the two stimuli separately. Both the latency and amplitude data were collapsed across the two test sessions and the eight electrodes for the CI and NH groups, respectively.

Fig. 2.

Grand average response waveforms to a 1000-Hz tone (a) and a speech syllable /da/ (b) across five interstimulus intervals (ISIs: 0.5 to 4 s) at each of the eight frontocentral channels for YCI (gray/thick lines) and OCI (red/thin lines) listeners. The waveforms were collapsed across the two recording sessions. The gray horizontal and vertical lines in each subplot represent the x-axis (time) and y-axis (amplitude) that meet at (0,0). YCI: younger to middle-aged adult cochlear-implant users, OCI: older adult cochlear-implant users

Fig. 3.

Grand average response waveforms to a 1000-Hz tone (a) and a speech syllable /da/ (b) across five interstimulus intervals (ISIs: 0.5 to 4 s) at each of the eight frontocentral channels for YNH (gray/thick lines) and ONH (red/thin lines) listeners. The waveforms were collapsed across the two recording sessions. The gray horizontal and vertical lines in each subplot represent the x-axis (time) and y-axis (amplitude) that meet at (0,0). YNH: younger to middle-aged adults with normal hearing, ONH: older adults with normal hearing

Fig. 4.

Grand average response waveforms to a 1000-Hz tone (top) and a speech syllable /da/ (bottom) across five interstimulus intervals (ISIs: 0.5 to 4 s) for (a) YCI (gray/thick lines) and OCI (red/thin lines) listeners and (b) for YNH (gray/thick lines) and ONH (red/thin lines) listeners. The waveforms were collapsed across the two recording sessions and the eight frontocentral channels. The shaded areas indicate 95 % confidence intervals around the mean. The gray horizontal and vertical lines in each subplot represent the x-axis (time) and y-axis (amplitude) that meet at (0,0). YCI: younger to middle-aged adult cochlear-implant users, OCI: older adult cochlear-implant users, YNH: younger to middle-aged adults with normal hearing, ONH: older adults with normal hearing

Fig. 5.

Mean (± 1 standard deviation) peak latencies across five interstimulus intervals (ISIs: 0.5 to 4 s) for younger (< 65 years; gray/thick lines) and older (> 65 years; red/thin lines) listeners. (a) N1 peaks of cortical responses to a 1000-Hz tone. (b) N1 peaks of cortical responses to a speech syllable /da/. (c) P2 peaks of cortical responses to a 1000-Hz tone. (d) P2 peaks of cortical responses to a speech syllable /da/. The latency data were collapsed across the two recording sessions and the eight frontocentral channels. CI: cochlear-implant users, NH: normal hearing

Fig. 6.

Mean (± 1 standard deviation) amplitudes across five interstimulus intervals (ISIs: 0.5 to 4 s) for younger (< 65 years; gray/thick lines) and older (> 65 years; red/thin lines) listeners. (a) N1 peaks of cortical responses to a 1000-Hz tone. (b) N1 peaks of cortical responses to a speech syllable /da/. (c) P2 peaks of cortical responses to a 1000-Hz tone. (d) P2 peaks of cortical responses to a speech syllable /da/. The amplitude data were collapsed across the two recording sessions and the eight frontocentral channels. CI: cochlear-implant users, NH: normal hearing

We report statistical results on latency and amplitude in the following sections. We also summarize the significant effects in Table 2. The addition of the sex variable into the optimal models for latency or amplitude data reported below did not significantly improve model fit (all ps > 0.118), suggesting that sex did not significantly affect latency or amplitude.

Table 2.

Summary of significant effects for latencies and amplitudes of N1 and P2

		Age-based effects	Non-age-based effects
Latency	N1	Age (p = 0.006); age × hearing status × stimulus type (p = 0.001) (Fig. 7a)	Hearing status (p < 0.001); stimulus type (p < 0.001); hearing status × stimulus type (p = 0.018); stimulus type × ISI (p = 0.032) (Fig. 7b)
	P2	Age (p < 0.001); age × ISI (p = 0.040) (Fig. 7c)	Hearing status (p < 0.001); stimulus type (p < 0.001); ISI (p < 0.001); hearing status × stimulus type (p = 0.036) (Fig. 7d)
Amplitude	N1	Age × hearing status (p = 0.047) (Fig. 8a)	Stimulus type (p < 0.001); ISI (p < 0.001); hearing status × ISI (p = 0.011) (Fig. 8b)
	P2	Age × ISI (p = 0.034); age × hearing status × stimulus type (p < 0.001)	Hearing status (p < 0.001); stimulus type (p = 0.029); ISI (p < 0.001); hearing status × stimulus type (p < 0.001); hearing status × ISI (p = 0.041) (Fig. 8c)

Latency

N1

Of our primary interest, the main effect of age was significant [F(1, 40.1) = 8.343, p = 0.006], and the age × hearing status × stimulus type interaction was significant [F(1, 40.3) = 11.972, p = 0.001]. In addition, the non-age-based main effects hearing status [F(1, 40.0) = 15.720, p < 0.001] and stimulus type [F(1, 39.9) = 38.461, p < 0.001] were significant. The non-age-based interactions hearing status × stimulus type [F(1, 40.3) = 6.028, p = 0.018] and stimulus type × ISI [F(4, 45.1) = 2.904, p = 0.032] were significant. The other main effect (ISI) and interactions (age × stimulus type or age × hearing status) were not significant (both ps > 0.092).

To understand the age × hearing status × stimulus type interaction, we replotted the N1 latency data from Fig. 5a, b and compared the age effects across the four conditions combined from hearing status and stimulus type. The comparisons are displayed in Fig. 7a. The estimated age effect for the speech stimulus in CI listeners was significantly larger than 0 (95% confidence interval: [0.261, 1.074]; p < 0.001), wherein N1 latencies were significantly delayed in older CI listeners in response to the speech stimulus (OCI: 131.8 ± 25.0 ms vs. YCI: 115.8 ± 28.2 ms). The estimated age effects for the other three conditions, however, were not significantly different from zero (95% confidence intervals: [−0.485, 0.639]; all ps > 0.394).

Fig. 7.

Significant interactions of fixed effects on the latency data. (a) Age, hearing status, and stimulus type interaction on N1 latency. (b) Stimulus type and interstimulus interval (ISI) interaction on N1 latency. (c) Age and ISI interaction on P2 latency. (d) Hearing status and stimulus type interaction on P2 latency. The violin plots show distributions of raw data. Horizontal lines inside the violin plots represent individual observations. The length of each of those horizontal lines represents the number of observations at a given latency relative to the maximum number of observations across all latencies. The line plots show means ± 1 standard deviation. For clarity, outliers were excluded from the violin plots but were included when deriving the line plots as well as in statistical analyses. Outliers were defined as values outside this range: [lower quartile – 1.5 × interquartile range, upper quartile + 1.5 × interquartile range]. Data for younger listeners (< 65 years; all YCI and YNH) are displayed in gray/thick lines/triangles; data for older listeners (> 65 years; all OCI and ONH) are displayed in red/thin lines/circles. Data for responses to the tone stimulus are displayed in black/thick lines/squares; data for responses to the speech stimulus are displayed in orange/thin lines/diamonds. Data for CI listeners (all YCI and OCI) are displayed in orange/thin lines/diamonds; data for NH listeners (all YNH and ONH) are displayed in black/thick lines/squares. YCI: younger to middle-aged adult cochlear-implant users, OCI: older adult cochlear-implant users, YNH: younger to middle-aged adults with normal hearing, ONH: older adults with normal hearing. *p < 0.05; **p < 0.01; ***p < 0.001

To understand the stimulus type × ISI interaction, we replotted the N1 latency data from Fig. 5a, b and compared the effect of stimulus type across ISIs. The comparisons are displayed in Fig. 7b. While the effect of stimulus type on N1 latency (i.e., prolonged latencies in response to speech vs. tone) was significant across all ISIs (all ps < 0.001), the magnitude of stimulus type effect was larger at the ISI of 1 s compared to the ISI of 4 s (p = 0.046). The magnitudes of stimulus type effect across other ISI comparisons were not significant (all ps > 0.333).

P2

Of our primary interest, the main effect of age was significant [F(1, 39.9) = 24.847, p < 0.001]. The age × ISI interaction was significant [F(4, 40.0) = 2.780, p = 0.040]. In addition, the non-age-based main effects hearing status [F(1, 40.0) = 13.796, p < 0.001], stimulus type [F(1, 40.0) = 26.372, p < 0.001], and ISI [F(4, 40.0) = 14.059, p < 0.001] were significant. The non-age-based interaction hearing status × stimulus type [F(1, 40.0) = 4.707, p = 0.036] was significant.

To understand the age × ISI interaction, we replotted the P2 latency data from Fig. 5c, d and compared the age effects across ISIs. The comparisons are displayed in Fig. 7c. The estimated age effects for all ISIs were significantly larger than 0 (95% confidence intervals: [0.145, 1.32]; all ps < 0.01), wherein P2 latencies were significantly delayed in older CI and NH listeners across all ISIs: 0.5 s (older: 200.4 ± 29.0 ms vs. younger: 178.9 ± 24.5 ms), 1 s (older: 196.1 ± 24.4 ms vs. younger: 179.4 ± 24.9 ms), 2 s (older: 205.8 ± 20.7 ms vs. younger: 185.6 ± 24.1 ms), 3 s (older: 208.7 ± 21.2 ms vs. younger: 189.6 ± 23.8 ms), and 4 s (older: 210.1 ± 21.8 ms vs. younger: 190.0 ± 24.1 ms). The interaction was driven by a larger age effect at the 0.5-s ISI than that at the 1-s ISI (p = 0.032); the other comparisons of age effects between ISIs were not significantly different (all ps > 0.390).

To understand the hearing status × stimulus type interaction, we replotted the P2 latency data from Fig. 5c, d, and compared the effect of hearing status across stimulus type and vice versa. The comparisons are displayed in Fig. 7d. The effect of hearing status on P2 latency (i.e., prolonged latencies in CI vs NH listeners) was driven by a significant group difference for the tone stimulus (CI: 196.4 ± 30.7 ms vs. NH: 175.4 ± 26.0 ms; p < 0.001), but not by the speech stimulus (CI: 206.2 ± 32.8 ms vs. NH: 199.6 ms ± 22.6 ms; p = 0.600). The effect of stimulus type on P2 latency (i.e., prolonged latencies in response to speech vs. tone) was driven by a significant difference in NH listeners (tone: 175.4 ± 26.0 ms vs. speech: 199.6 ms ± 22.6 ms; p < 0.001), but not in CI listeners (tone: 196.4 ± 30.7 ms vs. speech: 206.2 ± 32.8 ms; p = 0.136).

To summarize, we were primarily interested in aging effects; advancing age was associated with prolonged N1 latencies to the speech stimulus in CI listeners (Figs. 4a and 7a), and with prolonged P2 latencies in CI and NH listeners (Figs. 4 and 7c). The age effects appear to occur irrespective of the ISI. Finally, responses from CI listeners exhibited prolonged N1 and P2 latencies compared to NH listeners (Fig. 7a, d).

Amplitude

N1

Our primary interest in this study was the effects of age. The main effect of age was not significant [F(1, 43.8) = 2.250, p = 0.141], but the age × hearing status interaction was significant [F(1, 43.8) = 4.175, p = 0.047]. In addition, the non-age-based main effects stimulus type [F(1, 40.1) = 24.075, p < 0.001] and ISI [F(4, 43.2) = 63.805, p < 0.001] were significant. The non-age-based interaction hearing status × ISI [F(4, 43.3) = 3.723, p = 0.011] was significant. Finally, the non-age-based main effect of hearing status was not significant (p = 0.806).

To understand the age × hearing status interaction, we replotted the N1 amplitude data from Figs. 6a, b and compared the age effects between hearing status. The comparisons are displayed in Fig. 8a. The estimated age effect in NH listeners was significantly smaller than 0 (95% confidence interval: [−0.041, −0.00046]; p = 0.044), where N1 amplitude of NH listeners increased with advancing age (older: −2.47 ± 2.27 µV vs. younger: −2.34 ± 2.37 µV). The estimated age effect in CI listeners was not significantly different from 0 (95% confidence interval: [−0.015, 0.021]; older: −2.10 ± 1.95 µV vs. younger: −2.39 ± 1.93 µV; p = 1.000).

Fig. 8.

Significant fixed effects on the amplitude data. (a) Age and hearing status interaction on N1 amplitude. (b) Hearing status and interstimulus interval (ISI) interaction on N1 amplitude. (c) Hearing status and ISI interaction on P2 amplitude. The violin plots show distributions of raw data. Horizontal lines inside the violin plots represent individual observations. The length of each of those horizontal lines represents the number of observations at a given amplitude relative to the maximum number of observations across all amplitudes. The line plots show means ± 1 standard deviation. For clarity, outliers were excluded from the violin plots but were included when deriving the line plots as well as in statistical analyses. Outliers were defined as values outside this range: [lower quartile – 1.5 × interquartile range, upper quartile + 1.5 × interquartile range]. Data for CI listeners (all YCI and OCI) are displayed in orange/thin lines/diamonds; data for NH listeners (all YNH and ONH) are displayed in black/thick lines/squares. Data for younger listeners (< 65 years; all YCI and YNH) are displayed in gray/thick lines/triangles; data for older listeners (> 65 years; all OCI and ONH) are displayed in red/thin lines/circles. YCI: younger to middle-aged adult cochlear-implant users, OCI: older adult cochlear-implant users; YNH: younger to middle-aged adults with normal hearing, ONH: older adults with normal hearing. *p < 0.05

To understand the hearing status × ISI interaction, we replotted the N1 amplitude data from Fig. 6a, b, and compared the ISI effect between hearing status. The comparisons are displayed in Fig. 8b. For CI listeners, N1 amplitudes significantly increased as the ISI increased from 0.5 s up to 2 s and then to 4 s (i.e., 0.5 s < 1 s and longer ISIs; 1 s < 2 s and longer ISIs; 2 s < 4 s; all ps < 0.05); but there was no significant difference in N1 amplitude between ISIs of 2 s and 3 s (p = 0.142) or between ISIs of 3 s and 4 s (p = 0.058). For NH listeners, N1 amplitudes significantly increased for each increment in the ISI from 0.5 s up to 4 s (all ps < 0.05).

P2

Again focusing on the effects of age, the main effect of age was not significant [F(1, 40.0) = 0.254, p = 0.617], but the age × ISI interaction was significant [F(4, 40.1) = 2.902, p = 0.034]. The age × hearing status × stimulus type interaction was significant [F(1, 40.3) = 22.634, p < 0.001]. In addition, the non-age-based main effects hearing status [F(1, 39.7) = 17.899, p < 0.001], stimulus type [F(1, 39.5) = 5.166, p = 0.029], and ISI [F(4, 40.0) = 23.954, p < 0.001] was significant. The non-age-based interactions of hearing status × stimulus type [F(1, 39.5) = 25.929, p < 0.001], and hearing status × ISI [F(4, 40.0) = 2.752, p = 0.041] were significant. Finally, other interactions (age × hearing status or age × stimulus type) were not significant (both ps > 0.064).

To understand the age × hearing status × stimulus type interaction, we compared the age effects on P2 amplitudes across the four conditions combined from hearing status and stimulus type. The estimated age effect for any of the four conditions was not significantly different from 0 (95% confidence intervals: [−0.056, 0.059]; all ps > 0.244), which did not support an age effect on P2 amplitudes at any of the conditions. The interaction was possibly driven by the (insignificant) estimated age effect that was larger for the speech stimulus in NH listeners than that for the speech stimulus in CI listeners (p = 0.011) or than that for the tone stimulus in NH listeners (p < 0.001). The (insignificant) estimated age effects were not significantly different across all other comparisons (all ps > 0.171).

To understand the age × ISI interaction, we compared the age effects on P2 amplitudes across ISIs. The estimated age effect at any ISI was not significantly different from 0 (95% confidence intervals: [−0.051, 0.033]; all ps = 1.000), which did not support an age effect on P2 amplitudes at any ISIs. The interaction was possibly driven by the observation that the (insignificant) estimated age effect was smaller at the ISI of 2 s than that at the ISI of 3 s (p = 0.192 after multiple-comparison correction; p = 0.019 before multiple-comparison correction). The (insignificant) estimated age effects were not significantly different across all other comparisons (all ps > 0.05 even before multiple-comparison correction).

To understand the hearing status × ISI interaction, we replotted the P2 amplitude data from Fig. 6c, d and compared the ISI effect between hearing status. The comparisons are displayed in Fig. 8c. For CI listeners, P2 amplitudes were smaller at the ISI of 0.5 s than that at ISIs of 2 to 4 s (all ps < 0.01), and P2 amplitudes were smaller at the ISI of 1 s than that at the ISI of 4 s (p = 0.03). Other ISI comparisons were not significant (all ps > 0.062). For NH listeners, P2 amplitudes increased as the ISI increased from 0.5 s up to 2 s (i.e., 0.5 s < 1 s and longer ISIs; 1 s < 2 s and longer ISIs; all ps < 0.05), but P2 amplitudes were not significantly different across ISIs of 2, 3, or 4 s (all ps > 0.163).

To summarize, we were primarily interested in aging effects; we did not find support for aging effects on N1 and P2 amplitudes in CI listeners. But for NH listeners, advancing age may be associated with larger N1 amplitudes (Fig. 8a). Further, responses from CI listeners showed smaller P2 amplitudes compared to NH listeners (Fig. 8c).

Possible Role of Duration of Deafness in CI Listeners

We re-analyzed the N1 and P2 latencies and amplitudes for the CI data by including the duration of deafness (DoD) as an additional factor. The original models consist of four fixed effects: age at testing, DoD, stimulus type (tone or speech), and ISI (0.5, 1, 2, 3, or 4 s). Age and DoD were mean-centered and were treated as continuous variables. Stimulus type and ISI were treated as categorical variables. In the model, the reference levels were an ISI of 0.5 s and tone. For all models, the initial random effects were set as (1 | stimulus type × ISI | subject/electrode). The simplest and best-fitting models, found using the approach stated in the “METHODS” section, were reported below. For simplicity, we focused the descriptions of the results on the effects of age, DoD, and interactions with these factors.

Regarding N1 latencies, the main effects of age [F(1, 20.4) = 18.471, p < 0.001] and DoD [F(1, 20.4) = 5.998, p = 0.023] were significant. The age × stimulus type interaction [F(1, 20.2) = 8.880, p = 0.007], DoD × ISI interaction [F(4, 24.9) = 5.530, p = 0.003], age × DoD × stimulus type interaction [F(1, 20.2) = 5.176, p = 0.034], and age × DoD × ISI interaction [F(4, 24.9) = 7.447, p < 0.001] were significant.

To understand the age × DoD × stimulus type interaction, we compared the age effect across three levels of DoD (i.e., mean – SD, mean, mean + SD) at each stimulus type. For the tone stimulus, the estimated age effects were significantly larger than 0 only at longer DoD (i.e., mean + SD; 95% confidence interval: [0.237, 1.552]; p = 0.023), but not at shorter DoD (mean − SD and mean; 95% confidence interval: [−0.747, 0.682]; both ps > 0.203). For the speech stimulus, the estimated age effects were significantly larger than 0 across all DoD levels (95% confidence intervals: [0.186, 1.60]; all p < 0.05).

To understand the age × DoD × ISI interaction, we compared the age effect across three levels of DoD (i.e., mean − SD, mean, mean + SD) at each ISI. At an ISI of 0.5 s, the estimated age effects were significantly larger than 0 across all DoD levels (95% confidence intervals: [0.123 0.97]; all ps < 0.05). But at ISIs longer than 0.5 s, the estimated age effects were significantly larger than 0 at longer DoD levels (mean and mean + SD; 95% confidence intervals: [0.183, 1.837]; all ps < 0.05), but not at shorter DoD (mean − SD; 95% confidence intervals: [−0.27, 0.99]; all ps > 0.061).

Regarding P2 latencies, the main effects of age [F(1, 20.1) = 37.298, p < 0.001] and DoD [F(1, 20.1) = 7.417, p = 0.013] were significant. The age × ISI interaction [F(4, 20.4) = 3.134, p = 0.037] and DoD × stimulus type interaction [F(1, 20.0) = 5.345, p = 0.032] were significant. The age × DoD interaction [F(1, 20.1) = 1.235, p = 0.280] or DoD × ISI interaction [F(4, 20.6) = 2.752, p = 0.056] was not significant. But the age × DoD × ISI interaction [F(4, 20.4) = 3.770, p = 0.019] was significant.

To understand DoD × stimulus type interaction, we compared the DoD effects on P2 latency across stimulus types. The estimated DoD effect for the speech stimulus was significantly larger than zero (95% confidence interval: [0.633, 3.48]; p = 0.004), wherein P2 latencies were significantly delayed in CI listeners with longer DoD. But the estimated age effect for the tone stimulus was not significantly different from 0 (95% confidence interval: [−0.689, 2.02]; p = 0.503).

To understand the age × DoD × ISI interaction, we compared the age effect across three levels of DoD (i.e., mean − SD, mean, mean + SD) at each ISI. At ISIs of 0.5, 1, 3, and 4 s, the estimated age effects were significantly larger than zero across all DoD levels (95% confidence intervals: [0.145, 2.26]; all ps < 0.05). But at an ISI of 2 s, the estimated age effects were significantly larger than 0 at longer DoD levels (mean and mean + SD; 95% confidence intervals: [0.667, 2.52]; all ps < 0.01), but not at shorter DoD (mean – SD; 95% confidence interval: [−0.025, 1.36]; p = 0.176).

Regarding N1 amplitudes, the main effect of age [F(1, 20.2) = 3.683, p = 0.069] or DoD [F(1, 20.2) = 1.750, p = 0.201] was not significant. The age × DoD interaction [F(1, 20.2) = 0.500, p = 0.487], age × ISI interaction [F(4, 27.5) = 2.653, p = 0.054], and DoD × ISI interaction [F(4, 27.4) = 2.221, p = 0.093] were not significant. But the age × DoD × ISI interaction [F(4, 27.4) = 4.502, p = 0.006] was significant. To understand this three-way interaction, we compared the age effect across three levels of DoD (i.e., mean − SD, mean, mean + SD) at each ISI. The estimated age effect was significantly larger than 0 for longer DoD (i.e., mean + SD) at an ISI of 2 s (95% confidence interval: [0.003, 0.099]; p = 0.036 before multiple-comparisons correction; however, p = 0.109 after multiple-comparisons correction), and for the mean level of DoD at an ISI of 5 s (95% confidence interval: [0.005, 0.077]; p = 0.025 before multiple-comparisons correction; however, p = 0.075 after multiple-comparisons correction). The estimated age effect was not significant for any other conditions (all ps > 0.05 even before multiple-comparisons correction).

Regarding P2 amplitudes, the main effect of age [F(1, 20.0) = 0.864, p = 0.364] was not significant. The main effect of DoD [F(1, 19.9) = 32.396, p < 0.001] was significant, where longer DoD was associated with smaller P2 amplitudes. The age × stimulus type interaction [F(1, 20.0) = 12.072, p = 0.002]. To understand this interaction, we compared age effects on P2 amplitudes across stimulus type. The estimated age effect was not significantly different from 0 for the tone stimulus (95% confidence interval: [−0.006, 0.017]; p = 0.577). However, the age effect was significantly smaller than 0 for the speech stimulus (95% confidence interval: [−0.022, −0.002]; p = 0.014), where P2 amplitude for responses to the speech stimulus decreased with advancing age.

To summarize, after the inclusion of DoD into analyses, we still observed age-related delays of N1 and P2 latencies in CI listeners. But note that in certain conditions (e.g., N1 latencies of tone stimulus or longer ISIs), age effects may emerge only for listeners with relatively longer DoD.

DISCUSSION

Main Findings

This study examined the extent to which age influences auditory temporal processing as indexed by the CAEPs in adult CI and NH listeners. Consistent with our prediction, advancing age was associated with prolonged P2 latencies of CAEPs from CI and NH listeners (Figs. 4 and 7c), and with prolonged N1 latencies of CAEPs to the speech stimulus in CI listeners (Figs. 4a and 7a). Age was not found to have a significant effect on CAEP (N1 or P2) amplitudes in CI listeners but was associated with increased N1 amplitudes in NH listeners (Fig. 8a).

Age Effects on CAEPs in CI and NH Listeners: Implications for Mechanisms Underlying Age-Related Temporal Processing Deficits

Prolonged latencies of CAEPs with advancing age (N1 and P2 for CI listeners and P2 for NH listeners: Fig. 7a, c) in the current study generally concur with the existing literature on listeners with acoustic hearing (Bidelman et al. 2014; Harris and Dubno 2017; Roque et al. 2019a; Tremblay et al. 2004, 2003b). Only one study has directly examined the age effects on CAEPs in adult CI listeners (Mussoi and Brown 2019) and reported a lack of age effects on latency and amplitude measures of N1 and P2 between younger (18–40 years) and older (68–82 years) CI listeners. This is inconsistent with our findings with CI listeners, especially for the speech stimulus.

The discrepancy may lie in the methodological differences between our study and Mussoi and Brown (2019). For example, their study presented simple stimuli (trains of biphasic pulses) via direct stimulation to only one electrode in the middle of the electrode array. Our study presented pure tones and speech syllables via direct audio input to the sound processor. While the pure tone stimulus may be comparable to the stimulus used in Mussoi and Brown (2019), the speech stimulus is more complex and would utilize multiple electrodes of the electrode array. Multiple-channel stimulation may be more susceptible to the degradation of the electrode-to-neuron interface. Both animal and human works suggest a reduced number of auditory nerve fibers with aging (Otte et al. 1978; Sergeyenko et al. 2013; Wu et al. 2019). Poorer neural survival in older listeners may result in increased current levels required for stimulation (Mussoi and Brown 2020) and in turn higher channel interaction; hence, their electrode-to-neuron interfaces may be compromised. Consequently, older listeners may be more affected under multiple-channel stimulation. Consistent with this interpretation, the current study (Fig. 7a) and findings from prior work (Tremblay et al. 2004) suggest that age effects on CAEPs may be more apparent for relatively complex stimuli.

Unlike our study, Mussoi and Brown (2019) did not include middle-aged CI listeners, and therefore, there could have been an etiology confound between studies (in the other study, etiology was not reported, but younger CI listeners may lose their hearing for different reasons than middle-aged and older CI listeners). Thus, our results appear to support the existence of age-related alteration in CAEPs in CI listeners, and differences in methodology may reconcile different results from the existing literature.

Our final analyses factoring in age and DoD demonstrated age-related delays in N1 and P2 latencies in CI listeners irrespective of DoD. The age effects, however, might only be more evident for listeners with relatively longer DoD in specific instances (e.g., delayed N1 latencies to stimuli presented with longer ISIs). We interpreted these findings as further evidence to support the potential independent effect of aging on CAEP in CI listeners. However, considering the relatively small sample size of our study, these findings should be interpreted with caution.

As discussed above, age-related changes in CAEPs may originate from poorer neural survival with advancing age (Otte et al. 1978; Sergeyenko et al. 2013; Wu et al. 2019). Central changes such as prolonged latencies of onset responses in the auditory cortex with aging (Hughes et al. 2010) may also be associated with the delayed latencies of CAEPs in older listeners. Listening through a CI largely bypasses peripheral-cochlear encoding (Heeringa et al. 2020) and directly stimulates the auditory nerve. Hence, the demonstration of age-related temporal processing deficits of CI listeners in the current study and our prior work (Xie et al. 2019) provides evidence that changes beyond the cochlea (auditory nerve and above) may contribute to age-related temporal processing deficits. Note that our study and Mussoi and Brown (2019) have a relatively small CI sample size (N = 20), especially considering the number of testing conditions. Future studies with larger sample sizes are necessary to replicate the findings of changes in CAEPs with advancing age in CI listeners.

Age effects on CAEP amplitudes only emerged for the N1 peak in NH listeners, with larger amplitudes in older than younger listeners (Fig. 8a). The exaggerated responses for older adults may partially arise from the loss of neural inhibition in the auditory system associated with aging (Brodbeck et al. 2018; Caspary et al. 2008; Gao et al. 2015; Presacco et al. 2016) that compensates for degraded sensory inputs in older listeners (Caspary et al. 2008; Parthasarathy et al. 2019). This is generally consistent with previous findings on larger amplitudes in older NH listeners for earlier cortical peaks (e.g., N1, Bidelman et al. 2014; Herrmann et al. 2016; e.g., P1, McNair et al. 2019; Roque et al. 2019b). Unlike NH listeners, we did not observe age effects on N1 or P2 amplitudes in CI listeners, which might be related to plastic changes from deafness (Shepherd and Hardie 2001) as well as cochlear implantation (Pantev et al. 2006). It was unclear whether age affected P1 amplitude in CI listeners, as we were unable to analyze the P1 data because the P1 morphology of CAEPs was often greatly corrupted by the CI artifacts.

Age Effects on CAEPs in CI and NH Listeners: Dismissing Device- and Stimulus-Related Explanations

Device-related factors may change the morphology (latency and amplitude) of CAEPs in listeners with hearing devices (Billings et al. 2007; Chun et al. 2016) and thus may explain age effects on CAEPs of CI listeners. While possible, we believe it is less likely the case due to the following reasons: (1) We did not observe concomitant changes in CAEP amplitude and latency with age in our CI listeners, as may be inferred from previous studies (Billings et al. 2007; Chun et al. 2016). This finding indicates that the age effects in our CI listeners appear to be independent of potential changes in CI stimulation level with advancing age, because the changes in CI stimulation level should lead to concomitant changes in both CAEP amplitude and latency; (2) we are unaware of systematic device-related differences (e.g., programming approaches by clinicians) between the YCI and OCI groups (Table 1); (3) even if there were age-related differences in programming parameters and approaches, it may not significantly change the morphology (e.g., N1 and P2) of CAEPs (Lee and Bidelman 2017). Nevertheless, future studies should systematically investigate CI device-related factors to understand their influences on the morphology of CAEPs.

Stimulus level/audibility may affect the morphology of CAEPs in listeners with acoustic hearing. In quiet conditions, with increasing stimulus levels, N1 and P2 latencies shift earlier, and their amplitudes become larger (Billings et al. 2007). However, at around 70 dB SPL or above, level effects on latency and amplitude tend to plateau in NH listeners especially at fast stimulus presentation rates (Adler and Adler 1989, 1991; Picton et al. 1970) or become limited (e.g., latency and amplitude changes for N1, but no change for P2) in older hearing-impaired listeners (Jenkins et al. 2018; McClannahan et al. 2019). Hence, stimulus level/audibility may not be a dominant factor that accounts for the age effects in NH listeners in the current study, considering that (1) the stimulus level for NH listeners was set at 75 dB SPL, a level above the 70 dB plateau point, and (2) the average (from 0.25 to 4 kHz) pure-tone threshold differences between the two NH groups is only 4.2 dB HL (YNH: 8.0 dB HL versus ONH: 12.2 dB HL).

Comparison Between CI and NH Listeners

The CAEP N1 and P2 latencies in CI listeners were delayed compared to NH listeners (Fig. 7a, d). This may be related to the nature of CI listening, in that stimuli delivered to CI listeners are severely degraded. When NH listeners are presented with stimuli that simulate CI listening (vocoded stimuli), their CAEPs to those vocoded stimuli may be reduced in amplitude and delayed in latency (Anderson et al. 2020; Friesen et al. 2009). The effect of stimulus degradation on CAEPs might be more apparent in older listeners (Anderson et al. 2020). Further, the latency delays of CI listeners may be a result of maladaptive neuroplastic changes associated with severe-to-profound sensorineural hearing loss. Previous studies have shown cascading negative effects of severe-to-profound sensorineural hearing loss throughout the auditory system (e.g., Eggermont 2017), which may lead to prolonged latencies of CAEPs (e.g., Billings et al. 2015; Oates et al. 2002). Finally, CI sound processors may introduce delays in the delivery of signals to listeners (Zirn et al. 2015) that lead to delayed cortical response latencies in CI listeners. If this factor determines the latency differences between CI and NH listeners, we would expect to observe systematic latency delays in CI listeners. Such expectation, however, is not fully supported by our findings because we did not observe differences between CI and NH listeners for P2 latencies in response to the speech stimulus (Fig. 7d).

Age-Related CAEP Latency Delays were Largely Independent of Stimulation Presentation Rate

The current study partially replicates Tremblay et al. (2004) with an extended ISI range (0.5 to 4 s), demonstrating that age effects on P2 latency persisted across ISIs in CI and NH listeners (Fig. 7c). However, our finding of the lack of ISI influence on age effects on N1 latency is not aligned with the findings of Tremblay et al. (2004). The effect of stimulus presentation rate on CAEPs is thought to relate to neural refractory mechanisms (Pereira et al. 2014). Aging may negatively affect the refractory mechanisms, leading to impaired temporal processing in older listeners (Herrmann et al. 2016; Walton et al. 1998). Such a discrepancy in N1 latency may be related to methodological differences between our study and Tremblay et al. (2004). For example, compared to our ISI range (0.5 to 4 s), Tremblay et al. (2004) used a more limited range of ISIs (0.51, 0.91, and 1.51 s) that were skewed to faster rates. Because we included middle-aged adults into our younger groups (YNH: 29.1–59.7 years and YCI: 28.9–60.8 years), our groups were relatively older than the younger group tested in Tremblay et al. (2004; 21–33 years). Future research is needed to resolve the mixed findings of the stimulus presentation rate effect.

Limitations with the Current Study

In the current study, three CI listeners in the younger group had an early onset of deafness (< 2 years). Prior work suggests that prelingually deafened CI listeners with long DoD (which was the case for 1 YCI; see Table 1) may exhibit earlier CAEPs (e.g., N1) than postlingual deafened CI listeners (e.g., Lammers et al. 2015). Hence, age at onset of deafness might confound the effect of age (at testing) on CAEPs in our study. However, age at onset of deafness was highly correlated with age at testing in our sample (Spearman’s Rho = 0.96, p < 0.05). Because of this variable collinearity issue, we could not include age at onset of deafness in our analyses. Future studies should include larger sample sizes to determine the relative effects of age at testing and onset of deafness on CAEPs with larger sample sizes.

CONCLUSIONS

In CI and NH listeners, cortical responses to tone and speech stimuli (e.g., P2 latencies: Fig. 7c) appear to decline with advancing age. Such age-related changes in cortical responses seem to be largely independent of stimulus presentation rate. As the peripheral-cochlear processing is bypassed in CI listeners, these findings provide evidence that age-related changes beyond the cochlea may contribute to age-related temporal processing deficits. Degraded temporal processing may limit the ability of older listeners to restore speech understanding with CIs.

APPENDIX 1

Quantification of Test–Retest Reliability

The test–retest reliability was calculated as the maximal cross-correlation (lag range ± 5 sampling points) across 0 to 350 ms of the response waveforms between the two sessions, using the ccf function implemented in R version 3.6.2 (Team 2013). The time range (0 to 350 ms) encompasses the P1-N1-P2 complex. The correlation coefficients ranged from −1 to 1. The cross-correlation analysis was applied to response waveforms from each electrode for each condition in individual listeners. To improve data normality, the correlation coefficients were converted into Fisher’s Z scores, which were used for later analyses.

Statistical Analysis

A linear mixed-effects model implemented via the lme4 package (Bates et al. 2014) in R version 3.6.2 (Team, 2013) was used to fit the data for test–retest reliability. The model included the following fixed effects: age at testing, hearing status (NH and CI), stimulus type (tone or speech), and ISI (0.5, 1, 2, 3, or 4 s). Age was centered using the mean age of all CI and NH listeners and was treated as a continuous variable. Hearing status, stimulus type, and ISI were treated as categorical variables. In the model, the reference levels were CI, an ISI of 0.5 s, and tone. The initial random effects were set as (1 | stimulus type × ISI | subject) + (1 + stimulus type × ISI | electrode).

To reduce the risk of data overfitting, we systematically remove random and fixed effects that did not contribute significantly to the model (p > 0.05) using the step function in the lmerTest package (Kuznetsova et al. 2017). Results from the simplest, best-fitting model were reported in the “Results” section. We computed the significance values for fixed effects in the optimal model using the anova function in the lmerTest package (Kuznetsova et al. 2017). We conducted post hoc analysis for significant fixed effects, if necessary, with the emmeans (for categorical variables) and emtrends (for continuous variables) functions in the emmeans package (Lenth et al. 2018a). Multiple comparisons were corrected by the Bonferroni method.

Results

Figure 9a, b display grand-average waveforms for the four groups comparing the two test sessions. The response waveforms were highly consistent across sessions. The addition of the sex variable into the optimal model for test–retest reliability significantly improved model fit (p = 0.045), wherein CAEPs from male listeners exhibited lower test–retest reliability than female listeners [F(1, 40.0) = 6.080, p = 0.018].

Fig. 9.

Grand average waveforms to 1000-Hz tone (a) and syllable /da/ (b) for sessions 1 (gray thick lines) and 2 (black thin lines) averaged across the eight frontocentral electrodes for YCI, OCI, YNH, and ONH listeners across five inter-stimulus intervals (0.5, 1, 2, 3, and 4 s). Significant hearing status by stimulus type interaction (c) and hearing status by interstimulus interval (ISI) interaction (d) on cross-correlation coefficient (CC coef.). The CC coef. quantified the similarity of waveforms (0 to 350 ms) from the two sessions, which indexed test–retest reliability. The violin plots show distributions of raw data. Horizontal lines inside the violin plots represent individual observations. The length of each of those horizontal lines represents the number of observations at a given CC coef. relative to the maximum number of observations across all CC coefs. The line plots show means ± 1 standard deviation. Data for CI listeners (all YCI and OCI) are displayed in orange/thin lines/diamonds; data for NH listeners (all YNH and ONH) are displayed in black/thick lines/squares. YCI: younger to middle-aged adult cochlear-implant users, OCI: older adult cochlear-implant users, YNH: younger to middle-aged adults with normal hearing, ONH: older adults with normal hearing. *p < 0.05; **p < 0.01; ***p < 0.001

Of our primary interest, the main effect of age is not significant [F(1, 40.8) = 0.801, p = 0.376]. But the age × hearing status × ISI interaction was significant [F(4, 40.6) = 3.405, p = 0.017]. In addition, the non-age-based main effects hearing status [F(1, 39.8) = 16.918, p < 0.001] and ISI [F(4, 40.1) = 42.978, p < 0.001] were significant. The non-age-based interactions hearing status × stimulus type [F(1, 38.9) = 9.700, p = 0.003] and hearing status × ISI [F(4, 40.1) = 4.145, p = 0.007] were significant. Finally, other interactions (age × hearing status or age × ISI) were not significant (both ps > 0.260).

To understand the age × hearing status × ISI interaction, we compared the age effects across hearing status and ISIs. The interaction was probably driven by that the estimated age effect at the ISI of 0.5 s in NH listeners was significantly smaller than 0 (95% confidence interval: [−0.023, 0.0037]; p = 0.038, however, p = 0.374 after multiple-comparison correction), wherein responses from older NH listeners might exhibit lower test–retest reliability at the ISI of 0.5 s. But the estimated age effects at any other ISIs in NH listeners and any ISIs in CI listeners were not significantly different from zero (95% confidence intervals: [−0.019, 0.019]; all ps > 0.210 before multiple-comparison correction).

To understand the hearing status × stimulus type interaction, we compared the effect of hearing status across stimulus type and vice versa. The comparisons are displayed in Fig. 9c. The effect of hearing status on test-rest reliability (i.e., lower reliability in CI vs. NH listeners) was driven by the speech stimulus (p < 0.001) but not by the tone stimulus (p = 0.241). The effect of stimulus type on test-rest reliability (i.e., lower reliability in response to speech vs. tone) was driven by CI listeners (p = 0.036) but not NH listeners (p = 0.425).

To understand the hearing status × ISI interaction, we compared the effect of hearing status across ISIs and the effect of ISI across hearing status. The comparisons are displayed in Fig. 9d. The effect of hearing status on test-rest reliability (i.e., lower reliability in CI vs. NH listeners) was driven by ISIs ≥ 2 s (all ps < 0.05) but not by ISIs < 2 s (both ps > 0.23). For responses from CI listeners, the test-test reliability was lower for ISIs of 0.5 and 1 s compared to ISIs of 3 and 4 s (all ps < 0.05). The test-test reliability was not significantly different across other ISI comparisons (all ps > 0.068). For responses from NH listeners, the test–retest reliability significantly improved as the ISI increased from 0.5 s up to 2 s (0.5 s < 1 s < 2 s = 3 s = 4 s; all ps < 0.05).

Discussion

The test–retest reliability of CAEPs was lower in CI than NH listeners, particularly in response to the speech stimulus (Fig. 9c) and at longer ISIs (≥ 2 s: Fig. 9d). Simply, this may reflect that when processing stimuli, CI devices generate artifacts and contaminate the cortical responses, which leads to a reduction in test–retest reliability of CAEPs in CI listeners. But this explanation may not fully account for the lower test–retest reliability of CAEPs in CI listeners, considering that the test–retest reliability difference between CI and NH listeners is ISI-specific (i.e., at ISIs ≥ 2 s: Fig. 9d) and CAEPs at those ISIs (Fig. 9a, b) tend to show relatively high SNRs. The lower test–retest reliability of CAEPs in CI listeners may relate to biological changes associated with hearing impairments in CI listeners. For instance, CI listeners have undergone pathological changes in the auditory system, e.g., changes in the functioning of auditory nerve fibers, which can negatively influence their responses to CI electrical stimulation (Sly et al. 2007). As a result, cortical responses from CI listeners may become less reliable.

Furthermore, the test–retest reliability of CAEPs in CI and NH listeners generally improved with slower stimulus presentation rate (i.e., longer ISIs) and seemed to asymptote at an ISI around 2 s (Fig. 9d). Hence, practically, an ISI close to this value (i.e., 2 s) is recommended in future CAEP studies with relatively long stimuli (~500 ms) to ensure relatively high test-test reliability while maintaining a reasonable amount of test time (e.g., Friesen and Tremblay 2006).

APPENDIX 2

Waveforms across the 32 electrodes before (red lines) and after (blue lines) the removal of CI-related artifacts at an example condition from an example CI listener. For this listener, the stimuli were presented via the right ear. Note that, due to the presence of CI-related artifacts, response amplitudes before artifact removal are generally much larger than those after artifact removal. To help visualize them in the same plots, the amplitudes were scaled into the range of −1 to 1 by dividing individual amplitudes by the maximal absolute amplitude. The scaling was conducted separately for responses before and after artifact removal.

Author Contribution

Conceptualization: Zilong Xie, Olga Stakhovskaya, Matthew J. Goupell, Samira Anderson. Methodology: Zilong Xie, Olga Stakhovskaya, Matthew J. Goupell, Samira Anderson. Formal analysis and investigation: Zilong Xie, Olga Stakhovskaya, Matthew J. Goupell, Samira Anderson. Writing—original draft preparation: Zilong Xie. Writing—review and editing: Zilong Xie, Olga Stakhovskaya, Matthew J. Goupell, Samira Anderson. Funding acquisition: Matthew J. Goupell, Samira Anderson; Resources: Matthew J. Goupell, Samira Anderson; Supervision: Olga Stakhovskaya, Matthew J. Goupell, Samira Anderson.

Funding

Research reported in this publication was supported by the National Institute on Aging of the National Institutes of Health under Award Number R01 AG051603 (M.J.G.).

Availability of Data and Material

All relevant data are within the manuscript.

Code Availability

Codes are available upon request.

Declarations

Ethics Approval

All materials and procedures were approved by the Institutional Review Board at the University of Maryland.

Consent to Participate

Written informed consent was obtained from all listeners.

Conflict of Interest

The authors declare no competing interests.