Modeling the relationship between psychophysical perception and electrically evoked compound action potential threshold in young cochlear implant recipients: clinical implications for implant fitting

Clinical Neurophysiology 115 (2004) 2811–2824 www.elsevier.com/locate/clinph

Modeling the relationship between psychophysical perception and electrically evoked compound action potential threshold in young cochlear implant recipients: clinical implications for implant fitting* Hung Thai-Vana,b,*, Eric Truya,b, Basile Charassea, Florent Boutitiec, Jean-Marc Chanala, Nadine Cochardd, Jean-Pierre Pirone, Se´bastien Ribasd, Olivier Deguined, Bernard Frayssed, Michel Mondaine, Alain Uziele, Lionel Colleta,b b

a CNRS UMR 5020, IFR 19, Institut Fe´de´ratif des Neurosciences de Lyon, Universite´ Claude Bernard, Lyon I, France Unite´ d’implantation cochle´aire, CNRS UMR 5020 ‘Neurosciences et Syste`mes Sensoriels’, Pavillon U—Hoˆpital Edouard Herriot, Place d’Arsonval, 69437 Lyon Cedex 03, France c Service de Biostatistiques, Universite´ Claude Bernard, Lyon, France d Service d’Otologie et d’Otoneurologie, Hoˆpital Purpan, Toulouse, France e Unite´ d’implantation cochle´aire, Hoˆpital Gui de Chauliac, Montpellier, France

Accepted 17 June 2004 Available online 23 September 2004

Abstract Objective: In cochlear implant recipients, the threshold of the electrically evoked compound action potential (ECAP) has been shown to correlate with the perceptual detection threshold and maximum comfortable loudness levels (respectively, T- and C-levels) used for implant programming. Our general objective was to model the relationship between ECAP threshold and T/C-levels by taking into account their relative changes within each subject. In particular, we were interested in investigating further the validity of ECAP threshold as a predictor of psychophysical levels, depending on intra-cochlear electrode location and time of testing (from 1 to 18 months post-implantation). Methods: A total of 370 ECAP thresholds, measured in 49 children, using a Nucleusw 24 cochlear implant, were compared with the corresponding T- and C-levels obtained at the same visit, for the same electrode. Response profiles for the whole group of patients were modeled across four test electrodes spaced equally along the electrode array from base towards apex. A linear regression model was constructed and the quality of the ECAP threshold-based predictions was assessed by testing for correlation between measured and predicted psychophysics. Comparison was made with a more simplistic model (described here as the ‘parallel profiles method’) stipulating, within each subject, a 1 mA increase in psychophysical levels for every 1 mA increase in ECAP threshold. Results: Offset between ECAP threshold and psychophysics profiles was found to vary significantly along the electrode array for the T-, but not for the C-level. In contrast with the parallel profiles method, our regression model predicted, within each subject, an average increase of 0.23 mA (95% confidence interval: 0.18–0.28) in T-level for every 1 mA increase in ECAP threshold. This correction improved the quality of T-level prediction when our model was run using measured T-level and ECAP threshold from a reference electrode (rZ0.77 vs. rZ0.62). The shorter the distance between the electrode for which T-level was predicted and the one used as reference, the stronger the correlation between measured and predicted T-levels. In addition, poorer T-level predictions were obtained at the basal end of the array during the first 3 months post-implantation. In contrast to T-level, individual changes in C-level with ECAP threshold exhibited heterogeneous patterns across subjects so that no common coefficient could account for these changes. However, applying the parallel profiles method led to highquality C-level prediction. Conclusions and significance: The results suggest that covariation between ECAP thresholds and psychophysics plays a decisive role in the relationship of ECAP threshold with T-, but not with C-level. Therefore, our regression model and the parallel profiles method should

*

Part of the results reported in this paper were presented at the Seventh European Symposium on Paediatric Cochlear Implantation and also at the 11th NRTe research workshop, Geneva, Switzerland, May 2–5, 2004. * Corresponding author. Address: Unite´ d’implantation cochle´aire, CNRS UMR 5020 ‘Neurosciences et Syste`mes Sensoriels’, Pavillon U—Hoˆpital Edouard Herriot, Place d’Arsonval, 69437 Lyon Cedex 03, France. Tel.: C33-4-72-11-05-03; fax: C33-4-72-11-05-04. E-mail address: [email protected] (H. Thai-Van). 1388-2457/$30.00 q 2004 International Federation of Clinical Neurophysiology. Published by Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.clinph.2004.06.024

2812

H. Thai-Van et al. / Clinical Neurophysiology 115 (2004) 2811–2824

both be used for predicting, respectively, the T- and the C-levels. Although the predictability of our regression model seems to be better for middle and apical electrodes, its utilization should be extended to basal electrodes after 6 months’ implant use. q 2004 International Federation of Clinical Neurophysiology. Published by Elsevier Ireland Ltd. All rights reserved. Keywords: Cochlear implant; Compound action potential; Psychophysical levels; Neural response telemetry; Hierarchical regression analysis; Jack-knife validation

1. Introduction 1.1. Importance of cochlear implant programming: the psychophysical levels Multiple-channel cochlear implants overcome basilar membrane or cochlear hair cell damage by encoding acoustic information and providing electrical stimulation directly to the auditory nerve. Spectral information is transmitted to the auditory nerve by allocating to each stimulation electrode channel a frequency range according to the cochlea’s tonotopic organization. As a result, highfrequency sounds are conveyed by electrode channels located at the basal end of the cochlear implant electrode array, intermediate-frequency sounds by electrodes located at the middle of the array, and low-frequency sounds by electrodes located at its apical end. The perceived loudness of the electrical stimuli depends on the total charge delivered, with louder sounds elicited by higher intensity levels (Shannon, 1983). However, the amount of current that can elicit an auditory sensation greatly varies between subjects. It also varies with the intra-cochlear location of the electrode channel. As a consequence, electrical stimulation amplitudes need to be set individually for each cochlear implant recipient. This process, known as ‘programming’ or ‘mapping’ the cochlear implant speech processor, is performed at regular intervals during the post-operative course. The mapping process aims at determining the appropriate dynamic range of electrical stimulation for each electrode channel. The dynamic range is the difference between the perceptual detection threshold (T-level) and the maximum comfortable loudness (C-level). The T-level is the lowest intensity level that can consistently elicit an auditory sensation. It is determined, in adults and older children, by delivering increasing stimulation levels until the subject attests to perception. In younger children, its estimation usually requires behavioral techniques. The C-level represents the highest intensity level permitted for a particular electrode channel and is similar to the maximum power output level in a conventional hearing aid (Rance and Dowell, 1997). It is defined as the highest intensity level that does not elicit an uncomfortably loud sensation. Both psychophysical measurements are stored digitally for each electrode channel, allowing the speech processor to present the encoded sounds at intensity levels lying between the T- and C-levels.

1.2. Auditory nerve electrically evoked compound action potential In adults, cochlear implant mapping is generally an uncomplicated task, since the aim of the T- and C-level measurements is well understood by most of the patients. It is, however, more difficult and time-consuming in young children and infants, requiring the competence of skilled audiologists experienced in the behavioral techniques for hearing function assessment (Section 1.3). The success of such techniques depends not only on the child’s cognitive maturity, but also on his/her hearing status and degree of auditory experience prior to cochlear implantation. Furthermore, young children’s cooperation over long mapping processes may fluctuate. With the increasing numbers of young children and infants receiving a cochlear implant, audiologists face the challenge of optimizing their cochlear implant mapping techniques. Recent improvements in electrophysiological measurement via cochlear implants have opened up interesting horizons for the management of young cochlear implant recipients. With the latest generation of Nucleusw cochlear implants, it is possible to record the auditory nerve electrically evoked compound action potential (ECAP) in situ. The characteristics of this physiological measurement were initially studied in patients equipped with the Ineraid system—a cochlear implant with a percutaneous plug—(Brown et al., 1990), or during surgery by means of a temporary intra-cochlear electrode array (Gantz et al., 1994). The ECAP reflects the synchronous auditory nerve fiber activity that can be elicited by electrical stimulation through the implant. It can now be measured with the Neural Response Telemetry system (NRTe), a technology which allows successive stimulation of each of the 22 intra-cochlear electrode channels (Abbas et al., 1999; Brown et al., 1998). When the stimulus is large enough to elicit a synchronous neural response from local spiral ganglion cells, the ECAP is recorded from an adjacent electrode, then amplified, encoded and transmitted back via radio frequency code to the subject’s speech processor. Specific software communicates with the speech processor to capture, process, store and display the obtained traces on a personal computer. Several characteristics of the ECAP response, such as waveform shape, threshold and amplitude growth with stimulus level, have recently drawn the attention of a number of researchers. The most commonly encountered ECAP waveform shape consists of a distinct negative peak

H. Thai-Van et al. / Clinical Neurophysiology 115 (2004) 2811–2824

preceding a positive one of smaller amplitude (Lai and Dillier, 2000). The negative peak (N1) occurs around 0.2–0.4 ms after stimulus onset and the positive peak (P1) at around 0.6–0.8 ms (Cullington, 2000). In some cases, the N1 peak may be absent, occurring earlier than can be captured within the recording time window (Lai and Dillier, 2000). The ECAP threshold corresponds to the smallest amount of current that can elicit any of these physiological responses. 1.3. Relationship between ECAP threshold and psychophysical levels Recent studies have shown that ECAP threshold measured at post-operative test intervals with the NRTe system may correlate with psychophysical T- and C-levels in cochlear implant subjects (Brown et al., 2000; Franck and Norton, 2001; Hughes et al., 2000). However, the correlations observed were not strong enough to allow confident use of raw ECAP thresholds to estimate the psychophysical levels. Franck and Norton (2001) reported that absolute ECAP threshold in adults may lie at the T-level, C-level, or in between. Therefore, the most that can be ensured by using ECAP threshold directly for cochlear implant mapping is that a perceptible signal, probably not exceeding the maximum comfortable listening level, is at least being delivered. A positive ECAP test does not necessarily indicate that the stimulus delivered to the acoustic nerve will be centrally processed with the result of an auditory perception (Thai-Van et al., 2001). In addition, the relationship between raw ECAP threshold and psychophysical level may be affected by the refractory properties of the auditory nerve (Charasse et al., 2003). Different algorithms have been proposed for ECAP threshold-based estimation of psychophysical levels and implemented in clinical programming software (Almqvist and WillstedtSvensson, 1998; Brown et al., 2000; Smoorenburg et al., 2002). These automatic implant programming procedures may be of help during the first mapping sessions, especially in difficult-to-test children. All rely on the assumption that the offset between ECAP threshold and psychophysical level remains nearly constant from electrode to electrode within the array of a given subject. For instance, Brown et al. (2000) proposed calculating the offset for each subject between ECAP threshold and psychophysical levels measured on one electrode, then adding or subtracting the result from the ECAP threshold of each remaining test electrode. This method has the advantage of requiring only a small quantity of behavioral data (in this case, behavioral data were determined only for electrode 10). The present study aimed at investigating further the relationship of ECAP threshold to behavioral T- and C-levels within each subject, using measurements obtained at different times and for different electrodes. We assessed whether this relationship was common to all subjects in our series or whether it differed between them. The final goal

2813

was, for a given subject, to predict psychophysical levels from a measured ECAP threshold using the estimated relationship. The predictive quality of our model was assessed by testing for correlation between predicted and actual psychophysical levels. We analyzed ECAP thresholds and psychophysical data from three French pediatric cochlear implant centers: Lyon, Montpellier, and Toulouse. Data were collected from 1 to 18 months post-implantation at different intra-cochlear locations. Four stimulation electrode channels, spaced equally from the base towards the apex of the cochlea, were tested: one located at the basal end of the electrode array (electrode 5), two in the middle (electrodes 10 and 15) and one at the apical end (electrode 20). The validity of our model was examined according to the location of the tested electrode, and on the interval between cochlear implantation and time of testing.

2. Material and methods 2.1. Subjects Forty-nine children (30 girls, 19 boys) participated in the study. Four subjects had developed progressive sensorineural hearing loss. Otherwise, the cause of deafness was: congenital in 23 subjects, of infectious origin in seven subjects (meningitis: three cases; cytomegalovirus disease: three cases; toxoplasmosis: one case), hereditary in four subjects, genetic in one subject, and unknown in 10 subjects. Age at implantation ranged from 14 months to 12 years 4 months, with a mean age of 4 years 5 months. Thirty-six children were between 2 and 6 years of age at the time of implantation, 10 were above 6 years of age, and three were less than 2 years of age. All subjects met the following inclusion criteria: † children with profound hearing loss; † receivers of the Nucleusw 24 multichannel cochlear implant (CI24M system), with full insertion of the standard straight electrode array; † users of the same speech-coding strategy (‘spectral peak coding’: SPEAKe) throughout the study period.

2.2. Psychophysical levels Cochlear implant connection was performed at 10th day after surgery in one center (Lyon), and at 1 month after surgery in the others (Montpellier and Toulouse). The psychophysical levels to be compared with physiological data were collected during the 18-month period following surgery. To be included, each subject had to have been tested for psychophysical and ECAP thresholds in at least one of the following cochlear implant mapping sessions: 1–2 months post-connection (i.e. 1–3 months

2814

H. Thai-Van et al. / Clinical Neurophysiology 115 (2004) 2811–2824

post-surgery), 6–8 months post-connection (i.e. 6–9 months post-surgery), 12 months post-surgery and 18 months postsurgery. The cochlear implant speech processor (SPrinte) was programmed with the Cochlear Corporation’s diagnostic and programming system for Windows (Win-DPS software). This software let the audiologist stimulate at a level varying from 1 to 255 Cochlear Corporation programming units (p.u.) (increasing in approximately 2% steps per unit)—i.e. a non-linear progression from approximately 10 to 1.750 mA. The SPEAKe default parameters for monopolar stimulation were applied: 500 ms biphasic pulse trains delivered at a rate of 250 Hz, with pulse durationZ25 ms/phase, and inter-train intervalZ500 ms. The threshold and maximum comfort levels were obtained by audiometric testing methods adapted to the age of the child. The current gold standard for cochlear implant mapping process, adapted from Rance and Dowell (1997), was administered in the three pediatric cochlear implant centers: – In older children with adequate language skills, the methods used were similar to those used in adults. The T-level was determined with an ascending method of stimulation, adapted from the technique of Hughson and Westlake as modified by Carhart and Jerger (1959). Increasing stimulation levels were delivered in steps of 2 p.u. until the subject attested to perception, thereby giving a first approximation of the threshold. The experimenter then kept on stimulating until the signal was described as ‘clearly perceived’, before decreasing the stimulation by steps of 2 p.u. down to the extinction of any perception. The procedure was repeated for each test electrode channel, so as to check threshold reliability. C-level was also estimated by self-report, with stimulus levels increasing by 2 p.u. The subject was asked to indicate the maximum stimulus level that was comfortable for each test electrode. A visual loudness scale, covering the various degrees of perceived loudness from mild to uncomfortable, was used if the subject was able to make the symbolic transfer—a task which, however, was beyond the developmental abilities of most children under the age of 6. Here again, the procedure was gone through twice. – In younger children (ages ranging approximately between 2 and 6 years), psychophysical levels were obtained with play audiometry techniques, generally using a series of colored disks of different sizes, stacked on a peg. The child was instructed to hold a disk up to the ear and, on hearing a ‘bell’, to stack the disk on the peg. The T-level was thus determined as the lowest level eliciting a reliable response twice from the child. The audiologist kept on eliciting conditioned responses as the stimulus level was increased by steps of 2 p.u., until a loudness discomfort level was obtained. At this level, an aural palpebral reflex (eye blink) was often observed as well as signs of auditory intolerance: the child might stop

playing, throw the disks, or take refuge with a familiar person. The C-level was set at a point 70% of the way between T-level and loudness discomfort. – In children under 2 years of age, psychophysical levels were obtained with behavioral observation techniques using a similar ascending method of stimulation. Abrupt changes in facial expression, quieting, or handling of the stimulating coil were taken as behavioral indices of a signal being perceived. Loudness discomfort level was estimated by visual signs of discomfort or distress. The C-level was again set at 70% along the range from T- to loudness discomfort level. As with older children, two stimulation courses were performed, to verify reliability. Whatever the patient’s age, psychophysical levels were first determined separately for each test electrode then adjusted at the end of the mapping process, using the live speech mode. This precaution was taken to accommodate the loudness summation effect that may result from electrode interaction. At the time the speech processor was programmed, the experimenter was blind to the results of the ECAP recordings, which were performed afterwards. 2.3. ECAP recording 2.3.1. ECAP recording system The intra-cochlear array of the Nucleusw 24 multichannel cochlear implant consists of 22 electrode channels. The most basal channel is called Electrode 1 and the most apical Electrode 22. Two extra-cochlear electrodes, MP1 and MP2, act as ground for monopolar stimulation. MP1 is a small, platinum ball electrode placed by the surgeon under the temporalis muscle. MP2 is a platinum plate electrode located on the body of the internal receiver/stimulator. The NRTe system permits the Nucleusw 24 device to record the ECAP by using each of the 22 channels as stimulating or recording electrodes. In this study, ECAP measurements were obtained with version 2.04 of the NRTe software, initially developed by Dillier, Lai and Wyttenbach at Zurich University Hospital in 1995. The active recording electrode was located two electrodes apical to the stimulating one. The elicited signal was sampled at 16 points after the stimulus and, for each ECAP trace, two series of 16 points were superimposed. The ECAP response was obtained by means of forward masking paradigm first described by Brown et al. (1990). This paradigm allows cancellation of the stimulus artifact using the refractory period of the cochlear neural fibers. The method is described in more detail elsewhere (Abbas et al., 1999; Brown et al., 2000; Dillier et al., 2002). In short, a response to a probe pulse is first obtained (measurement A). In a second stimulus condition, the probe is delivered after a masker (measurement B). The AKB subtraction sequence represents the difference between auditory nerve responses in a non-refracted vs. a refracted state. Because this subtracted response is still contaminated by the masker

H. Thai-Van et al. / Clinical Neurophysiology 115 (2004) 2811–2824

artifact, response to the masker alone (measurement C) is also recorded. A last measurement (D) consists of the recording of the response with no applied stimulus. Finally, the ECAP response is given by the standard equation AKBCCKD. 2.3.2. ECAP recording procedures The ECAP was obtained with a monopolar stimulation delivered at 80 Hz. As for psychophysics measurements, the stimulus duration was set at 25 ms/phase. The stimulating electrode (electrode 5, 10, 15, or 20) was referenced to the ground electrode MP1 and the recording electrode (respectively, electrode 7, 12, 17, or 22) to MP2. The NRTe default recording parameters (gainZ60 dB, number of sweepsZ 100, delayZ60 ms) were applied initially, but could be optimized per subject. The masker advance—i.e. the interval between masker and probe—was set at 500 ms. Both probe and masker levels were initially set at between 200 and 220 p.u. (574–861 mA). To obtain the ECAP threshold, the probe level was decreased by steps of 3 p.u. (15–35 mA), the masker level being fixed. The ECAP threshold was determined by visual inspection and was defined as the lowest probe level at which a neural response could be identified (Fig. 1). This procedure has been reported as nearly equivalent to techniques using objective determination of ECAP threshold (Hughes et al., 2000). The numbers of tested subjects at each test interval, according to the number of electrodes tested and for each

Fig. 1. Determination of ECAP threshold in a 5-year-old child with 6 months of cochlear implant use. The probe level is decreased by steps of 20–35 mA, the masker level remaining at 574 mA (i.e. 200 p.u.). The ECAP waveform consists of an initial negative peak (N1) and a less prominent positive potential (P1). The ECAP threshold corresponds to the lowest probe level, which elicits an identifiable neural response. In this example, a neural response with a clear N1 peak is still seen at 353 mA (i.e. 176 p.u.).

2815

Table 1 Numbers of tested subjects at each test interval 1–3 months

12 months

18 months

(a) According to the number of electrodes tested (electrodes 5, 10, 15 and/or 20) Four electrodes tested 16 19 Three electrodes tested 3 4 Two electrodes tested 2 5 One electrode tested 2 3 Total of tested subjects 23 31

22 1 1 2 26

22 1 2 0 25

(b) For each electrode Electrode 5 Electrode 10 Electrode 15 Electrode 20

22 24 23 26

22 24 24 25

17 19 21 22

6–9 months

21 24 26 30

electrode, are summarized in Table 1(a) and (b). Overall, 12 subjects were tested at 1 mapping session, 21 at 2, 13 at 3, and 3 at 4 sessions. 2.4. Data analysis 2.4.1. Analysis of ECAP threshold and psychophysical levels across electrodes Mean response levels were modeled with respect to the test electrodes (5, 10, 15, or 20) and to the type of response (ECAP threshold, T-level, or C-level), using the measurements from all patients. A test for interaction between electrode location and type of response was performed to examine the inter-relationships between the three response profiles across electrodes. 2.4.2. Assessment of the relationship between psychophysics and ECAP threshold A total of 370 ECAP thresholds were compared with the corresponding T- and C-levels obtained at the same visit, for the same intra-cochlear electrode channel. We took advantage of the measurements across electrodes and time sessions to assess the relationship between ECAP threshold and psychophysical levels within subjects. Data were subjected to a linear regression analysis also known as the ‘hierarchical regression model’ (Ecochard and Clayton, 1998; Goldstein, 1995). In this regression analysis, the measurements were clustered within subjects in a two-level hierarchy where measurements at all test intervals and for all tested electrodes were at the lower level and subjects at the top of the hierarchy. The independent covariate was the ECAP threshold value and the dependent covariate was either the T- or the C-level. The ECAP threshold values were centered on 170 mA. T-level results were expressed using the following regression model T-levelij Z bj C ðinter-subject slope  mECAPj Þ C ðintra-subject slopej  ECAPij Þ

(1)

2816

H. Thai-Van et al. / Clinical Neurophysiology 115 (2004) 2811–2824

where ‘bj’ estimates the mean perceptual responsiveness specific to the jth patient (intercept), ‘mECAPj’ the mean ECAP threshold of the jth patient, and ‘ECAPij’ the deviation of the ith ECAP threshold value from the mean for the jth patient. The ‘inter-subject slope’ described changes in T-level across subjects as a function of mECAP. For a given patient, the ‘intra-subject slopej’ described the amount of expected change in T-level per 1 mA change in ECAP threshold. That is, the intra-subject slope actually reflected the intra-subject relationship between ECAP threshold and psychophysics. However, as no evidence of heterogeneity across the 49 intra-subject slope values was found (maximum likelihood ratio test: c2Z3.41, PZ0.15), an average intra-subject slope was considered. A similar regression analysis was performed for C-level values. Contrary to T-level, the intra-subject slopes here were found to differ across subjects (c2Z6.96, P!0.01). 2.4.3. Prediction of psychophysical level from ECAP threshold The validity of ECAP threshold as an indirect marker of the psychophysics was then tested. For each subject, the usefulness of a reference couple comprising ECAP threshold and psychophysical values for one electrode at one mapping session was assessed for predicting psychophysics according to ECAP threshold for other electrode locations and/or other test sessions. The predictions were obtained using a formula deduced from model (1), using a common intra-subject-slope ðT-leveln KT-level0 Þ Zintra-subject slopeðECAPthreshold n KECAPthreshold 0 Þ ð2Þ where ‘T-leveln‘ was the psychophysical value to be predicted, ‘ECAPthreshold n‘ the corresponding physiological measure, and ‘T-level0’ and ‘ECAPthreshold 0’ the reference couple of values. The validity of the predictions was estimated using the ‘jack-knife’ procedure (Rodgers, 1999). Each of the 49 subjects was successively excluded for determination of the ‘intra-subject slope’ value used in Eq. (2). That is, the value of the intra-subject slope was estimated on the remaining 48 subjects and used for prediction on the 49th subject following Eq. (2). The reference couple of T-level0/ ECAPthreshold 0 values were selected at random. The procedure was repeated for every subject. Correlation between predicted and observed T-levels was assessed by a parametric Pearson test. Eq. (2) was then rerun to assess the validity of the predictions according to intra-cochlear electrode location and time of testing. Each of the four electrode locations tested was successively used as reference (providing the reference couple of T-level0/ECAPthreshold 0 values),

T-leveln being predicted for the others. Again, r Pearson correlation coefficients were calculated between observed and predicted psychophysical values to assess the quality of the predictions. After appropriate transformation, these correlation coefficients were fitted with a linear regression to assess the effect of (1) time post-implantation; (2) the location of the electrode at which T-levels were predicted; and (3) the distance between the reference electrode and the electrode targeted for prediction. Effects were considered as significant for a P value !0.05. It was planned to use a similar procedure for the prediction of C-levels. 2.4.4. Comparison with ‘parallel profiles method’ Data were also subjected to a predictive model adapted from Brown et al. (2000), described here as the parallel profiles method. Based on the assumption of a constant offset between ECAP threshold and T/C levels across electrodes, this model stipulates that psychophysical levels for a given electrode (Electroden) can be predicted using the offset between ECAP threshold and psychophysical levels from a reference electrode (Electrode0) as follows ðT-leveln KECAPthreshold n Þ Z ðT-level0 KECAPthreshold 0 Þ 5ðT-leveln KT-level0 Þ ZðECAPthreshold n KECAPthreshold 0 Þ

ð3Þ

A similar formula was used for C-level predictions. Note that the parallel profiles method is equivalent to Eq. (2) for an intra-subject regression slope value equal to 11. Following Brown et al. (2000), offset values from electrode 10 served as reference, so that psychophysical predictions were made for electrodes 5, 15 and 20. Correlation coefficients between the predicted and actual psychophysical values were calculated for comparison with those obtained using our model. Statistical analyses were performed with the SAS 8.0 software package (SAS Institute, SAS Campus Drive, Cary, NC). 1 With the latest Nucleusw cochlear implants, the programming psychophysical levels are behaviorally measured by varying the amplitude of the electrical biphasic pulse (microampere) delivered by the implant, the pulse width being fixed. Although being logarithmically related to the actual stimulation current expressed in microampere, the Cochlear Corporation’s programming units (p.u.) used in routine are specific to the manufacturer’s programming software. They can be, however, easily converted to ‘microampere’ using the manufacturer’s conversion table (reference: Nucleusw Technical Reference Manual 3.38). In the context of this article, all results were expressed using the standard international unit (i.e. microampere). The ‘parallel profiles method’ we compared with Eq. (2) was accordingly tested using ‘microampere’. Consequently, it differed from the method proposed by Brown et al. (2000) which was established using Cochlear p.u.

H. Thai-Van et al. / Clinical Neurophysiology 115 (2004) 2811–2824

2817

3. Results 3.1. Modeling response profiles Response levels according to test electrodes (5, 10, 15, and 20) and type of response (ECAP threshold, T-level, C-level) are summarized in Table 2. Overall, no significant effect of age at implantation on ECAP threshold, T- or C-level was found. Total variance in T-level data was 3221, with 54% due to inter-subject variability. Total variance in C-level data was 29,229, with 47% due to inter-subject variability. Fig. 2 shows the modeling of response profiles across the four test electrodes. ECAP threshold was on average 130.8 mA higher and C-level 350.3 mA higher than T-level. All other electrodes showed significantly higher response levels than electrode 20 (used as reference in this model), the highest being observed for electrode 10. No statistical interaction between the three response profiles and the test electrode was found (PZ0.19). However, when the interaction analysis was limited to T-level and ECAP threshold profiles, a significant interaction was found (P!0.0001). The offset between T-level and ECAP threshold was significantly higher for all other electrodes than for electrode 20. In contrast, the offset between C-level and ECAP threshold was not found to differ significantly across electrodes. 3.2. Hierarchical regression analysis between ECAP threshold and T-level The slope of the relationship between mECAP and T-level across subjects was estimated as equal to 0.55 (standard error: 0.10). This means that the T-level increased by 0.55 mA when mECAP increased, across subjects, by 1 mA (Fig. 3A). The estimated slope of the linear function Table 2 Response level (mA) according to test electrode and type of response Type of response

Test electrode

Number of measurements

Mean

Standard deviation

C-level

5 10 15 20 All electrodes 5 10 15 20 All electrodes 5

82 91 94 103 370 82 91 94 103 370 82

620.5 650.4 634.1 599.9 625.6 274.1 283.3 277.1 267.5 275.3 405.1

162.5 168.7 164.6 184.2 171.2 55.5 57.2 60.3 57.0 57.6 67.8

10 15 20 All electrodes

91 94 103 370

428.4 422.8 372.0 406.1

75.7 77.9 84.6 80.2

T-level

ECAP threshold

Fig. 2. Modeling of T-level, C-level and ECAP threshold profiles across the four test electrodes. Mean responses (G2 standard errors of the mean) are plotted. Note the absence of parallelism between T-level and ECAP threshold contours.

linking individual changes in T-level with ECAP threshold values was also positive, but smaller (Fig. 3B). When the ECAP threshold increased, for a given subject, by 1 mA, the regression analysis predicted that T-level would increase by 0.23 mA on average (standard error: 0.03; 95% confidence interval: 0.18–0.28). Because no significant difference between individual intra-subject slopes was found, the results of the hierarchical regression analysis between ECAP threshold and T-level were modeled as follows T-levelij Z bj C 0:55 mECAPj C 0:23 ECAPij

(4)

3.3. Hierarchical regression analysis between ECAP threshold and C-level The estimated slope of the relationship between mECAP and C-level across subjects was 1.45 (standard error: 0.28). That is, the C-level increased by 1.45 mA when mECAP increased, across subjects, by 1 mA (Fig. 4A). Individual changes in C-level with ECAP threshold values are depicted in Fig. 4B. When the ECAP threshold increased, for a given subject, by 1 mA, the model predicted that C-level would increase by 0.47 mA on average (standard error: 0.14). However, many subjects significantly departed from this average value because of heterogeneity in individual slopes. 3.4. Prediction of psychophysical levels using the hierarchical regression model 3.4.1. T-levels The wide inter-subject variability in T-level data (accounting for more than the half of the total variance) had to be taken into account because it could preclude direct prediction of T-level from a given ECAP threshold value. We reasoned that this could be neutralized by combining the ECAP threshold data with one randomly selected T-level value per subject in Eq. (2) (Section 2). This enabled 321

2818

H. Thai-Van et al. / Clinical Neurophysiology 115 (2004) 2811–2824

Fig. 3. Hierarchical regression model linking ECAP threshold and T-level. (A) Linear regression, across subjects, between T-level and ECAP threshold. Each plot is an average per subject across electrodes and time. (B) Predicted patterns of individual changes in T-level with ECAP threshold value using the estimated common intra-subject slope for all subjects (in the absence of heterogeneity between individual slopes). Each line represents a single subject.

Fig. 4. Hierarchical regression model linking ECAP threshold and C-level. (A) Linear regression, across subjects, between C-level and ECAP threshold. Each plot is an average per subject across electrodes and time. (B) Predicted patterns of individual changes in C-level with ECAP threshold value. Each line represents a single subject. Note that no common intra-subject slope could be estimated by the model because of significant heterogeneity between individual slopes.

H. Thai-Van et al. / Clinical Neurophysiology 115 (2004) 2811–2824

2819

Table 3 Correlation coefficients between predicted and observed T-levels using the hierarchical regression model (depending on electrode location and the cochlear implantation to time of testing interval) Test session All sessions

1–3 months post-implant

6–9 months post-implant Fig. 5. Scatter plot showing the relationship between the observed T-levels and the respective values predicted by the hierarchical regression model (nZ321 comparisons).

T-level values to be predicted. As shown in Fig. 5, a strong correlation was found between predicted and observed T-levels (rZ0.77). Notably, a weaker correlation was found (rZ0.62) when the intra-subject regression value was taken as being equal to 1, rather than being estimated from model (1) (i.e.Z0.23). Eq. (2) was applied as follows: a reference couple of T-level and ECAP threshold values (e.g.Z280 and 430 mA, respectively) was taken. ECAP threshold was measured on another electrode and/or at another time session (e.g.Z380 mA), and then the corresponding T-level was predicted as T-leveln Z 280 C 0:23ð380 K 430Þ Z 268:5 mA 3.4.2. C-levels Since the patterns of individual change in C-level with ECAP threshold highly differed across subjects, no common intra-subject regression slope could be used in Eq. (2). Therefore, no prediction of C-level could be made using our hierarchical regression model. 3.4.3. Influence of electrode location and test session on T-level prediction Both intra-cochlear electrode location and time of testing were found to influence the accuracy of the predictions enabled by our model. This was shown by using successively each of the four tested electrodes as reference, with T-levels predicted for the others. The Pearson correlation coefficients found between predicted and measured T-levels are detailed in Table 3. Overall, by pooling the results obtained over time, the model was found to account for 40–81% of the variance in the data, depending on the location of the reference electrode (r ranging between 0.63 and 0.90). Looking at the changes with respect to time and to the electrode at which T-levels were predicted, correlation coefficients were on average

12 months post-implant

18 months post-implant

Reference electrode

Electrodes at which T-levels were predicted 5 (n)

10 (n)

15 (n)

20 (n)

5 10 15 20 5 10 15 20 5 10 15 20 5 10 15 20 5 10 15 20

– 0.83 (81) 0.74 (80) 0.63 (82) – 0.59 (17) 0.52 (16) 0.28 (17) – 0.79 (20) 0.65 (20) 0.55 (21) – 0.78 (22) 0.55 (22) 0.51 (22) – 0.91 (22) 0.88 (22) 0.81 (22)

0.85 (81) – 0.87 (85) 0.83 (91) 0.67 (17) – 0.77 (18) 0.83 (19) 0.87 (20) – 0.86 (21) 0.80 (24) 0.78 (22) – 0.80 (23) 0.83 (24) 0.92 (22) – 0.94 (23) 0.91 (24)

0.79 (80) 0.89 (85) – 0.90 (92) 0.63 (16) 0.79 (18) – 0.85 (20) 0.74 (20) 0.87 (21) – 0.93 (25) 0.57 (22) 0.81 (23) – 0.87 (23) 0.91 (22) 0.95 (23) – 0.92 (24)

0.70 (82) 0.86 (91) 0.90 (92) – 0.36 (17) 0.87 (19) 0.86 (20) – 0.67 (21) 0.82 (24) 0.93 (25) – 0.59 (22) 0.87 (24) 0.88 (23) – 0.85 (22) 0.92 (24) 0.92 (24) –

For each test session and each reference electrode, the bold number indicates the highest correlation coefficient and italic number the lowest. The number of predictions per condition is shown within brackets.

stronger with time post-implantation (linear trend: P!0.0001), and weaker on electrode 5 than on any of the other three (P!0.0004). Moreover, correlation coefficients became significantly weaker as the distance between the electrode targeted for prediction and the reference electrode increased (linear trend: P!0.001). Thus, feeding our predictive model with ECAP threshold and T-level from electrode 5 as reference could lead to sizeable errors when predicting T-levels at electrode 20 and vice versa, at least during the first year post-implantation. From 6 to 9 months post-implantation onwards, electrode 5 could, however, be used to predict T levels on electrode 10, and electrode 10 to predict T levels on electrode 5 (rR0.78). Finally, at 18 months post-implantation, predicted T-levels were found to be highly correlated with actual psychophysical measures whatever the respective electrode locations: at that session, the model accounted for 66–90% of the variance in the data (r ranging between 0.81 and 0.95). 3.5. Prediction of psychophysical levels using the ‘parallel profiles method’ Table 4 shows the correlation coefficients between the measured psychophysical levels and the respective values predicted with the parallel profiles method. For C-level predictions, a high correlation was found whatever the test electrode and time of testing. The parallel profiles method was able to account for about 55–88% of the variance in

2820

H. Thai-Van et al. / Clinical Neurophysiology 115 (2004) 2811–2824

Table 4 Correlation coefficients between predicted and observed psychophysical levels using the ‘parallel profiles method’ Test session

Electrodes at which T-levels were predicted 5 (n)

(a) T-level predictions All sessions 0.46 (82) 1–3 months 0.14 (17) post-implant 6–9 months 0.15 (21) post-implant 12 months 0.70 (22) post-implant 18 months 0.73 (22) post-implant (b) C-level predictions All sessions 0.84 (82) 1–3 months 0.79 (17) post-implant 6–9 months 0.83 (21) post-implant 12 months 0.74 (22) post-implant 18 months 0.91 (22) post-implant

15 (n)

20 (n)

0.77 (90) 0.64 (19)

0.77 (91) 0.73 (19)

0.63 (24)

0.68 (24)

0.67 (23)

0.80 (24)

0.93 (24)

0.88 (24)

0.92 (90) 0.94 (19)

0.88 (91) 0.92 (19)

0.86 (24)

0.77 (24)

0.90 (23)

0.90 (24)

0.94 (24)

0.92 (24)

C-level data (r ranging between 0.74 and 0.94). In contrast, for T-level predictions, correlation coefficients were found to be systematically lower than those allowed by our model with electrode 10 as reference. The lowest correlations were found for predictions on electrode 5 in early test sessions (r ranging between 0.14 and 0.15). 4. Discussion 4.1. Usefulness of ECAP threshold for predicting psychophysics in children with cochlear implants The results of the present study shed further light on the relationship between ECAP threshold and the psychophysics required for the cochlear implant mapping process. First of all, based on data collected in a unique series of 49 children with cochlear implants, a linear model can be proposed for predicting T-level from ECAP threshold. A distinctive characteristic of this model is to take account of the intrasubject relationship between ECAP threshold and T-level. Thirty-seven children out of the 49 were tested at two or more mapping sessions. The question may, however, be raised that not all subjects had all electrodes tested at every session, thus affecting the precision of the parameter estimates. This said, missing ECAP data are unlikely to have biased the parameter estimates. The main reason for ECAP data being missing related to the child’s fatigue and inability to undergo ECAP measurement on four electrodes after a long mapping process. As a consequence, unmeasured data are unlikely to differ widely from measured ones. Another topic requiring discussion is that psychophysical level and ECAP threshold

were determined using different stimulus presentation rates (250 and 80 Hz, respectively). This difference did not prevent a consistent link being established between ECAP threshold and T-level. Future studies should verify the validity of the model in users of other speech-coding strategies than SPEAKe: namely, the ‘advanced combination encoder’ (ACEe) or the ‘continuous inter-leaved sampling’ (CISe) strategy. Both allow even higher stimulation rates, ranging from 600 to 1800 Hz. Nevertheless, it should be underlined that ECAP threshold was not found to be influenced, at least in adults, either by the individual optimal strategy—whether SPEAKe, ACEe, or CISe—or speech comprehension differences between strategies (Kiefer et al., 2001). The observation that the T-level can be related to ECAP threshold is not surprising given the link described in the literature between both measures and structural characteristics of the peripheral auditory system. Histological studies conducted in animals have linked the ECAP amplitude measured from a single intra-cochlear electrode to the total number of surviving spiral ganglion cells (Shepherd and Javel, 1997). They also suggest that certain ECAP features—maximum P1 amplitude, and P1 and P1–N1 growth functions—may reflect auditory nerve status (Hall, 1990). Interestingly, psychophysical performance in animals with a cochlear prosthesis has also been found to depend on neural survival (Pfingst et al., 1981, 1985; Pfingst and Sutton, 1983). Examination of human implanted temporal bones has highlighted the link between spiral ganglion cell-count and psychophysical detection threshold. T-level was found to correlate positively with the presence of intra-cochlear fibrous tissue, whereas spiral ganglion cell survival was found to correlate negatively with the level of fibrous tissue and ossification of the cochlea (Kawano et al., 1998). These findings support the notion that T-level in human cochlear implant recipients can be related to peripheral auditory system physiological processes. Nevertheless, such a relationship remains limited by a large inter-subject variability in the responsiveness to stimulation via the cochlear implant. In our series, more than half of the total variance in T-level data was related to inter-subject variability. This clearly indicates the need, when estimating perceptual levels from ECAP thresholds, to consider each subject’s perceptual responsiveness.That C-level data exhibited even greater variance than T-level data may possibly be explained by the difficulty in determining the upper limits of electrical stimulation in children, unlike detection threshold setting. Visual loudness scale or pictorial representations of auditory comfort cannot generally be used before 7 years of age (Macpherson et al., 1991). The gold standard for C-level setting that was applied in all children under 6 years of age used predetermined variations from the loudness discomfort level. Whereas the T-level was determined directly, the C-level was deduced from a complex psychophysical percept giving

H. Thai-Van et al. / Clinical Neurophysiology 115 (2004) 2811–2824

rise to signs of auditory intolerance or distress (Hodges et al., 1997). Such reactions have been reported to be influenced by fluctuating criteria for comfort (and discomfort) in cochlear implantees (Spivak et al., 1994). This may explain the great variations in C-level within and between subjects. 4.2. Comparison with previous studies Previous studies have stressed the necessity of applying correction factors specific to each subject in order to predict psychophysical levels from ECAP threshold with reasonable accuracy. Franck and Norton (2001) hypothesized that the subject’s ECAP amplitude growth function may be of help for predicting both C-level and dynamic range. As a preliminary approximation, they recommended setting the T-level contour across electrodes as the ECAP threshold contour. The dynamic range was then mathematically approximated using the ECAP amplitude growth function, although a rather low correlation (rZ0.37) was found between that function and the dynamic range. The C-level contour was then obtained by adding the dynamic range approximated for each electrode to the T-level contour. Two constant shifts were finally determined per subject then applied, respectively, to the previous T- and C-level contours to closely approximate the actual psychophysics. Remarkably, these corrections were made in live-voice mode (i.e. using all electrodes), to be consistent with normal utilization of the implant. To our knowledge, the feasibility of this method has not been evaluated in pediatric populations (Franck, 2002; Franck and Norton, 2001). In adults, Brown et al. (2000) obtained highly significant correlations between actual psychophysics and those predicted from ECAP threshold by applying the offset between ECAP threshold and psychophysical levels measured on electrode 10 to the remaining test electrodes. These encouraging results were reproduced in young cochlear implant recipients (Gordon et al., 2002; Hughes et al., 2000; Mason et al., 2001). One of the main advantages offered by the model of Brown et al. (2000) is that only a small quantity of psychophysical data needs to be collected. Indeed, the behavioral data collected on the reference electrode (electrode 10) accounted for the subject-specific responsiveness to stimulation via the implant. The parallel profiles method adapted from the model of Brown et al. (2000) shared the same fundamentals, i.e. a constant shift between ECAP threshold and psychophysics across electrodes. It should be noted, however, that the parallel profiles method was applied to data expressed in standard international unit (i.e. mA) while the model of Brown et al. (2000) was established using Cochlear Corporation logarithmic p.u. Thus, these two methods cannot be directly compared. To what extent our model differs from the parallel profiles method merits discussion. In the hierarchical

2821

regression model (1) developed in the present study, the term ‘bj’ represents the mean perceptual responsiveness specific for the jth patient. The formula deduced from model (1) for T-level predictions also combines the results of ECAP threshold measurements with a behavioral measure, ‘T-level0’. Again, this behavioral measure acts as a reference value specific to each subject. One major difference between the two methods lies in the relative changes in T-level with ECAP threshold that are predicted within a single recipient. Whereas the parallel profiles method stipulates a 1 mA increase in T-level for every 1 mA increase in ECAP threshold, the formula deduced from model (1) predicts a variation that is nearly five times lower. The intra-subject regression slope used in our model (0.23 (95% confidence interval: 0.18–0.28), instead of 1 in the parallel profiles method) allowed a noticeable improvement in the quality of T-level prediction. Interestingly, a sensitivity analysis using the (0.18 and 0.28) confidence limits as the amount of compression yielded correlation coefficients between predicted and observed T-level values, which were similar to when the point estimate (0.23) was used. This suggests the possibility of picking the amount of compression from its 95% confidence interval without affecting the quality of T-level prediction. The modeling of the response profiles in our series highlighted an inconstant relationship between ECAP threshold and T-level contours across electrodes, in addition to overall differences in response level. This could possibly, in turn, affect the T-level predictions allowed by the parallel profiles method. In contrast, the fact that ECAP threshold and C-level contours were not found to differ across electrodes may explain the high correlations observed between measured C-levels and the respective predicted values using the parallel profiles method. Given the differences observed between T- and C-level values (with mean C-level being twice as high as mean T-level: 625.6 vs. 275.3 mA, respectively), another possible explanation would be that the intra-subject regression slope of the relationship between ECAP threshold and psychophysics may exert a proportionally greater influence on T- than on C-level predictions. 4.3. Influence of electrode location on T-level prediction: implications for cochlear implant mapping in children Another element brought out by the present study is that electrode location should be taken into account for T-level prediction based on ECAP recording. In our model, T-level and ECAP threshold obtained at the basal end of the electrode array (electrode 5) were found to be inappropriate for prediction of T-level measured at the apical end (electrode 20) and vice versa, at least during the first year post-implantation. This could not be explained by differences in ECAP recording or psychophysics estimation procedures across electrodes, since exactly the same

2822

H. Thai-Van et al. / Clinical Neurophysiology 115 (2004) 2811–2824

procedures were applied for all tested electrodes. One possible explanation would be that neural survival may differ greatly between basal and more apical cochlear regions within a single subject. Histological examination of human temporal bones has revealed that spiral ganglion cell counts are lower in the basal turn than in the apical turn of the cochlea in subjects with sensorineural hearing loss (Nadol, 1997). Thai-Van et al. (2002) have previously reported that the quality of EABR recordings may be better at the apical end and the middle of the electrode array than at the basal end, at least up to 9 months post-implantation. In children with a mean cochlear implant duration of about 8 months, Gordon et al. (2002) found higher ECAP amplitude for electrode 20 than for basal electrodes. Taken together, these findings suggest that electrical stimulation delivered by basal electrodes might involve a smaller number of spiral ganglion cells compared to that delivered by intermediate or apical electrodes. Since the T-level is thought to reflect spiral ganglion cell survival, the fact that the location of the electrodes used to run our model could affect the accuracy of its prediction is not surprising. Interestingly, at 18 months post-implantation, high correlation coefficients were found between predicted and actual T-levels whatever the respective location of the reference electrode and the one at which T-level was predicted. This finding pleads for the possibility of a better across-fiber synchrony in the basal region of the cochlea at that time. Such an improvement in basal spiral ganglion cells firing may act as a compensating process for the lower number of neurons stimulated by basal compared to intermediate and apical electrodes. In turn, this would help high-frequency perception at a more central level, and may contribute to the observed progress in speech perception scores experienced with time by children with cochlear implants (Kileny et al., 2001; Waltzman et al., 1994). This topic warrants further investigation, and opens up interesting horizons for a deeper understanding of the relationship between psychophysics, electrophysiological measurements and speech perception scores in young cochlear implant users. As suggested by Gordon et al. (2002), an alternative explanation for a smaller neural population being stimulated at the proximal end of the implant array compared to more apical segments could be the longer distance between the electrode array and neural elements at the basal end of the array. Besides the poorer predictions obtained taking the basal electrode as reference, we also observed an improvement in the predictive value of our model over time. This finding should be interpreted in the light of previous data reporting changes over time in the psychophysical levels of cochlear implantees. Such changes were described early in both humans (Eddington et al., 1978; Michelson, 1971) and animals (Pfingst, 1990). They may reflect physiological changes in the responsiveness of surviving spiral ganglion cells and/or central auditory pathways (Miller et al., 2000). In young cochlear implant recipients, T-levels have been

shown to increase up to 3 months post-implantation and then stabilize (Henkin et al., 2003), while ECAP threshold are known to fluctuate for up to 3–8 months (Hughes et al., 2001). The instability of T-level and ECAP threshold during the first post-implantation months may, in turn, affect the predictive value of our model. In young children, a period of adjustment to the implant is often necessary before reliable psychophysical levels can be obtained (Shapiro and Waltzman, 1995). During this period, behavioral responses to electrical stimulation are unlikely to reflect underlying physiological processes precisely. To minimize the risk of device rejection, care must be exercised by audiologists when programming basal electrodes at initial sessions, since children with profound deafness are unlikely ever to have experienced highpitched sounds before implantation (Thai-Van et al., 2001). This issue may lead to systematically lower psychophysical level values at the proximal compared to the apical end of the implant array, and may also contribute to the poorer predictive value of our model taking the basal electrode as reference. Whatever the exact mechanism underlying the influence of electrode location and post-implantation time on the predictability of our model, results of the present study may have important implications for cochlear implant mapping in difficult-to-test children. During the very sensitive period following cochlear implantation, any objective measure that could reliably confirm, or replace, behavioral assessment of Tlevel would be of great interest to specialized management teams. Except for basal electrodes in initial test sessions (i.e. 1–3 months post-implantation, in this study), a close approximation of T-level can be expected from our model when the distance between the T-level prediction and reference electrodes does not cover more than 5 electrodes. For example, if electrode 15 is selected as reference, the model should allow T-level to be accurately predicted for electrodes 10–20. The greatest interest of using ECAP measurements to determine psychophysical levels is in young subjects who cannot sustain the prolonged auditory attention required by behavioral mapping techniques. Since, during the first month post-implantation, the predictability of the model is better for middle and apical electrodes, a reasonable option would be to limit the use of the model during the early fitting sessions to predicting T-levels at middle and apical electrodes—e.g. using electrode 15 as the reference. At later fitting sessions, after 6 months post-implantation, the use of the model could be extended to basal electrodes. Further research is needed to determine whether the intra-subject regression coefficient found in the present study (roughly ranging from 0.2 to 0.3) can be safely applied in large cohorts of cochlear implant patients with various clinical characteristics and/or speech-coding strategies. There is also a need for assessing whether T-levels predicted with our model can replace observed T-levels when programming the implant without affecting progress in speech perception or production. If comparable clinical outcomes could be obtained with predicted

H. Thai-Van et al. / Clinical Neurophysiology 115 (2004) 2811–2824

and observed T-levels, this would definitely support the validity of our predictions.

5. Conclusion The results confirm that ECAP threshold bears a meaningful predictive relationship to the psychophysical T- and C- levels used for cochlear implant programming, provided that the child’s perceptual responsiveness to implant stimulation is taken into account. Both our linear regression model and the parallel profiles method fulfill this condition. As in the parallel profiles method, our regression model requires a T-level obtained behaviorally from a reference electrode. Additionally, our predictive procedure takes into consideration the relative relationship of T-level and ECAP threshold within each subject, using an intra-subject regression coefficient. This precaution was found to improve the accuracy of the T-level predictions. Given that individual changes in Clevel with ECAP threshold exhibited heterogeneous patterns across children, no common intra-subject regression coefficient could be proposed which would account for these changes. However, the high quality of the C-level predictions allowed by the parallel profiles method suggests that the relationship of C-level to ECAP threshold is less affected by the regression coefficient value. Hence, it would seem that our regression model and the parallel profiles method should be used in combination, as they complement one another for predicting, respectively, T- and C-level. Nevertheless, during the first 3 months post-implantation, neither was found to provide accurate approximations of T-level at the basal end of the electrode array. This underlines the fact that psychophysical level setting cannot be based solely on electrophysiological data. The cause of the poorer T-level predictions for basal electrode 5 than for intermediate or apical electrodes at the early fitting sessions is not fully understood, and it needs to be clarified whether this also applies to other basal electrodes.

Acknowledgements The authors wish to thank Tiphaine Bigeard, Muriel Kreiss and Sylvie Vancayseele for their assistance with data collection, and Professor Rene´ Ecochard, Head of Lyon Biostatistics Department, for supervising data analysis. The statistical analysis was partially funded by an unrestricted grant from Cochlear AG. The statistical analysis of this study has been partially funded by an unrestricted research grant from Cochlear AG.

References Abbas PJ, Brown CJ, Shallop JK, Firszt JB, Hughes ML, Hong SH, Staller SJ. Summary of results using the Nucleus CI24M implant to record the electrically evoked compound action potential. Ear Hear 1999;20:45–59.

2823

Almqvist B, Willstedt-Svensson U. The Lund Technique, a method to set an initial MAP based on objective measures. Personal Communication. In: Nucleus European Satellite Symposium Paediatric CI Conference, s’Hertogenbosh, June 1998. Brown CJ, Abbas PJ, Gantz BJ. Electrically evoked whole-nerve action potential: data from human cochlear implant users. J Acoust Soc Am 1990;88:1385–91. Brown CJ, Abbas PJ, Gantz BJ. Preliminary experience with neural response telemetry system in the Nucleus CI24M cochlear implant. Am J Otol 1998;19:320–7. Brown CJ, Hughes ML, Luk B, Abbas PJ, Wolaver AA, Gervais JP. The relationship between ECAP and EABR thresholds and levels used to program the Nucleus 24 speech processor: data from adults. Ear Hear 2000;21:151–63. Carhart R, Jerger JF. Preferred method for clinical determination of puretone thresholds. J Speech Hear Disord 1959;24:330–45. Charasse B, Thai-Van H, Berger-Vachon C, Collet L. Assessing auditory nerve recovery function with a modified subtraction method: results and mathematical modeling. Clin Neurophysiol 2003;114:1344–50. Cullington H. Preliminary neural response telemetry results. Br J Audiol 2000;34:131–40. Dillier N, Lai WK, Almqvist B, Frohne C, Muller-Deile J, Stecker M, Wallenberg E. Measurement of the electrically evoked compound action potential via a neural response telemetry system. Ann Otol Rhinol Laryngol 2002;111:407–14. Ecochard R, Clayton D. Multi-level modelling of conception in artificial insemination by donor. Stat Med 1998;17:1137–56. Eddington DK, Dobelle WH, Brackmann DE, Mladejovsky MG, Parkin JL. Auditory prostheses research with multiple channel intracochlear stimulation in man. Ann Otol Rhinol Laryngol 1978;87:1–39. Franck KH. A model of a nucleus 24 cochlear implant fitting protocol based on the electrically evoked whole nerve action potential. Ear Hear 2002; 23(Suppl):67–71. Franck KH, Norton SJ. Estimation of psychophysical levels using the electrically evoked compound action potential measured with the neural response telemetry capabilities of Cochlear Corporation’s CI24M device. Ear Hear 2001;22:289–99. Gantz BJ, Brown CJ, Abbas PJ. Intraoperative measures of electrically evoked auditory nerve compound action potential. Am J Otol 1994;15: 137–44. Goldstein H. The basic linear multilevel model and its estimation. In: Goldstein H, editor. Multilevel statistical models 3, 2nd ed. London: Kendall’ Library of Statistics; 1995. p. 15–37. Gordon KA, Ebinger KA, Gilden JE, Shapiro WH. Neural response telemetry in 12- to 24-month-old children. Ann Otol Rhinol Laryngol 2002;189(Suppl):42–8. Hall RD. Estimation of surviving spiral ganglion cells in the deaf rat using the electrically evoked auditory brainstem response. Hear Res 1990;45: 123–36. Henkin Y, Kaplan-Neeman R, Muchnik C, Kronenberg J, Hildesheimer M. Changes over time in electrical stimulation levels and electrode impedance values in children using the Nucleus 24M cochlear implant. Int J Pediatr Otorhinolaryngol 2003;67:873–80. Hodges AV, Balkany TJ, Ruth RA, Lambert PR, Dolan-Ash S, Schloffman JJ. Electrical middle ear muscle reflex: use in cochlear implant programming. Otolaryngol Head Neck Surg 1997;117: 255–61. Hughes ML, Brown CJ, Abbas PJ, Wolaver AA, Gervais JP. Comparison of ECAP thresholds with MAP levels in the Nucleus 24 cochlear implant: data from children. Ear Hear 2000;21:164–74. Hughes ML, Vander Werff KR, Brown CJ, Abbas PJ, Kelsay DM, Teagle HF, Lowder MW. A longitudinal study of electrode impedance, the electrically evoked compound action potential, and behavioral measures in Nucleus 24 cochlear implant users. Ear Hear 2001;22:471–86. Kawano A, Seldon HL, Clark GM, Ramsden RT, Raine CH. Intracochlear factors contributing to psychophysical percepts following cochlear implantation. Acta Otolaryngol 1998;118:313–26.

2824

H. Thai-Van et al. / Clinical Neurophysiology 115 (2004) 2811–2824

Kiefer J, Hohl S, Sturzebecher E, Pfennigdorff T, Gstoettner W. Comparison of speech recognition with different speech coding strategies (SPEAK, CIS, and ACE) and their relationship to telemetric measures of compound action potentials in the Nucleus CI 24M cochlear implant system. Audiology 2001;40:32–42. Kileny PR, Zwolan TA, Ashbaugh C. The influence of age at implantation on performance with a cochlear implant in children. Otol Neurotol 2001;22:42–6. Lai WK, Dillier N. A simple two-component model of the electrically evoked compound action potential in the human cochlea. Audiol Neurootol 2000;5:333–45. Macpherson BJ, Elfenbein JL, Schum RL, Bentler RA. Thresholds of discomfort in young children. Ear Hear 1991;12:184–90. Mason SM, Cope Y, Garnham J, O’Donoghue GM, Gibbin KP. Intraoperative recordings of electrically evoked auditory nerve action potentials in young children by use of neural response telemetry with the Nucleus C124M cochlear implant. Br J Audiol 2001;35:225–35. Michelson RP. Electrical stimulation of the human cochlea. A preliminary report. Arch Otolaryngol 1971;93:317–23. Miller AL, Morris DJ, Pfingst BE. Effects of time after deafening and implantation on guinea pig electrical detection thresholds. Hear Res 2000;144:175–86. Nadol Jr JB. Patterns of neural degeneration in the human cochlea and auditory nerve: implications for cochlear implantation. Otolaryngol Head Neck Surg 1997;117:220–8. Pfingst BE. Changes over time in thresholds for electrical stimulation of the cochlea. Hear Res 1990;50:225–36. Pfingst BE, Sutton D. Relation of cochlear implant function to histopathology in monkeys. Ann NY Acad Sci 1983;405:224–39. Pfingst BE, Sutton D, Miller JM, Bohne BA. Relation of psychophysical data to histopathology in monkeys with cochlear implants. Acta Otolaryngol 1981;92:1–13. Pfingst BE, Glass I, Spelman FA, Sutton D. Psychophysical studies of cochlear implants in monkeys: clinical implications. In: Schindler RA, Merzenich MM, editors. Cochlear implants. New York, NY: Raven Press; 1985. p. 305–21.

Rance G, Dowell RC. Speech processor programming. In: Clark GM, Cowan RSC, Dowell RC, editors. Cochlear implantation for infants and children: advances. San Diego, CA: Singular Publishing group; 1997. p. 147–70. Rodgers JL. The bootstrap, the jackknife, and the randomization test: a sample taxonomy. Multivariate Behav Res 1999;34: 441–56. Shannon RV. Multichannel electrical stimulation of the auditory nerve in man: I. Basic psychophysics. Hear Res 1983;11:157–89. Shapiro W, Waltzman S. Changes in electrical thresholds over time in young children implanted with the Nucleus cochlear prosthesis. Ann Otol Rhinol Laryngol Suppl 1995;166:177–8. Shepherd RK, Javel E. Electrical stimulation of the auditory nerve. I. Correlation of physiological responses with cochlear status. Hear Res 1997;108:122–44. Smoorenburg GF, Willeboer C, Van Dijk JE. Speech perception in Nucleus CI24M cochlear implant users with processor settings based on electrically evoked compound action potential thresholds. Audiol Neurootol 2002;7:335–47. Spivak LG, Chute PM, Popp AL, Parisier SC. Programming the cochlear implant based on electrical acoustic reflex thresholds: patient performance. Laryngoscope 1994;104:1225–30. Thai-Van H, Chanal JM, Coudert C, Veuillet E, Truy E, Collet L. Relationship between NRT measurements and behavioral levels in children with the Nucleus 24 cochlear implant may change over time: preliminary report. Int J Pediatr Otorhinolaryngol 2001;58: 153–62. Thai-Van H, Galle´go S, Truy E, Veuillet E, Collet L. Electrophysiological findings in two bilateral cochlear implant cases: does the duration of deafness affect electrically evoked auditory brain stem responses? Ann Otol Rhinol Laryngol 2002;111:1008–14. Waltzman SB, Cohen NL, Gomolin RH, Shapiro WH, Ozdamar SR, Hoffman RA. Long-term results of early cochlear implantation in congenitally and prelingually deafened children. Am J Otol 1994; 15(Suppl 2):9–13.