Multicenter U.S. Bilateral MED-EL Cochlear Implantation Study: Speech Perception over the First Year of Use

Transcription

1 Multicenter U.S. Bilateral MED-EL Cochlear Implantation Study: Speech Perception over the First Year of Use Emily Buss, 1 Harold C. Pillsbury, 1 Craig A. Buchman, 1 Carol H. Pillsbury, 1 Marcia S. Clark, 1 David S. Haynes, 2 Robert F. Labadie, 2 Susan Amberg, 2 Peter S. Roland, 3 Pamela Kruger, 3 Michael A. Novak, 4 Julie A. Wirth, 4 Jennifer M. Black, 4 Robert Peters, 5 Jennifer Lake, 5 P. Ashley Wackym, 6 Jill B. Firszt, 6 Blake S. Wilson, 7 Dewey T. Lawson, 7 Reinhold Schatzer, 7 Patrick S. C. D Haese, 8 and Amy L. Barco 8 Objective: Binaural hearing has been shown to support better speech perception in normal-hearing listeners than can be achieved with monaural stimulus presentation, particularly under noisy listening conditions. The purpose of this study was to evaluate whether bilateral electrical stimulation could confer similar benefits for cochlear implant listeners. Design: A total of 26 postlingually deafened adult patients with short duration of deafness were implanted at five centers and followed up for 1 yr. Subjects received MED-EL COMBI 40 devices bilaterally; in all but one case, implantation was performed in a single-stage surgery. Speech perception testing included CNC words in quiet and CUNY sentences in noise. Target speech was presented at the midline (0 degrees), and masking noise, when present, was presented at one of three simulated source locations along the azimuth ( 90, 0, and 90 degrees). Results: Benefits of bilateral electrical stimulation were observed under conditions in which the speech and masker were spatially coincident and conditions in which they were spatially separated. Both the head shadow and summation effects were evident from the outset. Benefits consistent with binaural squelch were not reliably observed until 1 yr after implantation. Conclusions: These results support a growing consensus that bilateral implantation provides functional benefits beyond those of unilateral implantation. Longitudinal data suggest that some aspects of binaural processing continue to develop up to 1 yr 1 Department of Otolaryngology/Head and Neck Surgery, University of North Carolina School of Medicine, Chapel Hill, North Carolina; 2 Vanderbilt Bill Wilkerson Center for Otolaryngology and Communication Sciences, Vanderbilt University School of Medicine, Nashville, Tennessee; 3 Department Otolaryngology/ Head and Neck Surgery, University of Texas Southwestern, Dallas, Texas; 4 ECHO program at Carle Foundation Hospital, Urbana, Illinois; 5 Dallas Otolaryngology Associates, Dallas, Texas; 6 Department of Otolaryngology and Communication Sciences, Medical College of Wisconsin, Milwaukee, Wisconsin; 7 Research Triangle Institute, Research Triangle Park, North Carolina; 8 MED-EL Corporation, Innsbruck, Austria. after implantation. The squelch effect, often reported as absent or rare in previous studies of bilateral cochlear implantation, was present for most subjects at the 1 yr measurement interval. (Ear & Hearing 2008;29;20 32) Under natural listening conditions, a normalhearing listener uses the sound streams reaching each ear to get a sense of the three-dimensional auditory scene. One important example of this binaural benefit is reception of speech in noise for two or more sound sources separated in space. In normal-hearing listeners, binaural presentation of a masked speech stimulus can improve speech reception threshold by 10 db when compared with monaural stimulus presentation (Plomp & Mimpen, 1981). Many hearing-impaired listeners obtain analogous binaural benefit when listening through binaural hearing aids (Byrne, 1981). A growing body of recent evidence suggests that bilateral cochlear implant recipients also receive benefit from bilateral stimulation when listening to speech in noise (Gantz, et al., 2002; Laszig, et al., 2004; Litovsky, Parkinson, Arcaroli, & Sammeth, 2006; Müller, Schön, & Helms, 2002; Ramsden, et al., 2005; Schleich, Nopp, & D Haese, 2004; Schön, Müller, & Helms, 2002; Senn, Kompis, Vischer, & Haeusler, 2005; Wackym, Runge-Samuelson, Firszt, Alkaf, & Burg, 2007). Although the existence of binaural benefits for bilateral implantation has been well documented, particularly for speech in noise, the cue or cues underlying these benefits are less clear. Some benefit has been ascribed to the fact that bilateral input allows a listener to base performance on the better of the two signals; in this case, binaural benefit is based on receiving two streams of information from which to choose the better representation of the signal. In contrast, true binaural benefit is often defined in terms of the use of interaural time and intensity differences to segre- 0196/0202/08/ /0 Ear & Hearing Copyright 2008 by Lippincott Williams & Wilkins Printed in the U.S.A. 20

2 BUSS ET AL. / EAR & HEARING, VOL. 29, NO. 1, gate sound sources in central stages of auditory processing; such use of interaural difference cues requires a combination of information from the two sides, similar to that which takes place in auditory localization. One factor that complicates interpretation of speech and speech in noise outcomes for bilateral cochlear implant users is the large variability in subject demographics, testing protocol, and device configuration in the existing data. For example, some research has limited the study population to patients with a relatively short duration of deafness (Ramsden, et al., 2005), whereas others included subjects with up to 67 yr of deafness (Müller, et al., 2002). In some cases, the studies were designed to assess the binaural benefit for patients with simultaneous implantation or implantation on the second side with a relatively short delay (Gantz, et al., 2002; Laszig, et al., 2004), whereas others have followed patients receiving a second implant after extensive unilateral implant experience (Ramsden, et al., 2005) or patients receiving mismatched implant technology bilaterally (Dorman & Dahlstrom, 2004). Despite these differences across studies, a few general conclusions can be drawn. The most widely reported and largest binaural advantage in bilateral cochlear implant listeners is the head shadow effect. This effect can be demonstrated when a target signal and the masking noise are presented on opposite sides of the subject. Under these conditions, the head and body can act as a baffle, introducing attenuation of the high-frequency components of the masker, such that the signal to noise ratio (SNR) at the ear more distant from the noise is increased. This effect can be quantified as the difference in speech recognition thresholds for two monaural listening conditions: the threshold listening with the implant closer to the noise source minus the threshold listening with the implant on the shadowed side. This difference ranges between 4 and 7 db (Gantz, et al., 2002; Müller, et al., 2002; Schleich, et al., 2004). A comparable difference of 8 db can be obtained in listeners with normal hearing (Bronkhorst & Plomp, 1988). Differences of 4 to 8 db are large and correspond roughly to improvements in percent correct scores of 28% to 56% for high-context sentences presented at a fixed SNR.* Aside from the head shadow effect, the next most commonly reported and robust effect is the benefit observed for diotic when compared with monotic presentation (bilateral versus unilateral presentation of the identical stimulus). This effect, called summation, is typically attributed to the increased *This assumes a 7%/dB approximation, as reported by Schön, et al. (2002) for German sentences. loudness associated with bilateral stimuli and to the redundancy of information in the stimuli at the two ears. In listeners with normal hearing, this effect is rather small, on the order of 1 db (Bronkhorst & Plomp, 1989). The summation effect in bilaterally implanted subjects tends to be somewhat larger, ranging from 10% to 20% or 1.5 to 2.9 db (Gantz, et al., 2002; Müller, et al., 2002; Schleich, et al., 2004). Schön, et al. (2002) speculated that an elevated summation effect in implant subjects may indicate greater benefit of bilateral redundancy in this population when compared with that in normalhearing listeners. An alternative interpretation is that variability across ears in implanted subjects leads to different representations of the diotic signal, such that new information is conveyed to the central auditory system by virtue of complementary representations encoded on the left and right sides (e.g., Lawson, Wilson, Finley, & Zebri, 1999; Lawson, et al., 1998; Loizou, Mani, & Dorman, 2003; Schatzer, Wilson, Wolford, & Lawson, 2003). The third binaural effect frequently examined in studies of bilateral implantation is the squelch effect. Squelch is based on similar listening conditions as those used to demonstrate the head shadow effect, with the signal and noise sources at different positions in the azimuthal plane. In this case, however, binaural benefit is defined as the improvement in performance obtained when the input to the ear farther from the masker source (i.e., with the better SNR) is supplemented with input from the contralateral side (i.e., with the worse SNR). This effect is quite sensitive to test conditions and is often reported to be small in normal-hearing listeners, on the order of 3 db (Bronkhorst & Plomp, 1988). Previous studies suggest that some bilateral implant subjects obtain this benefit to some degree, whereas half or more of the subjects tested do not (Gantz, et al., 2002; Laszig, et al., 2004; Müller, et al., 2002; Schleich, et al., 2004). In addition to these quantitative indicators of binaural benefit, there are other benefits noted in the literature. Several studies suggest that preoperative predictions regarding the better ear to implant are rather poor, and one consequence of bilateral implantation is that it ensures that the better ear has been implanted (Gantz, et al., 2002; Ramsden, et al., 2005). Other nonquantitative benefits have been reported for binaural listening in natural environments, including improved sound quality and ease of listening (e.g., Litovsky, et al., 2006; Wackym, et al., 2007). Similar effects have been described for bilateral when compared with unilateral hearing aid use (Balfour & Hawkins, 1992). Despite the growing literature touting the benefits of bilateral implantation, there are reasons to

3 22 BUSS ET AL. /EAR &HEARING, VOL. 29, NO. 1,20 32 believe that binaural processing per se may be less effective in cochlear implant subjects than in normal-hearing listeners under some conditions. One example is the small or absent squelch effect in many subjects, as mentioned above. Another example comes from data on the masking level difference (MLD). In the MLD paradigm, fine differences in interaural timing improve detection thresholds in normal-hearing listeners. In contrast, bilateral implant subjects show a very small MLD effect when the cue is based on temporal fine structure, on the order of 1.8 db when compared with 10 or 15 db in normal-hearing listeners (van Hoesel, 2004). This result is in contrast to the relatively normal MLD obtained when the cue is based on the temporal envelope of the stimulus (Long, Carlyon, Litovsky, & Downs, 2006). Two often mentioned impediments to binaural processing are the loss of temporal fine structure, as a consequence of most common coding strategies, and the mismatch in the place to frequency across ears, because of subtle differences in electrode insertion and neural survival (Moore, 2003). It has been argued that binaural diplacusis, a mismatch in perceived pitch across ears, is associated with a failure to obtain a binaural benefit with bilateral hearing aids (Markides, 1977). In addition, central masking has been shown to be a significant effect in bilateral implant subjects; van Hoesel and Clark (1997) showed that a contralateral masking stimulus elevated thresholds for a monotic signal by 20% to 40% of the dynamic range of hearing in their two subjects. A more thorough understanding of the mechanisms of binaural benefit in implanted listeners, including the conditions under which binaural input may be disadvantageous, is therefore of great importance. The current study reports speech and speech in noise performance for a cohort of bilateral cochlear implant subjects. Companion papers report localization performance, interaural difference thresholds and speech recognition under conditions of binaurally uncorrelated masking noise for the same cohort (Grantham, Ashmead, Ricketts, Labadie, & Haynes, 2007, 2008; Ricketts, Grantham, Ashmead, Haynes, & Labadie, 2006). Comparison of performance across these paradigms should shed light on the extent to which true binaural cues contribute to the functional binaural benefit observed. MATERIALS AND METHODS Subjects Study candidates were 18 yr of age or older and had a severe to profound sensorineural hearing loss bilaterally, with average pure-tone thresholds of 70 db HL or poorer at 0.5, 1, and 2 khz. None showed substantial benefit from hearing aids, with scores of 40% or poorer on Hearing in Noise Test (HINT) sentences in quiet under auditory-only testing conditions. As such, these subjects met standard criteria for unilateral cochlear implantation. In addition, study participants had a duration of deafness of no more than 15 yr in either ear and a mismatch in duration of deafness across ears of no more than 10 yr. Only postlingually deafened patients were considered for the study, and all were required to be literate and fluent in English. Exclusion criteria included (1) radiographic evidence of cochlear malformation that would prevent a full insertion of a standard electrode array, (2) indications of abnormality at the level of the auditory nerve or higher in the auditory processing pathways, (3) poor physical or mental health, or (4) prior experience with a cochlear implant. Subjects agreed at the outset of the study to undergo the outcome testing protocol. In exchange, each subject received one of the two implants free of charge. Study participants were recruited from five cochlear implant centers in the United States: The University of North Carolina at Chapel Hill, Vanderbilt University, The University of Texas Southwestern Medical Center, The Carle Clinic, Dallas Otolaryngology Associates, and The Medical College of Wisconsin. The initial enrollment included 29 subjects. After surgery, one subject suffered from an illness unrelated to cochlear implantation and was unable to provide postoperative performance data. Data from an additional two participants were collected but subsequently excluded from results reported below. In one of these cases, the subject experienced a unilateral device failure that resulted in a revision surgery. Because device failure can be progressive, with gradual degradation in performance over time, it is problematic to identify the date of failure before which the device could be assumed to be functioning properly. In the other case, a subject demonstrated a marked improvement in performance after revision of her map during the 6 mo follow-up visit. It was strongly suspected that the map before that point, which was used for the 3 mo testing interval, did not support optimal performance for this subject. The circumstances of these two exclusions device failure and inaccuracy of mapping are certainly relevant to clinical outcomes in cochlear implantation. However, it was reasoned that these factors are not restricted to bilateral implantation and might have obscured the binaural effects of primary interest here. The remaining 26 subjects were retained through the full year of outcome testing, and their data are reported below. This group included 8 men and 18 women, ages 25.3 to 76.6 (mean, 48.3) at the time of

4 BUSS ET AL. / EAR & HEARING, VOL. 29, NO. 1, TABLE 1. Subject demographics Subject Gender Age (yr) Suspected etiology Channels off at 12-mo 1 F 73.0 otosclerosis 11, 12 (AU) 2 M 71.0 genetic none 3 M 76.6 unknown none 4 M 26.4 unknown none 5 F 40.8 unknown none 6 F 25.3 unknown 12 (AU) 7 F 53.5 unknown none 8 F 53.9 unknown 6 (AU) 9 M 74.7 noise exposure none 10 F 62.3 head trauma none 11 F 42.6 meningitis 12 (L) 12 F 32.9 autoimmune none disease 13 F 55.6 genetic none 14 F 57.9 unknown none 15 M 32.5 genetic none 16 F 55.7 streptomycin none ototoxicity 17 F 41.1 unknown (L) 18 F 48.0 unknown none 19 F 53.8 scarlet fever/red none measles 20 F 34.8 unknown none 21 M 36.6 unknown 12 (AU); 11(L) 22 F 41.9 RH incompatibility none 23 F 33.5 rubella none 24 M 36.7 unknown (AU); 9 (R) 25 F 44.4 unknown (AU); 10 (R) 26 M 50.4 ear infection 1, 2, 11, 12 (AU) surgery. Additional information about the subjects appears in Table 1, including the gender, suspected etiology, and electrodes that were turned off (if any) for each subject. The entries in the table are ordered by subject number assigned according to order of enrollment in the study. In all but one of the cases (Subject 18), surgery was performed as a singlestage procedure. In the remaining case, the right side was implanted and activated 7 mo before the left side. Test intervals for this subject are measured relative to the implantation of the second (left) side. Procedures Subjects were implanted bilaterally with the MED-EL COMBI 40 internal device and the standard 31 mm electrode. They returned 2 to 4 weeks after surgery for their initial stimulation. At this visit, a map was generated for each of a pair of TEMPO speech processors, using the same audiological mapping procedures followed in standard, unilateral implantation. Detection thresholds and most comfortable loudness levels were obtained behaviorally for stimulation applied to each of 12 electrodes on the left and right sides to define the dynamic ranges of stimulation. These values are used by the processor to determine pulse amplitude for the continuous interleaved sampling processing strategy (Wilson, et al., 1991) used by the TEMPO. Electrodes were turned off for a variety of reasons, including facial stimulation and failure to attain sufficient loudness; in at least one case, this could be attributed to an incomplete electrode insertion. For some subjects, channels were turned off bilaterally to maintain comparable maps across ears, whereas in most cases, no such adjustments were made. Although the default of 26 s/phase pulse duration was used in most maps, this variable was increased as needed to achieve sufficient loudness. An increase in pulse duration results in a decrease in pulse rate. Both pulse duration and pulse rate were allowed to vary across ears, and no attempt was made to accommodate differences in perceived pitch. The default pulse rate was 1515 pps, and in no case did the rate drop below 1000 pps. After mapping each ear individually, the overall loudness between the two ears was balanced. The TEMPO speech processor stores up to three maps, and subjects may select from among the three maps based on preference and listening conditions. Volume and sensitivity controls on each processor provided further subject control. Processor maps were updated at each follow-up visit as needed to maximize performance. In most cases, speech testing was performed with the map favored by the subject before remapping, a procedure adopted to ensure familiarity with the stimulation parameters. Speech Testing Initial experience with the testing protocol suggested that variation in the configuration of the sound field test environments across centers could introduce substantial inconsistencies in testing procedures, so direct audio input (DAI) was used for presentation of all test stimuli. In this procedure, test materials were recorded onto a compact disk (CD) and later presented to the subjects using a battery-operated CD player. Each subject plugged one end of a specially designed cable into the stereo output of the CD player and the other (split) end into the direct connect jack of each of the two speech processors. The processor for the left ear was connected to the left stereo output of the CD player, and the processor for the right ear was connected to the right stereo output. Use of the input jack for each processor largely bypassed the microphone for that processor. The direct input was unattenuated, but the microphone input was attenuated by at least 30 db, and the automatic gain control circuit was disabled. To further guard against effects of ambient

5 24 BUSS ET AL. /EAR &HEARING, VOL. 29, NO. 1,20 32 noise, testing was performed in a quiet room or soundproof booth. To simulate the signals present in a free-field testing situation, stimuli presented using the direct input were preprocessed with head-related transfer function (HRTF) filters. These filters are based on KEMAR dummy-head recordings and incorporate the amplitude and phase characteristics of sounds originating from a specified position in space, as would be introduced when a sound interacts with a listener s head and body (Cox, Wolford, Schatzer, Wilson, & Lawson, 2001; Gardner & Martin, 1995; Wightman & Kistler, 1989). Different sets of filters have been designed to simulate conditions at the eardrum or at an ear level microphone, for sound sources at various locations in space around a listener (Gardner & Martin, 1995). The filters for the simulation of conditions at the ear level microphone were used for the present study, as this corresponds to the microphone position for a cochlear implant. Procedures described in section IV of the above-cited report by Cox et al. were used for automated filtering of all speech material. Although HRTF processing is a standard procedure in studies of binaural hearing, the current adaptation of the method was validated in a pilot experiment with the first five subjects recruited at The University of North Carolina. These subjects underwent CUNY (City University of New York) sentence testing in noise in both the sound field and using DAI; the data patterns obtained using these two methods were remarkably similar, supporting the claim that the simulated free-field presentation mode is a valid method for evaluating the use of spatial cues. Details of this pilot experiment appear in Appendix. Speech testing was performed at four postimplantation intervals: at 1, 3, 6, and 12 mo after the initial activation. At the beginning of a speech testing session, the subject was asked to select the map preferred for listening in a noisy environment and to adjust the levels on each speech processor so that bilateral speech stimuli were balanced between ears. Overall level could then be adjusted in both ears by manipulating the volume knob on the CD player. Once the subject had selected a comfortable listening level based on bilateral speech in quiet, he was encouraged to use this level over the entire test session; in some cases, however, the inclusion of masking noise in CUNY testing increased loudness beyond the comfortable limit, and in these cases the output level of the CD player was reduced. Because DAI stimulus presentation bypasses the compressive circuit of the processor, the sensitivity control had no effect in these listening conditions. Each test session lasted approximately 6 hr, including breaks, and typically began with consonantnucleus-consonant (CNC) words. Stimuli for CNC word tests were presented with a simulated position of 0 degrees azimuth. Percent correct was computed for three conditions, presented in random order: with only the left implant active (left only), with only the right implant active (right only), and with both implants active (bilateral). Separate lists of 50 words each were presented for the three different test conditions in each testing session. Lists were not repeated across sessions. As with CNC words, CUNY sentence testing was conducted in three listening conditions: with input from the left only, input from the right only, and with bilateral input. The signal was always presented from the front (simulated 0 degrees azimuth), whereas masking noise was presented at either a simulated location of 90, 0, or 90 degrees in the azimuthal plane. These stimulus conditions will be described with the notation N L S F,N F S F and N R S F, respectively, where the position of the noise is indicated by the subscript of N (either left, front or right), and the position of the target speech signal is indicated by the subscript of S (in this paradigm, always front). As such, there were a total of nine conditions (three implant configurations and three noise locations). No CUNY testing in noise was performed at the 1 mo session. Only three conditions were tested at the 3 mo session: N F S F (bilateral), N L S F (bilateral), and N R S F (bilateral). All nine conditions were performed at the 6 and 12 mo sessions. In all cases, masking noise was CCITT noise (Fastl, 1993), designed to reproduce the long-term spectrum of speech, averaged across many talkers. The first step in masked CUNY testing was to determine an appropriate SNR for that session, such that the subject could perform the task, but performance was not at ceiling. Each subject was tested at a fixed SNR, which was determined as follows: starting at an SNR of 1.8 db, percent correct was measured for two lists in each of two conditions, the N L S F (bilateral) and the N R S F (bilateral) conditions (i.e., both implants active and the noise from the simulated position of either 90 or 90 degrees). Masker level was adjusted if the percent correct score averaged across the four lists did not fall within the range of 40% to 80% correct. Adjustments were made in steps of 5 db, resulting in SNRs of 16.8, 11.8, 6.8, or 1.8 db. The target percent correct was achieved at one of these levels for each subject at all testing intervals. Because these adjustments were made separately for every subject at 3, 6, and 12 mo follow-up intervals, the SNR was not necessarily constant across time points within subjects. A total of four lists were tested in each condition, with 12 sentences per list. Each word omitted or

6 BUSS ET AL. / EAR & HEARING, VOL. 29, NO. 1, reported incorrectly was counted wrong, with a total of 102 or 104 words per list. The order in which subjects completed the test conditions in each session was not uniform across test sessions. In some cases, conditions were blocked, with all sentence lists presented in a particular condition before moving on to the next condition; sessions run in blocked order began with the bilateral listening conditions, but with noise conditions completed in random order. In other cases, thresholds were unblocked, with the four replicate thresholds in each condition randomly interleaved with those from the other eight conditions. At the 6 mo interval, about half of the subjects were run blocked and half unblocked. At the 12 mo interval, about one-third were run blocked and two-thirds unblocked. The unblocked order was preferred, to guard against practice effects, but the blocked order was used in some cases to reduce test time. The factor of test order is considered in detail below. RESULTS The following data points were unavailable because of either equipment failure, subject illness, or technical difficulties: one subject at the 1 mo interval, three subjects at the 3 mo interval, and one subject at the 6 mo interval. Data are available for all 26 subjects at the 12 mo interval. CNC Word Scores The distributions of CNC scores in quiet as a function of test interval are shown in Figure 1. Fig. 1. The distribution percent correct scores for CNC words in quiet are plotted as a function of postsurgery test interval. Horizontal lines indicate the median of each distribution, boxes span the 25th to 75th percentiles, vertical lines show the 10th to 90th percentile range, and stars indicate the minimum and maximum scores. Data for the worse of the two unilateral conditions are indicated with dark gray shading, those for the better unilateral condition with solid white shading, and those for the bilateral condition with gray and white hatching. The n at each time point is indicated in parentheses below each label on the abscissa. Horizontal lines show the median, boxes delineate the 25th and 75th percentiles, vertical lines span the 10th and 90th percentiles, and stars indicate the minimum and maximum scores. The gray boxes correspond to performance for the poorer of the two unilateral conditions, either left only or right only, determined independently for each subject at each test interval. White boxes correspond to the better of the two unilateral conditions, and hatched boxes indicate performance for the bilateral listening condition. The CNC data in Figure 1 illustrate two trends: an improvement in CNC scores over time, from 1 to 12 mo test intervals, and superior performance in the bilateral listening condition when compared with the better of the two unilateral conditions. Data were transformed using a rationalized arcsine transform (Studebaker, 1985), a procedure that converts percent correct data into units more closely approximating the normal distribution assumed in parametrical statistical tests. A repeated-measures analysis of variance (ANOVA) was performed on the transformed data for the 21 subjects for whom there were data in all cells. There were three levels of CONDITION (worse unilateral, better unilateral and bilateral) and four levels of test INTERVAL (1, 3, 6, and 12 mo). The ANOVA demonstrated a main effect of CONDITION (F 2, , p 0.001), a main effect of INTERVAL (F 3, , p ), and no interaction (F 6, , p 0.36). The between-subject factor was also significant (F 1, , p ), indicating reliable individual differences. Pairwise comparisons using Bonferroni adjustments were used to assess the significance of changes in performance as a function of time interval. Results indicated a significant improvement between 1 and 3 mo intervals and between 6 and 12 mo intervals (one-tailed, p 0.05); the apparent improvement in performance between 3 and 6 mo intervals approached significance (p 0.06). The three levels of CONDITION were all significantly different from one another (p 0.001). Ear dominance in CNC data was quantified as the difference in percent correct obtained in the left only and the right only unilateral listening conditions, determined separately for each subject at each test interval. The distribution of difference scores was assessed using a one-sample Kolmogorov-Smirnov test and found to be approximately normal (p 0.20). The consistency of ear dominance within subject and across test intervals was assessed with Although the arcsine transformation produces a more nearly normal distribution of data, this step did not play a central role in the pattern of results obtained. Repeating analyses on the nontransformed data produced similar results to those reported here.

7 26 BUSS ET AL. /EAR &HEARING, VOL. 29, NO. 1,20 32 Fig. 2. Percent correct for CNC words in the bilateral condition are plotted for individual listeners at 1 and 12 mo intervals, with filled and unfilled bars, respectively. Individual s data are arrayed along the abscissa, ordered according to performance at the 1 mo interval. Subject number appears at the top of the panel. repeated-measures ANOVA, with the dependent measure being the difference between left only versus right only and four levels of INTERVAL (1, 3, 6, and 12 mo). There were no significant effects in this analysis; of particular interest, there was no main effect of subject (F 1, , p 0.78). The CNC data alone do not allow a test of significance of ear dominance within test interval. However, the absence of a subject effect suggests that any difference that may exist between ears is not consistent over time within an individual. The significance of ear differences within interval will be considered below in relation to ear differences in CUNY data. As indicated by the spread of data in Figure 1, there were large differences in percent correct scores across subjects at each of the four test intervals, even in the bilateral condition. These data allow an assessment of the stability of individual differences over time, to see whether the same subjects performed well (or poorly) at each measurement interval. Pairwise correlations were computed for the bilateral listening condition at each of the four measurement intervals. Of the six possible comparisons, the smallest correlation was between results obtained at the 1 and 12 mo test intervals; that value was r 0.75 (p 0.001). Percent correct for these two time intervals are plotted in Figure 2, with individual subjects data arrayed on the abscissa and order determined by scores at the 1 mo data point. This result suggests that performance after 1 yr of practice can be roughly predicted based on 1 mo data, with about 55% of the variance across subjects in this dataset accounted for. It should be emphasized that this association is coarse, however, and that individual subjects performance may not be captured by this trend. For example, Subject 26 rose from 19th to 7th rank between the 1 and 12 mo intervals. CUNY Sentences in Noise Because testing with CUNY sentences involved the selection of SNR individually for each subject and at each test session, results cannot be compared across subjects or across sessions directly. The SNR at which the testing was performed, however, can serve as a rough estimate of the subjects sensitivity. Figure 3 shows a histogram of SNR values used in CUNY testing at each follow-up test interval. Recall that the SNR was selected based on the criterion that the average percent correct fell between 40% and 80% for the bilateral listening conditions with N L S F and N R S F stimulus presentations. Comparing the 3 and 6 mo test intervals, SNR dropped in seven cases, stayed the same in 14 cases, and went up in two cases; these are the same two cases where 3 mo testing was performed at the most difficult ( 1.8 db) SNR. No 3 mo data were available for the remaining 3 subjects. Between the 6 and 12 mo Fig. 3. Count of subjects run at each SNR at each test interval for masked CUNY sentence testing. Shading indicates SNR, either 1.8 db (white), 6.8 db (light gray), 11.8 db (dark gray) or 16.8 db (black).

8 BUSS ET AL. / EAR & HEARING, VOL. 29, NO. 1, intervals, SNR went down in seven cases and stayed the same in 19 cases, with no cases of increasing SNR (associated with a drop in performance). A Wilcoxon signed rank test indicates that SNR decreased (sensitivity improved) significantly between the 6 and 12 mo test intervals (Z 2.64, p 0.01), whereas the decrease between 3 and 6 mo test intervals represents a nonsignificant trend (Z 1.67, p 0.10). The purpose of adjusting SNR was twofold: (1) to ensure that changes in performance would not be obscured by floor or by ceiling effects, and (2) to provide comparable conditions under which to estimate the relative contributions of the various sources of binaural benefit. The distribution of percent correct scores for N R S F (bilateral) and N L S F (bilateral) conditions was found to be very similar across 3, 6, and 12 mo test intervals, with median values of 50% to 57%. This observation confirms that the pattern of effects obtained over time can be compared without confounding range effects. Because of nonuniformity in SNR across subjects and across measurement intervals, the most interesting results are based on derived measures. Following previous bilateral cochlear implant work, these are identified as the head shadow, summation, and squelch. There is some variation in the method for computing these derived measures across studies. For the most part, these differences are rooted in the use of percent correct as an indication of speech perception as opposed to speech reception threshold, which is reported in units of decibels. In the definitions below, bilateral effects are defined in terms of change in percent correct. In all cases, bilateral advantage is reflected in a positive value. The distributions of derived measures were assessed for the assumption of normality using a one-sample Kolmogorov-Smirnov test. In all cases, this test failed to reject the null hypothesis of normality (p 0.63), justifying the use of parametric statistics with these values in the absence of any transformation of the data. Head shadow is defined as the difference between percent correct obtained in a unilateral listening condition with noise presented ipsilateral to the active implant compared with noise presented contralateral to the active implant. As such, two estimates of head shadow can be computed for each test interval: Head Shadow (left) N R S F (left only) N L S F (left only). (1) Head Shadow (right) N L S F (right only) N R S F (right only). (2) where the subscript indicates the simulated spatial position of the noise (N) and speech signal (S) Fig. 4. Estimates of head shadow based on data from the 6 mo (top panel) and 12 mo (bottom panel) test intervals. Values are shown separately for individual subjects, distributed on the abscissa; values of HeadShadow (left) are indicated with leftpointing triangles [, Eq. (1)], and HeadShadow (right) are indicated with right-pointing triangles [, Eq. (2)]. relative to the midline, and the notation in parentheses indicates the implant or implants receiving stimulation. Figure 4 shows values of head shadow, with the top panel for the 6 mo and the bottom for the 12 mo test interval. Symbols show individuals results, with the left-pointing triangles corresponding to head shadow computed for the left side and the right-pointing triangles for the right side. The general pattern of results is very similar over time. Median values of head shadow are 37% and 38% for the 6 and 12 mo time intervals, respectively, and in all cases, head shadow is positive or near zero. A repeated-measures ANOVA was performed for these estimates of head shadow, with two levels of INTERVAL (6 and 12 mo) and two levels of SIZE (larger and smaller of the two values for each subject). The results indicate no main effect of INTERVAL (F 1, , p 0.12), a main effect of SIZE (F 1, , p ), and no interaction (F 1, , p 0.44). Each of the four data sets (2 intervals 2 effect sizes) was submitted to a t-test to determine whether the mean was significantly different from zero. In all cases, these values were significant at (two-tailed) after Bonferroni adjustment. Thus,

9 28 BUSS ET AL. /EAR &HEARING, VOL. 29, NO. 1,20 32 Fig. 5. Estimates of summation [Eq. (3)] based on data from the 6 mo (top panel) and 12 mo (bottom panel) test session. Values are shown separately for individual subjects, distributed on the abscissa. the benefit derived from head shadow was highly significant and remained stable across the two test intervals. Summation is the advantage associated with bilateral listening when compared with either ear alone when both ears receive the same signal, such as occurs in the N F S F conditions. Summation is computed here as Summation N F S F (bilateral) max N F S F (left only), N F S F (right only) (3) These data are shown in Figure 5, with results for the 6 mo interval in the top panel and those for the 12 mo interval in the bottom panel. A repeated measures ANOVA, with two levels of INTERVAL (6 and 12 mo), indicates no main effect of INTERVAL (F 1, , p 0.20). Data from each of the two measurement intervals was submitted to a t-test. Summation was significantly greater than zero for both the 6 mo (t , p 0.05) and 12 mo intervals (t , p 0.005), with a one-tailed test and Bonferroni adjustment for multiple comparisons. Similar to the head shadow effect, summation seems to be positive and stable over time in this dataset, with median values of 2.0% and 5.7% for the 6 and 12 mo time intervals. Fig. 6. Estimates of binaural squelch based on data from the 6 mo (top panel) and 12 mo (bottom panel) test session. Values are shown separately for individual subjects, distributed on the abscissa; values of Squelch (left) are indicated with left-pointing triangles [, Eq. (4)], and Squelch (right) are indicated with right-pointing triangles [, Eq. (5)]. Squelch is defined in terms of the advantage associated with bilateral listening when compared with the shadowed ear alone. This is computed as: Squelch (left) N R S F (bilateral) Squelch (right) N L S F (bilateral) N R S F (left only). (4) N L S F (right only). (5) Following conventions of Figure 4, values of squelch are shown in Figure 6 for the left and right side for each subject. Median values are 3.3% and 10.6% at 6 and 12 mo test intervals. This increase in squelch effect was assessed with a repeated-measures ANOVA using two levels of INTERVAL (6 and 12 mo) and two levels of SIZE (larger and smaller of the two values for each observer). The results indicate a significant main effect of INTERVAL (F 1, , p 0.001), a significant main effect of SIZE (F 1, , p ), and a statistical trend toward an interaction between INTERVAL and SIZE (F 1, , p 0.10), because the difference between the larger and smaller of the two values for each observer was smaller at the12 mo interval than

10 BUSS ET AL. / EAR & HEARING, VOL. 29, NO. 1, the 6 mo interval. Average squelch at the two intervals was submitted to a t-test to determine whether the mean was significantly different from zero. For the 12 mo interval, squelch was significant (t , p ). This was not the case at the 6 mo interval, where mean squelch was not significantly greater than zero (t , p 0.55). Inspection of Figure 6 suggests that turning on the implant closest to the noise source may have impaired performance (as reflected in a negative value of squelch) in some subjects and conditions, an effect that could be described as central masking. These results together suggest that the squelch effect increased between the 6 and 12 mo test intervals and that the effect may be more consistent across ears at the 12 than the 6 mo interval. However, recall that about half of the subjects were tested in the unblocked order at the 6 mo interval, and about two-thirds were tested in the unblocked order at the 12 mo interval. Therefore, it is necessary to consider whether differences in the estimates of squelch could be caused partly or wholly by these procedural inconsistencies. The effect of test order was assessed for the 6 and 12 mo data separately. The mean squelch was calculated for each subject, and data collected using the blocked and unblocked orders were compared using a one-tailed t-test. The results show no significant differences between values of derived squelch based on blocked versus unblocked data at the 6 mo interval (t , p 0.27). Values of squelch did differ for the two orders at the 12 mo interval, however (t , p 0.01); squelch was greater in the blocked when compared with the unblocked order, with median values of 21.2% and 8.7%, respectively. When the data that were collected blocked are excluded, median values of squelch dropped by 0.1% for the 6 mo test interval and by 1.9% for the 12 mo test interval. This bias fails to account for the change in squelch discussed above; because blocked trials were associated with larger values and because more blocked trials were run at the 6 mo interval, this bias would have had the opposite effect from the difference observed here. That is, it would have accentuated the squelch observed at the 6 mo interval. These findings suggest that the change in squelch between the 6 and 12 mo test intervals is not an artifact, but rather reflects a real change in the underlying phenomenon. Finally, the performance in the two unilateral N F S F conditions can be used to characterize the ear bias for sentence materials. The difference between N F S F (left only) and N F S F (right only) was compared with the difference in CNC scores for the comparable unilateral listening conditions. The difference scores for CUNY and CNC materials were significantly correlated at both the 6 and 12 mo test intervals (r 51, p 0.01; r 0.55, p 0.005). This result indicates that although the ear bias may not be stable across all test intervals, as discussed above with respect to CNC data alone, there is a reliable ear bias within test interval at the 6 and 12 mo intervals. DISCUSSION Results with CNC words presented in quiet indicate a significant bilateral advantage compared to the better of the two unilateral conditions. This effect is relatively constant across the year of follow-up testing, ranging from a mean of 5.8% at the 1 mo interval to 11% at the 12 mo interval. These results are generally consistent with those reported by others. For example, Müller, et al., (2002) report an 18.6% improvement in the recognition of German monosyllabic words for nine subjects in comparing bilateral stimulation with either implant alone. Dorman and Dahlstrom (2004) report a 10% to 20% difference with CNC words, whereas Laszig, et al. (2004) show a smaller effect of 0% to 6%. Results from the CNC word tests suggest that one ear may be better than the other, but that the better ear is not consistent over time within subjects. Unilateral CUNY data at 6 and 12 mo corroborate the ear differences observed with CNC words at these time intervals, suggesting that these differences represent reliable within interval ear dominance effects. A principal advantage of bilateral implants is that the better of the two ears will always be available for stimulation. This is important, because at present it is not possible to predict which ear will support better performance after implantation (Gantz, et al., 2002) and, as just noted, the better ear may not be constant across the life of an implant patient. Bilateral stimulation ensures use of the better ear. Data on CUNY sentences in noise suggest there is significant binaural benefit at the 6 and 12 mo test interval. Estimates of head shadow are relatively stable over the two measurement intervals, and at a median value of approximate 37.5% they are comparable to those reported in the implant literature (e.g., Gantz, et al., 2002; Müller, et al., 2002) and comparable to those of normal-hearing listeners. Similarly, estimates of binaural summation were statistically indistinguishable between the 6 and 12 mo test intervals, with median values of 2.0% to 5.7%. These values are somewhat smaller than the effects reported for comparable studies of bilateral cochlear implant subjects (Müller, et al., 2002; Schleich, et al., 2004). A novel finding of this study is the emergence of a squelch effect between the 6 and 12 mo test inter-

11 30 BUSS ET AL. /EAR &HEARING, VOL. 29, NO. 1,20 32 vals, with a highly significant increase in magnitude from 3.3% to 10.6%. Although there is some precedence in the literature for comparable magnitudes of squelch in implanted listeners (e.g., Laszig, et al., 2004), a change in the magnitude over time has not to our knowledge been reported previously. This new finding suggests that binaural processing capabilities may develop over time and therefore that assessments of performance with bilateral cochlear implants should include subjects with substantial experience, at least out to 12 mo and quite possibly well beyond that. This finding may be related to the effects observed by Hall and Grose (1993), who measured the MLD in otosclerosis patients before intervention, as well as 1 mo and 1 yr after middleear surgery. The MLD increased between the preoperative and 1 mo follow-up, but improvement was also observed between the 1 mo and 1 yr follow-up, suggesting that binaural hearing takes time to adapt to changes in peripheral input. Although squelch is typically interpreted as definitive evidence of true binaural processing, reflecting the use of interaural differences in central processing to parse the incoming signal, it is also possible that redundancy could play a significant role. It has been shown with normal-hearing listeners that complementary speech information presented separately to the two ears can be fused into a single binaural percept (Broadbent & Ladefoged, 1957). Similar results can be obtained in simulations of the spectrally sparse speech signal provided by a cochlear implant (Loizou, et al., 2003), supporting the idea that bilateral cochlear implant listeners could be combining complementary speech information across ears. If the information encoded at the two ears is different perhaps because of differences in physiology, in the hardware, or the interface between the two then the binaural advantage observed with the introduction of either ear could improve performance by virtue of conveying novel information, quite apart from binaural processing. An analogous suggestion has been made with respect to bilateral hearing aid data (Cox & Bisset, 1984). This alternative interpretation of the squelch effect leads to several predictions that can be tested with data available on this cohort. First, there should be a positive correlation between squelch and the binaural advantage obtained under diotic listening conditions. If introducing input from the more heavily masked side improves performance by way of contributing novel information about the target speech, then an analogous or greater benefit should be observed when the added input is at an equal SNR. This was assessed by calculating the bilateral advantage relative to each unilateral condition for a N F S F stimulus, a value analogous to computing summation based on just one unilateral condition rather than the maximum percent correct in either condition. Correlation between these values and comparable squelch values were r 0.47 for 6 mo and r 0.34 for 12 mo test intervals. These correlations suggest a small effect, accounting for only 22% and 12% of the variance in these values, respectively. Further, the correlation is no stronger at 12 than 6 mo, a finding contrary to expectation if improved ability to combine complementary information across ears were a late-emerging ability. These results fail to support the hypothesis that squelch and summation are both based on combination of complementary speech information in the left and right sides. Therefore, apparent squelch effects reported here are likely based, at least in part, on true binaural processing. Further insight into the underlying cues in the speech in noise task may be gained by comparing individual differences obtained in that task with those observed by Grantham, et al. (2007) in localization and lateralization. The only other study to take this approach, of comparing performance on spatial hearing tasks with that for speech in noise, was that of Schleich, et al. (2004). That study compared estimates of squelch with accuracy in a localization task for 16 bilaterally implanted subjects and found no significant relationship. Such a relationship would be predicted if the squelch effect were based on true binaural processes, such as those associated with use of interaural differences cues and inferences regarding relative positions of signal and noise sources. Of the cohort of 26 patients followed here, 21 also participated in a sound localization study approximately 5 mo after implant activation. Stimuli were either Gaussian noise or a male voice saying the word hey, presented from one of 17 loudspeakers spanning 80 degrees on the azimuthal plane; both types of stimuli were 200 msec in duration and bandpass filtered 100 to 4000 Hz (Grantham, et al., 2007). These stimuli introduced both interaural time and level difference cues to localization. The adjusted constant error for the localization task was taken as an indication of accuracy and compared with the derived measure of squelch based on 6 mo data in the current data set: the correlation was r 0.50 (p 0.015) and r 0.59 (p 0.004) for the speech and noise stimuli, respectively, indicating that small localization error was associated with large values of squelch. In contrast, the correlation between the localization statistic and the derived measures of head shadow and summation were not significant using a one-tailed test ( 0.05), with correlations ranging from r 0.32 to r This