Comparing the Fused Dichotic Words Test and the Intracarotid Amobarbital Procedure in children with epilepsy

Transcription

1 Neuropsychologia 38 (2000) 1216± Comparing the Fused Dichotic Words Test and the Intracarotid Amobarbital Procedure in children with epilepsy Myra A. Fernandes a, *, Mary Lou Smith a, b a Department of Psychology, University of Toronto at Mississauga, 3359 Mississauga Road North, Mississauga, Ontario, L5L 1C6, Canada b Hospital for Sick Children, Toronto, Ontario, Canada Received 25 February 1999; received in revised form 2 February 2000; accepted 14 February 2000 Abstract The validity of the Fused Dichotic Words Test (FDWT) in predicting the nature of speech representation, as determined by the Intracarotid Amobarbital Procedure (IAP), was examined in a sample of 28 children with epilepsy. Various methods of analysis (di erence score, l, and l ), for the FDWT data were calculated and compared. Results showed the validity of the FDWT did not change depending on the method of analysis. The di erence scores showed that 18 of the 19 patients with left hemisphere speech obtained right-ear advantages, while the patient with right hemisphere speech showed a left-ear advantage. As a group, patients with left-hemisphere speech obtained a statistically signi cant right-ear advantage for the l and l indices, while the patient with right-hemisphere speech showed a left-ear advantage that was also signi cant for both l measures. Patients with bilateral speech, as a group, displayed a non-signi cant ear advantage. Some of the scores from the left-hemisphere group overlapped with those from patients with bilateral speech representation. Controlling for stimulus dominance e ects using the l measure did not improve classi cation accuracy for nature of speech representation based on FDWT scores. Finally, comparison of our data using the laterality index from a similar study, showed scores smaller in magnitude than that found in adults with epilepsy Elsevier Science Ltd. All rights reserved. Keywords: Speech; Lateralization; Validity; Stimulus dominance; Wada 1. Introduction Dichotic listening tests are typically used to estimate lateralization of cognitive function, particularly speech representation [6]. The validity of the dichotic listening technique in determining hemisphere of speech representation was rst demonstrated by Kimura [20]. In her study, patients about to undergo surgery for relief from epileptic seizures, and who had undergone the intracarotid amobarbital procedure (IAP) to determine the hemisphere of speech representation [35], were given Broadbent's [4] digit-based free recall dichotic test. In this test, participants are presented with three * Corresponding author. Tel.: ; fax: address: myra@psych.utoronto.ca (M.A. Fernandes). pairs of numbers simultaneously to each ear (one member of a pair to each ear). Following each set of six numbers, participants were asked to recall, in any order, everything they heard. Kimura's results showed that patients classi ed as having right hemisphere speech representation by the IAP were more accurate in recalling items presented to the left ear on the dichotic test, while those classi ed with left hemisphere speech were more accurate on the right ear. These data, together with the rationale that contralateral projections from ear to brain are stronger than ipsilateral pathways [29], suggested that the dichotic listening technique could provide a noninvasive assessment of speech laterality, with right or left ear advantages indicating left and right hemisphere speech representation respectively. One of the drawbacks of her work, however, was /00/$ - see front matter Elsevier Science Ltd. All rights reserved. PII: S (00)00035-X

2 M.A. Fernandes, M.L. Smith / Neuropsychologia 38 (2000) 1216± that participants normally reported words from the right ear, followed by those in the left ear, thereby confounding the ear advantage by a short-term memory e ect and by the order in which items are reported [14]. Also, results from her dichotic listening test were presented as group ear advantages, which does not allow an evaluation of the validity of the test for individual patients. Furthermore, the study did not include patients with bilateral speech representation. Thus the usefulness of the dichotic listening test in a clinical setting where individual assessment is critical [30], was not addressed. Since then, there have been only a few studies investigating the validity of dichotic listening tests in determining the dominant hemisphere of speech representation. In a dichotic listening study, Strauss, Gaddes and Wada [32] tested young adult patients with epilepsy, whose speech dominance had been ascertained by the IAP, using a verbal dichotic listening test similar to that used by Kimura [20]. While most of the patients with left hemisphere speech showed the expected right ear advantage (REA), patients with right hemisphere speech, as a group, tended not to show a bias towards either ear. In fact half of the patients with speech dominance in the right hemisphere showed an REA! Furthermore, most of the patients with bilateral speech representation showed right-ear advantages. Strauss et al. [32] had expected that a reduced or reversed laterality e ect would be found in patients with bilateral or right hemisphere speech. The ear advantages reported from their dichotic test, however, misclassify several patients. It appears that the type of dichotic test used in this study is adequate at classifying those with left-hemisphere speech, but more equivocal for classi cation of those with right hemisphere or bilateral speech representation. In another study, Hugdahl et al. [18] used discriminant analysis to predict hemisphere of speech representation from consonant±vowel syllable dichotic listening data. Moreover, they investigated the validity of the test in a sample of children with epilepsy. Results showed that 12 of the 13 patients in their study were correctly classi ed by their analysis. However, the magnitude of ear advantage score was not considered in their study. There were no patients with bilateral speech in their sample, and so the possibility remains that the consonant±vowel dichotic test may not be adequate at di erentiating patients with unilateral speech from those with bilateral speech representation, based on the magnitude of asymmetry score on their dichotic test. Determination of a cut o score for ear advantage scores on dichotic tests may be of critical importance in classifying individual patients with bilateral versus unilateral speech representation [37]. Rather than using a cut-o score, Ge en and Caudrey [12] used Discriminant Function Analysis to classify patients in their study. They considered how performance on a dichotic monitoring test related to hemisphere of speech dominance in young adults, as determined by dysphasia following unilateral ECT treatments or IAP. Their results were impressive, and supported Kimura's initial ndings that detection of verbal material presented dichotically was better in the ear contralateral to the dominant speech hemisphere. Speci cally, their study showed right-ear advantages on the dichotic listening test for those with left hemisphere speech. However, some participants with right hemisphere and bilateral speech representation had right-ear advantages that did not appear to di er from patients with left hemisphere speech. Nevertheless, they were able to correctly classify 95% of their sample using a discriminant function analysis that included handedness, hit rates and reaction times on their monitoring test. Some researchers claim that there are extraneous variables, unrelated to hemispheric asymmetry, that may in uence ear asymmetry scores on some dichotic tests [2]. Attentional biases [8,22], strategies of information processing [5], and experience [13] have been shown to in uence ear asymmetry scores, leaving the construct validity of dichotic listening tests open to question. With these issues in mind, a somewhat better task is the fused dichotic words test (FDWT) developed by Wexler and Halwes [36]. In this test, pairs of monosyllabic rhyming words, di ering only on the initial consonant, are presented simultaneously to participants. The words are temporally aligned such that the participant experiences only a single percept localized in the center of the head. Participants are not told that words are presented dichotically, they are simply asked to report the word that they heard on each trial. This procedure is advantageous because it eliminates the order of report problem inherent in dichotic tests that require recall of items. Also, the results are not in uenced by attentional manipulations [1]. In fact, Asbjornsen and Bryden [2] have recently shown that the FDWT is much less likely to be a ected by shifts of attention (initiated by the participant) than other dichotic listening tests. On the negative side, the dichotic fusion procedure may be susceptible to stimulus dominance e ects [16,27], that contaminate the estimate of size of ear advantage in dichotic test results [14]. Stimulus dominance occurs when a participant responds with the same word regardless of ear of presentation. For example, if the stimulus pair boy/toy is presented an equal number of times to each ear, and the participant's response is always ``boy'', then ``boy'' is considered to be the dominant stimulus. As we shall see, however, these e ects can be estimated and separated

3 1218 M.A. Fernandes, M.L. Smith / Neuropsychologia 38 (2000) 1216±1228 from ear dominance, making the FDWT an ideal noninvasive test of speech lateralization. In 1989, Zatorre [37] provided the rst validation study of the FDWT. In Zatorre's study, the test was administered to 61 adult patients with epilepsy who had undergone the IAP to determine side of speech representation. His results were striking: 33/35 patients classi ed as having left hemisphere speech showed right-ear advantages. All of the patients (4/4) classi ed as having right hemisphere speech showed left-ear advantages (LEA). There were many patients classi ed as having bilateral speech representation. Some of them showed left-ear and others showed right-ear advantages, but these were for the most part small in magnitude. Thus, while there was some overlap between the distribution of scores for patients with bilateral and unilateral speech representation, the study showed that the FDWT could provide valid estimates of hemispheric speech representation for individual patients. The objective of the present study was to extend the ndings from Zatorre [37] by examining di erent methods of analysis for FDWT. One reason for examining various indices of lateralization was to determine whether controlling for stimulus dominance e ects improved classi cation accuracy for nature of speech representation based on FDWT scores. Another reason was to address a controversy which began with a study by Kimura [21] and continued with a study by Harshman and Krashen [17]. In these studies, the index of lateralization was right ear accuracy minus left ear accuracy (di erence score). Their results showed that the magnitude of the right-ear superiority decreased with increasing age. This led to the unlikely conclusion that ear asymmetries decrease as a child gets older, and if dichotic tests measure hemispheric asymmetry, that speech lateralization becomes less specialized with age. Since then, several methods of analysis of dichotic data were devised [7] which explain and eliminate the counterintuitive conclusions derived from use of the di erence score index. Bryden and Sprott [7] point out that young children nd dichotic tests di cult, and make many errors, thus permitting the di erence between performance on each ear to be quite large. Older children, on the other hand, make fewer errors and attain high levels of performance on each ear; hence they cannot obtain such large di erence scores. As such, the size of the di erence score is correlated with overall performance [7], and therefore may not be the best method of analysis for dichotic test data. In the present study we compared the conclusions reached by various laterality indices, to determine which index provided the most valid measure of speech laterality in children with epilepsy. Speci cally, we calculated three indices of laterality from the FDWT results: the di erence score, the traditional l index [7], and the l index which controls for stimulus dominance e ects [14]. This last index provides a statistically appropriate method for controlling for stimulus dominance e ects [14]. In Zatorre's [37] validation study of the FDWT, all responses that were attributable to stimulus dominance were discarded. The total number of non-stimulus dominant responses was then calculated for each ear. As Grimshaw et al. [14] point out, this technique of dealing with responses due to stimulus dominance e ects is inappropriate because it disregards the probabilistic nature of the data (see [14] for full discussion). We wished to determine whether controlling for stimulus dominance, using this more appropriate index, improved classi cation accuracy. Another goal of the study was to determine the validity of the FDWT for a group of children with epilepsy. In view of the claim that speech lateralization may not be as advanced in children as in adults [23,24], we were interested in comparing the scores obtained from our sample of children to those obtained in Zatorre's [37] study of adults with epilepsy. For this comparison, we used the same method of calculation for the laterality index as in Zatorre's [37] study. As Voyer [34] has recently pointed out, there is a need for further investigation of dichotic listening techniques as non-invasive measures of speech laterality. Several researchers have investigated the validity of various dichotic tests in adults [12,32,37]. The present study is the rst validation study of the FDWT for use with children. It is an arguably better dichotic test than others as it minimizes attentional confounds in asymmetry scores, and there now exists a method of analysis that controls for stimulus dominance. From a clinical perspective, a less expensive, less time-consuming method of determining speech lateralization than the IAP, would be preferred. If the validity of dichotic listening tests could be demonstrated for individual cases in addition to groups with known speech representation, it would lessen the need to put surgical candidates through such an invasive, sometimes traumatic procedure. Furthermore, a validity study of the FDWT with children provides researchers studying populations which are unlikely to undergo IAP testing, (e.g. children with learning disabilities) a sound method for assessing the nature of speech representation. 2. Method 2.1. Participants Twenty-eight children diagnosed with epilepsy were included in the study. All had been referred to the

4 M.A. Fernandes, M.L. Smith / Neuropsychologia 38 (2000) 1216± Hospital for Sick Children in Toronto for neuropsychological testing as part of a medical evaluation to determine whether they were suitable candidates for seizure surgery. The mean age of our sample was 12.9 (SD=2.9), with ages ranging from 6.3±16.9 years. All of the patients had an epileptogenic focus lateralized predominantly to one hemisphere. The epileptogenic focus was determined by prolonged EEG recordings of ictal and interictal electrographic abnormalities, and neuroimaging techniques (SPECT, PET, and/or MRI). Table 1 shows the characteristics of each patient in our sample. Handedness was determined by parent and self-report of writing hand. All of the patients were native English speakers. Hearing for each patient was evaluated using pure tone audiometry. It should be noted that patient RC displayed some bilateral hearing loss, but she was included in the sample since her FDWT scores did not indicate an unusual number of errors, and she showed little di culty following instruction and carrying out the dichotic test. All participants were on anticonvulsant medications at the time of testing. The mean number of medications taken by the group was two (range 1±4) Procedure Speech lateralization was determined IAP. Sodium amobarbital was injected, following catheterization of the internal carotid artery. The dose of sodium amobarbital was titrated to body weight. In accordance with standard practice for the IAP at the Hospital for Sick Children, the dose of sodium amobarbital administered was 1.5 mg/kg injected into each hemisphere. Both hemispheres were tested on the same day with a brief interval between the two tests. The speech protocol used was individualized to the child, taking into consideration the child's age, developmental level, and speech ability. Baseline testing was carried out prior to injection in order to compare performance to that when the drug was circulating. Whenever possible, the child was asked to count at the time of injection, and this was followed by tests of naming pictures and/or objects. Spelling, reading, and/or reciting the days of the week or the alphabet, were also assessed when the child was capable of such tasks. EEG monitoring for the presence of slow waves, as well as paralysis on the side contralateral to the injected hemisphere were taken as indicators that the injection was successful. Table 1 Characteristics of patient population and FDWT indices a ID Sex Hand Age (years) Method Seizure side and location Age of seizure onset (years) Zl Zl Wada ML m R R temporal L DB m L R temporal L JN m L R frontal-parietal L MP f R R frontal L VC f L L temporal L SW m R L temporal L LY f R L temporal L MW m R R temporal L KN f R R frontal L JL f R L temporal L AB f L L temporal L DC f R R temporal L RW m R L temporal L JR m R L temporal L MM f R L temporal L JI m R L frontal L SM m R L central L VG f L L temporal L KW f R R temporal L NR f R R frontal Bilat JF m L L frontal Bilat JO f L L temporal Bilat AT f R L temporal Bilat DA m R L temporal Bilat RC f R L occipital Bilat RB m R L temporal Bilat NB f R L temporal Bilat RM m R L temporal R a Zl calculation including stimulus dominant trials; m=male, f=female, L=left, R=right, Bilat=bilateral, 1=report aloud method, 2=reading method.

5 1220 M.A. Fernandes, M.L. Smith / Neuropsychologia 38 (2000) 1216±1228 Following injection, indications of speech representation within the hemisphere were speech arrest and/or errors on tasks the child was capable of performing perfectly at baseline. If the injection was deemed successful, a patient was classi ed as having bilateral speech representation if he/she demonstrated at least one of the following: a) no speech arrest or errors when either of the hemispheres was injected (i.e. duplication of speech representation), b) a similar number of errors and of comparable levels following injection of either hemisphere (i.e. speech representation shared between the hemispheres) or c) a qualitatively di erent pattern of speech errors following injection of either hemisphere (i.e. speech specialization distributed between the hemispheres). Each participant was also given the FDWT developed by Wexler and Halwes [36]. The test consists of 30 pairs of rhyming words, which di er only on the rst consonant (e.g. boy/toy). Words were presented via a Sony Walkman, and Telephonics TDH-39P headphones. In order to become familiar with the words, participants rst listened to each of the words binaurally. The same binaural list was then repeated and the participant was instructed to select the word they heard (reading method of data collection) from among four choices printed on a sheet of paper: the word presented to either ear plus three others that differed only in the rst consonant. In the case where a patient was an ine cient reader, or could not read (N=21), s/he was instructed to report aloud the word they heard for each trial, and the experimenter circled their response on the score sheet (report aloud method of data collection). 1 If the child reported a word other than the four choices listed on the score sheet, it was noted by the experimenter and later scored as an error. If the experimenter was unsure of what the patient reported, the FDWT tape was stopped, and the patient was asked to repeat the response. During the test phase, rhyming pairs of words were presented dichotically. Participants were not told that pairs of words were presented dichotically. To minimize the possibility of channel asymmetries, the headphones were switched between the ears after the rst 30 trials, and from then on after every 90 trials. Each 1 We did not expect that these two methods of collecting data for the FDWT, reading versus reporting aloud, for individuals who could and could not read, would alter the FDWT results. Roberts [28] compared scores obtained on the FDWT from a sample of young adults, half of which completed the test using the reading condition and the other half of which did the test using the report aloud condition. The two methods yielded comparable results, with a di erence of two correct responses for the right ear and one correct response for the left ear, across the two methods of data collection. participant heard each word pair four times, twice in each ear, for a total of 120 trials per randomization. Each participant was tested using two randomizations, for a total of 240 trials Calculations of laterality indices from the FDWT The number of correct right and left ear responses was tabulated, which represents the number of times the participant correctly identi ed the word presented to that ear (regardless of stimulus dominance). Several measures of laterality were calculated from the FDWT. The rst measure of laterality that was calculated was the di erence score between the number of right and left ear responses. A positive di erence score indicated a right-ear advantage (REA), and a negative di erence indicated a left-ear advantage (LEA). A second measure of laterality, the l index [7], was also calculated and provided another measure of relative ear advantage. This measure is the natural logarithm of the ratio between the right and left ear responses, which as Bryden and Sprott [7] point out, provides an index of lateralization that does not depend on overall accuracy, and that allows for various statistical tests. It should be noted that unlike the scoring method in Zatorre's study [37], all correct responses for each ear were included in the calculation of these aforementioned indices. We did not discard stimulus dominant trials, but instead compared these indices to one that controls for stimulus dominance e ects statistically (see below). A third measure of laterality was also calculated. Grimshaw, McManus and Bryden [14] outline a loglinear analysis of the responses from the FWDT which yields a l-type index (l ), that provides a measure of ear dominance independent of stimulus dominance. Trials that are subject to stimulus dominance do not provide any information about ear dominance, and simply create noise in the participant's data. This makes it more di cult to obtain a signi cant ear advantage and decreases the power of any statistical test of an individual's ear advantage [14]. Furthermore, there are likely to be individual di erences in stimulus dominance. As Grimshaw et al. [14] outline, laterality indices such as l measure a combination of ear and stimulus dominant e ects for a given individual, thus making comparisons between individuals' degree of lateralization, equivocal. The method of controlling for stimulus dominance e ects (l ) described in detail by Grimshaw et al. [14] is a conventional log-linear analysis using all correct responses from the FDWT. Their suggested analysis represents the FDWT data for each subject in a three way table: Word pairs Stimulus arrangements Responses. An interaction between Responses and Stimu-

6 M.A. Fernandes, M.L. Smith / Neuropsychologia 38 (2000) 1216± lus arrangement indicates that the participant's responses depend on which word was presented to which ear, a measure of ear dominance. An interaction between Responses and Word pairs indicates a participant's response varies depending on the word pair presented, a stimulus dominance e ect. The log-linear analysis is similar to an ANOVA in that one can evaluate main e ects, as well as interactions, independent of other e ects. As such, it is possible to obtain a measure of ear dominance that is independent of stimulus dominance e ects. Grimshaw et al. [14], describe how to test e ects in a log-linear analysis. Brie y, two models are t to each participant's data. Model 1 includes all possible e ects except Response Stimulus arrangement, whereas Model 2 adds in this interaction. Each model produces a likelihood ratio chi-square value, along with parameter estimates for main e ects and interactions in each model. If the Response Stimulus arrangement interaction is signi cant in Model 2, this model will be a better t to the data, and provide a parameter estimate for ear dominance independent of stimulus dominance e ects. The parameter associated with the Response Stimulus arrangement interaction is termed l. It represents the log-odds ratio of right to left ear responses, after stimulus dominance has been statistically controlled, and is analogous to Bryden and Sprott's [7] l index. l can be converted to a Z score by dividing it by its standard error. Zl is normally distributed, allowing for an easy test of signi cance. Zl scores greater or lesser than indicate a signi cant ear advantage (at the a=0.05 level), with a negative index denoting a LEA and positive index indicating a REA. We were also interested in comparing the laterality scores obtained from our sample of children, to those obtained in Zatorre's [37] study of adults with epilepsy. For this comparison we re-calculated the data for each patient, controlling for stimulus dominance using the same method as described in Zatorre [37]. That is, the number of non-stimulus-dominated responses was calculated for each ear, using an approach outlined by Halwes [16]. All responses that were attributable to Fig. 2. Distribution of Zl (l/se) scores on the FDWT according to side of speech representation as determined by intracarotid sodium amobarbital testing. Each symbol represents a patient. stimulus dominance were subtracted from the total. The number of non-stimulus dominated responses was totalled for each ear, and the natural logarithm of the ratio between the right and left ear responses was calculated. 3. Results In order to better compare the traditional l index with Zl, the former can also be converted to a Z score by dividing by the standard error. The distribution of scores from the FDWT is shown in Figs. 1± 3, grouped according to the nature of speech representation (left hemisphere, right hemisphere and bilateral) as determined by the IAP test. Di erence scores indicated 18/19 participants with left hemisphere speech showed an REA, and 1/1 participants with right hemisphere speech showed an LEA. Scores from the left hemisphere and bilateral group do overlap, for the most part within the range of 0±25 on the di erence score, and 0±1.5 on the Zl and Zl indices. Table 2 shows the mean values of each laterality index for each group. For comparison, also shown are the mean indices from a group of normal children taken from Grimshaw et al. [15]. The Zl measure classi es 12/19 of the patients in the left-hemisphere group as having a signi cant right- Fig. 1. Distribution of di erence scores (correct right±correct left) on the FDWT according to side of speech representation as determined by intracarotid sodium amobarbital testing. Each symbol represents a patient. Fig. 3. Distribution of Zl scores (l /SE) on the FDWT according to side of speech representation as determined by intracarotid sodium amobarbital testing. Each symbol represents a patient.

7 1222 M.A. Fernandes, M.L. Smith / Neuropsychologia 38 (2000) 1216±1228 ear advantage on the FDWT, and 5/8 of the patients in the bilateral group as having no signi cant ear advantage. Using the traditional l index, the same 12/19 patients in the left-hemisphere group had signi cant right-ear advantages, and the same 5/8 patients in the bilateral group had no signi cant ear advantage. In addition, one patient with bilateral speech showed a signi cant REA using Zl, but non-signi cant REA using Zl (see Table 1). For both measures, the righthemisphere patient was correctly classi ed as having a signi cant left-ear advantage. Thus for all but one case, both l-type indices led to the same conclusions. It should be noted that there were no signi cant di erences in the size of FDWT indices between patients who performed the report aloud versus reading method of data collection (t26=0.23, 0.09, and 0.08 for the di erence score, Zl, andzl respectively). Because of the small number of patients tested with the di erent methods of data collection, and the broader goals of this paper, all analyses were performed on data collapsed across method of data collection Homogeneity of FDWT scores within each group Bryden and Sprott [7] outline a method of statistically evaluating the homogeneity of a collection of traditional l scores from a dichotic listening test. The homogeneity of scores was examined for the group of patients with left-hemisphere and bilateral speech representation based on the IAP. This analysis revealed that although the sample size was relatively small, both groups could be considered homogeneous, w 2 (18)=3.4 for the left hemisphere group, and w 2 (7)=3.47 for the bilateral group, P > Analysis of errors on FDWT Because the FDWT was administered to children, an analysis of the error rate on test performance was important to demonstrate its appropriateness for this age-group. Errors are responses that do not correspond to either of the words presented dichotically. For the report aloud group, the mean error was 7.06% (SD=8.42), and for the reading group was 1.27% (SD=1.29). Despite the fact that the mean error rate for the report aloud group was larger than that of the reading group, and Zatorre's [37] adult population, it does not seem unusually large given that the test was carried out in its entirety (240 trials), in a population of children of varying ages. The error rates did not correlate signi cantly with IQ scores (r= 0.15). The correlation with age approached signi cance (r= 0.37, P = 0.054) Correlations between measures of laterality The main reason for calculating the Zl index was to control for stimulus dominance e ects, for which the traditional l index and di erence score do not take into account. The indices, however, were highly correlated. As such, stimulus dominance e ects do not greatly in uence the magnitude of the laterality index. Pearson product-moment correlations between the di erence score, l and Zl measures were calculated for each group. Results showed that all measures were highly intercorrelated for each group. For the lefthemisphere group, the correlation between the di erence score and l was r = 0.82, and with Zl was r = The traditional l score also correlated with Zl r = Similarly high correlations were obtained with the bilateral speech group; the correlation between the di erence score and l was r = 0.99 and with Zl was r = 0.99, and the correlation between these last two measures was Predictors of asymmetry scores In order to determine which factors contribute to the asymmetry in performance on the FDWT, analyses of variance were performed on the sample of children with left-hemisphere and bilateral speech representation only. Since the right-hemisphere group consisted of only one patient, he was not included in the analyses. Analysis of variance was carried out once for the di erence scores as the dependent variable and again with Zl, as the dependent variable (because Zl is a Table 2 Mean values for the di erence score, traditional Zl and Zl index according to side of speech representation (SD shown in parentheses) a Side of speech Di erence score Zl Zl Left (n = 19) (34.42) 2.61 (2.2) 2.61 (2.3) Bilateral (n =8) 3.38 (69.20) 0.03 (4.1) 0.10 (4.0) Right (n =1) Comparison group (from Grimshaw et al., 1995) [15] (n = 43) ± 2.59 a Mean Zl score and SDs from Grimshaw et al. [15] not available.

8 M.A. Fernandes, M.L. Smith / Neuropsychologia 38 (2000) 1216± Table 3 Mean values for the di erence score, traditional Zl and Zl index according to side of speech representation and handedness (SD shown in parentheses) Side of speech Handedness Di erence score Zl Zl Left (n = 14) right (32.65) 2.79 (2.2) 2.78 (2.2) Left (n = 5) left (42.53) 2.12 (2.6) 2.13 (2.6) Bilateral (n = 6) right (80.00) 0.60 (4.7) 0.32 (4.6) Bilateral (n = 2) left (2.12) 1.65 (0.2) 1.40 (0.0) Right (n = 1) right purer measure of ear advantage, we considered only this index in the analysis). Four factors were included for each ANOVA: Nature of speech representation as determined by IAP testing, Seizure side (hemisphere of epileptogenic focus), Sex, and Dominant hand. Because of the small number of observations in each cell, only main e ects, two-way and three-way interactions that included the IAP factor were analyzed, leaving the other sources of variation to be pooled in the error term. When all of these factors were included in the ANOVA for the di erence score and Zl, the only factor nearing signi cance was Nature of speech representation as determined by IAP, F(1,16)=4.39, P = and F(1, 16)=4.23, P = respectively. No other e ects or interactions approached signi cance. Although analyses were not conducted using the righthemisphere speech patient, his scores on the FDWT correctly predicted side of speech representation Comparison of the direction and degree of laterality indices based on handedness Table 3 shows the mean di erence score, traditional Zl and Zl score for each group, according to handedness. In the left-hemisphere group, right-handers had greater di erence scores, Zl, andzl scores, indicating a larger REA than left-handers. These greater scores, however, were not signi cant for any of the measures. While the laterality indices for those in the bilateral group were not indicative of signi cant ear-advantages, left-handers showed a slightly larger REA for all indices compared to right handers. The one patient with right hemisphere speech representation was right handed and obtained a signi cant LEA on all indices Distribution of patients by nature of speech representation Patients were grouped based on nature of speech representation (right, left, bilateral), as determined by the IAP. Analyses were conducted, independent of FDWT scores, to examine the distribution of handedness, sex, and seizure side in each group, and to determine whether the number of patients in each of these categories di ered according to the nature of speech representation. No signi cant di erences between the groups were found for any of the categories. Results from chi square analyses were as follows: for handedness w 2 (1)=0.005, for sex w 2 (1)=0.222, for seizure side w 2 (1)=1.600, P > 0.2. Thus, the number of patients in each group did not di er according to handedness, sex, or seizure side Comparison of di erent methods of controlling for stimulus dominance The mean di erence score and l score, using ear totals corrected for stimulus dominance as in Zatorre [37], are presented in Table 4, according to the nature of speech representation as determined by the IAP. Also shown is the total number of scorable responses (i.e. total non-stimulus dominated trials) for each speech group. 4. Discussion In our sample of children with epilepsy, FDWT results were related to the nature of speech representation, as determined by the IAP. The test revealed that 18 of the 19 patients with left hemisphere speech obtained REAs, while the patient with right hemisphere speech showed an LEA. Thus the test provides a valid, noninvasive assessment of speech laterality in children. This is consistent with Hugdahl et al.'s [18] Table 4 Mean values for the total number of non-stimulus-dominant responses, and laterality indices on the FDWT using ear totals corrected for stimulus dominance, as in Zatorre [37] (SD shown in parentheses) Side of speech Total score Di erence score l Left (n = 19) (26.85) (34.15) 1.04 (0.73) Bilateral (n = 8) (48.93) 3.50 (68.62) 0.24 (1.10) Right (n = 1)

9 1224 M.A. Fernandes, M.L. Smith / Neuropsychologia 38 (2000) 1216±1228 nding that a dichotic listening paradigm using consonant±vowel sounds was predictive of hemisphere of speech representation in children and adolescents. Insofar as the individual validity is concerned, large asymmetry scores on the FDWT were associated, for the most part, with unilateral speech representation. As a group, patients with left-hemisphere speech obtained a statistically signi cant REA for all indices, while the patient with right-hemisphere speech showed an LEA that was also signi cant. While patients with bilateral speech, as a group, displayed a non-signi cant ear advantage, some of the scores from the left-hemisphere group did overlap with those from patients with bilateral speech representation, though this region is constrained to scores near zero, or slightly larger than zero. The results from our validity study of FDWT in children, yielded a pattern of results strikingly similar to Zatorre's [37] study of adults. Both studies showed that the degree of ear asymmetry on the FDWT is important to consider in di erentiating patients with bilateral versus unilateral speech representation. What remains to be explained is why some patients with lefthemisphere speech obtained only small ear advantage scores, similar to those with bilateral speech representation. Because of the importance of di erentiating individual patients with bilateral versus unilateral speech representation, the use of a statistical criterion for determining signi cant ear advantages is necessary, and will be considered in more detail below. Although the sample in our study consisted of children with a neurological disorder, there is no reason to believe that their performance was unusual or biased. The mean size of the di erence score and Zl index obtained by patients in the left-hemisphere group is comparable to that found in Grimshaw et al.'s [15] study of normal children. Although the number of errors on the FDWT was larger for the report aloud group, these constituted less than 10% of the total number of trials on the test. Thus the FDWT lends itself well to a clinical population of children Does correcting for stimulus dominance alter the predictive validity of the FDWT? Zatorre [37] suggested that stimulus dominance e ects have an important in uence on FDWT results, which must be taken into account in the interpretation of dichotic listening test asymmetries. In our analyses, however, controlling for stimulus dominance e ects did not greatly alter the number of correct and incorrect classi cations of speech representation, as compared to a measure that does not control for this confound. In fact, using the Zl index led to one additional incorrect classi cation than the Zl method of analysis (see patient AT). This suggests that stimulus dominance e ects, which are controlled with the Zl measure, but not with the traditional Zl measure, did not a ect the size of the asymmetry obtained on FDWT substantially, nor did controlling for these e ects improve the number of correct classi cations. Consequently the predictive validity for determining hemisphere of speech representation was not a ected to a large extent by stimulus dominance e ects, at least in children. Zatorre [37] found that the degree of overlap between patients in the left-hemisphere and bilateral speech groups was smaller when he included only the data from patients who had at least 30 non-stimulusbound responses. He concluded from this that control of stimulus dominance e ects is crucial, and that if these e ects are ignored, many patients will not obtain su ciently asymmetric scores to permit classi cation. Yet in the present study, control of stimulus dominance e ects did not improve classi cation. In trying to explain the overlapping distribution of scores from left-hemisphere and bilateral speech groups, Zatorre [37] suggested that in his study, insu cient sampling could account for the overlap. That is, the small asymmetry score found in some patients with left-hemisphere speech occurred because there were relatively few non-stimulus-bound words available for calculating the asymmetry score. His results showed a strong correlation between the number of non-stimulus-bound words and the degree of asymmetry score, suggesting that some of the patients obtained small asymmetry scores because they were susceptible to a high number of stimulus dominant trials. He predicted that if the number of words available for scoring was increased, that patients with bilateral speech would continue to show small magnitude ear asymmetries, whereas patients with left-hemisphere speech whose score overlapped with the bilateral group, would show an increased REA. In the present study the Zl method of analysis did not involve discarding trials, but statistically controlled for the e ects of stimulus dominance. Nevertheless, there was still an overlap in the distribution of Zl scores obtained by those with bilateral and left-hemisphere speech. Thus there is no reason to suspect that those patients with left hemisphere speech who obtained small asymmetry scores, did so because of a high number of stimulus dominated trials Choosing a signi cance level Because of the importance of di erentiating bilateral and unilateral speech representation in clinical patients, the use of a cut-o criterion to determine signi cant ear advantages is an important issue. The choice of the signi cance level to use when assessing an individual ear-advantage determines the number of

10 M.A. Fernandes, M.L. Smith / Neuropsychologia 38 (2000) 1216± Type I and Type II errors that will be committed. When a fairly strict signi cance criterion (e.g., P < 0.05) for the l indices is used, as it was in the present study, the number of patients with small ear-advantages, who are misclassi ed as having a signi cant REA, is reduced. Correspondingly, the number of patients with true right-ear advantages, who are misclassi ed as having non-signi cant ear-advantages, increases. As Zatorre [37] suggested, and as our results indicate, non-signi cant ear advantages are meaningful because they may re ect bilateral speech representation, and as such one does not want to fail to notice them. For clinical purposes, it is important to determine if speech function resides unilaterally or bilaterally prior to surgical removal of a hemisphere; thus, it is better to set a fairly strict criterion such as P < 0.05 which again, would reduce the chances of misclassifying a patient with a non-signi cant ear-advantage as one with a REA (reduce the chance of committing a Type I error). For example, if the signi cance criterion in the present study were very liberal (e.g., P < 0.4), the overall accuracy for classifying patients based on laterality indices would not really change. That is, three fewer patients with left hemisphere speech would be misclassi ed as having bilateral speech, but there would be two more patients with bilateral speech that would be misclassi ed as having left hemisphere speech representation. Thus the optimal choice for a signi cance criterion for the present study, where identi cation of bilateral speech representation is of critical importance (prior to surgery), is a strict one such as P < There were nevertheless, several patients whose laterality indices from the FDWT were noticeably extreme given their IAP results. We re-examined data from the two patients with bilateral speech representation who obtained laterality indices indicative of strong unilateral speech, as well as the patient with left-hemisphere speech, whose FDWT data revealed an LEA. Patient NB showed the strongest LEA despite the fact that her IAP was indicative of bilateral speech representation. On IAP testing, NB exhibited a di erent pattern of errors following injection to each hemisphere. Leftsided injection produced errors on serial speech tasks (counting, spelling, days of the week), but naming and reading were not a ected. Right-sided injection produced errors in reading and naming, but not on serial speech tasks. Patient DA was also classi ed as having bilateral speech, yet showed a strong REA. DA's IAP showed that after left-hemisphere injection, naming and reading, as well as the ability to respond to questions was interrupted, although counting ability was preserved. Right-hemisphere injection resulted in a very brief interruption of serial speech (counting, spelling), but intact naming and reading. Moreover, speech function returned to baseline sooner than after the left-sided injection. These ndings were suggestive of bilateral representation, but with a greater contribution by the left than by the right hemisphere. Based on these discordant patient data, it may be that the FDWT assesses lateralization of speech functions responsible for reading and naming, rather than the hemisphere responsible for serial speech. However, such conclusions are speculative, and must await further investigation. The FDWT is a test of speech perception, whereas the IAP largely focuses on speech production, using a variety of tasks. It is possible that not all aspects of speech are similarly lateralized, and that the FDWT and IAP measure di erent manifestations of language. A closer look at the type of errors made on the IAP task may help delineate which aspect of language the FDWT is tapping into. Another discordant classi cation by the FDWT involved patient KW, who showed a non-signi cant LEA on the FDWT, but left-hemisphere speech on IAP testing. This child had a right temporal lobe lesion, and an atypical neuropsychological test pro le in that she had pronounced de cits in language skills and verbal abilities, with visuospatial abilities falling within normal limits. It was of interest that her mother had a verbal learning disability. Injection into the left hemisphere resulted in initial speech arrest; the gradual return of speech was marked by the initial production of jargon, and errors in counting, spelling, reading, picture naming, and reciting the days of the week in reverse order. Injection into the right hemisphere had no e ect on speech function. There is no obvious explanation for her discordant FDWT results Predictors of asymmetry scores The ANOVA results presented in this study indicate that one of the key factors in uencing the distribution of scores from the FDWT is the nature of speech representation. As in Zatorre's [37] study, sex, seizure side (side of epileptogenic focus), and handedness, did not a ect scores on the FDWT. Furthermore, the distribution of these factors across bilateral, left, right-hemisphere speech groups did not di er. In a review of studies investigating dichotic listening performance following unilateral brain damage, Bryden [6] points out that a lesion e ect is often found. That is, unilateral temporal-lobe damage produces a de cit on the contralateral ear. Left temporal damage leads to a de cit on the right ear, resulting in LEA scores for these patients. Similarly, right temporal damage results in an exaggerated REA [26,31]. Our results do not indicate a signi cant role for side of brain dysfunction (seizure side) in determining FDWT scores. Our sample consisted of patients with unilateral

11 1226 M.A. Fernandes, M.L. Smith / Neuropsychologia 38 (2000) 1216±1228 dysfunction in a variety of locations, not just the temporal lobes. This may have prevented us from nding a signi cant e ect of side of brain dysfunction. Thus the contribution of side of epileptogenic seizure, on the nature of speech representation, must await further research. A better understanding of the kinds of speech processes assessed by the FDWT, and the regions of the brain most involved with performance of this particular test seem necessary if we want to include side of brain dysfunction as a factor determining dichotic test scores. It is possible that when the epileptogenic focus is located unilaterally, in a brain region particularly associated with the FDWT, seizure side will account for much more variability in test scores. We hope to address some of these issues in a current fmri study that considers how language-elicited activity in various brain regions correlates with scores on the FDWT. On the other hand, because the sample in this study consisted of children, whose speech functions may still be developing [24], it is possible that seizure side matters little when the brain is in a relatively immature state. That is, recovery of function following brain damage may occur [9], making seizure side less of a factor in determining the nature of speech representation, and FDWT scores. To further complicate matters, there are many factors such as lesion size that have been shown to in uence the pattern and degree of brain plasticity. For example, small neocortical lesions are associated with compensation mediated by brain regions ipsilateral to the side of injury, whereas large lesions lead to compensatory changes in the contralateral hemisphere [19]. Another factor, shown to a ect the pattern and degree of recovery, is the presence and duration of intractable epilepsy [33], which limit compensation for lost function. Furthermore, the rate of speech development is highly variable across individuals [10]. Given that these factors in uence how the brain develops, it may be quite di cult to use side of brain dysfunction (epileptogenic focus) as a reliable and valid predictor of laterality scores on dichotic tests, and of speech representation in children. Moreover, these variables may explain why some patients with unilateral speech representation on IAP testing in our sample, did not obtain signi cant ear advantages on the FDWT. The contribution of handedness as a factor determining FDWT scores was shown to be non-signi cant. A closer look at the data, however, reveals that handedness may play a subtle role in determining degree of asymmetry. In the group of patients with left-hemisphere speech, left-handers had a smaller REA than right-handers, as indexed by the di erence score as well l indices. A similar result can be found in Kimura's [20] sample, which received a free recall dichotic test. A possible explanation for Kimura's ndings is that left-handers attend to their environment in such a way that they identify more items from the left ear than from the right ear in the free recall task [6]. If this is the case, Bryden [6] suggested the handedness e ect would disappear if a dichotic test was used which better controlled for attentional factors. Our results show that the FDWT, which o ers better control over attentional factors [2], still produced a di erence in size of asymmetry score, with left-handers obtaining a somewhat smaller ear-advantage than right-handers. A surprising nding in the bilateral group was that left-handers showed larger asymmetry scores than right-handers. This result is likely due to the outlier with bilateral speech who achieved an extreme LEA score. When excluded, right-handers showed larger asymmetry scores than left-handers for the bilateral group. Although the di erence in asymmetry scores between the handedness groups did not reach signi cance for the left-hemisphere and bilateral group, it appears that the e ect of handedness on FDWT test scores may be subtle, and in uence the degree, but not the direction of speech representation Comparing the di erence score and l index to a population of adults As we have already discussed, the choice of laterality indices does not change the validity of the FDWT within a sample of children. However, when the aim is to compare FDWT data obtained from children, with those obtained from adults, the measure used to describe the data can lead to di erent conclusions about the development of ear asymmetries and lateralization of speech across the life span. Our data were recalculated to control stimulus dominance in the same manner as in Zatorre [37] to permit a valid comparison with his sample of adults. It should be noted however, that the following comparisons should be interpreted with caution, as there were di erences in lesion location, duration of epilepsy and method of data collection across the studies. In Zatorre's study, the mean di erence score was 17.78, and 21.5 for the left, bilateral and right hemisphere speech groups respectively; the mean l index was 1.60, 0.96 and 1.81 for the left, bilateral and right-hemisphere speech groups respectively. The rst notable distinction from the present study is that the children with unilateral speech representation in our sample had larger di erence scores than the adults in Zatorre's [37] sample (see Table 4). Secondly, l indices are smaller in magnitude for the population of children from the present study, as compared to the adult groups from Zatorre [37]. At rst glance it appears that the di erence score and l index provide