Epilepsia, 49(1):22–32, 2008 doi: 10.1111/j.1528-1167.2007.01209.x
FULL-LENGTH ORIGINAL RESEARCH
Interictal and postictal language testing accurately lateralizes language dominant temporal lobe complex partial seizures ∗
Maya J. Ramirez, ∗ †Bruce K. Schefft, ∗ Steven R. Howe, ‡Hwa-Shain Yeh, and †Michael D. Privitera
Departments of ∗ Psychology and †Neurology, University of Cincinnati, and ‡Department of Neurosurgery, University of Cincinnati Medical Center, Cincinnati, Ohio, U.S.A.
SUMMARY
ficity, positive predictive value and negative predictive values were computed. Results: PILD, PostPE and InterPE production were equally effective and accurate in lateralizing DOM seizure onset. Patients with DOM TLE had a longer PILD and committed more PostPE and InterPE than those with nondominant (NDOM) TLE. Respective sensitivity and specificity values were as follows: PILD (84%, 86%), PostPE (94%, 64%), and InterPE (97%, 86%). No single predictor was significantly better but a combination model yielded enough incremental utility to collectively outperform each separate predictor model. Conclusions: Interictal language testing is as accurate as postictal language testing in predicting DOM lateralization of TLE. Clinicians should also attend to the quality of errors produced during interictal and postictal language testing. KEY WORDS: Paraphasic error, postictal language delay, Epilepsy, Confrontation naming, Confrontation naming, Postictal language testing.
Purpose: Noninvasive tests that accurately localize seizure onset provide great value in the presurgical evaluation of patients with intractable epilepsy. This study examined the diagnostic utility of three expressive language disturbances in lateralizing language-dominant (DOM) temporal lobe complex partial seizures: (1) the postictal language delay (PILD; time taken to correctly read a test phrase out loud immediately following seizures); (2) the production of postictal phonemic paraphasic errors (PostPE); and (3) interictal phonemic paraphasic errors (InterPE). Methods: All 60 subjects underwent inpatient video/EEG monitoring and had surgically confirmed temporal lobe epilepsy (TLE). We determined the presence and number of PostPE and, PILD times (in s) for 212 seizures, and InterPE on the Boston Naming Test (BNT). Each technique’s diagnostic usefulness was evaluated via logistic regression and ROC curve analysis. Sensitivity, speci-
Noninvasive methods that can localize or lateralize the epileptogenic focus are preferred in the evaluation of epilepsy surgery candidates. Prior studies have shown that postictal and interictal language disturbances can accu-
rately differentiate language dominant (DOM) from language nondominant (NDOM) temporal lobe complex partial seizures (Mayeux et al., 1980; Koerner and Laxer, 1988; Gabr et al., 1989; Privitera et al., 1991; Schefft et al., 2003; Busch et al., 2005; Fargo et al., 2005). Since the initial findings of Koerner and Laxer (1988) and Gabr et al. (1989), a handful of subsequent studies have further established the diagnostic value of postictal language disturbance in temporal lobe epilepsy (TLE) (Privitera et al., 1991; Devinsky et al., 1994; Fakhoury et al., 1994; Adam et al., 2000; Leutmezer, 2002; Loddenkemper and Kotagal, 2005). Specifically, Privitera et al. (1991) reported that having a postictal language delay (PILD; the amount of time that elapses from the end of an ictal discharge until a
Accepted May 9, 2007; Online Early publication July 21, 2007. This work was part of the first author’s Master’s Thesis, chaired by the second author in the Department of Psychology, University of Cincinnati, Cincinnati, Ohio. Address correspondence and reprint requests to Bruce K. Schefft, Department of Psychology, Dyer Hall, ML 376, University of Cincinnati, Cincinnati, OH, 45221-0376. E-mail:
[email protected] Blackwell Publishing, Inc. C 2008 International League Against Epilepsy
22
23 Interictal and Postictal Language Testing patient first correctly reads a test phrase aloud) greater than or equal to 60 s provided greater diagnostic accuracy of DOM TLE than MRI, interictal EEG, ictal EEG and overall neuropsychological assessment. Moreover, 33% of all NDOM TLE patients in their sample could read the test phrase during the seizure itself, whereas none of the DOM TLE patients was able to so. Studies conducted by Ficker et al. (2001) and Goldberg-Stern et al. (2004) further specified this. They found that NDOM temporal lobe seizures and DOM frontal lobe seizures only produced a prolonged PILD when seizure activity also spread to the DOM temporal lobe. Thus, prolonged PILD (≥60 s) occurrence may be specific to the DOM temporal lobe and not just indicative of DOM hemispheric involvement. Phonemic paraphasic errors are unique expressive language disturbances in which there is an incorrect phoneme or morpheme substitution (i.e., when an individual is shown a target picture of a “muzzle” and he or she instead calls it a “muzzler” (Kaplan et al., 1983)). Postictal paraphasic errors were first reported as a lateralizing sign by Jackson (1898) in the nineteenth century, but the introduction of video/EEG seizure recording has allowed more consistent detection (McKeever et al., 1983; Koerner and Laxer, 1988; Gabr et al., 1989). The diagnostic utility of postictal paraphasias is limited by the finding that most patients with dominant temporal lobe complex partial seizures do not speak spontaneously in the ictal and early postictal states. Moreover, the repetitive presentation of the test phrase in the postictal period markedly increases the elicitation of postictal paraphasic errors (Privitera et al., 1991). Longer PILD elicits more frequent presentation of the sign, which provides greater opportunity to commit postictal paraphasias. However, Privitera and colleagues (1991) observed that for their sample, phonemic paraphasic errors were the only dysphasic language errors produced during the postictal period, and that they occurred exclusively in patients with DOM temporal lobe seizure onset. Schefft et al. (2003) investigated the diagnostic utility of interictal paraphasic errors (InterPE) on two measures of confrontation naming ability, the Boston Naming Test (BNT) (Kaplan et al., 1983) and the Visual Naming Test (VNT) (Benton and Hamsher, 1983), which are sensitive to patient production of InterPE. Schefft and colleagues found that one InterPE was 6.8 times more likely in individuals with a left (DOM) seizure focus compared with individuals with a right (NDOM) focus. They also found a robust effect of global intelligence on BNT performance. The overall predictive ability of BNT performance and InterPE production to accurately lateralize seizure focus improved among patients with Full Scale IQs ≥ 90. Using the BNT, Fargo et al. (2005) analyzed the overall frequency and type of paraphasic naming errors produced by individuals with confirmed DOM and NDOM TLE as well as those without epilepsy (individuals with psychogenic non-
epileptic seizures). Other types of paraphasic errors (e.g., verbal-semantic paraphasic errors, circumlocutions, etc.) that were examined did not yield the degree of diagnostic differentiation between DOM and NDOM TLE patients as did the presence of at least one phonemic paraphasic error. Overall, Fargo et al. (2005) found that DOM TLE patients produced significantly more InterPE’s than both the NDOM and control groups. Although Schefft et al. (2003) and Fargo et al. (2005) suggested that the quality of expressive language errors that occur between seizures are diagnostically useful, neither study addressed the potential diagnostic value of postictal language disturbances (i.e., during the PILD). Privitera et al. (1991) demonstrated the presence and importance of the PILD as it relates to seizure origin in the DOM temporal lobe; however, they performed limited analysis of the method’s diagnostic utility (i.e., no ROC curve analyses, etc). Additionally, no follow-up analyses were performed examining the usefulness of the PILD as compared with other language measures. Moreover, Ficker et al. (2001) and Goldberg-Stern et al.’s (2004) findings that prolonged PILD can also occur in individuals with secondary seizure propagation cast doubt on the established diagnostic utility (i.e. specificity) of the PILD. Prolonged PILD due to secondary seizure generalization is clearly diagnostically different from prolonged PILD due to DOM temporal lobe seizure onset. Lastly, interictal language measures provide a distinct advantage over postictal measures as testing can occur at any time, whereas the PILD can only be measured once a seizure has been detected. In light of the limitations of previous studies, the present study used a larger, surgically confirmed sample to conduct a comprehensive analysis of the diagnostic utility of PILD and both InterPE and PostPE for DOM seizure lateralization. It was hypothesized that, as compared with individuals with NDOM TLE, individuals with DOM TLE would: (a) have a significantly longer PILD, (b) commit more PostPEs, and (c) commit more InterPEs on a confrontation naming task. This study also reevaluated the effect of global intelligence on the predictive accuracy of these three methods.
M ETHOD Participants We conducted a retrospective study of 60 participants (32 DOM TLE, 28 NDOM TLE) who were evaluated on the Epilepsy Monitoring Unit (EMU) at University Hospital in Cincinnati, OH. This research was fully approved by the University of Cincinnati Institutional Review Board and all participants provided informed written consent prior to participation at the time of their admission. The EMU is a medical inpatient facility where all patients undergo prolonged 24-hour video/EEG monitoring Epilepsia, 49(1):22–32, 2008 doi: 10.1111/j.1528-1167.2007.01209.x
24 M. J. Ramirez et al. to differentiate epileptic from nonepileptic seizures, and to localize the epileptogenic region for presurgical evaluation in patients with focal epilepsy. Individuals on the EMU also participate in comprehensive neuropsychological assessment to measure interictal neurocognitive functioning. This study included sequential patients who met all of the following criteria: (a) surgically confirmed diagnosis of either language dominant (DOM) or language nondominant (NDOM) temporal lobe epilepsy (TLE); patients were considered accurately localized if surgical outcome included at least a 90% reduction in seizure frequency (Engel Class I and II seizure outcome) at approximately 1 year of follow-up, (b) confirmed language dominance by intracarotid amobarbital test or intraoperative electrocorticography testing; (c) Wechsler Adult Intelligence Scale-III or WAIS-R Full Scale IQ of 70 and above; (d) normal MRI or structural brain lesions limited to mesial temporal areas such as mesial temporal sclerosis, low grade tumor or vascular malformation (e) age 17or older; (f) received the 60-item version of the Boston Naming Test (BNT); (g) minimum of eight years of education; (h) English as the native language; (i) no prior brain surgery; and (j) absence of a comorbid neurological or serious psychiatric disorder as screened via medical records and psychological self-report measures including the Minnesota Multiphasic Personality Inventory, second edition, and the Beck Depression Inventory, second edition (e.g., Multiple Sclerosis, Schizophrenia). Serious psychiatric disorder was defined as schizophrenia, autism, or severe bipolar disorder. Participants in this study were part of a larger sample of over 1,000 patients who were evaluated on the EMU between January 1994 and January 2005. Of these patients, approximately 40% (400) were diagnosed as having epileptic seizures and 186 met the inclusion criteria for a non-surgically confirmed TLE diagnosis (i.e., via videoEEG monitoring). Of these 186 TLE patients, 101 underwent temporal lobectomy and 80 (36 left, 44 right) met the seizure outcome criteria of at least a 90% reduction in seizure frequency. Three patients were excluded because of missing data. Postictal language testing could not be completed for 10 patients because of the following reasons: secondary generalized seizures (n = 4), no response to presentation of the sign during postictal language testing (n = 1), no seizures recorded during monitoring (n = 2), did not undergo postictal language testing (n = 2), or data was not available (n = 1). In addition patients with independent seizure onset from both temporal lobes were excluded (n = 1). All patients included in the study had examinations on the EMU that yielded all three language disturbance indices. In addition, individuals for whom language dominance could not be determined (n = 5) were excluded as were those with mixed language dominance (n = 6) according to intracarotid amobarbital test (IAT) results. Standard IAT procedures at this epilepsy center include inEpilepsia, 49(1):22–32, 2008 doi: 10.1111/j.1528-1167.2007.01209.x
jection of both hemispheres of the brain. One patient was included who was right hemisphere language dominant, according to IAT. This patient’s data were consistent with those patients who were left hemisphere language dominant. Language dominance for 11 patients was confirmed via intraoperative language mapping, whereas the rest underwent IAT. Language results for all IATs were conclusive. Application of these inclusion criteria resulted in a sample of 60 (32 DOM TLE, 28 NDOM TLE). Procedures Since 1988, all video/EEG monitoring reports at this center include descriptions of ictal and postictal behavior and document the duration of the PILD. All available video/EEG-monitoring reports were reviewed for PILD (in seconds) and presence of PostPE production. These language disturbances were assessed by the attending physicians (all fellowship trained epileptologists) who reviewed the video-monitoring footage to determine the PILD and presence of PostPE. Since most patients had multiple seizures during prolonged video/EEG monitoring, an average of all recorded PILD times was calculated for each patient and used in the analyses. Average paraphasic error production was not used as a number of people did not commit any language errors during testing and the range was limited from 0-6. Measures The following is a description of EMU procedures for assessing postictal language disturbance. The EEG technician entered the patient’s room upon identifying a seizure based on EEG and patient behavior changes. During the seizure (defined by ictal discharge), the EEG technician presented the patient with a test phrase from the Boston Diagnostic Aphasia Examination (Goodglass and Kaplan, 1972) to read aloud (“They heard him speak on the radio last night.”). Throughout the ictal and postictal period, the EEG technician continued to instruct the patient to read the sign until the patient correctly read it aloud. The time from the end of the EEG ictal discharge until the patient read the sign correctly was the PILD for that seizure. Both the PILD (in seconds) as well as the presence of PostPEs were evaluated with these procedures. In computing the number of PostPEs per patient the current study used the total number of errors committed after all recorded seizures for each patient. The Boston Naming Test (BNT) (Kaplan et al., 1983) is a well-known confrontation naming measure. During this test the patient is presented with a stimulus card and asked to name aloud the stimuli, line drawings of common objects, pictured on the card (Kaplan et al., 1983). The BNT was administered using standard clinical procedures. Medical charts were reviewed to obtain BNT scores, the presence and number of interPEs and language dominance information. Presence of paraphasic errors on the BNT was rated according to previously established criteria
25 Interictal and Postictal Language Testing (Goodglass and Kaplan, 1983; Ardila and Rosselli, 1993; Goodglass et al., 1997; Fargo et al., 2005). The current study only used the categorical distinction of phonemic paraphasic errors as empirical investigation has demonstrated very limited diagnostic utility for any other types of paraphasic errors in lateralizing TLE (i.e., verbal-semantic, circumlocution, etc.) (Fargo et al., 2005; Schefft et al., 2003). Phonemic paraphasic errors occur when there is an incorrect phoneme or morpheme substitution, such as saying the word “hap” when the target word is “hat.” These error ratings were reviewed by the staff clinical neuropsychologist (B.K.S.) to ensure accuracy. Analysis Correlations, independent-sample t-tests and χ 2 tests were conducted to assess between-group differences and relationships among the various demographic and neuropsychological variables. Logistic regression was used to model the probability of belonging to the DOM group (when compared with the NDOM group). Logistic regression models the logit, which is the natural logarithmic transformation of the odds that an observation is in one group as opposed to the other (i.e., the odds of DOM TLE group membership). The utility of each predictor is measured by the regression coefficient β, and exp(β) is the odds ratio, an estimate of the change in the odds of group membership for the target group as the predictor variable increases by one unit. Separate logistic regression analyses were performed for PILD and for the number and presence of InterPEs and PostPEs to evaluate the utility of each as a predictor of DOM seizure focus. Every point on a receiver-operator characteristic (ROC) curve represents a combination of a sensitivity and specificity value. Receiver-operator characteristic (ROC) curves were constructed for each model; the area under each curve (AUC, or c) was calculated as an index of the ability of each predictor to lateralize TLE. The software program MedCalc was used to assess the AUCs. To compare AUCs the standard error of the difference between the two AUCs must be calculated. MedCalc does this while employing Hanley and McNeil’s parametric method in which the AUC is equivalent to the Wilcoxon statistic and the standard error is computed using exponential approximation (Hanely and McNeil, 1983). Multiple logistic regression was also used to investigate whether combinations of these predictors might significantly improve diagnostic accuracy. Sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV), and ROC curve analysis were used to assess the diagnostic utility of each model. Because of the moderating effect of global intelligence (Full Scale Intelligence Quotient (FSIQ) ≥ 90) on BNT performance (Schefft et al., 2003), additional analyses were performed on all predictors using FSIQ ≥90 as a cut point. FSIQ ≥90 was used as these analyses were conducted to evaluate the utility of the predictors when intel-
lectual functioning is largely intact. FSIQ ≥90 is the standard clinical demarcation between intact and below average intellectual functioning.
R ESULTS Preliminary analyses Independent-sample t-tests and χ 2 tests of independence (Table 1) revealed no significant differences between the DOM and NDOM TLE groups in terms of age, education, (global) intelligence, sex, handedness or race. The NDOM group had a higher FSIQ mean and was slightly older than the DOM group, although the differences were not significant (α = .05). As expected, t-tests of neuropsychological variables revealed significant group differences for mean PILD, mean number of PostPE’s, and mean number of InterPEs (Table 1). Follow-up outcome for all participants was assessed between 1 and 10 years after surgery. Twohundred and twelve seizures were assessed overall with 98 in the DOM and 114 in the NDOM TLE group. In addition, PILD ranged from 0–1560 s (∼26 min). A correlation matrix of demographic and neuropsychological variables is presented in Table 2. The following significant relationships were found: (a) FSIQ and years of education were positively correlated; (b) FSIQ and InterPE production were negatively correlated; (c) negative correlations were found among the three language indices; (d) there was a positive correlation between PILD and phonemic error production; (e) all three language measures were positively related to DOM seizure onset. Logistic regression models Logistic regression was used to evaluate the predictive utility of the PILD to lateralize epileptogenic seizure focus to the DOM hemisphere. Separate regressions were also run to evaluate the utility of the number as well as the presence of InterPEs and PostPEs for the same task. Table 3 presents the individual parameter estimates for each model as well as the estimated odds ratios for each predictor. Overall, each predictor was able to accurately lateralize DOM seizures to the DOM temporal lobe. For PILD, the odds of belonging to the DOM TLE group were greater for higher PILD values. Therefore, the longer the PILD time the more likely that the patient was categorized as a member of the DOM TLE group. However, the relatively small odds ratio associated with the PILD is deceptive since PILD was measured in seconds. Thus, the odds ratio indexed the change in the odds associated with each 1-s increase in PILD. The PILD had the widest range of all the predictors spanning 0 to >1000 s. The average PILD for the DOM and NDOM TLE group were 288.76s and 37.82s, respectively. Because of the measurement of the PILD in seconds the odds ratio is small. Had the PILD been measured in minutes, it would have increased the size of the odds ratio. Therefore, the phenomenon is of clinical significance despite the small odds ratio. Epilepsia, 49(1):22–32, 2008 doi: 10.1111/j.1528-1167.2007.01209.x
26 M. J. Ramirez et al. Table 1. Comparisons of DOM (N = 32) and NDOM (N = 28) groups FSIQ Age Education
Sex Race Handedness
InterPE PILD PostPE
DOM NDOM DOM NDOM DOM NDOM
M
SD
90.78 97.11 32.72 37.57 12.56 13.43
13.21 13.69 10.99 11.22 2.09 2.36
DOM NDOM DOM NDOM DOM NDOM
DOM NDOM DOM NDOM DOM NDOM
t
df
p
−0.47
d
1.82
58
0.074
−0.44
1.69
58
0.096
−0.39
1.51
58
0.138
%
df
χ2
p
Odds ratio
56% female 46% female 91% white 96% white 88% right 82% right
1
0.58
0.448
0.674
1
0.81
0.367
0.358
1
0.34
0.562
0.657
M
SD
d
t
df
p
1.59 0.25 288.76 37.82 2.47 0.68
0.98 0.7 304.64 60.31 1.65 1.09
1.56
6.17
55.91
