Language testing in adolescents with brain injury: A consideration of the CELF-3
Turkstra, Lyn S
As the incidence of traumatic brain injury (TBI) peaks in adolescence, many survivors receive rehabilitation and other post-acute services through the schools. Whether these services include speech-language therapy depends in part on the results of standardized language tests. Test scores may also be used to plan intervention and measure progress. However, few tests have been developed specifically with adolescents in mind, and none have been standardized on adolescents with TBI as a unique group. Thus, school clinicians with a need to evaluate communication ability in adolescent students with brain injury have a paucity of tools available to them.
Of necessity, clinicians may turn to instruments that have been established for other purposes (e.g., aphasia tests and tests designed to identify learning disabilities), including the Clinical Evaluation of Language FundamentalsThird Edition (CELF-3, Semel, Wiig, & Secord, 1995).
The experienced speech-language pathologists who participated in this study worked in sites across the United States and in Canada. These participants affirmed the widespread use of the CELF-3 in school settings to determine service eligibility for students with communication disorders, including students with TBI. Their colleagues working in rehabilitation settings stated that the Clinical Evaluation of Language Fundamentals-Revised (CELF-R, Semel, Wiig, & Secord, 1987) and CELF-3 were used commonly for adolescents with TBI who were returning to school (Susan Grey, personal communication, January 5, 1998; Carlene McBride, personal communication, January 12, 1998), in part because the tests are among the few that have been designed and well-standardized for this age group, and in part because the CELF tests are the standard for qualifying students for speech pathology services in most school districts. Subtests of the CELF tests have also been recommended for assessment in publications concerning pediatric TBI (e.g., Blosser & DePompei, 1994; Ylvisaker, 1998).
The CELF-3 has several strengths that support its use for adolescents with TBI. First, the standardization sample included a substantial number of adolescents, with 200 subjects at each of ages 13 through 16, and 250 subjects at ages 17 to 21. This feature is a particular asset given the paucity of published, standardized measures of adolescent communication.
Second, the CELF-3 was developed for purposes that are highly relevant to students with TBI. The purposes of the CELF-3, stated in the technical manual (Semel et al., 1995, p. 1), are to:
identify individuals with language disabilities,
provide diagnostic information and a description of the nature and degree of the disability,
identify areas of relative strength and weakness,
identify areas of extension testing to establish priorities for treatment, and
plan follow-up intervention.
These are important goals for clinicians working with students with brain injury.
A third positive feature of the CELF-3 is that it is norm-referenced, but appears to permit some criterionreferenced interpretation. In other words, the test is designed to permit the comparison of children and adolescents with language impairments with their peers (i.e., detect a difference), but it also purports to identify strengths and weaknesses in “skills that comprise the foundation of mature language use” (i.e., describe the problem) (Semel et al., 1995, p. 2). The latter is a criterion-referenced aspect of the test, and criterion-referenced testing has been suggested to help with intervention planning (McCauley, 1996).
Conversely, the CELF-3 has properties that potentially limit its application to subjects with TBI. First, it was not designed to be administered to individuals with acquired language impairments. This limits the test’s ability to identify individuals with language disabilities due to TBI, to provide diagnostic information concerning these students, and to describe the nature and degree of the disability.
Second, relative to identification, the standardization data suggest that the CELF-3 may not be a highly sensitive indicator of language disorder (LD). In the standardization sample, CELF-3 standard scores misclassified 43% of the students with LD as non-LD and 15% of non-LD students as LD.
Using these sensitivity and specificity figures, together with the prevalence of LD in the general population, it is possible to calculate the positive predictive value of the test (Fletcher, Fletcher, & Wagner, 1988).’ Given an estimated prevalence of LD of 5% in the United States, the positive predictive value of the CELF-3 in a general population would be 15%. In other words, as a screening test, the CELF-3 has a 15% success rate for discriminating individuals with LD. However, it is important to keep in mind that the sensitivity and specificity were not calculated on individuals with TBI, who may have very different communication profiles than those of individuals with developmental disorders (Obrzut & Hynd, 1987; Towne & Entwisle, 1993).
A third potential limitation of the CELF-3 is the relatively high variability in normal subjects’ standard scores across subtests. In the standardization sample, there was an average range of five standard score points between the lowest and highest subtest scores for each subject.2 This affects the test’s ability to identify strengths and weaknesses, areas of extension testing, and priorities for treatment, because a large difference in standard scores across subtests cannot be interpreted to reflect a clinically significant pattern of relative strengths and weaknesses. The test’s ability to achieve these purposes is also limited by the fact that, according to the information in the technical manual (Semel et al., 1995, p. 60), the CELF-3 is a onefactor test. That is to say, although it was designed to reveal strengths and weaknesses in several individual aspects of language (i.e., content, form, use, memory, and receptive and expressive language), instead, it measures one global aspect of language, referred to by the authors as “general language ability” (Semel et al., 1995, p. 60).
Finally, the test-retest reliability coefficients for individuals aged 13 (the oldest participants in the test-retest phase of development) are modest, ranging from .53 to .80. This is of particular relevance to individuals with TBI, whom one might expect to be even more variable than their uninjured peers (Stuss, Pogue, Buckle, & Bondar, 1994).
Given the potential strengths and limitations just discussed, it was considered worthwhile to further examine the validity of the CELF-3 for adolescents with TBI. Test validity is a function of the purpose for which the clinician employs a test and the inferences the clinician intends to draw from the results (Plante, 1996). Thus, the question posed here was whether the test was valid for the specific purpose of identifying and describing language impairments in adolescents with TBI, with a view to providing speechlanguage pathology services.
Validity was considered relative to the test’s stated purposes, that is:
Did the test identify language impairments in a group of adolescents with TBI, some of whom were identified by other measures (e.g., speech-language pathologists’ reports) as having language impairments, and some of whom were identified as having cognitive impairments that would influence language comprehension and use?
This question was addressed in two ways. First, the number of subjects having Total Language standard scores that were more than one standard deviation (SD) below the mean was determined. Second, the number of subjects with Receptive-Expressive cluster score discrepancies that were greater than 1 SD was compared to the number in the standardization sample with this discrepancy (15%) (Semel et al., 1995, p. 57).
Although the test authors did not recommend using Receptive-Expressive discrepancies to justify intervention, this analysis was included because individuals with TBI often have inconsistent performance across different cognitive-communication skills (Hartley, 1995; Prigatano, 1987; Stuss et al., 1994), and unusually large differences here might be clinically significant yet obscured in the Total Language score.
Did the test permit the identification of strengths and weaknesses and suggest areas for extension testing?
This question was difficult to address, given the factor structure and subtest variability in the standardization sample. The identification of strengths and weaknesses is possible only if the test has multiple components that can be distinguished reliably. However, an attempt was made using two approaches. First, the withinsubject variability across subtests in subjects with TBI was compared to that in the standardization sample, with the idea that within-subject variability beyond what would be expected in uninjured subjects might point to directions for future testing. Second, correlations among subtests were made for comparison with inter-subtest correlations in the standardization sample. If the subtests were less correlated in the TBI group than in the standardization sample, it would support the assertion that differences among subtest scores were meaningful. In contrast, higher correlations among subtests would suggest that the single factor measured by the test was exerting a greater influence on subjects with TBI than on their uninjured peers.
If the test was measuring a single factor in adolescents with TBI, was working memory that factor?
Working memory was suspected to be a major contributor to test performance in individuals with TBI for two reasons. First, several of the test procedures appeared to place a considerable load on both working memory storage and working memory processing. Second, verbal memory impairments have frequently been described after TBI (Goldstein, Levin, Boake, & Lohrey, 1990; Hartley, 1995; Ponsford & Sloan, 1995; Turkstra & Holland, 1998), and were documented in all of the present subjects. Such impairments have been shown to significantly influence standardized test performance in adolescents with TBI (Turkstra, 1998).
The question of whether memory influenced test performance was addressed by first establishing an a priori ranking of subtests according to their hypothesized demands on working memory storage and processing, measuring the working memory storage and processing ability of each subject using normreferenced working memory tests, then determining if the subtest scores were a function of the ranked subtest memory demand and the subjects’ memory ability. In other words, subjects with low scores on a standardized test of working memory storage capacity might have lower standard scores on the CELF-3 subtests that tapped storage most and better scores on subtests that required less storage. A similar finding was expected in regard to processing.
Eleven adolescents, five females and six males, who had sustained a TBI within 3 years preceding the test were included in the TBI group. Demographic information for the subjects is presented in Table 1. No subject had a history of pre-injury learning disability, although subject 4 had received remedial reading therapy in elementary school. All but two subjects (subjects 5 and 6) had qualified for and received special education services. Subject 5 had completed high school by correspondence and was enrolled in regular classes at a community college. Subject 6, previously an honor student, failed her returning semester and left school.
Five subjects (subjects 1, 4, 5, 8 and 9) had communication impairments that had been documented by speechlanguage pathologists using other tests and observations and had received speech-language therapy on discharge from acute care. For all of these subjects except subject 5, the therapy was aimed at the remediation of cognitive-communication impairments. For subject 5, therapy focused on dysarthria only.
Six subjects (subjects 2, 3, 6, 7, 10, and 11) showed impairments in verbal information processing that would be expected to affect communication performance, but were not referred for speech-language therapy when they were discharged from acute care. Subjects 2, 3, 6, and 10 had abnormal scores on neuropsychological tests of auditory memory, learning, or auditory verbal attention. Impairments in these domains would be expected to affect the ability to process complex verbal information, particularly in the classroom. All four of these subjects reported a substantial drop in grades and an increase in effort on returning to school. Subject 7 had impaired verbal fluency and confrontation naming on neuropsychological testing, which would be expected to affect performance on tasks requiring generative language. This subject required one-on-one support to complete language-based school tasks post-TBI.
Subject 11 reported post-TBI impairments in short-term memory, although neuropsychological test results were not available. This subject had requested and received special education services to return to school.
There were no regional, financial, or insurance factors that distinguished participants who were referred for speech-language therapy from those who were not. In fact, several of the participants who were not referred for speech-language therapy received physical therapy or psychotherapy. From the reports of these individuals, it appeared that speech-language therapy was not considered necessary at the time they were discharged, presumably because communication impairments were not detected. Nonetheless, the possibility that these individuals had communication disorders that merited intervention cannot be excluded.
Subtest descriptions. The CELF-3 is composed of six core subtests for individuals aged 9 and older. The six subtests are grouped into two clusters, Receptive and
Expressive. In the Receptive cluster, the Concepts and Directions subtest evaluates the “ability to interpret, recall, and execute oral comments of increasing length and complexity that contain concepts of logical operations” (Semel et al., 1995, p. I1). The Word Classes subtest evaluates the “ability to perceive associative relationships between words” (Semel et al., p. 15). The Semantic Relationships subtest assesses the “ability to interpret different semantic relationships” (Semel, et al., p. 19).
In the Expressive cluster, the Formulated Sentences subtest measures the “ability to formulate compound and complex sentences with given semantic and syntactic constraints” (Semel et al., 1995, p. 13). The Recalling Sentences subtest assesses the “ability to recall and reproduce sentence surface structures of varying length and syntactic complexity” (Semel et al., p. 16). The Sentence Assembly subtest evaluates the “ability to assemble syntactic structures into syntactically and semantically acceptable sentences” (Semel et al., p. 17).
Rankings by memory load. Subtests were ranked on the dimensions of storage and processing demands by 12 certified speech-language pathologists. Eleven of these raters had more than 1 year of experience with the CELF-3 or CELF-R. According to their report, two had administered the CELF-3 between 10 and 29 times, and nine had administered the CELF-3 or CELF-R more than 50 times. The remaining rater was a clinical professor with over 10 years of experience in language and memory research.
Participants were asked to evaluate each CELF-3 subtest on two working memory dimensions: storage (i.e., the amount of information to be held in working memory while performing the task) and processing (i.e., the demand for mental operations to be performed on that information). They were told that the storage requirement could be determined by factors such as the number of pieces of information to be held in short-term memory in order to respond to an item, and the processing requirement could be determined by factors such as the number of mental operations that needed to be performed in order to respond correctly. Examples were provided of tasks other than the CELF-3 subtests that had varying storage and processing demands. The median rank assigned for each subtest was used in subsequent calculations. Participants were also asked to share any comments regarding the CELF-3 item structure or administration procedure.
Working Memory Tests
Working memory storage capacity was measured with the Word Sequences subtest of the Detroit Tests of Learning Ability (DTLA, Hammill, 1991; Hammill & Bryant, 1991). The DTLA was normed on more than 2,500 individuals ages 6 to 17 in the United States. The Word Sequences subtest of the DTLA is designed to test shortterm auditory memory for words, and requires the subject to repeat, verbatim, a series of words in strings of varying length. The subtest was administered, basal and ceiling levels were determined in accordance with the test manual, and standard scores were calculated for each subject.
Working memory processing capacity was assessed using the Competing Language Processing Task (CLPT) for children (Gaulin & Campbell, 1994), a modified version of the Listening Span task that was developed by Daneman and Carpenter (1980) to measure verbal working memory ability in adults. Preliminary norms were available for individuals aged 6 to 12 (Gaulin & Campbell) and further normative data were collected by the author for individuals aged 13 to 21 (Turkstra & Noga, 1997).
This test requires subjects to listen to groups of short sentences, respond “true” or “false” to each sentence as it is presented, then, after all sentences in the group are presented, recall the final word of each sentence in the group. There are two items for each group size, from single sentences to groups of six. The test was administered according to the authors’ instructions, and the percentage of words recalled was calculated. Each subject’s CLPT score was then corrected for simple storage span, defined as the longest list length on the Word Sequences subtest at which the subject recalled 60% or more of the words in the correct order. CLPT items that required a longer simple span than the subject had demonstrated on the Word Sequences subtest were not included in the calculation of the subject’s final score. To illustrate, if a subject’s simple span was five words on the Word Sequences subtest, only those CLPT items with five or fewer sentences were scored.
Subjects were seated in a quiet room for the duration of testing. Each subject was first administered the two memory tests. Next, subjects were administered the six subtests of the CELF-3 in the order prescribed by the test authors (Concepts and Directions, Formulated Sentences, Word Classes, Recalling Sentences, Semantic Relationships, Sentence Assembly).
For the DTLA Word Sequences subtest, standard scores were obtained for each subject. For the CLPT, the percentage of items recalled was calculated. CELF-3 protocols were scored according to the instructions in the test manual and standard subtest and cluster scores were derived.
The question of identification of language disorders was addressed by calculating the number of study participants scoring more than 1 SD below the Total Language score mean of 100, the cutoff score for language disorders recommended by the test authors. This number was compared to the number of participants who were receiving speech-language services and the number of participants who had information processing disorders but were not receiving services. The number of subjects with a discrepancy of more than 1 SD (15 standard score points) between Receptive and Expressive standard scores was also calculated, and was compared to the percentage of subjects in the standardization sample with this discrepancy (15%).
The comparison of variability across subtest standard scores was accomplished by calculating the range of subtest scores for each subject (i.e., subtracting the lowest from the highest score) in both the TBI and the standardization groups; then, as the standardization data were not normally distributed, using a Mann Whitney Rank Sum Test to compare the range values between the two groups. The correlations among subtests for the TBI group were calculated using Pearson correlation coefficients, and displayed in a matrix along with the corresponding coefficients from the standardization sample.
The role of memory in subtest performance was evaluated using a Spearman rank order correlation of median subtest ranks, assigned by experienced raters, with the coefficients derived from the correlation of CELF-3 subtest scores for subjects in the TBI group with scores from the DTLA Word Sequences subtest (Hammill, 1991) (storage) and the CLPT (Gaulin & Campbell, 1994) (processing). The criterion alpha level for significance for all statistical tests was set at .05.
Subject subtest standard scores and cluster standard scores are displayed in Tables 2 and 3, respectively. Word Sequences standard scores and the percentage of items recalled on the CLPT for each subject are shown in Table 4.
Identification of Language Impairments
Total Language standard scores. The Total Language standard scores of subjects 1, 4, 8, 9, and 10 fell more than I SD below the mean for the standardization sample of the CELF-3. Four of these subjects had been identified for speech-language pathology services. The fifth subject identified by the CELF-3 had received no services at the time of participation in the study.
Subject 5 received services but was not identified by the CELF-3. However, this individual had received therapy for dysarthria only. Thus, the CELF-3 closely approximated the determination of eligibility for speech-language therapy by other means. No subject with verbal information processing impairments who was not referred for therapy was identified as disordered by the CELF-3.
Receptive-Expressive cluster standard scores. Only subject 4 had a discrepancy of more than 1 SD between Receptive and Expressive cluster scores. In the standardization sample, 15% of normally developing individuals had a standard score difference greater than I SD. Thus, the discrepancy in scores for subject 4 is within what might be expected from the test’s inherent variability.
Identification of Strengths and Weaknesses
Variability across subtests. The mean range of subtest scores was 5.58 for subjects in the CELF-3 standardization sample, SD = 2.03, and 5.73 for TBI subjects, SD = 1.79. The median values were 5 and 6, respectively. This difference was not significant, T= 6364.00, P = .61.
Correlations among subtests. The matrix of correlations in the CELF-3 standardization sample and the TBI group is presented in Table 5. Twelve of the fifteen correlations among subtests are as high in the TBI group as in the standardization sample, supporting the notion that the test is measuring a single skill. Although these correlations were not evaluated statistically, three intercorrelations appear lower in the TBI group. These three involve the Recalling Sentences subtest.
Working Memory Influence on CELF-3 Performance
The median ranks for each subtest on the dimensions of working memory storage and processing are shown in Table 6. Also shown are the correlations of subject subtest scores with scores on the Word Sequences subtest and the CLPT. Participating clinicians stated that subtest ranking was a difficult task, particularly in regard to working memory processing load. This is evidenced by the range in ranks for each subtest, and the variability in inter-rater agreement across subtests. The range of assigned ranks for processing load was 1-6 for Concepts and Directions, Word Classes, Semantic Relations, and Formulating Sentences; 2-6 for Sentence Assembly; and 1-3 for Recalling Sentences. The range of assigned ranks for storage load was 1-2 for
Concepts and Directions, 2-5 for Word Classes, 3-6 for Recalling Sentences, 1-2 for Sentence Assembly, 2-6 for Semantic Relations, and 1-4 for Formulating Sentences.
In regard to inter-rater agreement, the median ranks for storage, shown in Table 6, were assigned by eight of the 12 raters for Recalling Sentences, seven raters for Sentence Assembly and Formulating Sentences, six raters for Concepts and Directions, five raters for Word Classes, and four raters for Semantic Relationships. Median ranks for processing were assigned by six raters for Recalling Sentences; five raters for Word Classes; four raters for Semantic Relationships, Sentence Assembly, and Formulating Sentences; and only three raters for Concepts and Directions.
There was no significant relation of the subtest ranks to the correlation of CELF-3 subtest scores with scores on the Word Sequences subtest, Spearman correlation coefficient = .60, p = .24. There also was no significant relation of the subtest ranks to the correlation of subtest standard scores with scores on the CLPT, Spearman correlation coefficient = .77, p = .10. Contributing to the discrepancy between scores and ranks were the lower than expected correlation of the Semantic Relationships subtest with Word Sequences scores and the higher than expected correlation of the Formulated Sentences subtest with CLPT scores.
The CELF-3 classified all of the participants with TBI who were receiving language-based therapy services as LD. The one individual who had received therapy but who scored within the normal range on the CELF-3 had therapy only for dysarthria. One subject was classified by the CELF-3 but had not received speech therapy services. This individual had been injured during the summer break from school, and despite her own repeated complaints regarding her change in cognitive status, had received no therapy or accommodations upon returning to school. This individual volunteered for testing by the investigator after failing her returning semester in school, and ultimately was placed in a special needs classroom. Thus, this individual may have been a good candidate for speech therapy in school but was missed.
The six subjects without identified communication disorders but with verbal memory, learning, and fluency impairments were not identified as LD by the CELF-3. The CELF-3 does not purport to measure these aspects of communication function, so this finding is consistent with the stated goals of the test. However, all of these students experienced difficulty with the listening, reading, writing, and speaking demands of school, and required academic modifications and assistance. Thus, a score within the normal range on the CELF-3 does not eliminate the potential for handicap in aspects of communication.
Consideration of Receptive-Expressive cluster scores did not provide any additional information. Only one subject had a cluster score discrepancy that was greater than 1 SD, and this individual’s communication would have been considered disordered by his Total Language score alone.
CELF-3 scores did not permit the evaluation of strengths and weaknesses. Individual subtest scores varied across the six subtests, at times by as many as eight standard score units. Although these variations may appear meaningful, they were not pathological in reference to the standardization sample. This was an important finding given that several of the participating clinicians reported administering individual subtests for the evaluation of specific aspects of language, and the administration of individual subtests is recommended in a text on pediatric TBI (Blosser & DePompei, 1994). Clearly, the statistical properties and structure of the test, together with the results from individuals with TBI, strongly argue against the interpretation of results from individual subtests.
The moderate intercorrelations among subtest scores in the TBI group also suggest that the test is not measuring distinguishable strengths and weaknesses. Most of the intercorrelations were higher in the TBI group than in the standardization sample, although this was not tested statistically, supporting the assertion that the CELF-3 is measuring a single construct in individuals with TBI as it is in their uninjured peers. Exceptions were the intercorrelations between the Recalling Sentences subtest and three others: Word Classes, Sentence Assembly, and Semantic Relationships. All three of these intercorrelations were not as high in the TBI group.
One possible explanation relates to relative impairments in storage versus processing among participants with TBI. The Recalling Sentences task places a high load on verbatim storage, with minimal demand for processing. In contrast, the Word Classes, Sentence Assembly, and Semantic Relationships subtests all require mental comparisons and, therefore, have a relatively higher processing demand than the Recalling Sentences subtest. In the participants with TBI, processing and storage ability were not correlated. Hence, one might expect a low correlation between subjects’ scores on the processing-demanding tests and scores on the storage-demanding tests.
To illustrate, subject 3 had the highest processing score of the group, although his storage score was below the average for his age. His scores on the Word Classes and Semantic Relationships subtests were more than I SD above the mean, whereas his score on the Recalling Sentences subtest was more than I SD below the mean. By contrast, subject 11 had a storage score that was I SD above the mean. whereas his processing score was below the average for his age. Correspondingly, this subject’s Recalling Sentences score was above average, and his Semantic Relationships score was below average. This is a tentative interpretation, particularly in such a small group, and there are notable exceptions. However, it is interesting to consider in light of the hypothesized relationship between working memory and CELF-3 subtests.
Although the CELF-3 appears to measure a single construct, it was not clear that this construct was memory. Although the hypothesized memory load of the subtests was related to the actual correlation of memory scores with subtest scores, this relationship was not statistically significant. In part, this reflects the use of a conservative statistical measure. The wide range of ranks assigned by experienced clinicians, and their reported difficulty with assigning ranks, suggests that the training of raters also may have been a factor. Nonetheless, there were unexpected correlations.
Although raters considered the Formulating Sentences subtest to have a relatively low storage load, it was more positively correlated with storage test scores than were subtests such as Word Classes, which appeared to have more of a storage demand. In retrospect, the storage requirement for the Formulating Sentences subtest may have been underestimated because this subtest required subjects to retrieve information from long-term memory and hold it in mind while constructing a grammatically and pragmatically appropriate sentence for each stimulus item. Conversely, the contribution of memory storage ability to performance on the Semantic Relationships subtest was overestimated. This subtest was thought to be relatively storage-demanding because of the length of the questions and the fact that they were spoken rather than written. However, it is possible that the influence of memory load was outweighed by limitations in semantic knowledge and reasoning, so that storage accounted for only a small amount of variance in scores.
In the present subjects, the CELF-3 appeared to be a valid indicator of language disorders in the domains tested-that is, most of the subjects who scored in the abnormal range were receiving speech-language pathology services. It did not aid in the detection of cognitive impairments that would have an impact on communication ability, and this was not a stated goal of the test. The CELF-3 did not permit the identification of strengths and weaknesses, and there was evidence that a single factor was influencing test performance. Whether this factor was memory or some other cognitive function, such as attention, remains to be seen, although there was a trend for scores on subtests with a higher perceived memory load to be more correlated with memory measures. The contribution of memory to subtest performance is a subject for further study, and currently is being addressed in a study of normally developing adolescents.
This work was supported in part by a small grant from the Office of the Vice President for Research at the University of Arizona to the author. The author wishes to thank Erwin B. Montgomery, Jr. for his statistical assistance, and the following clinicians for their contribution to the rankings: Pelagie Beeson, Yvon Blais, Lila Carson, Marianne Casey, Pamela Dengos, Kristine Edahl, Linda Foskey, Maureen Johnston, Kathy Kennedy, Patricia McMahon, Elizabeth Mercer, Heather Miller, and Denise Simon.
Copyright American Speech-Language-Hearing Association Apr 1999
Provided by ProQuest Information and Learning Company. All rights Reserved.