- Email: [email protected]
Decision strategies for cerebral dysfunction IV: Determination of cerebral dysfunction
Archives o/Clinical Neumpsychology, Vol. 6. pp. 25%DO, F’rinted in the USA. All rights reserved.
1991 Copyright
08E7-6!77/9! $3.00 + .@I @ 1991 Nations! Acadm!, of Newc+ycholo~y
Decision Strategies for Cerebral Dysfunction IV: Determination
of Cerebral Dysfunction
Murry G. Mutchnick,’ Memphis
State University
Leslie K. Ross and Charles J. Long and University of Tennessee, Memt&tis
The present study examined the contribution of tests that compose the Im+knent Index with regard to their ability to predict brain im_oairment. The investigation further examines the ability of various other tests, chosen because of their observed usefulness in detecting brain impairment. Subjects composing ths brain damaged group (n = 298) were found to be impaired on both CT and EEG ezaminations. The pseudo-neurological control group (n = 193) consisted o-f patients referred,for testing yet all non-neuropsychological tests were normal. Discr5nkant analyses were conducted to determine the weightings of each test as well as to determine the overall prediction accuracies of three groupings of tests. These analyses demonstrate that tests, not comprising the Impcirment It&x, are ofpredictive value in determining dysfunction: Thursrone Word Fhsncy and a 00 ?ninute delayed recall from the WMS. Overall prediction accuracies of the various test groupings rangedfrom 73.52% to 78.02%. No statistically significant reduction of accuracy resulted with cross validation. All tests of statistically predictive value and all prediction results with their corresponding discriminant formulas are reported as well as a discussion of the ~pplicetion of thesefir,zlings.
In recent years the basic emphasis of neuropsychological testing has tended to shift from the determination of the presence of brain dysfunction toward the description of its consequences (Boll, 1986). While this has, in lxu+t,resulted from the medical profession’s increasing reliance on various scanning techniques for diagnosis, it also obtains from the increasing applications of neuropsychological test data to the development and implementation of treatment *Department of Psychology, Memphis State University, Memphis, TN, 35152. Requests for reprints should be sent to Charles J. Long, Ph.D., Department Psychology, Memphis State University, Memphis, TN 38152. 259
of
260
M. G. ~~ic~~ick,
L. K. Ross, and C. .I. Long
plans for cognitive retraining. Yet the determination of the presence and extent of brain dysfunction remains a vital component of medical diagnosis since the assessment of the functional consequences of brain dysfunction can almost always augment the findings of other neurological tests. Certainly where head injury is involved, there are often no equivalent measures of similar sensitivity (Long & Gouvier, 1984). Another developing area involves the dete~ination of dysfunction and its relationship to disability. The effective use of such neuropsychological test findings is exerting an increasing impact on court decisions (Incagnoli, 1985; Anchor, Rogers, Solomon, Barth, Peacock, & Martell, 1987; Malone, 1987). Therefore, regardless of new developments in medicine and new directions of neuropsychological practice, a major component of the neuropsychological evaluation will remain the detection of the presence, and the dete~ination of the extent, of brain dysfunction. The determination of brain dysfunction remains difficult because both lesion size and location produce a broad range of behavioral deficits. This variability renders single tests of limited value and suggests that effective use of neuropsychological test batteries must include not only an understanding of the differential sensitivity of individual tests but the pattern of performance across the test battery. Since individual subtests from the battery are sensitive to damage to different areas of the brain, the neuropsychologist must be concerned about the disto~ions occurring when data are combined in some fixed or otherwise prearranged manner to make such decisions. Reitan (1964) explained that patterns of data are often very subtle and when gross summaries are made from the data the subtle patterns are “washed out”. He further notes that individual classification of cases is a fundamental compliment to analysis of level of performance. While both procedures must be used with each patient, it is important to undertake a comprehensive level of analysis in reaching a clinical decision, While it is quite likely that no single formula can be derived that will work equally well for all cases, investigation of the various decision strategies will aid in our understanding and enhance our clinical decisions regarding brain dysfunction. The Impairment Index (Halstead, 1947) was not originally designed for clinical determination of subtle dysfunction. Rather, it was a psychobiological approach to making gross distinctions between individuals with different leveIs of brain function. Nevertheless, it has subsequently been employed and has remained the most frequently used and one of the most sensitive indicators of brain dysfunction. In some instances, it is easier to obtain the Impairment Index than it is to obtain other indices such as percent of Impaired Ratings within the Impaired Range and the Average Impairment Rating (Boucher et al., 1986). On the other hand, the Impairment Index’s methods inherently allow for the loss of potentially usefu1 data. Key predictor test scores obtained from an individual are compared to normative cut-off scores and given a one if over the cut-off or a zero if not. The degree to which they deviate from the
Determination of Cerebral Dysfunction
261
cut-off scores is ignored. A proportion is then computed, again ignoring the range of data. Using this method, little data is available to clarify the difference in severity represented by two different indices in the impaired range, as exemplified by an impairment index of .5 or .8. This weakness is discussed by Vega and Parsons (1967) who demonstrate that examining the mean T score (across tests) for each subject does offer some improvement in the classification of subjects, especially in reducing false positive errors, when compared to the Impairment Index. The Impairment Index also assumes that each subtest contributes the same amount of information to the overall prediction of brain damage, when, in fact, they are not equivalent (Goldstein & Shelly, 1972). The Impairment Index’s limitations regarding the range of scores would seem to negate its ability to make subtle distinctions between patients. It has also been demonstrated that reliance on the Impairment Index can pro duce error in detecting subtle lesions specific to the temporal lobes. Long and Hunter (1981) found that the Impairment Index produced less accurate predictions than those determined by “ ...careful analyses of the characteristic behavioral profile of the patients...“, and concluded that dysfunction “...cannot effectively be reduced to a simple cutoff score...” Using a clinical decision process including the patient’s test taking behavior, self-reported history, the Halstead battery, the WAIS, the Wechsler Memory Scale and the MMPI, 87% of the patients with left hemisphere lesions and 82% of the patients with right hemisphere lesions were correctly classified. It should be noted that these percentages are limited to subtle or reasonably focal lesions. These prediction accuracies are contrasted to 40% and 56% respectively as classified by the Impairment Index. These findings stimulate questions as to the ability of the Impairment Index to determine other subtle or focal lesions. In order to differentiate between subtle patient characteristics in the determination of brain damage one needs an understanding of tests that possess the greatest differentiating ability. Discriminant analysis techniques allow for differentially weighting of test scores in order to enhance predictive accuracy and such procedures have been shown to be equally effective as the Impairment Index in differentiating brain damaged from controls (Stuss & Trites, 1977). Using the neurological examination as criteria measurement, Wheeler, Burke and Reitan (1963) performed discriminant analyses in order to predict differences between controls and brain impaired patients. Results from 104 patients indicated accuracy in prediction at 90.7%. Variables for prediction consisted of eleven WechslerBellevue Adult Intelligence Scale subtests, eleven subtests from the Halstead battery and Parts A and B of the Trail Making Test. This 24 measure discriminant function was compared, and found to be superior, to the Halstead Impairment Index (correctly classifying 87.2%). More recently, discriminant analysis techniques were compared to the key approach (Russell, Neuringer, & Goldstein 1970) for prediction and localization of brain damage. Swiercinsky and Wamock (1977) compared the predic-
262
M. G. Mutchnick, L. K. Ross, and C. J. Long
tion accuracies of three sets of variables from a modified Halstead battery as analyzed by the key approach and discriminant analyses. Two hundred and sixty veterans were used as subjects with complete neuropsychological examinations as criteria for brain impairment. Total percentage of correct identification using the neuropsychological keys, the key variables in a discriminant function analysis, and a maximum set of variables in a discriminant function analysis were found to be 75.4, 69.6, and 72.7 respectively. The “maximum set” of variables includes the twelve variables used in the key approach in addition to the remaining subtests of the Wechsler Adult Intelligence Scale, age arKi education. Further comparisons of these methods were conducted with regard to their ability to make localization distinctions, but these findings are beyond the scope of this paper. The authors note that these results do not support one approach as being more accurate than the other in all situations. However, the key approach did yield 23.1 percent more false positives than the same variables used in a discriminant function. In discussing their findings, the authors suggest the possibility of using the key and discriminant functions in the clinical setting. They explain that, “When the maximum set of variables is included, additional information could be examined for improving predictions. The aim is to consider other variables that discriminate well between the groups that may not be included among the key variables.” Furthermore, discriminant analysis techniques allow for the differential weightings of test scores as opposed to the use of dichotomous scoring as used with the Impairment Index. There is growing evidence that many tests, not included in the Impairment Index, contribme to more accurate diagnosis. Matthews, Shaw, and Klove (1966) discussed tnat the ability of Halstead’s tests to differentiate patients with cerebral deficits from control patients has been effectively demonstrated. However, the ability of these tests to discriminate brain impaired individuals from patients presenting with symptoms suggestive of brain damage, but who do not have confirmation of damage, must be understood. It is this more sensitive discrimination, between neurologic and “pseudo-neurologic” patients, that must be made by the neuropsychologist in the clinical setting. Matthews et al. (1966) found that subjects with unequivocal brain damage could be discriminated wim statistical significance from pseudo-neurologic patients with fourteen of twenty-nine variables from Halstead’s battery, the Trail Making Test, and the MMPI (p < .Ol). Four additional variables from me same group of tests also discriminated between the two groups at significant levels (p < .05). Such findings suggest that an understanding of how individual tests contribute to the overall preaication accuracy of an index can enhance the index’s flexibiliry and provide rno;e nnformation to the overall clinical decision. Furthermore, the addition of test findings from tests not currently included in the impairment index need further investigation as to their utility. The purpose of this study will first be to investigate the contribution of
Determination of Cerebral Dysfunction
263
tests composing the Impairment Index in differentiating brain impaired from pseudo-neurological control patients. This will afford a better understanding of the differential weighting of each test. Secondly, this investigation will be extended to include other tests which, in the authors’ opinion, will be of value in enhancing the decision regarding the presence or absence of cerebral dysfunction. Finally, the development of predictive formulas providing differential weighting to each test will be investigated.
METHOD Subjects
The subjects in this study consisted of 491 patients selected from a larger pool of 2500 patients who were primarily non-institutionalized medical patients referred for neuropsychological testing by neurosurgeons and other physicians. Their actual performance on the Halstead-Reitan Test battery was not considered in group composition. Of the selected sample, 49.49% were male and 50.51% were female. The mean age of all subjects was 39.8 years and the mean education was 12.0 years. Patients found to be impaired on both computerized tomography (CT) and electroencephalography (EEG) tests composed the brain damaged group (n = 298). Patients referred for testing yet evidencing no such impairment composed the pseudo-neurological control group (n = 193). Patients with known head injury, dementia as diagnosed by their neurosurgeon, or documented psychaitric disturbance were excluded from the study. The data base analyzed in this study has been analyzed in other research, however, it has been significantly updated with new patient records. It should be noted that group composition did not consider neuropsychological performance and, in fact, the group is very difficult to accurately classify using such data and/or more global clinical procedures. Procedure All data from subjects receiving complete neuropsychological evaluation were stored into a microcomputer data base. Raw scores were then standardized by age correction (Long and Klein, 1987). Patients from both the impaired and the pseudo-neurological controls, as described above, were combined and divided, by odd and even numbered cases, into two heterogeneous groups of near equal size. See Table 1 for descriptive statistics of selected demographic characteristics for all patient groups. The Impairment Index. For comparison purposes, all subjects were analysed using the Impairment Index. As will be noted from the tests employed, the cat-
M. G. Mutchnick, L. K. Ross, and C. J. Long
264
Descriptive
TABLE 1 Statistics of Selected Demographic
Characteristics
for All
Patient Grouos Patient Group
Initial Cross-Validation Total
N
Education
Age
242 249 491
Mean
SD
Mean
SD
39.80 41.93 40.88
17.21 16.24 16.77
12.04 12.12 12.08
4.00 3.67 3.84
Age and education means for the initial and cross-validation groups were not found to be significantly different (p > .05).
egory test is not used and does not contribute to the analysis. Raw scores from the following tests were used in calculating the Impairment Index: Tapping right (TR), Speech Perception (SP), Seashore Rhythm (RY), Trail Making A (TA), and Trail Making B (TB), TPT - total time (TT), TPT - Memory (TM), and TPT location (TL). If a subject was determined to be impaired on a test, using cutoff scores, a 1 was assigned. The number of tests determined to be impaired were divided by the total number of tests. Therefore, the greater value of the ratio obtained the greater the assumed impairment. Analysis of Tests Contributing to the Impairment Index. Discriminant analyses were performed on one of the two heterogeneous groups using Wilks method (Norusis, 1986) for each of three sets of variables. The first set of variables consisted of a seven measure modified Impairment Index: Speech Perception (SP), Seashore Rhythm (RY), Tapping - Right (TR), Trail Making B (TB), TPT - Total Time (TT), TPT - Memory (TM) and TPT - Location (TL). Analysis of Modified Impairment Index. The second set of variables consisted of a nine measure modified Impairment Index: Speech Perception (SP), Seashore Rhythm (RY), Tapping - Right (TR), Trail Making A (TA), Trail
TABLE 2 Results of Classification of Subjects in the Initial and Cross-validation Samples Using Seven Variables From the Impairment Index
Number of Subjects
Actual Group Membership
86 N 111 93
Impaired
Controls
65 (67.7%) 10 (11.6%)
31 (32.3%) 76 (88.4%)
Initial
N 96
Predicted Group Membership
Impaired Controls
Cross-Validation Impaired Controls
69 (62.2%) 19 (20.4%)
42 (37.8%) 74 (79.6%)
265
Determination of Cerebral Dysfunction TABLE 3 Results of Classification of Subjects in the Initial and Cross-validation Samples Using a Nine Variable Modified Impairment Index Number of Subjects
N 96 86 N 111 93
Predicted Group Membership
Actual Group Membership
Impaired
Controls
66 (68.8%) 10 (11.6%)
30 (31.3%) 76 (88.4%)
Initial Impaired Controls
Cross-Validation Impaired Controls
70 (63.1%) 19 (20.4%)
41 (36.9%) 74 (79.6%)
Making B (TB), TPT - Total Time (TT), TPT - Memory (TM), TPT - Location (TL) and Greek Crosses (CR). Analysis of 18 Clinically Significant Variables. The third set of variables con-
sisted of eighteen measures: Speech Perception (SP), Seashore Rhythm (RY), Tapping- Right (TR), Trail Making A (TA), Trail Making B (TB), TPT Total Time (TT), TPT - Total Blocks (TX), TPT - Memory (TM), TPT Location (TL), Greek Crosses (CR), Thurstone Word Fluency (WF), Aphasia Screening (AS), WAIS-R Verbal IQ (VQ), WAIS-R Performance IQ (PQ), Wechsler Memory Scale MQ (MQ) and Delayed Memory Total Percent (DR). Delayed Memory Total Percent is the percentage of material from Logical Memory, Visual Reproduction, and Associate Learning of the Wechsler Memory Scale recalled after a one hour delay. Cross-validation Analyses. Cross-validation analyses were then carried out for each set of variables using the second heterogeneous group. These cross validation procedures were performed using the Fisher’s Linear Discriminant Functions (Norusis, 1986) determined by the initial discriminant analyses. Use of Fisher’s Linear Discriminant Functions allow for the interpretation of individual cases. The data is processed through both the “impaired” and “control” formulas. The largest discriminant score indicates group membership (Norusis, 1986). Application of the formulas derived in this study to incoming new data indicates accuracies that are observably consistent with the accuracies presented here. These new samples were statistically interpreted using data from the second group. Significance tests for independent proportions (Ferguson, 1981) were the n conducted between the initial and cross validation “correct” classifications. This allows one to determine if the proportion of subjects classified correctly significantly changes when “shrinkage” occurs. “Independent” refers to the fact that these proportions are from samples drawn independently, and not of the same sample.
266
M. G. Mutchnick, L. K. Ross, and C. J. Long TABLE 4 Results of Classification of Subjects in the Initial and Crossvalidation Samples Using Eighteen Variables
Number of Subjects
Actual Group Membership
N II
47 N 93 44
predicted Group Membership Impaired
Controls
56 (72.7%) 07 (14.9%)
21 (27.3%) 40 (85.1%)
Initial Impaired Controls
Cross-Validation Impaired Controls
66 (71.0%) 06 (13.6%)
27 (29.0%) 38 (86.4%)
RESULTS The Impairment Index. The Impairment Index correctly classified 73.52% of the total sample; 67.4% of the impaired group and 82.9% of the pseudo-neurological group. Analysis of Tests Contributing to the Impairment Index. The first set of analy-
ses used a seven measure modified Impairment Index. After the removal of nondiscriminating variables in the initial discriminant analysis, six variables remained. These variables correctly classified 77.47% of the sample. Wilks’ Lambda was equal to .6660 and was significant (p c .Ol). The predication rate using this set of variables are shown in Table 2. As may be seen in this table, a cross validation, using Fisher’s Linear Discriminant Functions on a new sample resulted in the correct classification of 70.10% of the subjects. Prediction formulas using Fisher’s Linear Discriminant Functions from seven measures of the Impairment Index yield the following formulas: Brain Impaired = -34.3814 + O.l0778*(SP) + O.l9465*(RY) + 0.27668*(TR) + O.O2332*(TB) + 0.25249*(TT) + 0.27634*(TL). Controls = -24.5980 + O.O7853*(SP) + O.l5894*(RY) + 0.25885*(TR) O.O0921*(TB) + 0.21472*(TT) + 0.25216*(TL). Analysis of Modified Impairment Index. The second set of analyses used a
nine measure modified Impairment Index. After the removal of nondiscriminating variables in the initial discriminant analysis, six variables remained. These variables correctly classified 78.02% of the sample. Wilks’ Lambda was equal to .6680 and was significant (p c .05). Predication rates using this set of variables are shown in Table 3. Discriminant Functions on a new sample resulted in the correct classification of 70.59% of the subjects. Predication Formulas Using Fisher’s Linear Discriminant Functions From Nine Measure Modified Impairment Index yielded the following formulas:
Determination of Cerebral Dysfunction
267
Brain Impaired = -33.9289 + O.l0481*(SP) + O.l9247*(RY) + 0.27470*(TR) + O.O1862*(TB) + 0.24722*(TT) + 0.27882*(TL). Controls = -24.2285 + O.O7637*(SP) + O.l5708*(RY) + 0.25658* (TR) O.O1250*(TB) + 0.20911 *(TT) + 0.25408*(TL). Analysis of I8 Clinically Significant Variables. The final set of analyses used
eighteen measures. After the removal of nondiscriminating variables in the initial discriminant analysis, five variables remained. These variables correctly classified 77.42% of the sample. Wilks’ Lambda was equal to .5662 and was significant (p < .Ol). Predication rates using this set of variables are shown in Table 4. A cross validation, using Fisher’s Linear Discriminant Functions on the new sample resulted in the correct classification of 75.91% of the subjects. Predication formulas using Fisher’s Linear Discriminant Functions from eighteen measures yielded the following formulas: Brain Impaired = -31.2271 + O.l7747*(RY) + O.l7327*(TT) + O.O9071*(TM) + 0.30882*(WF) + 0.23593*(DR). Controls = -20.3517 + O.l4068*(RY) + O.l2396*(TT) + O.l2592*(TM) + 0.25843*(WF) + O.l4806*(DR). All tests of differences between initial correct classifications and their cross validations for both “impaired” and “control” groups were determined not to be significant (p > .05). Although the reduction in classification accuracy was not statistically significant, the clinical significance for the reduction in accuracy remains to be determined.
DISCUSSION This study demonstrates that three groupings of variables analyzed with discriminant analyses are not substantially superior to the Impairment Index in terms of correctly classifying individuals as brain impaired or not. This may relate to the considerable overlap of the variables comprising the Impairment Index. However, tests not comprising the Impairment Index offer potential classification utility. Using the seven and nine variable groupings in discriminant analyses, it was determined that Speech Perception Test, Seashore Rhythm, Tapping - Right, Trail Making B, TPT - Total Time and TPT Location were the tests with the best predictive ability for both the impaired and control groups. Using the eighteen variable combination of tests in a discriminant analysis, it was found that Seashore Rhythm, TPT - Total Time, TPT - Memory, Thurstone Word Fluency, and Delayed Memory Total Percent were the tests with the best predictive ability for both the impaired and control groups. This analysis is interesting in that five tests from the original impairment index formula along with the Thurstone Word Fluency and Percent recall from the
268
M. G. Mutchnick.
L. K. Ross, and C. J. Long
WMS classify subjects with accuracy equivalent to the impairment index. These findings suggest that these tests are deserving of consideration in the evaluation of test battery results as well as in the design of test batteries in experimentation. One inherent weakness in the present and other studies investigating decision strategies regarding brain dysfunction is the assumption that one formula will effectively classify patients as impaired or not impaired. As previously mentioned, the pattern of performance changes as a function of etiology, location and size of the lesion. To base decision strategies upon one formula will necessarily place the emphasis on tests that are sensitive to more generalized impairment. There may well be tests which are sensitive to more select lesion locations. Performance on these tests would be impaired if the lesion involves this brain area but may not if the lesion is located elsewhere. While most neuropsychological tests are sensitive to broad areas due to their complexity, it is easy to see that more complex tests are more sensitive to the presence of generalized dysfunction and thus are more likely to be selected as general predictors of brain dysfunction. Quantitative analysis of data in research can alert the clinician to tests of importance allowing for better decision strategy. Delayed Memory Total Percent, as described above, for example, is shown to be a sensitive measure of brain impairment and should be considered by the neuropsychologist as a test of clinical value. The clinician should always consider quantifications as a tool for integrating information into a final clinical decision. A seemingly useful approach to analysing data begins with an analysis of level of performance. This may include the interpretation of group normative data, such as the Impairment Index and/or discriminant analysis data. Then the score profile of the patient must be interpreted. The present research has offered an alternative to commonly used tools, such as indices of impairment. By presenting a closer examination of the tests that comprise such indices as well as other tests, the clinician is provided with more specific information which can be used in a flexible, more interpretable way. If specific tests are understood in terms of their ability to differentiate groups of patients, the neuropsychologist has choices as to their integration. One alternative to the single formula method is to use several available formulas to determine brain dysfunction and then make the clinical decision based upon the overall pattern. An obvious alternative is to investigate predictive formulas for more focal lesions. The clinical judgment could be based on the pattern of such predictive formulas. Research involving discriminant analyses is commonly conducted using linear discriminant functions, also known as canonical discriminant function coefficients, to classify their subjects. A serious limitation of this method disallows for the further classification of additional patients for cross-validation or for the determination of an individual’s brain dysfunction in the clinical set-
Determination of Cerebral Dysfunction
269
ting. These linear discriminant functions (as routinely presented by SPSS/PC+ [Norusis, 19861) are not presented with all algorithms needed for their computation and thus, cannot be directly used for classification. Therefore, use of these discriminant functions for additional subjects would yield incorrect results. The Fisher’s Linear Disc~min~t Functions (Norusis, 1986) do provide ali information necessary for further application and are identical to the discriminant functions with regard to accuracy of prediction. The present study utilizes Fisher’s Linear Discriminant Functions for cross-validation in an attempt to rectify the errors of previous research using discriminant statistical manipulations. Finally, discriminant analysis may not provide a convenient alternative but such procedures can clarify the decision strategies involved in the determination of brain dysfunction. This research suggests that there is no single good predictive formula. Various data analysis methods can at best only synthesize the data and provide the clinician with better information upon which to make a judgment. Greater gains may be obtained in future research that focuses upon predictive formulas sensitive to Iateralized or more specific focal lesions.
REFERENCES Anchor, K. N., Rogers, 3. P., Solomon, G. S., Barth, I. T., Peacock, C., & Martell, D. A. (1987). Fu~men~ls of disability dete~ination and mhabili~tion: A higher ground for the applied neurobehavioml sciences. Defense Law .lourrral, 36.363-402. Boll, T. J. (1986). Nontraditional and threshold considerations in neuropsychological assessment. In S. B. Filskov & T. J. Boll (Eds.), Iiundbook of clinical neuropsychology (pp. 103-120). New York: John Wiley & Sons. Boucher, M. L., Dewan, M. J., Donnelly, M. P., Pandurangi, A. K., Bartell, K., Diamond, T., & Major, L. F. (1986). Relative utility of three indices of neuropsychologic~ irn~~ent in a young, chronic schizophrenic population. The Journal of Nervous and Mental Disease, 174, 44116. Ferguson, G. A. (1981). Statisticalanalysis in psychology and education. New York: McGraw Hill. Goldstein, G., & Shelly, C. H. (1972). Statistical and normative studies of the Halstead Neuropsychological Test Battery relevant to a neuropsychiatric hospital setting. Perceptual and Motor Skills, 34, W-620. Halstead, W. C. (1947). Bruin and intelligence. Chicago: University of Chicago Press. Incagnoli, T. I. (1985). How much brain damage? The clinical neuropsychologist. Trial Lawyer’s Quarterly, 17,30-34. Long, C. J. (1985). Neuropsychology in private practice: Its changing focus. Psychotherapy in Private Practice, 3,45-S.
Long, C. J., & Gouvier, W. D. (1982). Neuropsychological assessment of outcome following closed head injury. In R. N. Malatesha & L. C. Ha&age (Eds.), Ne~rop~c~togy and cognition, Volmne Ii. The Hagne: Mattinus Nijhoff. Long, C. J., L Hunter, S. E. (1981). Analysis of temporal cortex dysfunction by neuropsychological techniques. Clinical Neuropsychology, 3.16-24. Long, C. I., & Klein, K. (1987, October). Analysis of age effects upon neuropsychological functioning. Poster presented at seventh annual meeting of the National Academy of Neuropsychologists. Chicago, IL.
270
M. G. Mutchnick, L. K. Ross, and C. J. Long
Personal Injury Review, 777-806. Matthews, C. G., Shaw, D. J., & Klove, H. (1966). Psychological test performances in neurologic
Malone, T. W. (1987). Direct examination of a neuropsychologist.
and “pseudoneurologic” subjects. Cortex, 2.244-253. Norusis, M. J. (1986). Advanced Stafisfics: SPSSIPC+ [Computer Program and Manual]. Chicago: SPSS, Inc. Reitan, R. (1964). Psychological deficits resulting from cerebral lesions in men. In J. M. Warren & K. Akert (Eds.), The frontal grundular cortex and behaviors (pp. 295-312). New York: McGraw-Hill. Reitan, R. M. (1966). A research programme on the psychological effects of brain lesions in human beings. In N. R. Ellis (Ed.), International review of research in mental retardation, Vol. 1. New York: Academic Press. Russell, E. W., Neuringer, C., & Goldstein, Cl. (1970). Assessment of brain damage - A neuropsychological key approach. New York: Wiley-Interscience. Stanczak, D. E., (1984). Assessment of level of performance in neuropsychological diagnosis: A comparison of the discriminant validity of alternative statistical methods. Unpublished doctoral dissertation. Memphis State University. Stuss, D. T., & Trites, R. L. (1977). Classification of neurological status using multiple discriminant function analysis of neuropsychological test scores. Journal of Consulting and Clinical Psychology, 45,145.
Swiercinsky, D. P., & Warnock, J. K. (1977). Comparison of the neuropsychological key and discriminant analysis approaches in predicting cerebral damage and localization. Journal of Consulting and Clinical Psychology, 45808-814.
Vega, A., Jr., & Parsons, 0. A. (1967). Cross-validation of the Halstead-Reitan tests for brain damage. Journal of Consulting Psychology, 31,619-625. Wheeler, L., & Reitan, R. M. (1963). Discriminant functions applied to the problem of predicting cerebral damage from behavioral tests: A cross-validation study. Perceptual and Motor Skills, 16.681-701. Wheeler, L., Burke, C. J., & Reitan, R. M. (1963). An application of discriminant functions to the problem of predicting brain damage using behavioral variables. Perceptual and Motor Skills, 16.417-440.
1991 Copyright
08E7-6!77/9! $3.00 + .@I @ 1991 Nations! Acadm!, of Newc+ycholo~y
Decision Strategies for Cerebral Dysfunction IV: Determination
of Cerebral Dysfunction
Murry G. Mutchnick,’ Memphis
State University
Leslie K. Ross and Charles J. Long and University of Tennessee, Memt&tis
The present study examined the contribution of tests that compose the Im+knent Index with regard to their ability to predict brain im_oairment. The investigation further examines the ability of various other tests, chosen because of their observed usefulness in detecting brain impairment. Subjects composing ths brain damaged group (n = 298) were found to be impaired on both CT and EEG ezaminations. The pseudo-neurological control group (n = 193) consisted o-f patients referred,for testing yet all non-neuropsychological tests were normal. Discr5nkant analyses were conducted to determine the weightings of each test as well as to determine the overall prediction accuracies of three groupings of tests. These analyses demonstrate that tests, not comprising the Impcirment It&x, are ofpredictive value in determining dysfunction: Thursrone Word Fhsncy and a 00 ?ninute delayed recall from the WMS. Overall prediction accuracies of the various test groupings rangedfrom 73.52% to 78.02%. No statistically significant reduction of accuracy resulted with cross validation. All tests of statistically predictive value and all prediction results with their corresponding discriminant formulas are reported as well as a discussion of the ~pplicetion of thesefir,zlings.
In recent years the basic emphasis of neuropsychological testing has tended to shift from the determination of the presence of brain dysfunction toward the description of its consequences (Boll, 1986). While this has, in lxu+t,resulted from the medical profession’s increasing reliance on various scanning techniques for diagnosis, it also obtains from the increasing applications of neuropsychological test data to the development and implementation of treatment *Department of Psychology, Memphis State University, Memphis, TN, 35152. Requests for reprints should be sent to Charles J. Long, Ph.D., Department Psychology, Memphis State University, Memphis, TN 38152. 259
of
260
M. G. ~~ic~~ick,
L. K. Ross, and C. .I. Long
plans for cognitive retraining. Yet the determination of the presence and extent of brain dysfunction remains a vital component of medical diagnosis since the assessment of the functional consequences of brain dysfunction can almost always augment the findings of other neurological tests. Certainly where head injury is involved, there are often no equivalent measures of similar sensitivity (Long & Gouvier, 1984). Another developing area involves the dete~ination of dysfunction and its relationship to disability. The effective use of such neuropsychological test findings is exerting an increasing impact on court decisions (Incagnoli, 1985; Anchor, Rogers, Solomon, Barth, Peacock, & Martell, 1987; Malone, 1987). Therefore, regardless of new developments in medicine and new directions of neuropsychological practice, a major component of the neuropsychological evaluation will remain the detection of the presence, and the dete~ination of the extent, of brain dysfunction. The determination of brain dysfunction remains difficult because both lesion size and location produce a broad range of behavioral deficits. This variability renders single tests of limited value and suggests that effective use of neuropsychological test batteries must include not only an understanding of the differential sensitivity of individual tests but the pattern of performance across the test battery. Since individual subtests from the battery are sensitive to damage to different areas of the brain, the neuropsychologist must be concerned about the disto~ions occurring when data are combined in some fixed or otherwise prearranged manner to make such decisions. Reitan (1964) explained that patterns of data are often very subtle and when gross summaries are made from the data the subtle patterns are “washed out”. He further notes that individual classification of cases is a fundamental compliment to analysis of level of performance. While both procedures must be used with each patient, it is important to undertake a comprehensive level of analysis in reaching a clinical decision, While it is quite likely that no single formula can be derived that will work equally well for all cases, investigation of the various decision strategies will aid in our understanding and enhance our clinical decisions regarding brain dysfunction. The Impairment Index (Halstead, 1947) was not originally designed for clinical determination of subtle dysfunction. Rather, it was a psychobiological approach to making gross distinctions between individuals with different leveIs of brain function. Nevertheless, it has subsequently been employed and has remained the most frequently used and one of the most sensitive indicators of brain dysfunction. In some instances, it is easier to obtain the Impairment Index than it is to obtain other indices such as percent of Impaired Ratings within the Impaired Range and the Average Impairment Rating (Boucher et al., 1986). On the other hand, the Impairment Index’s methods inherently allow for the loss of potentially usefu1 data. Key predictor test scores obtained from an individual are compared to normative cut-off scores and given a one if over the cut-off or a zero if not. The degree to which they deviate from the
Determination of Cerebral Dysfunction
261
cut-off scores is ignored. A proportion is then computed, again ignoring the range of data. Using this method, little data is available to clarify the difference in severity represented by two different indices in the impaired range, as exemplified by an impairment index of .5 or .8. This weakness is discussed by Vega and Parsons (1967) who demonstrate that examining the mean T score (across tests) for each subject does offer some improvement in the classification of subjects, especially in reducing false positive errors, when compared to the Impairment Index. The Impairment Index also assumes that each subtest contributes the same amount of information to the overall prediction of brain damage, when, in fact, they are not equivalent (Goldstein & Shelly, 1972). The Impairment Index’s limitations regarding the range of scores would seem to negate its ability to make subtle distinctions between patients. It has also been demonstrated that reliance on the Impairment Index can pro duce error in detecting subtle lesions specific to the temporal lobes. Long and Hunter (1981) found that the Impairment Index produced less accurate predictions than those determined by “ ...careful analyses of the characteristic behavioral profile of the patients...“, and concluded that dysfunction “...cannot effectively be reduced to a simple cutoff score...” Using a clinical decision process including the patient’s test taking behavior, self-reported history, the Halstead battery, the WAIS, the Wechsler Memory Scale and the MMPI, 87% of the patients with left hemisphere lesions and 82% of the patients with right hemisphere lesions were correctly classified. It should be noted that these percentages are limited to subtle or reasonably focal lesions. These prediction accuracies are contrasted to 40% and 56% respectively as classified by the Impairment Index. These findings stimulate questions as to the ability of the Impairment Index to determine other subtle or focal lesions. In order to differentiate between subtle patient characteristics in the determination of brain damage one needs an understanding of tests that possess the greatest differentiating ability. Discriminant analysis techniques allow for differentially weighting of test scores in order to enhance predictive accuracy and such procedures have been shown to be equally effective as the Impairment Index in differentiating brain damaged from controls (Stuss & Trites, 1977). Using the neurological examination as criteria measurement, Wheeler, Burke and Reitan (1963) performed discriminant analyses in order to predict differences between controls and brain impaired patients. Results from 104 patients indicated accuracy in prediction at 90.7%. Variables for prediction consisted of eleven WechslerBellevue Adult Intelligence Scale subtests, eleven subtests from the Halstead battery and Parts A and B of the Trail Making Test. This 24 measure discriminant function was compared, and found to be superior, to the Halstead Impairment Index (correctly classifying 87.2%). More recently, discriminant analysis techniques were compared to the key approach (Russell, Neuringer, & Goldstein 1970) for prediction and localization of brain damage. Swiercinsky and Wamock (1977) compared the predic-
262
M. G. Mutchnick, L. K. Ross, and C. J. Long
tion accuracies of three sets of variables from a modified Halstead battery as analyzed by the key approach and discriminant analyses. Two hundred and sixty veterans were used as subjects with complete neuropsychological examinations as criteria for brain impairment. Total percentage of correct identification using the neuropsychological keys, the key variables in a discriminant function analysis, and a maximum set of variables in a discriminant function analysis were found to be 75.4, 69.6, and 72.7 respectively. The “maximum set” of variables includes the twelve variables used in the key approach in addition to the remaining subtests of the Wechsler Adult Intelligence Scale, age arKi education. Further comparisons of these methods were conducted with regard to their ability to make localization distinctions, but these findings are beyond the scope of this paper. The authors note that these results do not support one approach as being more accurate than the other in all situations. However, the key approach did yield 23.1 percent more false positives than the same variables used in a discriminant function. In discussing their findings, the authors suggest the possibility of using the key and discriminant functions in the clinical setting. They explain that, “When the maximum set of variables is included, additional information could be examined for improving predictions. The aim is to consider other variables that discriminate well between the groups that may not be included among the key variables.” Furthermore, discriminant analysis techniques allow for the differential weightings of test scores as opposed to the use of dichotomous scoring as used with the Impairment Index. There is growing evidence that many tests, not included in the Impairment Index, contribme to more accurate diagnosis. Matthews, Shaw, and Klove (1966) discussed tnat the ability of Halstead’s tests to differentiate patients with cerebral deficits from control patients has been effectively demonstrated. However, the ability of these tests to discriminate brain impaired individuals from patients presenting with symptoms suggestive of brain damage, but who do not have confirmation of damage, must be understood. It is this more sensitive discrimination, between neurologic and “pseudo-neurologic” patients, that must be made by the neuropsychologist in the clinical setting. Matthews et al. (1966) found that subjects with unequivocal brain damage could be discriminated wim statistical significance from pseudo-neurologic patients with fourteen of twenty-nine variables from Halstead’s battery, the Trail Making Test, and the MMPI (p < .Ol). Four additional variables from me same group of tests also discriminated between the two groups at significant levels (p < .05). Such findings suggest that an understanding of how individual tests contribute to the overall preaication accuracy of an index can enhance the index’s flexibiliry and provide rno;e nnformation to the overall clinical decision. Furthermore, the addition of test findings from tests not currently included in the impairment index need further investigation as to their utility. The purpose of this study will first be to investigate the contribution of
Determination of Cerebral Dysfunction
263
tests composing the Impairment Index in differentiating brain impaired from pseudo-neurological control patients. This will afford a better understanding of the differential weighting of each test. Secondly, this investigation will be extended to include other tests which, in the authors’ opinion, will be of value in enhancing the decision regarding the presence or absence of cerebral dysfunction. Finally, the development of predictive formulas providing differential weighting to each test will be investigated.
METHOD Subjects
The subjects in this study consisted of 491 patients selected from a larger pool of 2500 patients who were primarily non-institutionalized medical patients referred for neuropsychological testing by neurosurgeons and other physicians. Their actual performance on the Halstead-Reitan Test battery was not considered in group composition. Of the selected sample, 49.49% were male and 50.51% were female. The mean age of all subjects was 39.8 years and the mean education was 12.0 years. Patients found to be impaired on both computerized tomography (CT) and electroencephalography (EEG) tests composed the brain damaged group (n = 298). Patients referred for testing yet evidencing no such impairment composed the pseudo-neurological control group (n = 193). Patients with known head injury, dementia as diagnosed by their neurosurgeon, or documented psychaitric disturbance were excluded from the study. The data base analyzed in this study has been analyzed in other research, however, it has been significantly updated with new patient records. It should be noted that group composition did not consider neuropsychological performance and, in fact, the group is very difficult to accurately classify using such data and/or more global clinical procedures. Procedure All data from subjects receiving complete neuropsychological evaluation were stored into a microcomputer data base. Raw scores were then standardized by age correction (Long and Klein, 1987). Patients from both the impaired and the pseudo-neurological controls, as described above, were combined and divided, by odd and even numbered cases, into two heterogeneous groups of near equal size. See Table 1 for descriptive statistics of selected demographic characteristics for all patient groups. The Impairment Index. For comparison purposes, all subjects were analysed using the Impairment Index. As will be noted from the tests employed, the cat-
M. G. Mutchnick, L. K. Ross, and C. J. Long
264
Descriptive
TABLE 1 Statistics of Selected Demographic
Characteristics
for All
Patient Grouos Patient Group
Initial Cross-Validation Total
N
Education
Age
242 249 491
Mean
SD
Mean
SD
39.80 41.93 40.88
17.21 16.24 16.77
12.04 12.12 12.08
4.00 3.67 3.84
Age and education means for the initial and cross-validation groups were not found to be significantly different (p > .05).
egory test is not used and does not contribute to the analysis. Raw scores from the following tests were used in calculating the Impairment Index: Tapping right (TR), Speech Perception (SP), Seashore Rhythm (RY), Trail Making A (TA), and Trail Making B (TB), TPT - total time (TT), TPT - Memory (TM), and TPT location (TL). If a subject was determined to be impaired on a test, using cutoff scores, a 1 was assigned. The number of tests determined to be impaired were divided by the total number of tests. Therefore, the greater value of the ratio obtained the greater the assumed impairment. Analysis of Tests Contributing to the Impairment Index. Discriminant analyses were performed on one of the two heterogeneous groups using Wilks method (Norusis, 1986) for each of three sets of variables. The first set of variables consisted of a seven measure modified Impairment Index: Speech Perception (SP), Seashore Rhythm (RY), Tapping - Right (TR), Trail Making B (TB), TPT - Total Time (TT), TPT - Memory (TM) and TPT - Location (TL). Analysis of Modified Impairment Index. The second set of variables consisted of a nine measure modified Impairment Index: Speech Perception (SP), Seashore Rhythm (RY), Tapping - Right (TR), Trail Making A (TA), Trail
TABLE 2 Results of Classification of Subjects in the Initial and Cross-validation Samples Using Seven Variables From the Impairment Index
Number of Subjects
Actual Group Membership
86 N 111 93
Impaired
Controls
65 (67.7%) 10 (11.6%)
31 (32.3%) 76 (88.4%)
Initial
N 96
Predicted Group Membership
Impaired Controls
Cross-Validation Impaired Controls
69 (62.2%) 19 (20.4%)
42 (37.8%) 74 (79.6%)
265
Determination of Cerebral Dysfunction TABLE 3 Results of Classification of Subjects in the Initial and Cross-validation Samples Using a Nine Variable Modified Impairment Index Number of Subjects
N 96 86 N 111 93
Predicted Group Membership
Actual Group Membership
Impaired
Controls
66 (68.8%) 10 (11.6%)
30 (31.3%) 76 (88.4%)
Initial Impaired Controls
Cross-Validation Impaired Controls
70 (63.1%) 19 (20.4%)
41 (36.9%) 74 (79.6%)
Making B (TB), TPT - Total Time (TT), TPT - Memory (TM), TPT - Location (TL) and Greek Crosses (CR). Analysis of 18 Clinically Significant Variables. The third set of variables con-
sisted of eighteen measures: Speech Perception (SP), Seashore Rhythm (RY), Tapping- Right (TR), Trail Making A (TA), Trail Making B (TB), TPT Total Time (TT), TPT - Total Blocks (TX), TPT - Memory (TM), TPT Location (TL), Greek Crosses (CR), Thurstone Word Fluency (WF), Aphasia Screening (AS), WAIS-R Verbal IQ (VQ), WAIS-R Performance IQ (PQ), Wechsler Memory Scale MQ (MQ) and Delayed Memory Total Percent (DR). Delayed Memory Total Percent is the percentage of material from Logical Memory, Visual Reproduction, and Associate Learning of the Wechsler Memory Scale recalled after a one hour delay. Cross-validation Analyses. Cross-validation analyses were then carried out for each set of variables using the second heterogeneous group. These cross validation procedures were performed using the Fisher’s Linear Discriminant Functions (Norusis, 1986) determined by the initial discriminant analyses. Use of Fisher’s Linear Discriminant Functions allow for the interpretation of individual cases. The data is processed through both the “impaired” and “control” formulas. The largest discriminant score indicates group membership (Norusis, 1986). Application of the formulas derived in this study to incoming new data indicates accuracies that are observably consistent with the accuracies presented here. These new samples were statistically interpreted using data from the second group. Significance tests for independent proportions (Ferguson, 1981) were the n conducted between the initial and cross validation “correct” classifications. This allows one to determine if the proportion of subjects classified correctly significantly changes when “shrinkage” occurs. “Independent” refers to the fact that these proportions are from samples drawn independently, and not of the same sample.
266
M. G. Mutchnick, L. K. Ross, and C. J. Long TABLE 4 Results of Classification of Subjects in the Initial and Crossvalidation Samples Using Eighteen Variables
Number of Subjects
Actual Group Membership
N II
47 N 93 44
predicted Group Membership Impaired
Controls
56 (72.7%) 07 (14.9%)
21 (27.3%) 40 (85.1%)
Initial Impaired Controls
Cross-Validation Impaired Controls
66 (71.0%) 06 (13.6%)
27 (29.0%) 38 (86.4%)
RESULTS The Impairment Index. The Impairment Index correctly classified 73.52% of the total sample; 67.4% of the impaired group and 82.9% of the pseudo-neurological group. Analysis of Tests Contributing to the Impairment Index. The first set of analy-
ses used a seven measure modified Impairment Index. After the removal of nondiscriminating variables in the initial discriminant analysis, six variables remained. These variables correctly classified 77.47% of the sample. Wilks’ Lambda was equal to .6660 and was significant (p c .Ol). The predication rate using this set of variables are shown in Table 2. As may be seen in this table, a cross validation, using Fisher’s Linear Discriminant Functions on a new sample resulted in the correct classification of 70.10% of the subjects. Prediction formulas using Fisher’s Linear Discriminant Functions from seven measures of the Impairment Index yield the following formulas: Brain Impaired = -34.3814 + O.l0778*(SP) + O.l9465*(RY) + 0.27668*(TR) + O.O2332*(TB) + 0.25249*(TT) + 0.27634*(TL). Controls = -24.5980 + O.O7853*(SP) + O.l5894*(RY) + 0.25885*(TR) O.O0921*(TB) + 0.21472*(TT) + 0.25216*(TL). Analysis of Modified Impairment Index. The second set of analyses used a
nine measure modified Impairment Index. After the removal of nondiscriminating variables in the initial discriminant analysis, six variables remained. These variables correctly classified 78.02% of the sample. Wilks’ Lambda was equal to .6680 and was significant (p c .05). Predication rates using this set of variables are shown in Table 3. Discriminant Functions on a new sample resulted in the correct classification of 70.59% of the subjects. Predication Formulas Using Fisher’s Linear Discriminant Functions From Nine Measure Modified Impairment Index yielded the following formulas:
Determination of Cerebral Dysfunction
267
Brain Impaired = -33.9289 + O.l0481*(SP) + O.l9247*(RY) + 0.27470*(TR) + O.O1862*(TB) + 0.24722*(TT) + 0.27882*(TL). Controls = -24.2285 + O.O7637*(SP) + O.l5708*(RY) + 0.25658* (TR) O.O1250*(TB) + 0.20911 *(TT) + 0.25408*(TL). Analysis of I8 Clinically Significant Variables. The final set of analyses used
eighteen measures. After the removal of nondiscriminating variables in the initial discriminant analysis, five variables remained. These variables correctly classified 77.42% of the sample. Wilks’ Lambda was equal to .5662 and was significant (p < .Ol). Predication rates using this set of variables are shown in Table 4. A cross validation, using Fisher’s Linear Discriminant Functions on the new sample resulted in the correct classification of 75.91% of the subjects. Predication formulas using Fisher’s Linear Discriminant Functions from eighteen measures yielded the following formulas: Brain Impaired = -31.2271 + O.l7747*(RY) + O.l7327*(TT) + O.O9071*(TM) + 0.30882*(WF) + 0.23593*(DR). Controls = -20.3517 + O.l4068*(RY) + O.l2396*(TT) + O.l2592*(TM) + 0.25843*(WF) + O.l4806*(DR). All tests of differences between initial correct classifications and their cross validations for both “impaired” and “control” groups were determined not to be significant (p > .05). Although the reduction in classification accuracy was not statistically significant, the clinical significance for the reduction in accuracy remains to be determined.
DISCUSSION This study demonstrates that three groupings of variables analyzed with discriminant analyses are not substantially superior to the Impairment Index in terms of correctly classifying individuals as brain impaired or not. This may relate to the considerable overlap of the variables comprising the Impairment Index. However, tests not comprising the Impairment Index offer potential classification utility. Using the seven and nine variable groupings in discriminant analyses, it was determined that Speech Perception Test, Seashore Rhythm, Tapping - Right, Trail Making B, TPT - Total Time and TPT Location were the tests with the best predictive ability for both the impaired and control groups. Using the eighteen variable combination of tests in a discriminant analysis, it was found that Seashore Rhythm, TPT - Total Time, TPT - Memory, Thurstone Word Fluency, and Delayed Memory Total Percent were the tests with the best predictive ability for both the impaired and control groups. This analysis is interesting in that five tests from the original impairment index formula along with the Thurstone Word Fluency and Percent recall from the
268
M. G. Mutchnick.
L. K. Ross, and C. J. Long
WMS classify subjects with accuracy equivalent to the impairment index. These findings suggest that these tests are deserving of consideration in the evaluation of test battery results as well as in the design of test batteries in experimentation. One inherent weakness in the present and other studies investigating decision strategies regarding brain dysfunction is the assumption that one formula will effectively classify patients as impaired or not impaired. As previously mentioned, the pattern of performance changes as a function of etiology, location and size of the lesion. To base decision strategies upon one formula will necessarily place the emphasis on tests that are sensitive to more generalized impairment. There may well be tests which are sensitive to more select lesion locations. Performance on these tests would be impaired if the lesion involves this brain area but may not if the lesion is located elsewhere. While most neuropsychological tests are sensitive to broad areas due to their complexity, it is easy to see that more complex tests are more sensitive to the presence of generalized dysfunction and thus are more likely to be selected as general predictors of brain dysfunction. Quantitative analysis of data in research can alert the clinician to tests of importance allowing for better decision strategy. Delayed Memory Total Percent, as described above, for example, is shown to be a sensitive measure of brain impairment and should be considered by the neuropsychologist as a test of clinical value. The clinician should always consider quantifications as a tool for integrating information into a final clinical decision. A seemingly useful approach to analysing data begins with an analysis of level of performance. This may include the interpretation of group normative data, such as the Impairment Index and/or discriminant analysis data. Then the score profile of the patient must be interpreted. The present research has offered an alternative to commonly used tools, such as indices of impairment. By presenting a closer examination of the tests that comprise such indices as well as other tests, the clinician is provided with more specific information which can be used in a flexible, more interpretable way. If specific tests are understood in terms of their ability to differentiate groups of patients, the neuropsychologist has choices as to their integration. One alternative to the single formula method is to use several available formulas to determine brain dysfunction and then make the clinical decision based upon the overall pattern. An obvious alternative is to investigate predictive formulas for more focal lesions. The clinical judgment could be based on the pattern of such predictive formulas. Research involving discriminant analyses is commonly conducted using linear discriminant functions, also known as canonical discriminant function coefficients, to classify their subjects. A serious limitation of this method disallows for the further classification of additional patients for cross-validation or for the determination of an individual’s brain dysfunction in the clinical set-
Determination of Cerebral Dysfunction
269
ting. These linear discriminant functions (as routinely presented by SPSS/PC+ [Norusis, 19861) are not presented with all algorithms needed for their computation and thus, cannot be directly used for classification. Therefore, use of these discriminant functions for additional subjects would yield incorrect results. The Fisher’s Linear Disc~min~t Functions (Norusis, 1986) do provide ali information necessary for further application and are identical to the discriminant functions with regard to accuracy of prediction. The present study utilizes Fisher’s Linear Discriminant Functions for cross-validation in an attempt to rectify the errors of previous research using discriminant statistical manipulations. Finally, discriminant analysis may not provide a convenient alternative but such procedures can clarify the decision strategies involved in the determination of brain dysfunction. This research suggests that there is no single good predictive formula. Various data analysis methods can at best only synthesize the data and provide the clinician with better information upon which to make a judgment. Greater gains may be obtained in future research that focuses upon predictive formulas sensitive to Iateralized or more specific focal lesions.
REFERENCES Anchor, K. N., Rogers, 3. P., Solomon, G. S., Barth, I. T., Peacock, C., & Martell, D. A. (1987). Fu~men~ls of disability dete~ination and mhabili~tion: A higher ground for the applied neurobehavioml sciences. Defense Law .lourrral, 36.363-402. Boll, T. J. (1986). Nontraditional and threshold considerations in neuropsychological assessment. In S. B. Filskov & T. J. Boll (Eds.), Iiundbook of clinical neuropsychology (pp. 103-120). New York: John Wiley & Sons. Boucher, M. L., Dewan, M. J., Donnelly, M. P., Pandurangi, A. K., Bartell, K., Diamond, T., & Major, L. F. (1986). Relative utility of three indices of neuropsychologic~ irn~~ent in a young, chronic schizophrenic population. The Journal of Nervous and Mental Disease, 174, 44116. Ferguson, G. A. (1981). Statisticalanalysis in psychology and education. New York: McGraw Hill. Goldstein, G., & Shelly, C. H. (1972). Statistical and normative studies of the Halstead Neuropsychological Test Battery relevant to a neuropsychiatric hospital setting. Perceptual and Motor Skills, 34, W-620. Halstead, W. C. (1947). Bruin and intelligence. Chicago: University of Chicago Press. Incagnoli, T. I. (1985). How much brain damage? The clinical neuropsychologist. Trial Lawyer’s Quarterly, 17,30-34. Long, C. J. (1985). Neuropsychology in private practice: Its changing focus. Psychotherapy in Private Practice, 3,45-S.
Long, C. J., & Gouvier, W. D. (1982). Neuropsychological assessment of outcome following closed head injury. In R. N. Malatesha & L. C. Ha&age (Eds.), Ne~rop~c~togy and cognition, Volmne Ii. The Hagne: Mattinus Nijhoff. Long, C. J., L Hunter, S. E. (1981). Analysis of temporal cortex dysfunction by neuropsychological techniques. Clinical Neuropsychology, 3.16-24. Long, C. I., & Klein, K. (1987, October). Analysis of age effects upon neuropsychological functioning. Poster presented at seventh annual meeting of the National Academy of Neuropsychologists. Chicago, IL.
270
M. G. Mutchnick, L. K. Ross, and C. J. Long
Personal Injury Review, 777-806. Matthews, C. G., Shaw, D. J., & Klove, H. (1966). Psychological test performances in neurologic
Malone, T. W. (1987). Direct examination of a neuropsychologist.
and “pseudoneurologic” subjects. Cortex, 2.244-253. Norusis, M. J. (1986). Advanced Stafisfics: SPSSIPC+ [Computer Program and Manual]. Chicago: SPSS, Inc. Reitan, R. (1964). Psychological deficits resulting from cerebral lesions in men. In J. M. Warren & K. Akert (Eds.), The frontal grundular cortex and behaviors (pp. 295-312). New York: McGraw-Hill. Reitan, R. M. (1966). A research programme on the psychological effects of brain lesions in human beings. In N. R. Ellis (Ed.), International review of research in mental retardation, Vol. 1. New York: Academic Press. Russell, E. W., Neuringer, C., & Goldstein, Cl. (1970). Assessment of brain damage - A neuropsychological key approach. New York: Wiley-Interscience. Stanczak, D. E., (1984). Assessment of level of performance in neuropsychological diagnosis: A comparison of the discriminant validity of alternative statistical methods. Unpublished doctoral dissertation. Memphis State University. Stuss, D. T., & Trites, R. L. (1977). Classification of neurological status using multiple discriminant function analysis of neuropsychological test scores. Journal of Consulting and Clinical Psychology, 45,145.
Swiercinsky, D. P., & Warnock, J. K. (1977). Comparison of the neuropsychological key and discriminant analysis approaches in predicting cerebral damage and localization. Journal of Consulting and Clinical Psychology, 45808-814.
Vega, A., Jr., & Parsons, 0. A. (1967). Cross-validation of the Halstead-Reitan tests for brain damage. Journal of Consulting Psychology, 31,619-625. Wheeler, L., & Reitan, R. M. (1963). Discriminant functions applied to the problem of predicting cerebral damage from behavioral tests: A cross-validation study. Perceptual and Motor Skills, 16.681-701. Wheeler, L., Burke, C. J., & Reitan, R. M. (1963). An application of discriminant functions to the problem of predicting brain damage using behavioral variables. Perceptual and Motor Skills, 16.417-440.