Objective methods for matching neuropsychological patterns: Formulas and comparisons.

INTRODUCTION: Objective neuropsychology test score pattern matching methods can help to identify data similarities and differences with comparison groups which can help the clinician in diagnosis and in identifying treatment options. MATERIALS AND METHODS: The current study examines five methods of matching a data set: Correlation, Configuration, Kullback-Leibler (KL) Divergence, Pooled Effect Size (Cohen's d), and a new method called MNB (Meyers Neuropsychological Battery) Code. Thirty data sets diagnosed with Traumatic Brain Injury (TBI) were compared with four Comparison Group data sets consisting of TBI, Depression, Anxiety and Attention Deficit/Hyperactivity Disorder. RESULTS: The Correlation Method was correct 90% (27/30) and Configuration was correct 86% (26/30). The KL Divergence was correct 76% (23/30) and the MNB Code was correct 73% (22/30). The Effect Size Method was correct 70% (21/30). When using a simple majority of all the matching methods, the classification rate was 90+ percent. CONCLUSIONS: The results of this study demonstrate that there are statistical methods that can identify patterns of cognitive strengths and weaknesses. Multiple matching methods and a simple majority of agreement between the different comparisons suggests the best matching profile for diagnosis. In some cases, more than one pattern may be present.

Ecological validity of the Meyers Neuropsychological Battery.

Neuropsychologists are increasingly being asked to apply neuropsychological test results to real world functioning; however, neuropsychological tests are not usually constructed to do so, but instead are more concerned with diagnostic accuracy than with prediction of daily functioning. Using samples of 5,460 patients that did self-ratings and 2791 patients that had family ratings plus the Meyers Neuropsychological Battery (MNB), it was found that the family ratings were better predicted by neuropsychological test data than were self-ratings on the 38 item Patient Competency Rating Scale (PCRS). The R values for family ratings on the 36 regression equations ranged from .236 to .763. The results show that the ratings given patients by family members could be predicted by the neuropsychological test results. These findings can help the clinician to make broad statements regarding likely real-life functioning and also support the ecological validity of the tests that make up the MNB.

Premorbid estimates of neuropsychological functioning for diverse groups.

One of the basic tasks performed by a neuropsychologist is to identify the difference between current performance and the premorbid expected performance. Baseline expected performance for Intellectually Impaired (n = 21), Developmentally Delayed (n = 40), Attention Deficit Disorder (n = 98), Learning Disability (n = 42), and "Normal" groups (n = 75) were developed along with a demographically corrected prediction of premorbid functioning and a word reading based prediction of premorbid functioning. We utilized a subset of this data pool for development (n = 107) and validation (n = 108) of premorbid functioning estimates. Findings show that a combination of three methods (baseline, demographic, and reading) were superior to any individual method. The effect size (Cohen's d) calculations show that differences in the prediction of domain level performances were small and likely not clinically meaningful, indicating that the premorbid estimates would be usable as a prediction of expected performance at the domain level. However, the motor domains were not well predicted.

Normative Data for the Neurobehavioral Symptom Inventory.

The demographically diverse populations served by large health care systems (Veterans Affairs, Department of Defense, Medicare, Medicaid) are routinely screened with the Neurobehavioral Symptom Inventory (NSI). The extent to which a patient's report of symptoms either initially and/or across time is affected by demographic variables-gender, ethnicity, age, or education-has not been investigated despite widespread use of the NSI. In practice, the effectiveness of this tool might be improved with demographically based norms. A large data set of normal community-dwelling individuals was collected using the NSI. Emphasis was made to collect data from individuals of diverse ethnic backgrounds. It was hypothesized that ethnic/cultural backgrounds would have an impact on NSI scores. The results provide normative data for the NSI applicable to a wide variety of individuals of various ages and ethnic backgrounds. An analysis of variance indicated there was no significant difference in NSI responses based on ethnic/cultural background; however, age and gender were found to contribute significantly to the variance associated with symptom endorsement. The NSI appears to be a reliable measure of self-report postconcussive symptoms. Age is a variable associated with differential symptom endorsement on the NSI. Follow-up studies are needed to provide a measure of the sensitivity and specificity of this measure.

Is Co-norming Required?

Researchers who have been responsible for developing test batteries have argued that competent practice requires the use of a "fixed battery" that is co-normed. We tested this assumption with three normative systems: co-normed, meta-regressed norms and a system of these two methods. We analyzed two samples: 330 referred patients and 99 undergraduate volunteers. The T scores generated for referred patients using the three systems were highly associated with one another and quite similar in magnitude, with an Overall Test Battery Means (OTBMs) using the co-normed, hybrid, and meta-regressed scores equaled 43.8, 45.0, and 43.9, respectively. For volunteers, the OTBMs equaled 47.4, 47.5, and 47.1, respectively. The correlations amongst these OTBMs across systems were all above .90. Differences among OTBMs across normative systems were small and not clinically meaningful. We conclude that co-norming for competent clinical practice is not necessary.

Using pattern analysis matching to differentiate TBI and PTSD in a military sample.

Distinguishing between traumatic brain injury (TBI) residuals and the effects of posttraumatic stress disorder (PTSD) during neuropsychological evaluation can be difficult because of significant overlap of symptom presentation. Using a standardized battery of tests, an artificial neural network was used to create an algorithm to perform pattern analysis matching (PAM) functions that can be used to assist with diagnosis. PAM analyzes a patient's neuropsychological data and provides a best fit classification, according to one of four groups: TBI, PTSD, malingering/invalid data, or "other" (depressed/anxious/postconcussion syndrome/normal). The original PAM was modeled on civilian data; the current study was undertaken using a database of 100 active-duty army service personnel who were referred for neuropsychological assessment in a military TBI clinic. The PAM classifications showed 90% overall accuracy when compared with clinicians' diagnoses. The PAM function is able to classify detailed neuropsychological profiles from a military population with a high degree of accuracy and is able to distinguish between TBI, PTSD, malingering/invalid data, or "other." PAM is a useful tool to help with clinical decision-making.

An adaptation of the MMPI-2 Meyers Index for the MMPI-2-RF.

Using an overall sample of 278 individuals who had taken the Minnesota Multiphasic Personality Inventory-Second Edition (MMPI-2) and who had clear diagnostic information available in their medical records, the Meyers Index (MI) for the MMPI-2 (Meyers, Millis, & Volkert, 2002 ) was calculated for each individual, and a new version of the MI created for the MMPI-2 Restructured Form (MMPI-2-RF) was calculated. The MI is a method of combining multiple MMPI-2 validity scales into a single weighted index to assess exaggerated self-report on the MMPI-2. The new index is intended to provide the same type of global assessment of validity but for the MMPI-2-RF (MI-r). The MI and the MI-r were compared at both individual and group levels and were found to correlate well (r = .87). Diagnostic groups of litigants and nonlitigants of traumatic brain injury, chronic pain, and posttraumatic stress disorder were also examined; and the performance of the MI and the MI-r were similar. Similarly, the pass and fail agreement rate for the two scales was 93%. The results indicate that the MI and MI-r perform very similarly and are good methods of assessing overall validity of MMPI-2 and MMPI-2-RF test performance.

Using likelihood ratios to detect invalid performance with performance validity measures.

Larrabee (2008) applied chained likelihood ratios to selected performance validity measures (PVMs) to identify non-valid performances on neuropsychological tests. He presented a method of combining different PVMs with different sensitivities and specificities into an overall probability of non-validity. We applied his methodology to a set of 11 PVMs using a sample of 255 subjects. The results of the study show that in various combinations of two or three PVMs, a high reliability of invalidity can be determined using the chained likelihood ratio method. This study advances the ability of clinicians to chain various PVMs together and calculate the probability that a set of data is invalid.