Efficacy of several statistical methods in differentiating TBI and co-occurring conditions: A replication study

Abstract

Much historical controversy has surrounded TBI criteria especially associated with milder injuries. Lack of sufficient injury documentation underlies problems in the diagnosis of TBI as does variability across multiple expert consensus criteria for diagnosing TBI. The efficacy of diagnostic classification utilizing test data sets with comparison group methodology increases diagnostic accuracy. Five statistical comparisons utilized here improved differential diagnostic accuracy compared with PCS and somatization with neuropsychological testing. Objectives (1) Cross-validation of neuropsychological test data sets of moderate-severe TBI (N = 30) with test data from moderate-severe (N = 74); somatization (N = 24) and PCS (N = 22) cases in a database, (2) Determine if cognitive test data sets alone differentiated TBI from other groups, and (3) Evaluate the efficacy of measures in comparisons: Kullback-Leibler, Correlation, Patterns, Cohen’s d, and MNB. Materials and methods Meyer’s Neuropsychological System; Comparison groups -TBI sample with structural evidence of brain injury (CT/MRI); comparison of 5 statistical measures’ efficacy in test data analysis comparing a community sample of moderate TBI (N=30) with a data base containing moderate-severe TBI (N = 74) + co-occurring groups (PCS N = 22) + Somatization (N = 24). Measures utilized: Correlation, Kullbeck-Leibler divergence, Cohen’s d, MNB code, Configuration. Results Combining the five measures most accurately matched the TBI sample (30/30 cases) with MNB comparison groups of similar TBI severity while differentiating those cases from PCS and Somatoform cognitive testdata. Both Kullback Leibler & Cohens’ d reduced false positive errors in comparison with the other measures.