Innovations In Clinical Neuroscience

NOV-DEC 2017

A peer-reviewed, evidence-based journal for clinicians in the field of neuroscience

Issue link: https://innovationscns.epubxp.com/i/924986

Contents of this Issue

Navigation

Page 46 of 83

47 ICNS INNOVATIONS IN CLINICAL NEUROSCIENCE November-December 2017 • Volume 14 • Number 11–12 O R I G I N A L R E S E A R C H misfit include small unmodeled correlated residuals. Finally, the fit of a unidimensional model yielded a c 2 of 37,872 (df=405), CFI=0.917, RMSEA=0.159, and SRMR=0.123, all indicating that the unidimensional model significantly worsens the fit relative to the bifactor. That is, there is statistically significant multidimensionality in the PANSS ratings that needs to be accounted for by the bifactor model. The last column in Table 2 displays the ECVI index values with an asterisk (*) next to the 10 items with the largest ECVI. Inspection of these values along with the loadings on the general and specific factors indicate a great diversity in the PANSS items in terms of measurement properties. Specifically, while many items are highly discriminating (i.e., high factor loadings) and relatively "pure" indicators of general symptom severity (Delusions, Lack of Judgment, Conceptual Disorganization), others are better indicators of specific dimensions (Blunted Affect, Mannerisms and Posturing, and Depression). Moreover, the relatively small loadings (<0.50) on specific domain Factors F4 (positive) and F5 (excited) indicate these factors are not well determined. Finally, ω reliability was 0.96 and ω H was 0.88 indicating that 88 percent of (unit-weighted) total score variation can be attributed to variation in general symptom severity, and only eight percent reflected secondary domain factors. This implies that the total scores are an excellent indicator of symptom severity and interpretation is not overly compromised by multidimensionality. PANSS IRT bifactor model. For efficiency, conditional and marginal IRT parameter estimates are not shown. Instead, Table 3 displays the final "pseudo-unidimensional" IRT model parameters derived from the marginal IRT bifactor model. The column labeled a contains the discrimination parameters reflecting how well each PANSS item functions as an indicator of global symptom severity. The columns labeled b contain the threshold parameters indicating the trait level necessary to respond above a given category. We note that some of these values are extremely high, suggesting problems in category labeling that are beyond the scope of discussion here. To further evaluate the remission set, Figure 4 displays the TIC for the entire 30 items and for the four 8-item subsets, R, D, I, and U. Figure 4 also displays the corresponding conditional standard errors of measurement. It is clear from these figures that the 30 items provide much better measurement precision relative to any eight-item subset. More importantly, however, it appears that the remission subset is the least informative and thus the least precise subset in terms of scaling individuals on a global symptom severity dimension. Generally speaking, the best subset to judge symptom relief in terms of information and conditional standard error is the set of the most discriminating items. The set based on the information criterion, by design, performs slightly better in the lower symptom severity/ symptom relief ranges (θ≤ -1). Also, shown in Table 3 are four 8-item subsets. These were derived as follows. The R subset contains the eight items used to judge remission as suggested by Andreasen et al. 2 The U subset are eight items with the highest ECVI in Table 2. These items are the most univocal or pure indicators of symptom severity. The D subset are the eight items with largest discrimination parameters. Finally, the TABLE 3. Item discrimination and threshold parameters from the pseudo-unidimensional IRT model SYMPTOM a b1 b2 b3 b4 b5 R U D I Blunted Affect 0.65 -3.63 -1.41 1.47 4.29 7.13 X Emotional Withdrawal 1.05 -2.51 -1.01 0.91 2.86 4.42 Poor Rapport 1.09 -1.45 -0.01 1.77 3.29 4.65 Passive Apathetic Social Withdrawal 1.03 -2.4 -0.92 0.99 2.56 4.12 X Lack of Spontaneity Conversation 0.77 -2.04 -0.22 1.97 3.88 5.73 X Motor Retardation 0.57 -0.69 1.67 4.76 7.7 10.43 Active Social Avoidance 1.35 -1.71 -0.46 1.15 2.47 3.75 X Delusions 1.92 -1.3 -0.36 0.55 1.46 2.3 X X X X Hallucinatory Behavior 1.51 -0.63 0.15 0.89 1.64 2.77 X Grandiosity 1.08 0.05 1.12 2.19 3.43 4.87 Suspiciousness/Persecution 2 -1.11 -0.24 0.71 1.63 2.59 X X X Stereotyped Thinking 1.4 -1.61 -0.39 1.11 2.5 4.1 X X Somatic Concern 0.77 -0.8 1.08 2.9 4.5 6.44 Unusual Thought Content 1.75 -1.28 -0.34 0.9 1.93 2.82 X X X X Lack of Judgment and Insight 1.17 -2.23 -0.85 0.65 2.21 3.35 X X Conceptual Disorganization 1.58 -1.45 -0.32 0.89 1.95 3.07 X X X X Difficulty in Abstract Thinking 0.99 -2.78 -1.32 0.51 2.14 3.91 Mannerisms and Posturing 0.82 -0.81 0.59 2.9 4.96 6.98 X Poor Attention 1.27 -1.26 0.04 1.71 3.08 4.51 Disturbance of Volition 0.96 -1.66 -0.24 1.79 3.91 5.76 Preoccupation 1.75 -1.35 -0.27 0.98 2.1 3.28 X X Disorientation 0.93 0.22 1.52 3.69 5.39 7.21 Excitement 1.95 -0.44 0.41 1.42 2.42 3.24 X Hostility 1.71 0.04 0.96 1.97 2.85 3.63 X X Uncooperative 1.57 -0.02 1.01 1.94 2.91 4.04 Poor Impulse Control 1.57 -0.32 0.71 1.96 3.07 4.18 Anxiety 1.42 -1.05 0.13 1.43 2.42 3.68 Guilt Feelings 0.74 0.35 1.94 3.77 5.6 7.54 Tension 1.83 -0.93 0.15 1.39 2.44 3.52 X X Depression 0.85 -0.52 0.91 2.54 4.18 5.88 NOTE. Items in the Remission set are bolded and italicized. a =item discrimination; b =item threshold; R=remission set; U=most unidimensional as judged by ECVI (explained common variance per item); D=most discriminating as judged by discrimination; I=most informative at low symptomology levels

Articles in this issue

Archives of this issue

view archives of Innovations In Clinical Neuroscience - NOV-DEC 2017