Innovations In Clinical Neuroscience

NOV-DEC 2017

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

Issue link:

Contents of this Issue


Page 42 of 83

43 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 the bifactor is a multidimensional model, we can ultimately "collapse" the model down into a single dimension in order to study how well PANSS items are reflecting the general trait of symptom severity, while simultaneously controlling for the biasing effects of secondary or nuisance dimensions. Although the specific dimensions are controlled for in the bifactor model, we can derive CRCs and IICs in a bifactor model that are analogous to their unidimensional IRT model counterparts. METHODS Data and demographics. A total of 3,647 subjects with SZ from 11 different trials were included in this study and are detailed further in Table 1. The Item Category 7 was infrequently endorsed, so was recoded to Item Category 6. A total of five different medications (paliperidone ER [extended release], paliperidone palmitate, olanzapine, quetiapine, and risperidone) were compared with a placebo intervention. All subjects were off antipsychotic medications at the baseline assessment. All subjects provided written informed consent after receiving a complete description of the study, which was conducted in accordance with the latest version of the Declaration of Helsinki. The length of time varied from 1 day to 5 days depending on the study so not all subjects were strictly medication-free because of washout variability. Benzodiazepines were allowed to certain limits to control agitation. All except one SZ trial used PANSS scores of 60 to 120 or 70 to 120 for inclusion criteria. Further details are published in the original articles (Table 1). 19–29 Fitting the bifactor IRT model. Factor analysis. Our ultimate goal is to fit a bifactor IRT model with one general factor (representing global symptom severity) and five specific domain factors (positive, negative, disorganized, excited, and anxiety/depression symptoms). Prior to fitting the IRT model, we first conducted a set of exploratory and confirmatory factor analyses to 1) judge the viability of a bifactor structure for the PANSS data, 2) determine the degree to which each item loads on the general factor and each of five specific domain factors, and 3) identify items with sizeable cross-loadings on multiple factors. Such items are known to bias bifactor solutions depending on the degree of violation. 30 Specifically, we conducted an exploratory bifactor factor analysis using the Schmid-Leiman technique available in the psych library 31 in R 3.41 (R Core Team, 2017) using minres estimation and oblimin rotation of polychoric correlations. We then fit both unidimensional and bifactor confirmatory factor models using diagonally weighted least square (DWLS) estimation available in the lavaan package 32 in R. Our goal of these preliminary analyses was to judge the fit of a bifactor structure using standard indices (CFI, RMSEA, and SRMR) and to test the superiority of a bifactor model relative to a unidimensional, single trait, model. FIGURE 2. Item information curves for 4 Positive and Negative Syndrome Scale (PANSS) items that vary in discrimination FIGURE 1. Category response curves for 4 Positive and Negative Syndrome Scale (PANSS) items that vary in discrimination—Blunted Affect and Difficulty in Abstract Thinking would be considered weak items, while Conceptual Disorganization and Delusions are strong items with scores that provide much information about the latent trait.

Articles in this issue

Archives of this issue

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