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 49 of 83

50 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 analytic and IRT, to discern how well the PANSS total scores reflect global symptom severity and to what degree meeting the Remission criteria identifies individuals who have low levels of symptom severity. To address the first question, the bifactor models shows that 88 percent of total symptom score variation ( ω H coefficient) can be attributed to variation in general symptom severity, and only eight percent reflected secondary domain factors; thus, there is four-percent error variance. We can conclude from these values that the 30-item PANSS is highly reliable and that total scores predominantly reflect a general latent factor that we termed symptom severity. The results suggest that interpretation of PANSS scores is not overly complicated or compromised by multidimensionality. On the other hand, not all PANNS items are "good" or "pure" indicators of general symptom severity. In fact, there was large variation with some items being relatively better measures of the general factor than of specific domain factors. This fact needs to be considered in any future short-form versions of PANSS. Ideally, in short form creation, items that have high discrimination on the general factor, are "univocal" measures of the general factor (i.e., with high ECVI), and are balanced across the content domains (e.g., selecting two items from each domain) should be selected. The rationale behind this scale construction strategy and how it yields the most interpretable scores is beyond the present scope, but Stuckey et al 16 and Edelen et al 36 provide lengthy discussions and examples. Our analysis of the eight-item Remission set revealed two important findings. First, individuals who were judged Remitted based on scoring 3 or below on the remission set, scored about 0.8 SD lower in the general latent trait θ than non-remitted patients, with 43.5 percent overlap between subjects who were and were not remitted. However, being judged as remitted using these criteria is not associated with symptom relief, where "relief " is defined by low levels of the latent trait reflecting general symptom severity (i.e., trait levels less than θ= -1 corresponding to average PANSS item scores between 1 and 2). Post-treatment, 534 individuals (or about 15% of the whole sample) scored below θ= -1 in symptom severity, but 1,351 subjects satisfied Remission criteria (37%). Second, our analysis of the test information curves for item subsets revealed that the Remission set is not ideal in terms of discriminating individuals along the symptom severity dimension as defined by the bifactor IRT model. In fact, all tested alternative subsets, especially the most discriminating items, outperformed the Remission set in the amount of information measured (Figure 6). The discrimination items were most valuable for testing along the entire symptom range (including high severity), while the Relief set items were strongest for testing in the low-symptom range. One advantage of IRT models is that scores are derivable based on any subset of items, and the metric of the latent variable is easily related to the metric of raw scores. Embretson et al 17 provides further discussion on this topic. FIGURE 7. The "best" item subset depends on the illness level of the patient being assessed. The items providing the most information over the entire latent trait range are those with the largest discrimination parameters, which differ slightly from items providing the most information in the Symptom Relief range of θ =(-4,-1). Item scores ≤2 would be associated with trait levels below the mean range. Items marked with a * were part of the original remission subset. ECVI (explained common variance per item), Discrimination, and Information are rescaled between 0-1 for comparison purposes.

Articles in this issue

Archives of this issue

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