The relationship, structure and profiles of schizophrenia measurements: a post-hoc analysis of the baseline measures from a randomized clinical trial

Schizophrenia is a complex, multidimensional disorder. Patients diagnosed with schizophrenia exhibit positive, negative, and mood symptoms as well as experience cognitive and functional impairments. To fully assess the various dimensions affected by schizophrenia, clinical trials often include multiple scales measuring various symptom profiles, cognition, quality of life, subjective well-being, and functional impairment.

While the different measurement scales generally assess diverse aspects of schizophrenia, these scales may have some overlap in the constructs they measure. For example, the Schizophrenia Objective Functioning Instrument was designed to assess functioning, and the Quality of Life Scale was designed to assess a patient's quality of life; however, both have domains that assess occupational and psychosocial functioning. Therefore, it seems reasonable to propose that both scales would show some similarity (i.e., conceptual overlap of functioning). The use of multiple scales that result in the collection of redundant information may lead to increased study burden (e.g., multiple scales, multiple items per scale, and a 30- to 45-minute time requirement to administer each scale). Incorporating multiple assessment scales may require more time and effort from patients, additional staff time at study sites to administer measures and record data which could cause potential data quality issues, and an overall increased cost of conducting clinical trials.

Recently, we published data from a prospective clinical study in patients diagnosed with schizophrenia or a related disorder showing that patients who exhibited early response to antipsychotic treatment experienced early and consistent improvement across multiple symptom domains, subjective well-being, and health outcomes [1, 2]. In this exploratory study, we used baseline data from this clinical trial to: 1) characterize the relationship among different symptomatic and functional measures; and 2) identify a more parsimonious set of measures that minimize conceptual overlap.

Pearson correlation coefficients among the various measurement scales are shown in Table 2. The greatest correlation was observed between the QLS and SOFI (r = 0.669). The second greatest correlation was observed between the PANSS and SOFI (r = -0.407), followed closely by the correlation coefficients for PANSS and QLS (r = -0.387), and the MADRS and SWN-K (r = -0.354).

In this exploratory study, we aimed to quantitatively characterize the relationships among clinical, functional, cognitive, and quality-of-life measures, and to identify a more parsimonious set of measurements. In the optimal path model, the SWN-K was selected as the ultimate outcome, although the overall model explained only 15% of variance in the SWN-K. The strongest correlation was observed between the SOFI and the QLS. In the path model, the effect of the PANSS (an exogenous variable) on the QLS was mediated primarily by the SOFI. The factor analysis suggested four factors: "Functioning" (loading by SOFI, QLS, and PANSS negative), "Daily Living" (loading by SOFI), "Depression" (loading by MADRS, PANSS anxiety/depression factor, and SWN-K negatively-asked questions), and "Psychopathology" (loading by PANSS negative and positive symptoms, disorganized thoughts, and hostility/excitement factors). In addition, the measurement profile analysis revealed three classes that generally followed a pattern from less severely impaired on measures of functioning (QLS, SOFI) and less severely ill (PANSS) (Class 1) to moderately impaired on functioning (Class 2), and most severely impaired on functioning and psychopathology (Class 3).

Our findings suggest that redundancy exists among the measures studied, particularly among the clinician-rated functional and quality of life measures. The Pearson correlation coefficients among measurements revealed the strongest correlation between the SOFI and the QLS (r = .669). Previously, during development and validation of the SOFI, psychometric properties also revealed a moderate correlation between the SOFI and QLS, similar in magnitude to the one observed in the current study with a correlation coefficient of r = .61 for the patient-rated version of the SOFI and r = .52 for the informant version [8]. In addition, the path model explained 47% of the variance in the QLS, and a majority of this effect was mediated by SOFI, suggesting overlap between QLS and SOFI. Furthermore, while the QLS loaded on "Functioning," the SOFI loaded on both the "Functioning" and "Daily Living" factors. While both the SOFI and QLS provide, to some degree, a measure of social and occupational functioning, the SOFI provided a broader measure of outcomes that included functioning and daily living.

The best-fit model for the path analysis, which selected the SWN-K to be the ultimate outcome, revealed that only 15% of the variance in SWN-K could be explained by the model, which included measures of symptoms, functioning, and cognition. This finding may suggest that the SWN-K is a unique measure capturing potential treatment effects not captured by the other measurement scales. The SWN-K negatively worded statements loaded on the "Depression" factor, while the SWN-K positively worded statements did not meet the factor-loading criteria. This latter finding may suggest important differences between these two components of the SWN-K, a finding that was consistent with a recent factor and item response theory analysis on the English version of SWN-K [7]. Additional work will be necessary to further understand the relationship of the positively and negatively worded statements to the psychometric properties of the scale as a whole, and the unique qualities of the SWN-K to overall treatment responsiveness.

The SWN-K total score has been demonstrated previously to be associated with dopaminergic D2 receptor blockade [11], medication adherence [12], and the likelihood of achieving enduring symptomatic remission [13]. Subjective well-being has also been associated with depression. In patients with schizophrenia, depressive symptoms were significantly associated with subjective well-being in newly admitted patients [14] and during the course of acute treatment with atypical antipsychotics [15]. In these studies, a significant negative correlation was observed between the SWN-K score, the PANSS depression factor score, and the subjectively-rated Beck Depression Inventory (BDI) [14, 15], although only the correlation between the SWN-K and BDI was significant following 8 weeks of treatment [15]. These findings are consistent with the factor analysis in which the SWN-K negatively worded statements loaded on the "Depression" factor. In the path analysis, the physician-rated MADRS was predictive of the SWN-K in a direct fashion, with a one SD change of the MADRS leading to a change of -0.33 SD in the SWN-K. In addition, the correlation analysis had revealed a small to moderately sized negative correlation between the MADRS and SWN-K (r = -0.35). These findings collectively highlight the important role that depressive symptoms may play in low subjective well-being, and the importance of a patient's subjective well-being to treatment outcomes, including medication adherence and remission.

The measurement profiles of the study population detected heterogeneity primarily in measurements of social and occupational functioning and daily living activities via the SOFI and QLS, whereas the study population was generally homogeneous in psychopathological symptoms, as defined by the study inclusion criteria. Patients in Class 3 stood out as having the worst functioning and daily living and the worst or most severe symptoms. Patients in Classes 1 and 2 had moderately severe symptoms, with patients in Class 1 having the best functioning and daily living, while patients in Class 2 had moderately impaired functioning and daily living. Patients in Class 1 with the best functioning and daily living also seemed to show somewhat higher scores on SWN-K positively worded statements.

The BACS-SCT, or symbol coding test, is a measure of attention and speed of information processing [5]. A recent meta-analysis of 37 studies comparing digit symbol coding tasks to other cognitive measures in schizophrenia demonstrated a significantly larger mean effect size for impairment in digit symbol coding compared with the effects of impairment in episodic memory, executive function, and working memory, suggesting that information processing inefficiency is a central feature of the cognitive deficit in schizophrenia [16]. A subsequent study examining the predictive relationships between neuropsychological domains, functional competence, social competence, symptoms, and real-world behavior demonstrated that only processing speed had both direct and indirect effects on all three real-world behaviors including domains of work skills, interpersonal relationships, and community activities [17]. Reduced processing speed has been associated with functional disability observed in patients with schizophrenia [18, 19]. In previous research, we found information processing speed had both direct and indirect effects via negative symptoms on three domains of functioning, as measured by the QLS at baseline and following 24 weeks of antipsychotic treatment [20].

In the current analysis, the BACS-SCT did not play a major role in any of the current analyses including the path-modeling and factor analysis. Previously, we used a composite measure of processing speed that included an average of two subscales including digit symbol coding and the verbal fluency scale, and that focused on QLS domains of functioning as the ultimate outcome [20]. Perhaps the use of only the digit symbol coding test underlies the different findings. It would be legitimate to argue that the limited role of the BACS-SCT observed in these analyses suggests that this test may also be capturing unique information. However, in contrast to the SWN-K, which showed a mild to moderate correlation with the QLS, SOFI, and MADRS, the BACS-SCT was not significantly related to any of the clinical or functional measures evaluated in this study.

Our findings from the path analysis using structural equation modeling, the factor analysis of the measurement structure, and the measurement profiles from the latent class analysis complement each other in understanding the measurements. Though each model was implemented under different assumptions, the findings that QLS and SOFI measures were highly correlated was consistent. The MADRS and SWN-K were also correlated, while the BACS-SCT was not significantly related with any of the other measures. This study may contribute to the effort to better understand schizophrenia measurements with the goal of identifying a parsimonious data set.

There were several limitations to the current analyses. First, patients had to have a particular level of acuity to enter the study, and this likely restricted the possible range of baseline scores on the PANSS. Second, for the path analysis, an assumption was made that the PANSS, MADRS, and BACS-SCT were "exogenous variables" assessing symptoms and attention and processing speed deficits assumed to be more proximal to disease manifestation. Additionally, it was assumed that the SWN-K, SOFI, and QLS were "endogenous variables" assessing subjective well-being, functioning, and quality of life thought to be the consequences of the proximal symptoms. However, our previous work has demonstrated that subjective well-being, functioning, and quality of life can change as early as 2 weeks into treatment and seemingly mirror improvements in symptoms [2]. Thus, the temporal relationship of change among these variables is not fully understood, and the outcomes observed are thereby limited by the proposed relationships set forth by the specifications of the statistical models. Third, we incorporated the SWN-K total score in the correlation and path analyses and the SWN-K positively worded and negatively worded statements in the factor analysis and measurement profiles. Therefore, comparisons for the SWN-K total score cannot be made across all of the analyses.

This study was exploratory in nature, with the results being driven by both statistics and knowledge of the disease and population. Further, even a perfect fit of the model would not prove that the inferences are causal, but merely suggest that the model fits the data well. It would be helpful to attempt to replicate the results for a similar population at a different time, and/or to replicate the results in a different patient population with similar or varying disease characteristics. It is important to realize that results of this study reflect a chronically ill patient population moderately to severely ill with an exacerbation of symptoms, and the observations made are dependent upon the scales incorporated in the study design and assessed at baseline.

Researchers should consider collecting PANSS, SOFI, and SWN-K in their trials. This would allow a broad spectrum of assessments that would have the ability to capture a wide range of treatment outcomes and allow for a rich characterization of the subgroups involved. Additional research is needed to identify the critical cognitive measures.