Abstract
Background: The Positive and Negative Syndrome Scale (PANSS) is widely used as an instrument to assess symptoms. The 5-factor model is the most replicated model in studies analyzing the subjacent structure underlying symptoms profiles. However, there is no confirmatory factor analysis (CFA) study with good fit indexes. The CFA hypothesizes that an item loads on only 1 factor (no items have cross-loadings). More flexible approaches are necessary to fit the true nature of the scale. The bifactor CFA and Bayesian CFA are more flexible because they allow direct and/or indirect correlation between factors or items. The aim of this study is to perform a bifactor CFA and a Bayesian CFA as alternatives to CFA to evaluate the dimensions’ symptoms of PANSS in patients with schizophrenia.
Methods: We assessed 700 patients with schizophrenia of 4 different centers with PANSS. A Confirmatory Factor Analysis (CFA), bifactor CFA (bi-CFA), and Bayesian CFA were performed. The fit of the bi-CFA and CFA models were assessed by considering: Comparative Fit Index (CFI) and Non-Normed Fit Index (NNFI) > 0.95, the Root Mean Square Errors of Approximation (RMSEA) < 0.06 and Weighted Root Mean Square Residual (WRMR) < 1.0. The Bayesian CFA use the estimator bayes and has only 1 fit index, which is the Posterior Predictive P value (PPP) and its confidence interval (CI) of 95%. PPP values near 0.5 and a CI centered on zero suggests a good fit index.
Results: The mean (SD) age was 34.9 (10.3) years, and 64.3% were male. Age of onset and duration of illness means were 21.7 (7.5) and 13.2 (9.6), respectively. The CFA model had a poor fit index as follow: RMSEA = 0.102 (90% CI: 0.097–0.107; Cfit was <0.001), CFI = 0.921 and NNFI = 0.906, and WRMR = 1.952. The bi-CFA had fit indexes worse than the CFA: RMSEA = 0.120 (90% CI: 0.115–0.125; CFit 0.05 was <0.001), CFI = 0.898, NNFI=0.871, and WRMR=2.301. The Bayesian CFA was the last approach and presented a PPP value of <.001 and CI 57.878–203.999, which indicates a poor fit index.
Conclusion: All the approaches have failed. This suggests that a methodological issue is not the biggest problem, since even the most flexible approach (Bayesian CFA) had a poor fit. Considering the heterogeneity of the different centers, additional analysis may be necessary to evaluate the influence of clinical staging in the factor structure of PANSS.