Navegando por Palavras-chave "Latent Class Analysis"

Agora exibindo 1 - 1 de 1
  • Nenhuma Miniatura disponível
    Item
    Somente Metadadados
    Explorando novas abordagens para compreensão da heterogeneidade clínica da esquizofrenia por meio da modelagem de equações estruturais
    (Universidade Federal de São Paulo (UNIFESP), 2019-04-26) Higuchi, Cinthia Hiroko [UNIFESP]; Araripe Neto, Ary Gadelha De Alencar [UNIFESP]; Universidade Federal de São Paulo (UNIFESP)
    Introduction: Schizophrenia is a heterogeneous disease clinically, therapeutically and biologically. This scenario is considered one of the greatest challenges to achieve real transformation in the field. Current proposals to reduce heterogeneity attempt to delimit dimensions, subtypes, or models of clinical staging. However, these models, for the most part, do not reach psychometric validity (dimensional models), do not have robust biological validation (classical subtypes or staging) or, still, do not reach greater clinical utility than the current constructs. We will use techniques based on the structural equation modeling to evaluate the Positive and Negative Syndromes Scale (PANSS) items. This instrument is widely used and address the possibilities of clinical presentation of schizophrenia. Objective: To explore the potential of PANSS to generate dimensions of symptoms and subgroups of patients with schizophrenia through models with psychometric validity (clinical models) and biological models (neuroimaging biomarkers) generated by structural equation modelling. Specific objectives: Study 1: a) To identify the best 5-factor dimensional model of PANSS; b) To evaluate the impact of clinical staging and other clinical variables in PANSS dimensions. Study 2: a) To use the PANSS as generator of more homogeneous groups regarding the profile of symptoms through a latent class analysis (LCA); b) Validate the final model of classes with external and biological variables (cortical thickness). Methods: Data from 700 patients diagnosed with schizophrenia from four different centers were analyzed. Study 1: CFA models were compared with Bayesian CFA, the latter considered to be more flexible. The multilevel structure was then included. In addition, Multiple Indicators Multiple Causes (MIMIC) modeling evaluated the impact of clinical staging of schizophrenia on the formation of factor mean. Study 2: The best LCA model was chosen based on the comparison of AIC, BIC and Log likelihood values and according to the evaluation of the items probabilities of response applied to the clinical utility of the model. The LCA derived class variable was used in univariate general linear models (GLM) to verify its effect on the cortical thickness of 143 patients, xi controlling the result for sex and age. The frontal and temporal regions of cortical thickness were selected according to the Desikan-Killiany atlas. The p-values were corrected for multiple comparisons (FDR and bonferroni). Results: Study 1: the PANSS CFA factorial solution achieves good fit indices when a multilevel structure is added. The clinical staging of schizophrenia can predict a higher mean of the factors according to the stage of the disease. Study 2: The six-class model best represents patient profiles. The class variable has effect on the cortical thickness of two regions: right superior temporal gyrus (pvalue = 0.012) and right temporal pole (p-value = 0.007), but such p-values did not remain significant after correction by multiple comparisons. Conclusions: The final models have psychometric validity and present: 1) The best dimensional model of PANSS is the CFA with multilevel structure; 2) There is impact of clinical staging in the formation of PANSS mean factors; 3) The sixclass model of PANSS indicated more homogeneous groups that indicate relation to measurements of cortical thickness in temporal regions.