Navegando por Palavras-chave "ROC Curve"

Agora exibindo 1 - 1 de 1
  • Nenhuma Miniatura disponível
    Item
    Somente Metadadados
    Acurácia do teste de linguagem expressiva para identificar alterações morfossintáticas em crianças pré-escolares
    (Universidade Federal de São Paulo (UNIFESP), 2021) Kayanoki, Fabiana Yumi Sato [UNIFESP]; Puglisi, Marina Leite [UNIFESP]; Universidade Federal de São Paulo
    Introduction: The evaluation of preschool children's grammatical aspects is essential to identify the areas of major language impairments, in addition to often directing the speech rehabilitation of these children. Thus, it is necessary to develop simple, specific protocols and with reference standards. In Brazil, however, there are few instruments used to assess preschool children using statistical and psychometric analyses, especially with regard to validation in protocols in speech therapy practice. Objective: To validate, compare forms of analysis of the grammatical modality and verify the diagnostic accuracy of the Expressive Language Test (ELT) to identify children with low language performance. Methods: The ELT is an instrument created to assess the grammatical skills of preschoolers and was applied in 618 children aged 3 years and 6 months to 6 years and 1 month enrolled in Years I or II of Municipal Schools of Early Childhood Education in São Caetano do Sul. Children were individually evaluated and encouraged to describe the content of 12 scenes, based on guide questions (picture-elicited language samples). Three objective forms of grammatical analysis of responses were proposed, in which only the sentences pertinent to the content of the test figures were analyzed, pondering the obligatory context of the scenes: 1) Mean Length Utterances – words (MLU-w): mean number of words produced in a sentence; 2) Length Utterances (LU): sum of number of words produced in a sentence e 3) Percentage Grammatical Utterances (PGU): ratio of the number of grammatically correct sentences to the total number of sentences produced in the test. Then, the MLU-w_II and LU_II were computed again, but considering the entire production of the child (both the pertinent sentences and those that extrapolated the content of the figures) for all utterances produced in each item in order to verify whether these measures would also assess the child's complete response. All forms were computed and tested for construct validity by Confirmatory Factor Analysis and only those with the most fitting indicators were considered to test the accuracy. To investigate whether the TLE is sensitive to identify children with Language Disorder (LD), the vocabulary, receptive grammar and phonological working memory tests were considered as the gold standard. And the accuracy of the ELT was assessed using the ROC Curve and the sensitivity and specificity values. Results: The LU, MLU-w and PEG analyzes were considered valid to assess the grammatical skills of children. However, EME-p_II and EE_II did not show indexes of adequacy as robust as the previous ones and therefore were not validated for the ELT. In addition, the test showed 66% Positive Predictive Value (probability of a child presenting a positive result and LD) and 67% Negative Predictive Value (probability of a child presenting a negative result and not LD) for the diagnosis of LD in preschoolers. Conclusions: LU and PEG are the two recommended measures for grammatical analysis for ELT. The cutoff point of the ROC curve -0.429 showed the best prognostic accuracy for the test with a sensitivity of 70.4% and specificity of 66.6%.