Modelo de decisão para predição da disfonia a partir de dados autorreferidos

Abstract

The self-assessment tools have been recurrent and reliable strategies to detect and evaluate general health conditions in the population, because they represent robust alternatives to measure the impact of a health condition on individual well-being. In the context of vocal disorders, these questionnaires have gained wide recognition in the last decades, however, they still require more modern and rigorous analyzes from the psychometric point of view, to reinforce their validity, robustness and reliability. The detailed evaluation of the relationship between each factor investigated and the presence of dysphonia is important for the interpretation of the results of the instrument and for the elaboration of fast and efficient decision-making methods in the identification of this disorder. Thus, the goal of this research was to elaborate a statistical decision model for predicting dysphonia based on information from the main questionnaires of vocal self-assessment. For that, a documentary research was done from the database of the Integrated Laboratory of Voice Studies (LIEV) of the Universidade Federal da Paraíba. The sample consisted of 139 individuals over 18 years old, of both genders, professionals and non-professionals voice, with and without vocal complaint. Participants were allocated to the group with dysphonia (GCD) or in the vocally healthy group (GVS), according to medical diagnostic and evaluation performed by speech-language pathologists. The items of the Voice-Related Quality of Life (V-RQOL), Voice Handicap Index (VHI) and Voice Symptom Scale (VoiSS) were collected for the adjustment of several logistic regression models, with the aim of investigate the most significant set of items in decision making for prediction of dysphonia. The statistical treatment was performed using Software R, version 3.5.1. In the exploratory analysis, comparative tests between GCD and GVS indicated that total scores and domains of VHI and VoiSS were higher in GCD, with the exception of the emotional domain for both questionnaires. No differences were observed between the groups regarding the V-RQOL. In the regression analysis, model 1, adjusted with the V-RQOL items, was not considered valid by the global adequacy tests. The model 2, composed of 3 items of VHI and model 3, composed of 2 VoiSS items, were considered valid and with high accuracy level (model 2 = 80,2% and model 3 = 81,9%). A global model, adjusted with the most significant variables of the previous models, resulted in a structure containing only the item 14 of the VHI ("I feel as though I have to strain to produce voice") and the item 4 of the VoiSS ("My voice is hoarse? "- adapted), with the highest accuracy in relation to the others (83,4%), representing the most efficient model in the identification of dysphonic individuals. The results of this study allow the conclusion that the use of a decision rule to identify dysphonia, based on only two questions self-referenced by the patient, represents an alternative and efficient resource for population screening, which can be applied and analyzed in future research.