Triagem de pacientes com depressão na atenção primária usando ferramentas de aprendizado de máquina

Abstract

Depression is an omnipresent disease, with a high prevalence in the world, high economic burden and generally associated as a risk factor for absenteeism at work and suicide. On the other hand, the advent of the COVID-19 pandemic brought a myriad of challenges and obstacles to be overcome, such as unemployment, social isolation, the obligatory use of a mask, the need for a respiratory tag, the impossibility of attending different places for leisure and entertainment, the reduction of contact with family members, the loss of loved ones. All of them are examples of these important challenges regarding the promotion of mental health problems: this panorama that has increased (three times) the prevalence of depressive symptoms. The treatment of this disease involves the use of several methods, such as the use of antidepressants and psychotherapy. However, many individuals do not receive adequate treatment, simply because they remain undiagnosed. In this context, in this work Machine Learning tools were used to assess their performance regarding the diagnostic screening of patients with undiagnosed depression. Sociodemographic, clinical and laboratory data collected from 2016 to 2018 by the Cardiovascular Disease Research Network were used and eight different models were tested, namely, Logistic Regression, K-Nearest Neighbors, Classification and Regression Trees, Adaptive Boosting, Gradient Boosting, Extreme Gradient Boosting, CatBoost, Support Vector Machine and Random Forest. Support Vector Machine had the best performance, having obtained an area below the receiver operating characteristic curve equal to 0.74 and accuracy of 0.77. From the treatment point of view, it represents a new screening tool, which can assist in reducing the number of undiagnosed cases of the disease, in addition to facilitating early treatment initiation.