Análise preditiva de variáveis de sono e biomarcadores de comorbidades na detecção de sintomas depressivos em adultos de meia-idade e idosos: uma abordagem de aprendizagem de máquinas

Abstract

Introduction: studies have shown high incidences of depression worldwide and its cooccurrence with several important medical conditions, especially in middle-aged and elderly subjects. In this multimorbidity scenario, depression is commonly associated with diseases related to metabolic syndrome, such as obesity and diabetes. Chronic alterations in circadian sleep-wake rhythm represent a relationship with the development of depression and its associated comorbidities, as they favor the breakdown of the internal temporal organization of essential physiological and metabolic processes. Currently, making accurate clinical diagnoses and screenings have been a persistent challenge in mental health, due to the use of limited traditional tools that do not include additional characteristics of important clinical data of the patient, including objective observations of disease biomarkers. Objective: Thus, the objective of the present study was to detect depressive symptomatology from general biomarkers of obesity and diabetes, as well as variables related to sleep and physical activity, in middle-aged and elderly adults, through a learning approach of machines. Method: Data from the Global Physical Activity Questionnaire (GPAQ - physical activity level), from the Patient Health Questionnaire (PHQ-9), and from the sleep habits questionnaire were extracted from the National Health and Nutrition Examination Survey database (NHANES) in the period 2015-2016. Other variables were accessed and used as predictive resources, such as anthropometric measurements and plasmatic biomarkers of obesity and diabetes. A total of 2907 middle-aged and elderly adult subjects were eligible for the study. Three supervised learning algorithms were implemented: Penalized Logistic Regression with Lasso (RL), Random Forest (RF) and Extreme Gradient Boosting (XGBoost). Results: The XGBoost model provided greater accuracy and precision (87%), with a proportion of correct answers in cases with depressive symptoms above 80%. In addition, daytime sleepiness was the predictor variable that best contributed to predicting depressive symptoms. Conclusions: Sleep and physical activity variables, in addition to obesity and diabetes biomarkers, together assume significant importance in predicting, with an accuracy and precision of 87%, the occurrence of depressive symptoms in middle-aged and elderly individuals.