Predição de síndrome metabólica em adolescentes com obesidade utilizando técnicas de aprendizado de máquinas

Abstract

The prevalence of obesity and overweight among children and adolescents has increased significantly over the last four decades. Since excess weight is considered the main risk factor for the development of metabolic syndrome (MS), more and more children and adolescents are being diagnosed with this condition. To prevent metabolic syndrome in this age group, the development of predictive models to identify potential high-risk individuals is of great value. The objective of this study is to develop an intelligent system based on machine learning capable of stratifying the risk of adolescents with metabolic syndrome through clinical and laboratory parameters. This research is a clinical, descriptive, and cross-sectional study. The sample consisted of 104 adolescents, including 52 with MS and 52 without MS, selected from a database of 235 electronic medical records of adolescents monitored at the Pediatric Endocrinology outpatient clinic at the University Hospital of the Federal University of Ceará (HU UFC) from January 2018 to October 2023. Sociodemographic data, dietary habits, lifestyle behaviors, clinical and laboratory indicators, and anthropometric parameters were assessed. The study was approved by the Research Ethics Committee of the University Hospital Walter Cantídio with CAAE number: 70563423.0.000.5045. The study showed that 58% (n = 30) of the male participants had MS, while 42% (n = 22) were female. Regarding BMI classification, 55% had severe obesity. The exclusive breastfeeding rate was 67%. Screen time exposure was over 2 hours in 88% of the participants. Only 27% of the patients engaged in physical activity. Obesity was present in both parents in 44% of the cases. The average Abdominal Circumference (AC) and AC-to-E ratio were 104 ± 14 cm, and the neck circumference was 39.0 ± 3.9 cm. It was observed that 84% had acanthosis, 41% had hypertension, and more than 70% of the participants had dyslipidemia. The Linear SVM algorithm proved to be a good predictor for MS screening, with gender (male and female), age, BMI, abdominal circumference, hypertension, systolic blood pressure (SBP), and diastolic blood pressure (DBP) as the most relevant variables. The algorithm's performance was evaluated based on sensitivity (0.6), specificity (0.73), accuracy (0.67), and ROC curve (AUC = 0.72). The computational model employed in this study, the Linear SVM, showed better performance in identifying metabolic syndrome, and it was used to build a mobile application for Android operating system devices.