Predictive analysis of metabolic syndrome based on 5-years continuous physical examination data

Abstract

Metabolic syndrome (MetS) represents a complex group of metabolic disorders. As MetS poses a significant challenge to global public health, predicting the occurrence of MetS and the development of related risk factors is important. In this study, we conducted a predictive analysis of MetS based on machine learning algorithms using datasets of 15,661 individuals. Five consecutive years of medical examination records were provided by Nanfang Hospital, Southern Medical University, China. The specific risk factors used included WC, WHR, TG, HDL-C, BMI, FGLU, etc. We proposed a feature construction method using the examination records over the past four consecutive years, combining the differences between the annual value and the normal limits of each risk factor and the year-to-year variation. The results showed that the feature set, which contained the original features of the inspection record and new features proposed in this study yielded the highest AUC of 0.944, implying that the new features could help identify risk factors for MetS and provide more targeted diagnostic advice for physicians.