P03-222 - Predictive Models of Metabolic Syndrome for Patients with Schizophrenic or Bipolar Disorders

Abstract

IntroductionMetabolic syndrome is a frequent, severe, undiagnosed physical comorbidity in patients with severe mental disorders.AimTo develop a predictive model of metabolic syndrome for patients with schizophrenic or bipolar disorders, useful for both clinical practice and research.MethodsNaturalistic, one-year follow-up study conducted in Asturias, Spain. A total of 172 patients with schizophrenic (Sch-P) or bipolar (BD-P) disorders (ICD-10 criteria), under maintenance treatment, who gave written informed consent were included. Metabolic syndrome was defined according to the modified NCEP ATP-III criteria. Multivariate Adaptive Regression Splines (MARS), Genetic Algorithms (GA), and Support Vector Machine (SVM) analysis were performed.ResultsStarting from a large set of demographic and clinical variables, and by means of intermediate MARS and GA models, an SVM model able to classify if a patient with schizophrenia or bipolar disorder suffers from metabolic syndrome with an accuracy of 98.68% (sensitivity 100%, specifity 94.4%) was obtained. The final model only needs 6 variables: Sch-P:(1)Low HDL-cholesterol,(2)Fasting glucose level,(3)Family history of obesity,(4)Triglyceride level,(5)Family history of dyslipidemia, and(6)Use of antidepressants; BD-P: (1), (2), (3),(7)Use of lipid-lowering medication,(8)Use of antipsychotics, and(9)Use of mood stabilizers.ConclusionWe developed a simple and easy to use predictive model to identify metabolic syndrome in patients with schizophrenic or bipolar disorders.