908-P: Predictive Model for Glycemic Level of People with Diabetes Mellitus Using Data Extracted from Electronic Medical Records in Primary Care

Abstract

Background: In Brazil, the use of electronic medical records in primary care has been expanding, improving the supply of information and data storage, and allowing the analysis of various aspects and prediction of future outcomes. Aim: To develop a predictive model for the glycemic level of people with type 2 diabetes mellitus based on data from electronic medical records in primary care. Methods: Data mining techniques were applied to choose response variables and potential predictors. Afterwards, data modelling was performed using Artificial Neural Networks (ANN), which are mathematical models inspired by the neural structure of intelligent organisms that acquire knowledge through experience. They detect non-linear relationships between the response variable and the explanatory variables without these variables having been defined. Results: The highest probability of model accuracy was observed in the capillary blood glucose range between 100 and 300mg/dL. Model predictors and relative importance of each variable are shown in Figure 1. Conclusion: Applying predictive modelling to data available in primary care electronic medical records can help the early identification of individuals with difficulty in glycemic control, and increase the efficiency in the allocation of efforts to treat diabetes. Disclosure S. Mistro: None. T.V.O. Aguiar: None. V.V. Cerqueira: None. K.O. Silva: None. J.A. Louzado: None. C.N. Kochergin: None. D.A. Soares: None. W.W. Amorim: None. D.S. Medeiros: None. V.M. Bezerra: None. V.H. Carvalho: None. E. Amaro: None. M.G. Oliveira: None. M.L. Cortes: None. Funding SUS Institutional Development Support Program; Brazil Ministry of Health; Israelite Hospital; Albert Einstein (25000.028646/2018-10); Medtronic Foundation