Smart and personalized sysmtem to predict blood glucose levels and recommend preventive actions to reduce glycemic variability in DM1 and DM2 patients

• Grant number: 18/08056-9
• Support Opportunities: Research Grants - Innovative Research in Small Business - PIPE
• Start date: April 01, 2019
• End date: March 31, 2021
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computer Systems
• Principal Investigator: Yuri Kendji Matsumoto
• Grantee: Yuri Kendji Matsumoto
• Company: Glucogear Serviços em Tecnologia Ltda. - ME
• CNAE: Desenvolvimento e licenciamento de programas de computador customizáveis ; Tratamento de dados, provedores de serviços de aplicação e serviços de hospedagem na internet
• City: São Paulo
• Associated researchers: Camilo Rodrigues Neto ; Gabriel de Souza Pereira Moreira ; Maria Cristina Foss de Freitas ; Rafael Ferraz Bannitz ; Rebeca Antunes Beraldo
• Associated scholarship(s): 20/01765-4 - Smart and personalized system to predict blood glucose levels and recommend preventive actions to reduce glycemic variability in DM1 and DM2 patients, BP.TT; 19/15061-1 - Smart and personalized system to predict blood glucose levels and recommend preventive actions to reduce glycemic variability in DM1 and DM2 patients, BP.TT; 19/15062-8 - Smart and personalized system to predict blood glucose levels and recommend preventive actions to reduce glycemic variability in DM1 and DM2 patients, BP.TT

Abstract

Diabetes Mellitus (DM) is considered one of the main current pandemics, the International Diabetes Federation (IDF) estimates more than 424 million adults living with DM in 2017 and about 12.5 million in Brazil. In cases where hyperglycemic condition remains for long periods, patients may present other complications such as retinopathies, nephropathies and neuropathies (27). Recently, the analysis of glycemic variability (VG) using physical activity sensors and continuous monitoring of glycemic rates has shown good potential for the development of predictive models for blood glucose levels, so that patients can not only treat diabetes, but also prevent adverse events (10, 11). Due to the continuous inflow of large data volume and the effect of several random factors such as idiosyncrasies, physical activity, meal choice, among others, it is necessary to develop customized models for this type of prediction (8,9). In this sense, some projects have arisen using artificial intelligence and mathematical prediction models (10). The objective of this project is to develop and validate an intelligent system of personalized glycemic prediction, as well as to develop a system of recommendations to reduce VG, using as input data glycemia, diet and physical activity values, among other clinical indicators. Phase 1 of the research has already been carried out in full, in which an application called GlucoTrends (www.glucotrends.com.br) was developed for data collection and usability learning and also a predictive model of glycemia based on Recurrent Neural Networks - RNN) (21) was developed and tested. For the technical-scientific feasibility of the proposed solution, an in vitro pilot and another in vivo pilot were performed to test the accuracy of the predictive model resulting from RNN. Although the accuracy of the in vitro model was higher, the predictive model, even in vivo, for a time horizon of up to 1 hour, presents a relevant accuracy with a high potential for application in clinical practice. Thus, Phase 1 of technical-scientific feasibility is completed and was considered satisfactory by the project team, and Phase 2 of the project can be started. For Phase 2, an observational, cross-sectional study will assess the glycemic variability of patients with DM1 and DM2 in different food intake and physical activity situations. The research will be coordinated by Professor Maria Cristina Foss-Freitas and held at the Clinical Hospital of the Ribeirão Preto Medical School (HC / FMRP-USP). Continuous glycemic monitoring sensor in 60 individuals with DM will undergo clinical and nutritional evaluation, should register their daily activities for 28 days using the application GlucoTrends. Blood glucose monitoring will be performed with a continuous glucose monitoring device. The results obtained will be applied to improve the RNN developed by the team. After the accuracy test of the predictive models, a personalized and automated recommendation system for insulin dose calculation for DM1 with a focus on reducing glycemic variability and an automated and personalized nutritional plan recommendation system for DM2 will be developed. After development and validation, the systems will be integrated into a smartphone application developed by the company that will be offered as a platform for access to predictive and preventive technology for patients with DM. (AU)