DiaPreReApp

Main Objective

Diabetes is a major global health problem: 592 million adults and a cost of USD 490 billion are estimated by 2030. Portugal had in 2011, 11.7% of the population from 20 to 79 years with the disease, with numbers for 2015 indicating 13.3% prevalence. Correct management of diabetes leads to a normal life, without complications. Unfortunately, if glycaemic values are not controlled, diabetes can seriously decrease the quality of life. In the worst cases, the disease will lead in the long term to amputations, blindness or heart problems. Type 1 diabetic patients (insulin dependent) require a very strict control of their disease. They should collect data on glucose levels, carbohydrate intake, insulin, exercise, stress, illness and other factors that influence glycaemia. However, real data may ignore features and context, such as life-style habits or health history. The most problematic short-term issue with glycaemic control is hypoglycaemia, i.e., a value of blood glucose that is below the acceptable health threshold (70 mg/dL). This may lead to blurred vision, confusion, erratic behaviour, garbled speech, clumsiness, seizures and even loss of consciousness. Our project aims to predict possible future hypoglycaemias for a type 1 diabetic patient, using the records registered in a mobile application. Using machine-learning models, trained with available data sets, we can build a generic model. However, using the specific data from the patient and its context the model can be more accurate, with a higher correct prediction ratio. We aim to embed a predictive framework using data imputation, data fusion and data mining techniques on a smartphone application. We will base our work on the currently developed MyDiabetes application that uses a rule-based system (RBS) to provide diabetics with management advices. We will also use this RBS system to, when a hypoglycaemia is predicted, infer the possible causes and advise the user. This inference will be built on logic rules from medical knowledge and evolve to data mining approaches. This work will involve researching the appropriate ensemble of predictive models to use in a mobile environment. Testing and validating the framework will encompass real-data from selected cohorts of French and Portuguese patients. The project involves researchers from Covilhã with expertise on the machine learning aspects related to prediction and researchers from Porto with work on the mobile application and RBS systems. Both groups have connections with medical experts, needed to provide the knowledge regarding the disease.