lookAfterRisk - Dynamic risk assessment using home-mobile technologies for the management of myocardial infarction patients

Description

======== GOALS: This project combines the development of prediction risk assessment models with home-mobile technologies to improve the management of myocardial infarction (MI) patients. The scientific goal is the use of computational intelligence methodologies for the development of personalized, interpretable and dynamic models in the cardiovascular risk assessment of acute events, namely death and re-hospitalization. The models are applied at hospital admission, in patients with a first episode of acute MI and are continuously updated when the patient returns home. The result is an integrated tele-monitoring platform, merging the advantages of the continuous monitoring provided by low cost mobile technologies with prediction models, achieving the implementation of preventable strategies in supporting the decision of professionals. The project lasts 36 months, and the team is composed by two institutions Universidade Coimbra (UC) and Centro Hospitalar Leiria (CHL), and have the commitment of Sociedade Portuguesa Cardiologia (SPC). ======== MOTIVATION: Risk prediction tools are decisive for cardiovascular (CV) risk assessment, although they present some limitations. They are average models, derived for a general population, hindering the assessment of a specific patient. Personalized models, with the ability to combine clinical evidence (embedded in existent risk models) with new information (recent datasets) and patient’s specific information would represent a significant impact on the CV risk evaluation. Moreover, the interpretability of the models is critical to increase physicians’ confidence, while facilitating the integration of the different sources of information. On the other hand, current risk tools are static, i.e. they do not allow the incorporation of new risk factors and do not consider the risk factors’ time evolution. Dynamic strategies, supported on home monitoring solutions, able to continuously incorporate new information into the initial risk assessment, will contribute to improve the accuracy of such models. Additionally, the evolution over time may reflect different disease progressions, indicative of risk evolution. ======== METHODS: The extraction of knowledge from datasets is based on computational intelligence techniques, namely clustering based methodologies (patients with similar behavior/characteristics). The integration of different sources of information is explored considering ensemble models and rule based approaches, while guaranteeing interpretability. Concerning dynamical characteristics, regression models, able to incorporate time, and similarity strategies capable to compare and discover in the historical conditions patterns similar to the current condition, are explored. ======== VALIDATION: The models significance is assessed in a first phase (hospital admission) using the largest MI Portuguese dataset (N=16000), in collaboration with the SPC. The second phase is based on a home telemonitoring observational study, involving 50 patients (admitted with a first episode of acute MI) that left the CHL. The study (9 months), collects blood pressure, heart rate, cholesterol and glycaemia on a regular basis, through commercial devices and laboratory exams.

Researchers

Funded by

FCT - Portuguese Science Foundation

Partners

UC, Centro Hospitalar de Leiria; Sociedade Portuguesa de Cardiologia

Total budget

232 836,00 €

Local budget

213 386,00 €

Keywords

Computational Inteligence; Data fusion, interpretability and prediction strategies; Cardiovascular models

Start Date

2018-07-16

End Date

2021-07-16