Smart-GISSA : um Sistema para Governança em Saúde Digital Baseado em Aprendizado de Máquina

• Ano de defesa: 2021
• Autor(a) principal: Costa Filho, Raimundo Valter
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Palavras-chave em Português: sistemas de saúde. mineração de dados. aprendizado de máquina. medição de risco. epidemias.; Mineração de dados; Aprendizado de máquina; Medição de risco; Epidemias
• Link de acesso: http://www.repositorio.ufc.br/handle/riufc/60257

Decision making using Machine Learning mechanisms can incur considerable improvement in the public health system. For example, predictions that allow managers to adopt preventive measures, mitigating, if not preventing, outbreaks of classic diseases. The system Governança Inteligente em Serviços de Saúde (GISSA) extracts, transforms and loads data into dashboards, from Brazilian Ministry of Health databases, for decision-making. GISSA, which was implemented by the Atlantic Institute, with the support of FINEP, is still being developed by researchers at UFC, Fiocruz and IFCE and is operational in several municipalities in the Brazil. In this context, this work presents the Smart-GISSA , a system for governance in Digital Health based on Machine Learning, which is an evolution of the GISSA architectural model. Smart-GISSA implements a layered architecture model following the data chain from capture to availability in databases to enable the emergence of applications in Machine Learning. This research describes new features added to GISSA from the analysis of its first solution built upon ontology and linked data techniques, proposing a new architecture that is more efficient and effective in health governance. Two new Data Mining methodologies are proposed with a focus on risk of death analysis and on epidemiological surveillance to predict epidemics. As a first case study, to better illustrate the adoption of the practices advocated by the methodology, it is built and validated two models for the risk of death, one maternal and the other infant, useful in identifying risky pregnancies accompanied by family health teams . Maternal and infant risk analyzers demonstrate the ability to alert risk of death in 97.50% of cases, considering 15 features, and 99.82% of cases, considering 27 features, respectively. For the second case study, it is built a model for predicting dengue epidemics for the city of Fortaleza, CE, inferring the number of cases of infection in the metropolitan region. The results show that it is possible to detect the trend in the number of new infections with a forecast horizon of 15 weeks. Additionally, the proposed methodological process can model any other type of epidemic, e.g., tuberculosis, cholera, COVID-19, among others.