Loading...
| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 36085 | 1.27 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
O presente trabalho, propõe desenvolver e validar um modelo de risco de condições de saúde para prever a probabilidade de um indivíduo desenvolver uma ou múltiplas Doenças Crônicas Não Transmissíveis (DCNT), antes que elas se manifestem. Inúmeras pessoas numa escala global correm o risco de desenvolver DCNT. Em concreto, os principais grupos de DCNT, incluem: doenças cardiovasculares, câncer, respiratórias crônicas e diabetes. Os grupos de doenças crônicas com características multifatoriais e não-infecciosas em sua origem têm sido motivo de crescente preocupação da sociedade e governos de todo o mundo, por colocar as pessoas em maior risco de complicações, chegando até mesmo a óbito, e colocando os sistemas de saúde em crise sistémica. A crise provocada pela COVID-19 acelerou o setor da saúde não só no sentido de repensar e reorientar a prestação de cuidados, mas também priorizar a integração da predição, prevenção e gestão de DCNT na população. Essa crise, mostrou também que os modelos preditivos e de prevenção e gestão de doenças crônicas em uso são insuficientes nas respostas as situações de saúde de condições crônicas. Uma abordagem multifatorial com fatores de riscos desencadeadores e verticalizados: Fatores de Risco Modificáveis (FRM) e Fatores de Risco Não Modificáveis (FRNM) é preferível para reverter essas situações de saúde e fornecer respostas específicas da probabilidade de um indivíduo desenvolver determinadas doenças crônicas. Nesse contexto, esse trabalho tem como objetivo principal “Desenvolver uma Plataforma Inteligente de Predição do Risco de Doenças Crônicas (PIPRDC) com o propósito de apoiar os profissionais da saúde”. O PIPRDC tem como fundo pesquisas na literatura, consultas a médicos especialistas e um modelo Preditivo de Risco de DCNT. Utilizamos classificadores de modelos multi-label ET, RF e DT para prever dez tipos de DCNT simultaneamente. Entre os modelos do experimento, o MLC RF apresentou o melhor desempenho de precisão e F1-score com 96,16% e 90,48%, respectivamente.
The present work proposes the development and validation of a health conditions risk model to predict the probability of a specific individual to develop one or multiple Chronic Non-Communicable Diseases (CNCD), before they manifest themselves. Countless people on a global scale are at risk of developing CNCD. Specifically, the main groups of CNCD include: cardiovascular diseases, cancer, chronic respiratory diseases and diabetes. The groups of chronic diseases with multifactorial and non-infectious characteristics in their origin have been a reason for growing concern of society and governments around the world, as they put people at greater risk of complications, even leading to death, and putting systems at risk. Both related with health and economic systemic crisis. The crisis caused by COVID-19 has accelerated the health sector not only to rethink and reorient the provision of care, but also to prioritize the integration of prediction, prevention and management of CNCD in the population. This crisis also showed that the predictive models and the prevention and management of chronic diseases in use are insufficient in responding to health situations of chronic conditions. A multifactorial approach with triggering and verticalized risk factors (Modifiable Risk Factors (FRM) and Non-Modifiable Risk Factors (FRNM) is preferable to revert these health situations and provide specific responses to the probability of an individual developing certain chronic diseases. As such, this work has as its main objective to “Developing an Intelligent Platform for Predicting the Risk of Chronic Diseases (PIPRDC) with the purpose of supporting health professionals.” The PIPRDC is based on research in the literature, consultations with medical specialists and a model CNCD Risk Predictive. We used ET, RF and DT multi-label model classifiers to predict up to nine types of CNCD simultaneously. Among the experimental models, the MLC RF showed the best performance of accuracy and F1-score with 96.16% and 90.48%, respectively.
The present work proposes the development and validation of a health conditions risk model to predict the probability of a specific individual to develop one or multiple Chronic Non-Communicable Diseases (CNCD), before they manifest themselves. Countless people on a global scale are at risk of developing CNCD. Specifically, the main groups of CNCD include: cardiovascular diseases, cancer, chronic respiratory diseases and diabetes. The groups of chronic diseases with multifactorial and non-infectious characteristics in their origin have been a reason for growing concern of society and governments around the world, as they put people at greater risk of complications, even leading to death, and putting systems at risk. Both related with health and economic systemic crisis. The crisis caused by COVID-19 has accelerated the health sector not only to rethink and reorient the provision of care, but also to prioritize the integration of prediction, prevention and management of CNCD in the population. This crisis also showed that the predictive models and the prevention and management of chronic diseases in use are insufficient in responding to health situations of chronic conditions. A multifactorial approach with triggering and verticalized risk factors (Modifiable Risk Factors (FRM) and Non-Modifiable Risk Factors (FRNM) is preferable to revert these health situations and provide specific responses to the probability of an individual developing certain chronic diseases. As such, this work has as its main objective to “Developing an Intelligent Platform for Predicting the Risk of Chronic Diseases (PIPRDC) with the purpose of supporting health professionals.” The PIPRDC is based on research in the literature, consultations with medical specialists and a model CNCD Risk Predictive. We used ET, RF and DT multi-label model classifiers to predict up to nine types of CNCD simultaneously. Among the experimental models, the MLC RF showed the best performance of accuracy and F1-score with 96.16% and 90.48%, respectively.
Description
Relatório apresentado à Universidade Fernando Pessoa como parte dos requisitos para o cumprimento do programa de Pós-Doutoramento em Ciências da Informação
Keywords
Doenças crônicas Predição Prevenção Covid-19 Risco Modelo Chronic diseases Prediction Prevention Risk Model