Study of machine learning techniques applied to the health area to minimize negative hospital outcomes in a Brazilian public hospital: Hospital de Clínicas de Uberlândia. Funded by Fundação de Amparo à Pesquisa de Minas Gerais (FAPEMIG) -- a public agency for funding research.
The decline in nutritional status during hospitalization, as well as the long hospital stay and high rates of mortality and hospital readmission, are still a reality in many Brazilian hospitals. Artificial intelligence through the Machine Learning (ML) technique can be a potential prediction tool of greater scope than the classic investigation of risk factors in observational studies, and can contribute to minimizing negative outcomes, reducing hospital costs and possibly improving the quality of service provided.
To develop predictive algorithms and models for factors associated with hospital mortality and readmission, length of stay and malnutrition through the use of artificial intelligence via ML, using data available in the Hospitalization System of a university hospital (SIH). Methods: A retrospective documentary-based study will be carried out, including medical records of patients admitted to a university hospital. Demographic and nutritional data, medical diagnoses, biochemical tests, admission and discharge dates, as well as death records will be automatically extracted from the SIH. The data will be submitted to pre-processing and data processing, with identification of predictor variables for each model, and subsequent construction of prediction algorithms. Expected results: It is estimated that 89,000 admissions were obtained during the collection period. More accurate prediction algorithms will be developed for the aforementioned negative outcomes, which will support clinical decisions, allowing the health team to early identify the patient at higher risk of mortality, readmission, prolonged hospitalization or malnutrition. In this way, the proposal may help to recognize patients who are more likely to develop such unfavorable outcomes. Public health system resources are known to be limited, so identifying and prioritizing care at higher risk of negative events can improve quality and reduce the costs of services provided.
Nayara Cristina Da Silva - http://lattes.cnpq.br/5510573560249365
Geórgia das Graças Pena - http://lattes.cnpq.br/4569169833604734
André Ricardo Backes - http://lattes.cnpq.br/8590140337571249
Marcelo Keese Albertini - http://lattes.cnpq.br/1404596833493304