SUPPORTING DECISION MAKING IN HEALTH SCENARIOS WITH MACHINE LEARNING MODELS

Machine Learning (ML) techniques have been employed in data analysis for generating models able to support health care and management decisions, but in Brazil there is still a lack of adequate data collection procedures in hospital environments, preventing a proper data analysis. Here we discuss some approaches adopted to deal with these challenges and carry out analyzes of hospital data to support reliable decision-making. From data stored in the COVID Data Sharing/BR
repository we set up two datasets for severity prediction of COVID-19 cases upon admission parameters. We explain decisions made in data preprocessing and discuss some results of the ML models, including the possibility of generalizing these models across different hospitals. The best AUC values achieved were around 0.75, a value lower than that reported in related literature, which can be explained by the absence of patient clinical information from the raw public repositories.