Classificação de risco neonatal usando aprendizado de máquina e dados dos sistemas de informação de saúde pública e de censo demográfico brasileiro

Abstract

The infant mortality rate is considered one of the most critical indicators of society. From it, it is possible to portray the living conditions of a given country or region, where high indices may reflect a low social and economic development level of the assessed population. Despite the improvement in recent years, Brazil still needs to improve in reducing this index, especially in some regions of the country that lack infrastructure and face more severe socioeconomic problems. Inserted in the composition of infant mortality, studies show that the largest share of deaths come from the neonatal component, requiring greater attention from the public authorities in this segment. Thus, predicting the risk of a baby dying in its first days of life can positively impact the public health system and, consequently, Brazilian society. This work uses machine learning techniques and data from health information systems and the Brazilian demographic census to generate classifiers that make it possible to issue an alert to the health system in case of risk of neonatal death, directing attention to maternal and newborn follow-up. The classifiers were trained using a large dataset by linking records from five databases. In addition, the data set was enriched with new variables allowing a broader study of the factors that influence neonatal mortality. The results reveal f1-score and AUC rates above 89% in the neonatal death risk classification, showing that the proposed classifier is viable as one more resource for preventing neonatal deaths and improving the health system.