Maternal And Assistance Profile Associated With Neonatal Mortality Preventable Underlying Causes Of Death Using Machine Learning And Administrative Health Data

Infant mortality is a reflection of analysis including biological, socioeconomic, and assistance factors and an analytical analysis of this problem implies the processing of large datasets from different areas. Data Science approaches have become increasingly widespread to deal with problems that require large datasets to perform deep analysis. Machine learning methods have become popular due to their efficiency and efficacy in discovering knowledge by identifying patterns in feature interactions of large datasets. This work proposes the use of a machine learning approach to evaluate the association between maternal and assistance features, with neonatal mortality by preventable underlying causes of death. For this, demographic and epidemiological data from the Brazilian public health births and mortality information systems were used (SINASC and SIM, respectively). Using the unsupervised clustering algorithm K-Modes, clusters are created based on patterns identified in the dataset, and then, an analysis is done to evaluate the socio-demographic profile of each cluster. In this way, it is possible to evaluate the differences between the profiles of each cluster. The profile defined in this paper includes the following features: maternal age, maternal years of schooling, race, public or private assistance, number of consultations, and date of first prenatal consultation. The analysis was performed using data from the period between 2012 and 2017, for the city of São Paulo, one of the wealthiest regions of the country. The data quality for this region in this specific period is considered to be very high, so there is no need of applying demographic methods to correct births or deaths. Besides that, the dataset has only categorical data, the method selected does not require hard data assumptions and is suitable for categorical data. Considering that this is a data-driven approach, preliminary results indicate that just a few assumptions can be made on the profile using the selected features. However, some associations between variables and neonatal mortality by preventable underlying causes can be identified. We hope to encourage reflection on the newborn in the socioeconomic environment and contribute to public health policies.