Banca de QUALIFICAÇÃO: SAYONARA CRISTINA DE OLIVEIRA MAGALHÃES

Uma banca de QUALIFICAÇÃO de MESTRADO foi cadastrada pelo programa.
STUDENT : SAYONARA CRISTINA DE OLIVEIRA MAGALHÃES
DATE: 11/09/2026
TIME: 09:30
LOCAL: meet.google.com/pys-gbtx-rrm
TITLE:

Predictive Model for Mortality in Preterm Infants

KEY WORDS:

Prematurity; infant mortality; machine learning; predictive model; XGBoost; temporal validation; SINASC; SIM.

PAGES: 50
BIG AREA: Engenharias
AREA: Engenharia Elétrica
SUMMARY:

Prematurity is one of the leading causes of infant mortality worldwide and remains an important challenge for Brazilian public health, reflecting regional inequalities, socioeconomic conditions, and disparities in access to prenatal and perinatal care. The early identification of preterm infants at higher risk of death may contribute to intensive monitoring, care prioritization, and the rational allocation of resources within the Brazilian Unified Health System (SUS). In this context, this study develops and evaluates an XGBoost-based predictive model to estimate the risk of mortality among preterm infants up to 365 days after birth. The study uses a national cohort obtained by linking data from the SINASC and SIM information systems, comprising 3,064,338 birth records from 2014 to 2022, with a mortality prevalence of 1.78%. The methodology includes class imbalance handling, stratified cross-validation, prospective temporal validation, and decision-threshold optimization guided by clinical sensitivity. Threshold selection was analyzed according to three operational criteria: recall maximization subject to a minimum precision constraint, F1-score maximization, and Youden’s index. Model performance was assessed using discriminative metrics, such as AUC-ROC and AUPRC; operational metrics, including precision, recall, F1-score, and MCC; and prioritization metrics, such as Lift and cumulative gain. Temporal stability was investigated through annual walk-forward validation using six expanding windows from 2017 to 2022, including the COVID-19 pandemic years. A systematic data-drift analysis was also conducted across the 26 predictor variables using the Kolmogorov-Smirnov test and the Population Stability Index (PSI). The results indicate that the model has the potential to support the identification of preterm infants at higher risk of mortality, contributing to the guidance of healthcare interventions and the prioritization of public health resources.

COMMITTEE MEMBERS:
Externo à Instituição - GABRIEL BEZERRA MOTTA CÂMARA - UFRN
Externa à Instituição - LUCILEIDE MEDEIROS DANTAS DA SILVA - IFRN
Presidente - 1837240 - MARCELO AUGUSTO COSTA FERNANDES