Development and validation of an epidemiological risk score for neonatal death in a middle-income country

Kelsy N. Areco^1*Paulo Bandiera-Paiva¹Rita C. X. Balda¹Daniela Testoni¹Ana Sílvia Scavacini Marinonio¹Adriana Sanudo¹Mandira Daripa Kawakami¹Milton H Miyoshi¹Carina Nunes Vieira¹Rosa M V Oliveira²Monica La Porte Teixeira²Bernadette Teixeira²Carlos Kiffer¹Maria Fernanda de Almeida¹Ruth Guinsburg¹Tulio Konstantyner¹

• ¹Paulista School of Medicine, Federal University of São Paulo, São Paulo, São Paulo, Brazil
• ²Fundacao Sistema Estadual de Analise de Dados, São Paulo, Brazil

Introduction. Despite global reductions in neonatal mortality, significant disparities remain between regions and population groups. Identifying newborns at higher risk at birth may help direct preventive actions and enhance health planning. Objective. To develop an epidemiological risk score for neonatal death based on individual and contextual factors. Methods. A cohort study was conducted using data from over 5.6 million live births in 645 municipalities of São Paulo State, Brazil, between 2009 and 2018. The outcome was neonatal death. Risk weights were calculated from adjusted odds ratios obtained through multilevel logistic regression, with coefficients transformed using the natural logarithm and scaled from 1 to 10. Internal validation was performed within the cohort; external validation used data from 2008. Results. Points were assigned to congenital anomalies (4, 7, or 10 depending on severity), preterm vaginal birth (4), preterm cesarean (4), birthweight <2500 g (4), and fewer than seven prenatal visits (3). Conditions assigned 1 point included male sex, maternal age <17 or ≥40 years, term cesarean, birth in spring/summer, multiple pregnancy, low municipal nurse density in public services, and low municipal health insurance coverage. The area under the ROC curve (AUC) was 0.83 for internal and 0.81 for external validation. Risk stratification thresholds were proposed based on total points. Conclusion. This score combines routinely collected individual and municipal-level data in Brazil to classify neonatal death risk. It may support clinical prioritization, resource allocation, and identification of low-risk deaths, complementing individualized clinical assessment.