A two-tiered latent class and spatial analytical approach to identify clusters of neonatal mortality among very low birth weight infants: A population-based cohort study.

Abstract

To identify and analyze patterns of neonatal deaths among very low birth weight (VLBW) infants in the most socioeconomically developed state of Brazil, from 2004 to 2020, using a two-tiered probabilistic approach that combines Latent Class Analysis (LCA) and spatial analysis. This historical population-based cohort study included 137,224 live births with birthweight of 400-1499g to mothers residing in São Paulo State, using linked birth and death certificate data. Among 42,230 neonatal deaths, five distinct latent classes were identified: infection-dominant, intrapartum event-dominant, malformation-dominant, respiratory-dominant, and other. Survival analysis showed differences in timing of death across classes, with intrapartum-related deaths concentrated in the first hours of life, and infection-related deaths occurring later. Spatial analysis revealed geographic clustering especially for infection, malformation, and respiratory-related deaths, primarily in southern municipalities of the State. The combined use of LCA and spatial analysis identified distinct patterns of neonatal mortality. LCA differentiated clinically meaningful profiles with specific timing of death, while spatial analysis revealed municipal-level clustering and overlap of these patterns. These findings showed how neonatal mortality is shaped by both diagnostic profiles and territorial context, providing actionable evidence to guide targeted improvements in perinatal and neonatal care to reduce preventable deaths among VLBW infants.