Beyond single cause-of-death labels: A secondary analysis of outcome-specific latent-class phenotyping and marker-supported reclassification of stillbirths and neonatal deaths in Jordan.

Abstract

ObjectivesConventional perinatal death review assigns a principal cause despite overlapping pathways. We examined whether outcome-specific latent classes could complement conventional classification and whether structured markers could inform review of unexplained deaths.MethodsIn this secondary analysis, 588 expert-adjudicated deaths from 18 Jordanian hospitals (261 stillbirths, 327 neonatal deaths) were analyzed. Gestational age was modeled as mutually exclusive categories (<32, 32-36, ≥37 weeks) with eight outcome-specific binary markers. Model selection considered Bayesian information criterion (BIC), classification quality, parsimony, and stability to hospital omission and one-death-per-pregnancy resampling. Expert care-quality and preventability judgments were auxiliary outcomes. Marker hierarchies were applied to 192 unexplained/unclassifiable deaths.ResultsStillbirths supported one class (BIC 2357.4), stable in all sensitivity analyses. For neonatal deaths, two- and three-class BIC values were nearly identical (3398.8 vs 3397.6); the two-class solution had better classification quality and stability and was retained. It comprised a preterm respiratory/multiple-birth phenotype (n = 204) and a later-gestation congenital-asphyxial/sepsis phenotype (n = 123); definite preventability was 20.1% and 46.7%, respectively. Marker hierarchies assigned 149/192 deaths (77.6%) to marker-supported groups, leaving 43 (22.4%) unexplained.ConclusionsAvailable stillbirth indicators did not support latent subclasses. Neonatal deaths showed clinically interpretable heterogeneity, although class enumeration remained uncertain. Marker-supported grouping can recover residual clinical information but is order-dependent and does not establish cause or validate preventability.