Mediational analysis of severe retinal injury causation in children with acute closed head injury

Author Department

Pediatrics

Document Type

Article, Peer-reviewed

Publication Date

6-2025

Abstract

Background: While generally regarded as discriminating findings, elevating the probability of abusive head trauma, the association of severe retinal injuries with inertial biomechanics, shaking, and abuse has been challenged.

Objective: Evaluate five causational models of severe retinal injury utilizing multivariable mediational analysis.

Participants: Children, under age three-years, admitted to 18 pediatric intensive care units with acute, symptomatic, closed head injury who had not been injured in a car-crash and who had received an ophthalmologist's retinal examination.

Methods: Multivariable logistic regression, and multiple mediational analysis were applied to five causational models of severe retinal injury. Primary causes evaluated were inertial injury biomechanics, severe parenchymal brain injury, and provision of cardio-respiratory resuscitation. Results were reported as direct and mediated log-odds (ꞵ), with standard errors and p-values.

Results: The most explanatory model, having the greatest total effect (ꞵ = 3.768, p ≤ 0.001) and direct effect (ꞵ = 2.431, p ≤ 0.001), identified isolated inertial injuries as the primary cause of severe retinal injury. Additional indirect effects were mediated by radiographically evident diffuse parenchymal brain injury (ꞵ = 0.834, p ≤ 0.001) and the presence of extra-axial blood on imaging (ꞵ = 0.589, p ≤ 0.001).

Conclusions: Causational modeling of cross-sectional data in this population best supports inertial head injury biomechanics, above a threshold to cause diffuse parenchymal brain injury and extra-axial hemorrhage, as the most likely cause of severe retinal injury. Support for other causal pathways is statistically significant, but less robust.

Keywords: Abusive head trauma; Etiology; Mediational analysis; Retinal hemorrhage.

PMID

40466513

Share

COinS