Prediction of post-hemorrhagic ventricular dilatation trajectory using a growth mixture model in preterm infants.

Abstract

Early intervention for post-hemorrhagic ventricular dilatation (PHVD), guided by ventricular size measurements from cranial ultrasound (cUS), is associated with improved neurodevelopmental outcomes in preterm infants but benefits must be balanced against intervention risks. Anterior horn width (AHW) and ventricular index (VI) were measured from cUS for preterm infants (<29 weeks) with intraventricular hemorrhage admitted from 2010-2018. PHVD was defined as AHW > 6 mm or VI >97th percentile for postmenstrual age. Individual ventricular size trajectories were plotted, and a growth mixture model (GMM) used to identify latent trajectory classes and compare these to predetermined outcome of neurosurgical intervention. Measurements were obtained from 1543 cUS in 249 infants, of whom 39 had PHVD without and 17 PHVD with neurosurgical intervention based on signs of raised intracranial pressure. The GMM predicted trajectory identified: 93.3% of infants without PHVD, 88.2% and 30.8% of infants with PHVD with and without intervention using AHW; 100% of infants without PHVD, 52.9% and 59.0% of infants with PHVD with and without intervention using VI. The AHW GMM identified a significant proportion of infants with severe PHVD. Model refinement offers a promising approach for identifying differences in PHVD trajectory at an early stage to guide management. It is difficult to distinguish the trajectory of PHVD in the early stage of development, in particular PHVD that spontaneously arrests from slowly progressive PHVD which eventually requires intervention. We report the first modeling-based evaluation of PHVD trajectory for the prediction of short-term outcome of PHVD progression and neurosurgical intervention. With additional clinical validation and optimization to increase accuracy, predictive modeling has the potential to identify important differences in PHVD trajectory at an early stage in the clinical course, allowing for more individualized data-driven risk-benefit assessments to guide decisions on early intervention.