81 - Neonatal Peripheral Microvascular Imaging as a Biomarker for Retinopathy of Prematurity

Poster Number: 81
Publication Number: 81.43

Daniel J. York, University of Utah School of Medicine, Cottonwood Heights, UT, United States; Gustave H. Falciglia, Northwestern University, Chicago, IL, United States; Isabelle G.. De Plaen, Northwestern University The Feinberg School of Medicine, Chicago, IL, United States

Background: Retinopathy of Prematurity (ROP), a retinal vascular disease and leading cause of childhood blindness worldwide, has no cure or prevention. While preterm birth < 32 weeks gestational age (GA), low birth weight < 1500 grams and postnatal oxygen exposure independently influence ROP risk, screening paradigms informed by these factors lack specificity. As a result, we are unable to predict infants at greatest disease risk thereby exposing roughly twice the number of preterm infants to stressful ophthalmic screening exams than those who will develop ROP. We identified nailfold peripheral microvascular density (PVD) within the first postnatal month as an imaging biomarker for subsequent development of severe ROP. PVD may increase screening specificity and improve infant outcomes by better targeting ophthalmic care.

Objective: Here we evaluate the utility of PVD as a solitary biomarker for the purpose of predicting severe ROP and compare with current ROP screening models.

Design/Methods: We prospectively assessed nailfold microvascular vessel length density (VLD), a metric of PVD, using a machine-learning based, automated quantification platform in a cohort of 26 premature infants (average GA 27.9) within 2 weeks of life (T1) and two weeks later (T2). VLD estimates vascular density by summing the lengths of all microvessel centerlines within an image’s region of interest (µm/mm2). ROP phenotype was determined by the screening ophthalmologist and ROP activity severity score calculated according to the 2018 guidelines from the International Neonatal Consortium ROP workgroup. Severe ROP was defined as a score of ≥7/22 on this scale; Receiver Operator Characteristic (ROC) analysis was used across VLD cut-points to assess the ability of VLD to predict severe ROP.

Results: A VLD cutoff > 14,100 at T1 or a VLD cutoff > 11,900 at T2 was associated with an ROP severity score ≥ 7 in 10/10 infants (100% sensitivity) with a specificity of 75%. The ROC curve evaluating VLD thresholds for predicting ROP ≥ 7at each timepoint showed an area under the curve (AUC) of 0.85. A VLD cutoff of > 15,100 at T1 or at T2 correctly detected 3/3 infants requiring ROP therapy with ROC-AUC of 0.84.

Conclusion(s): Neonatal peripheral microvascular density is a promising non-invasive imaging biomarker for development of severe ROP. These test characteristics are comparable to those reported for the well-validated G-ROP algorithm. Addition of VLD to current ROP screening models may increase specificity and sensitivity thereby improving patient outcomes.