463 - Early cerebral desaturation events in extremely preterm infants are associated with intraventricular haemorrhage in the first week of life

Poster Number: 463
Publication Number: 463.342

Minoo Ashoori, University College Cork, Cork, Cork, Ireland; John M. O'Toole, University College Cork, Cork, Cork, Ireland; Fiona B. McDonald, University College Cork, Cork, Cork, Ireland; Ken D. O'Halloran, UCC, Cork, Cork, Ireland; Eugene Dempsey, University Colleeg Cork, Cork, Cork, Ireland

Background: Extremely preterm infants are often hemodynamically unstable and conditions such as hypotension, patent ductus arteriosus, and apnea of prematurity are common. The brain relies on an adequate continuous supply of oxygen and glucose and controlled perfusion to maintain homeostasis. Development of intraventricular haemorrhage (IVH) in preterm infants is a risk factor for brain injury resulting in long-term negative outcomes. Cerebral near infrared spectroscopy provides an insight into cerebral oxygenation status.

Objective: To test the potential clinical utility of cerebral near infrared spectroscopy (NIRS) extracted short-duration (transient) oxygen desaturations in predicting IVH in the first week of life.

Design/Methods: We analyzed a subset of infants enrolled in the Management of Hypotension in Preterm infants (HIP) trial (n=46). All infants were < 28 weeks of gestational age, had continuous regional cerebral oxygen measurement using NIRS over the first 72 hours and clinical ultrasounds performed during the first week of life. The NIRS signals were pre-processed, and transient cerebral oxygen desaturations were extracted from the NIRS signal using a recently developed algorithm. Extracted transient cerebral oxygen desaturations were characterized using 14 signal features and then combined using a machine-learning approach. The machine-learning model was trained and tested to predict a binary IVH injury using a leave-one-out cross validation approach. IVH injury was defined as IVH grade II-IV and each infant was graded according to the highest grade within the first week of life. The predictability value of eight clinical features was also determined using another machine-learning model (random forest).

Results: The mean (sd)  birthweight was 767 (155)g and  gestational age  25.7(1.2) weeks.  We find a significant association between our extracted transient cerebral oxygen desaturations signal features and IVH. The area under the receiver operating characteristic curve (AUC) for the NIRS machine-learning model was 0.789 with a 95% confidence interval (CI) of 0.639 to 0.915 compared to AUC for the clinical features machine learning model of 0.575 (95% CI: 0.39 to 0.739).Conclusion(s): Automated analysis of transient desaturations of the cerebral NIRS signal in the first days of life may aid prediction of IVH in extremely preterm infants. Further investigation of these transients, their association with peripheral saturations and an understanding of the physiology underlying the frequent cerebral desaturations observed in some patients is warranted.