Predicting the Neurodevelopmental Outcome in Extremely Preterm Newborns Using a Multimodal Prognostic Model Including Brain Function Information

• Published in: JAMA network open 2023-03, Vol.6 (3), p.e231590-e231590
• Main Authors: Routier, Laura, Querne, Laurent, Ghostine-Ramadan, Ghida, Boulesteix, Julie, Graïc, Solène, Mony, Sandrine, Wallois, Fabrice, Bourel-Ponchel, Emilie
• Format: Article
• Language: English
• Subjects: Brain - diagnostic imaging; Child, Preschool; Decision trees; Female; Gestational Age; Human health and pathology; Humans; Infant; Infant, Extremely Premature; Infant, Newborn; Life Sciences; Online Only; Original Investigation; Pediatrics; Pregnancy; Prognosis; Retrospective Studies; Risk factors; Ultrasonic imaging; Variables
• ISSN: 2574-3805; 2574-3805
• DOI: 10.1001/jamanetworkopen.2023.1590

Early assessment of the prognosis of preterm newborns is crucial for accurately informing parents and making treatment decisions. The currently available prognostic models rarely incorporate functional brain information from conventional electroencephalography (cEEG). To examine the performance of a multimodal model combining (1) brain function information with (2) brain structure information (cranial ultrasonography), and (3) perinatal and (4) postnatal risk factors for the prediction of death or neurodevelopmental impairment (NDI) in extremely preterm infants. Preterm newborns (23-28 weeks' gestational age) admitted to the neonatal intensive care unit at Amiens-Picardie University Hospital were retrospectively included (January 1, 2013, to January 1, 2018). Risk factors from the 4 categories were collected during the first 2 weeks post delivery. Neurodevelopmental impairment was assessed at age 2 years with the Denver Developmental Screening Test II. No or moderate NDI was considered a favorable outcome. Death or severe NDI was considered an adverse outcome. Data analysis was performed from August 26, 2021, to March 31, 2022. After the selection of variables significantly associated with outcome, 4 unimodal prognostic models (considering each category of variable independently) and 1 multimodal model (considering all variables simultaneously) were developed. After a multivariate analysis for models built with several variables, decision-tree algorithms were run on each model. The areas under the curve for decision-tree classifications of adverse vs favorable outcomes were determined for each model, compared using bootstrap tests, and corrected for type I errors. A total of 109 newborns (58 [53.2% male]) born at a mean (SD) gestational age of 26.3 (1.1) weeks were included. Among them, 52 (47.7%) had a favorable outcome at age 2 years. The multimodal model area under the curve (91.7%; 95% CI, 86.4%-97.0%) was significantly higher than those of the unimodal models (P