Skip to main content

  • Quantifying Early Neonatal Oxygen Exposure in ROP

    By Jean Shaw
    Selected by Emily Y. Chew, MD
    Add to My Bookmarks

    Journal Highlights

    Ophthalmology Science, December 2021

    Download PDF

    Chen et al. evaluated oxygen’s nuanced role in the development of retinopathy of prematurity (ROP), particularly treatment-requiring ROP (TR-ROP) and aggressive ROP (A-ROP). They found that data on early neonatal oxygen exposure can be extracted from the electronic health record (EHR) and quantified as a risk factor for TR-ROP and A-ROP.

    For this proof-of-concept study, the researchers had three lines of inquiry: 1) Is it possible to develop quantitative variables for oxygen exposure from the EHR? 2) What is the relationship between these oxygen variables and incident TR-ROP and A-ROP? 3) Using machine learning, does the quantifica­tion of oxygen exposure add predictive value for incident TR-ROP?

    The 244 premature infants in this retrospective analysis were screened for ROP at Oregon Health & Science Uni­versity in Portland. All 244 were born at or before 30 weeks’ gestational age. In addition, oxygen data for each infant was complete from birth to 30 weeks’ postmenstrual age (PMA) in the EHR. The researchers manually extracted data on oxygen saturations and fraction of inspired oxygen (FiO2), and four oxygen variables were calculated on a weekly basis. Random forest models were trained with fivefold cross-valida­tion using gestational age and cumula­tive FiO2 at 30 weeks’ PMA to identify infants who developed TR-ROP.

    For TR-ROP, the main outcome was cross-validation performance, assessed using area under the receiver operating curve (AUC) and preci­sion-recall curve (AUPRC) scores. For A-ROP, there were not enough cases to build machine learning models; thus, the researchers calculated AUC and evaluated sensitivity and specificity at a high-sensitivity operating point.

    Of the 244 infants, 33 developed TR-ROP, and 28 of the 33 were diagnosed with A-ROP. For TR-ROP, random forest models trained on gestational age plus cumulative minimum FiO2 were not significantly better than those trained on gestational age alone. How-ever, models using oxygen alone had an AUC of .80 ± .09. In the secondary anal­ysis for A-ROP, the AUC was .92.

    In the future, the oxygen variables evaluated in this study could be made available at the time of first ROP screening along with standard demo­graphic risk factors, the researchers noted. “Future work may build upon this study by examining oxygen con­centrations at a more granular level to improve modeling for TR-ROP and other diseases related to oxygen expo­sure,” they concluded.

    The original article can be found here.