Machine Learning to Improve Accuracy of Transcutaneous Bilirubinometry

Abstract

Plain Language SummaryNeonatal jaundice is a common condition that might occur in the first week of life. Severe high bilirubin levels can lead to long-term problems like cerebral palsy, hearing impairment, and developmental delay. Therefore, newborns are usually evaluated for jaundice every day during their first week. To accurately assess jaundice, we need to measure the total bilirubin in the blood. However, daily blood draws for bilirubin measurements can be uncomfortable and may cause anemia in newborns. Thus, transcutaneous bilirubinometry is devised. It is a device that measures bilirubin through the skin and is commonly used because it is non-invasive. However, transcutaneous bilirubinometry has some accuracy issues. In this study, we successfully improved the accuracy of bilirubin measurement by combining machine learning with transcutaneous bilirubinometry. Being used alone, the transcutaneous bilirubinometer had an error of 1.08 mg/dL. However, combining transcutaneous bilirubinometry with machine learning, the error decreased to 0.80 mg/dL, which is a 25% of improvement. Using this approach, unnecessary blood draws could be reduced by up to 78%. If we incorporate this algorithm into transcutaneous bilirubinometry, this novel method has the potential to improve prediction accuracy and reduce the burden on babies.