Validation of published rebound hyperbilirubinemia risk prediction scores during birth hospitalization after initial phototherapy: a retrospective chart review

Background: Hyperbilirubinemia commonly affects newborns and may lead to neurotoxicity if untreated. Neonates can experience rebound hyperbilirubinemia (RHB), defined as elevated bilirubin levels requiring re-initiation of treatment. Although studies have formulated risk prediction scores, they lack external validation. In this study, we examine the discrimination and calibration performance of risk prediction scores for RHB, to provide external validation.

Methods: We reviewed charts of neonates born ≥35 weeks of gestation between January 2015 and December 2019 receiving phototherapy at birth hospitalization. We plotted predicted probabilities against observed outcome proportions to assess model calibration and evaluated discrimination using area under the receiver operating characteristic (AUROC) curves. Odds ratios (ORs) were estimated to evaluate variables associated with RHB.

Results: Of the 271 infants identified, 24% developed RHB. Two- and three-variable prediction scores had lower discrimination in our cohort with AUROC of 0.662 (95% CI 0.590-0.735) and 0.691 (95% CI, 0.619-0.763) compared to 0.876 (95% CI 0.854-0.899) and 0.881 (95% CI 0.859-0.903), respectively, in the published studies. Estimated ORs confirm associations between RHB and variables included in prediction scores.

Conclusions: Current prediction models for RHB have unclear clinical utility in our patient population. Additional studies are required to further validate these scores.

Impact: Describes performance characteristics of two- and three-variable risk prediction scores that lack external validation beyond the initial study cohort. Our findings suggest unclear clinical utility in our clinical population of neonates during birth hospitalization, with lower performance of these prediction scores than observed in the derivation cohort. Odds ratios estimated by logistic regression in our study cohort provide further evidence that variables in published risk prediction scores are associated with rebound hyperbilirubinemia. Further studies are required to externally validate these risk prediction scores and to assess their generalizability.