Abstract
BACKGROUND: Unbound bilirubin (UB) was measured on day 5 ± 1 in 1101 ELBW newborns in the Aggressive vs Conservative Phototherapy randomized controlled trial. We accessed this dataset to quantify the UB-mediated risk of severe neurodevelopmental impairment (sNDI) in extremely low birthweight (ELBW) newborns. METHODS: UB levels were standardized within laboratories as z-score percentiles. A dichotomous UB exposure variable was based on sharply increased risk of sNDI at the 85th percentile (UBz>85th). Targeted maximum likelihood estimation with ensemble machine learning (TMLE + ML) and logistic regression (LR) were used to estimate sNDI risk attributable to UBz>85th. RESULTS: UB - total bilirubin correlation was poor: R(2) 0.24, 95% CI: 0.19, 0.30. The UB level corresponding to the UBz>85th cut-point in this cohort was estimated to be between 0.56 and 0.79 µg/dL. In 825 survivors with known sNDI status the crude risk difference (RD) associated with UBz>85th was 19.5%. With TMLE + ML, the RD was 14.4% (95% CI: 5.7%, 23.1%) corresponding to a number need to harm (NNH) of 7 (95% CI: 4, 18). LR results were similar: RD 12.4% (95% CI: 4.7%, 20.1%), NNH 8 (95% CI: 5, 21). CONCLUSIONS: Exposure to UBz>85th was associated with a large increase in risk of sNDI in ELBW newborns. IMPACT: Accessing the only large dataset with measures of unbound bilirubin (UB) and neurodevelopmental outcomes in extremely low birthweight (ELBW) newborns we found poor correlation between UB and total bilirubin (TB). The risk of severe neurodevelopmental impairment (sNDI) associated with UB rose non-linearly with an inflection point at UB z-score 85th percentile (UBz>85th), increasing the risk of sNDI by an estimated 14% (number need to harm: 7). The current incidence of exposure to harmful UB levels in very premature newborns is unlikely to be rare. TB levels provide a false sense of security to clinicians caring for very premature newborns.