Abstract / Summary
Abstract Bilirubin is measured routinely in neonates, but whether its trajectory over the first postnatal week adds risk-stratification value beyond a single measurement is unclear. In this preregistered two-database retrospective cohort study (MIMIC-III, n = 8,059; PIC, n = 2,953 neonates admitted at < 28 days of age), we estimated latent-class trajectories of log-transformed total bilirubin on postnatal days 0–7 (neonates with ≥ 3 measurements; n = 2,951 and 1,817) and related them to a composite of in-hospital death, necrotizing enterocolitis, or hospital-acquired late-onset sepsis. Four phenotypes were identified (typical 62.6%, delayed high peak 9.4%, persistently low 6.8%, rapid-rise early peak 21.2%). The rapid-rise early-peak class had higher composite-outcome risk in both databases (OR 2.65, MIMIC-III; 1.90, PIC), attenuated to non-significance after birth-weight adjustment in MIMIC-III. The persistently low class had the lowest bilirubin values but the highest risk (OR 7.45; 2.55), largely explained by birth weight. The trajectory model improved discrimination over the Bhutani risk zone in MIMIC-III (corrected AUC 0.820 vs 0.768) but matched the peak value (0.820 vs 0.822); discrimination in PIC was modest. Trajectory phenotypes identified high-risk neonates consistently, yet incremental value over single-point measures was modest and associations partly reflected immaturity and severity.