Predicting the Need for Phototherapy After Discharge: Update for 2022 Phototherapy Guidelines.

Abstract

In 2021, Kuzniewicz et al1 reported that the difference (Δ-total serum bilirubin [TSB]) between last TSB before discharge and the 2004 American Academy of Pediatrics (AAP) phototherapy threshold was an excellent predictor of a postdischarge TSB above the 2004 AAP phototherapy threshold (area under the receiver operating characteristic curve = 0.93). The 2004 AAP hyperbilirubinemia guidelines have since been updated with thresholds for phototherapy that now increase with each week of gestation and are about 1 to 3 mg/dL higher than those from 2004. We sought to quantify how well the predischarge Δ-TSB predicts a postdischarge TSB exceeding the new 2022 AAP phototherapy thresholds.2This retrospective cohort study used the same cohort as our 2021 publication: 163 930 infants born at ≥35 weeks’ gestation July 1, 2012, to December 31, 2017, at 11 Kaiser Permanente Northern California hospitals employing universal TSB screening with at least 1 predischarge TSB. We excluded 9800 for prolonged hospitalization, 4 for direct hyperbilirubinemia, 476 with a predischarge TSB before 12 hours of age, and 2425 who received phototherapy or had a TSB above the 2022 AAP threshold before discharge from all analyses.For all analyses except those used to calculate inverse-probability of censoring weights, we also excluded 4546 infants who, after discharge, received phototherapy before having a TSB above the 2022 AAP threshold. Thus, the final cohort included 146 679 infants. The Kaiser Permanente Northern California institutional review board approved the study (CN-17-3051 01).Our primary predictor was Δ-TSB, defined as the difference between the 2022 AAP phototherapy threshold (for infants of the same age, gestational age and direct antiglobulin test result) and the last inpatient TSB level. (Note in our previous article, our definition was the negative of the above: the TSB minus the threshold.) We created a 5-category Δ-TSB variable to match the AAP 2022 follow-up guideline, as well as an 8-category Δ-TSB variable. We did not have data on transcutaneous bilirubin levels.Our outcomes were a TSB level above the 2022 AAP phototherapy threshold at <24 hours, <48 hours or <30 days from the predischarge TSB. We used linear interpolation to estimate the time that the TSB crossed the phototherapy thresholds.To avoid bias from censoring those who received phototherapy before exceeding the threshold (“subthreshold phototherapy”), we used inverse probability of (non)censoring weights.3 We created models for the probability P of receiving subthreshold phototherapy, based on data available at the time of discharge. We then excluded those infants who had received subthreshold phototherapy and weighted the remaining observations by 1/(1-P) after replacing weights above the 99th percentile with 99th percentile weights.Results by either in 1-mg/dL categories of Δ-TSB or in the 5 categories used in the 2022 guideline (0–<2; 2–<3.5; 3.5–<7; or ≥7 mg/dL) were consistent with those previously reported,1 showing excellent prediction and discrimination (Table 1). Eighty-four percent of the infants had Δ-TSB ≥5.5 mg/dL, with a risk of crossing the phototherapy threshold ≤0.3% (Table 1).We have documented that a single number available at the time of hospital discharge, the Δ-TSB, was a strong predictor of the probability of exceeding the AAP's 2022 phototherapy threshold. The maintenance of discrimination with the higher thresholds is not surprising: Although the slightly raised phototherapy thresholds reduced the risk of the outcome, for an infant with the same TSB level, the Δ-TSB would also be slightly increased, thereby reflecting that lower risk.The excellent discrimination we observed (area under the receiver operating characteristic curve ≥0.93) is largely a result of 84% of the infants having a Δ-TSB that correctly identified them as being at very low risk.Strengths of this study include the large sample size and population with near universal TSB screening. As previously described,1 prediction could be improved by including additional variables, including the most recent rate of rise, formula feeding, age, and gestational age. Nonetheless, our results suggest that a single number, the Δ-TSB divided into 5 categories, provides considerable prognostic information. This supports its use in the 2022 AAP guideline, which provides explicit follow-up recommendations based upon the Δ-TSB in 5 categories.