Predicting Risk of Readmission in Newborns Treated with Phototherapy

Background: Hyperbilirubinemia requiring phototherapy is a common problem in newborns impacting length of stay and need for readmission. Guidelines exist for phototherapy initiation, but there is less guidance on when to discontinue treatment to optimize outcomes. We sought to compare the validity of several proposed methods for predicting risk of readmission.

Methods: We performed a retrospective chart review of all newborns ≥ 35 weeks gestation treated with phototherapy during their birth hospitalization from 1/2018 to 12/2019. Patients were excluded if they were discharged from the neonatal intensive care unit. We collected data related to hyperbilirubinemia risk and used serum bilirubin levels and timing of phototherapy initiation and termination to calculate predicted risk of rebound hyperbilirubinemia by previously developed 3-variable and 2-variable models. We compared these calculated risks (dichotomized at 10%) to the difference (dichotomized at 3mg/dl) between the treatment threshold at phototherapy completion and the neonate’s serum bilirubin. We analyzed bivariate relationships between neonatal characteristics, including feeding type, which we dichotomized into breast/expressed breast milk without additional supplement versus partial or exclusive donor breast milk/formula. Our main exposures were the dichotomized risk percentages and bilirubin differences, and our outcome was readmission for phototherapy. Using covariates that were significantly associated with risk percentages of bilirubin difference and readmission, we performed adjusted logistic regression to assess adjusted odds of readmission. Finally, we assessed goodness of fit of predictive models with the Hosmer-Lemeshow goodness of fit test and compared discrimination with ROC curves.

Results: We had available readmission data on 76% of neonates (n=370), of which 69% were ≥ 38 weeks gestation, and 49% were female. Mean birthweight was 3.22 kg (SD 0.5kg). Phototherapy was initiated at a mean of 39 hours (SD 17), and average length of stay was 3.1 days (SD 0.8). 6% of neonates had rebound hyperbilirubinemia (n=23). Among covariates, only sex and feeding type were significantly associated with readmission, with 78% of readmissions being male. In logistic models adjusting for feeding type and sex, odds of readmission were similar between two calculated risks (aOR 5.4; 95% CI: 2, 14 and aOR 5.3; 95% CI: 2,14), but adjusted odds of readmission were much higher if the difference between serum bilirubin and phototherapy threshold was less than 3mg/dl (aOR 8.2; 95% CI: 3, 22). All methods had good fit, with the simple difference in final bilirubin having slightly higher AUC (0.82) than the 3-variable (0.77) or 2-variable (0.78) model.

Conclusion: Newer mathematical models for predicting risk of rebound hyperbilirubinemia are helpful tools for determining timing of discontinuation of phototherapy. Using our cutoffs, proximity to the phototherapy threshold may be slightly more predictive of readmission for phototherapy than calculated risk scores.