Predicting Nonhemolytic Neonatal Hyperbilirubinemia

Discussion

This study provides solid and contemporary estimates of neonatal hyperbilirubinemia in term newborn infants in relation to a set of perinatal risk factors. Our main finding was that a simple collection of common maternal and obstetric risk factors, as well as infant birth weight, predicted >100-fold variation in the incidence of neonatal hyperbilirubinemia already at birth. By disentangling the selective contributions from these risk factors, predictions of hyperbilirubinemia specific for GA and mode of delivery could be constructed for a given maternal ethnicity, BMI, and parity and for birth weight. This new information enables individualized risk prediction in developed health care systems, and perhaps in clinical setting with limited or no resources for blood type or bilirubin screening programs, if the bilirubin epidemiology in these settings is similar to that of Sweden.

We found that 1.9% of term infants in Sweden were treated for neonatal nonimmune hyperbilirubinemia. Given that national Swedish guidelines recommend starting treatment at a total serum bilirubin level >20.6 mg/dL (>350 µmol/L) in infants born at ≥37 weeks’ gestation and without other complicating morbidity,²⁷ our findings are in line with US data suggesting that a total serum bilirubin level of ≥20 mg/dL (≥342 μmol/L) is seen in ∼1% to 2% of infants born at a GA of ≥35 weeks.^12,28

The risk factors for development of severe hyperbilirubinemia identified herein add new knowledge to current standards. In the AAP guidelines from 2004, a GA of 35 to 36 weeks was identified as a major risk factor for severe hyperbilirubinemia, whereas a GA of 37 to 38 weeks was listed as a minor risk factor.¹⁷ In the update from 2009,²⁹ infants born after 35 to 37 weeks were characterized as being at increased risk for severe hyperbilirubinemia. However, infants born at 38 weeks’ gestation were categorized as low-risk infants,²⁹ possibly because of observations (n = 48 patients treated for hyperbilirubinemia) of no significant difference in rates of hyperbilirubinemia between infants born at 38 to 39 weeks and infants born at ≥40 weeks’ gestation.³⁰ Our study confirms several smaller (n = 73–270 patients) US studies^12,31–33 documenting an impact of each decreasing week of gestation on the risk of developing severe hyperbilirubinemia. Given that infants born not only after 37 but also after 38 weeks’ gestation were, in most combinations of other perinatal risk factors, found to be at moderate to high risk for hyperbilirubinemia (Figure 1), we suggest that a GA of 38 weeks should be listed as a major risk factor in coming updates of neonatal hyperbilirubinemia guidelines.

Whereas the previously known risks associated with a cephalohematoma, probably in the causal pathway of the association between VE and hyperbilirubinemia, and with an East Asian mother could be confirmed, our findings suggest that maternal obesity, primiparity, and LGA (also identified by Newman et al¹²) and SGA infants should be added to the list of major risk factors.¹⁷ In line with AAP recommendations, maternal diabetes and infant male gender were associated with increased risks, but the effect sizes were small. And we could not find support for the recommendation that advanced maternal age (defined in AAP guidelines as ≥25 years) is a risk factor for neonatal hyperbilirubinemia:¹⁷ although maternal age ≥30 years was significantly associated with hyperbilirubinemia in our study, the effect size (aOR = 1.05) was very small and without clinical significance. Finally, the evidence behind the recommendation that GA ≥41 weeks is associated with a decreased risk for hyperbilirubinemia¹⁷ was only partially supported in our study. In Asian women and in SGA infants delivered with VE, the risk for significant hyperbilirubinemia remained high to very high across the GA range, underlining the need to take interactions between different risk factors into account.

Even though a majority of infants in our study did not reach a bilirubin level that necessitated treatment, a >10% incidence of the need for acute treatment within hours or days to detect and prevent a potentially disabling disease is very high (depicted dark red in Figure 1). By convention, side effects from drugs are classified as very common (≥10%), common (≥1%, <10%), and less common (≥0.1%, <1%). We used a slightly more conservative classification.

Today, most infants born at term will have their peak bilirubin concentration after discharge from hospital, and hyperbilirubinemia has become one of the most common reasons for readmission to hospital in the neonatal period.²² The problem is not readmission per se; given that hyperbilirubinemia peaks at home after early discharge from hospital, increased readmission rates are expected in a health care system focused on patient safety. The concern is that some readmitted infants have been reported to exhibit dangerously elevated bilirubin levels, >25 mg/dL (>425 µmol/L).³⁴ This development calls for an even more thorough predischarge risk assessment of all newborn infants. Knowledge about maternal and obstetric risk factors is an important part of such an assessment, and the use of checklists may be helpful.^10,17 An individualized score sheet available at birth, as suggested herein, may take hyperbilirubinemia risk assessment several steps forward in terms of accuracy of risk prediction. Such an individualized risk prediction can be useful for parental counseling, planning of optimal timing for hospital discharge and follow-up visits, and timing of bilirubin determinations, before and after hospital discharge.

Besides the risk factors studied herein, a history of older sibling with neonatal hyperbilirubinemia (especially if intervention was needed), a family history of hemolytic disease, and maternal blood group (information not available in the Swedish Medical Birth Registry) will add important information to the overall risk assessment at birth.

Most of the newborn infants diagnosed with hyperbilirubinemia in our study (and not mediated by hemolytic disease) will have been treated with phototherapy. As judged from our data, phototherapy to reduce bilirubin appears to be more frequently used in the United States³⁵ and other countries³⁶ than in Sweden. Besides our exclusion of hemolytic diseases and late preterm infants, higher rates of phototherapy may be explained by lower bilirubin thresholds for initiating treatment in other countries than in Sweden.^17,27 Population differences in the distribution of some maternal and obstetric risk factors, such as ethnicity (2.4% of mothers in our study had an Asian origin vs 6.7% of all births in the United States³⁷), overweight and obesity (21% of the women in our study vs >50% in the United States³⁸), and rates of delivery in weeks 37 to 38 (19% among term deliveries in our study vs 28% among term deliveries in the United States³⁷), may also contribute to the explanation of higher rates of phototherapy in the United States than in Sweden. In addition, the risk of neonatal hyperbilirubinemia is affected by postnatal management. The most important management issue relates to feeding practices. In Sweden, all women are instructed and encouraged to breastfeed their infants. Early and frequent breastfeeding decreases the risk, whereas inadequate breastfeeding can contribute to severe hyperbilirubinemia.¹⁵

We excluded patients with hemolytic hyperbilirubinemia because we assumed that they would have a different risk factor profile (determined mainly from maternal and fetal blood groups) than infants with nonhemolytic hyperbilirubinemia. We also excluded patients with a registry diagnosis of kernicterus because the registry did not provide data on the underlying cause (hemolytic or nonhemolytic hyperbilirubinemia) in most cases and because we did not have access to any case records to validate the kernicterus diagnoses. Nevertheless, in the database, 26 term newborn infants had received a diagnosis of kernicterus, giving a crude estimation of the kernicterus incidence of 26 out of 1 261 974, or 20.6 infants per million. Although not the aim of this study, we note with great concern that our crudely estimated incidence is high compared with estimates of kernicterus in term infants reported previously.^7,14,39 Our observation warrants additional validation of the kernicterus diagnosis in the Swedish Medical Birth Register and more information on background, clinical course, and outcome in these patients.

Exclusion of the patients with registry diagnosis kernicterus does not invalidate the results or interpretations of this study. The patients diagnosed with kernicterus constituted a minute fraction of the whole cohort treated for hyperbilirubinemia, and there is no obvious reason to anticipate that the maternal and obstetric risk factor profile for patients with kernicterus should differ from that of the included patients (all treated for severe hyperbilirubinemia and at risk for kernicterus). Applying the prediction sheet created herein (Figure 1) to the 3 excluded infants documented in the birth registry with a diagnosis of kernicterus and nonhemolytic hyperbilirubinemia would have categorized all 3 of them at birth as being at moderate risk (incidence >1% but <5%) for developing postnatal hyperbilirubinemia in need of treatment. This observation has several important implications. First, it shows that prediction of a nonexisting risk for kernicterus cannot be made. Second, one has to keep in mind that our grading of predicted incidences of severe hyperbilirubinemia, from low to very high, is not a grading of the severity of hyperbilirubinemia, once this complication has occurred. Third, this observation clearly indicates that a clinical risk assessment at birth, based on the maternal and obstetric risk factors studied herein or on other sources of information, can never be foolproof alone. Successful hyperbilirubinemia prediction and kernicterus prevention build on a system approach in which accurate risk assessment immediately after birth could serve as a good starting point that must be combined with other actions in the neonatal period. Finally, our study design, including a large (>20 million of data points) registry database, can be a valuable tool for creating precise risk factor estimates for as accurate prediction as possible. But it is unsuitable for analysis of rare and serious adverse events such as kernicterus; our data set did not provide information on the postnatal course of the infants with kernicterus, on the duration of their hospital stay, on how follow-up was organized and executed, whether these infants suffered from additional significant neonatal morbidity (besides severe hyperbilirubinemia) before or after discharge from the hospital, and whether kernicterus in these patients represented a system failure or occurred despite proper detection and treatment of hyperbilirubinemia.

The major strengths of this study are the large study population and the high quality of the Swedish Medical Birth Register (98% of all pregnant women in Sweden included). The size of the sample (23 711 cases of hyperbilirubinemia) played a crucial role in avoiding errors, in estimating effect size and significant interactions between different risk factors, that may occur in smaller studies or by intuitive prediction based on case series. The major limitation is that the register lacks information on important covariates such as family or sibling history of neonatal hyperbilirubinemia, blood group, number and duration of tractions using VE, and bruising. In addition, we do not have information on neonatal outcomes such as maximal serum bilirubin levels and clinical picture.

Finally, the proposed sheet of hyperbilirubinemia prediction based on risk factors should optimally be cross-validated, assessing how the results of our study would generalize to an independent data set or in other settings. Unfortunately, other Swedish population-based registers do not contain the set of risk factors studied herein, excluding the possibility of validation using other Swedish data sets. We see 2 possibilities for future validation: to perform a similar study to ours using a sufficiently large data set generated in another country or population or to use the proposed prediction sheet and prospectively study to what extent the prediction of significant hyperbilirubinemia mirrors the outcome.