Current approaches to identifying infants at risk of developing severe neonatal hyperbilirubinemia include use of an hour-specific bilirubin nomogram that employs predischarge bilirubin measurements and clinical risk factor assessment that employs multiple factors in clinical prediction rules. Determining the hour-specific total serum bilirubin before discharge has been shown to be the most accurate method for assessing risk of severe hyperbilirubinemia. Combining clinical risk factors and predischarge bilirubin values may offer additional predictive performance above either approach used alone. Current risk assessment strategies need to be validated prospectively in a large and diverse newborn population, and the risk assessment strategies should be paired with recommended actions. Finally, transcutaneous bilirubin and end-tidal carbon monoxide measurements and screening for specific genetic markers of neonatal hyperbilirubinemia have the potential to refine risk assessment strategies further.
After completing this article, readers should be able to:
Explain how to use existing strategies for assessing the risk of severe neonatal hyperbilirubinemia in newborns.
Discuss the methodological and practical issues involved in evaluating and comparing alternative risk assessment strategies.
Identify the limitations of current risk assessment strategies and next steps and future research priorities for developing, implementing, and evaluating risk assessment strategies.
Identifying infants at risk of developing severe neonatal hyperbilirubinemia and kernicterus is a problem that clinicians have faced since the condition of neonatal jaundice initially was recognized more than 100 years ago. (1) Historically, clinicians have relied on the visual assessment of jaundice and consideration of known clinical risk factors to make decisions about which infants need bilirubin concentrations measured and which need close follow-up. However, such assessments were not necessarily systematic or quantitative; rather, they relied on the clinical experience and subjective judgment of the clinician. In the last decade, this informal approach to estimating the risk of severe neonatal hyperbilirubinemia and determining follow-up plans has been challenged by a number of developments, including research demonstrating the limited accuracy of visual assessment of jaundice, (2) shortened postpartum lengths of stay (3)(4)(5) that result in discharge prior to bilirubin peak, and increasing reports of kernicterus in otherwise well term and near-term infants. (6)(7)(8)
In response to these developments, researchers have designed and evaluated several strategies for formally assessing the risk of severe neonatal hyperbilirubinemia, and the American Academy of Pediatrics (AAP) has recommended formal hyperbilirubinemia risk assessment for all newborns. (9) Risk assessment strategies that provide accurate estimates of risk can be used to target preventive care, such as further testing and closer follow-up, for newborns at the greatest risk of developing severe neonatal hyperbilirubinemia and kernicterus, while avoiding the cost and inconvenience of testing and follow-up of low-risk infants. The two strategies that have been studied and recommended by the AAP include: 1) measuring bilirubin concentration and plotting the results on a nomogram that displays bilirubin percentiles with respect to postnatal age in hours and 2) systematically identifying clinical risk factors associated with severe hyperbilirubinemia.
In this article, we review the evidence regarding the predictive performance of these two approaches for estimating the risk of severe neonatal hyperbilirubinemia. We discuss methodological and practical considerations in evaluating and comparing alternative risk assessment strategies. Finally, we suggest next steps and future research priorities for developing, implementing, and evaluating risk assessment strategies.
Risk Assessment Strategies
Predischarge Bilirubin Measurement
An infant’s total serum bilirubin (TSB) concentration represents the net accumulated bilirubin in circulation. This bilirubin load is determined by the balance of bilirubin production, conjugation, elimination, and enterohepatic circulation. Biologic factors that influence production of bilirubin, a breakdown product of hemoglobin, include the initial hematocrit; extravascular hemolysis (of collections of blood in bruises or cephalohematomas); and intravascular hemolysis, either immune-mediated (eg, ABO, Rh, or minor blood group incompatibility) or nonimmune-mediated (eg, glucose-6-phosphate dehydrogenase [G6PD] deficiency, congenital spherocytosis, pyruvate kinase deficiency). Bilirubin conjugation is impaired to some extent in all infants, but the degree of impairment is determined by gestational age and genetics, as evidenced by variable conjugation rates in infants of different sexes, races, and ethnicities. Enterohepatic circulation rates are largely influenced by the frequency of stooling, which, in turn, is related to the volume and frequency of feedings and the consequent triggering of the gastrocolic reflex. Delayed passage of meconium and deconjugation of bilirubin by intestinal glucuronidases also may contribute to increased enterohepatic circulation.
The hour-specific bilirubin nomogram (10) (Fig. 1) that has been recommended as one approach for predicting neonatal hyperbilirubinemia was developed based on the hypothesis that early bilirubin values, expressed as a percentile with respect to the infant’s age in hours, are predictive of the infant’s later bilirubin values, also expressed as a percentile with respect to the infant’s age in hours (hereafter referred to as the hour-specific TSB). More specifically, Bhutani and colleagues (10) reasoned that if the biologic factors that determine bilirubin production and elimination when TSB levels are peaking on day 3 to 5 of extrauterine life also are present in the first 2 days after birth, then the hour-specific TSB value measured prior to discharge should be closely correlated with the hour-specific value when TSB is peaking. The concept was that the hour-specific bilirubin nomogram could be used like a pediatric growth chart, with the expectation that children starting out on a certain percentile track will stay on or near that percentile track as time passes.
The bilirubin nomogram, developed from bilirubin values of 2,840 infants followed in an early discharge program at Pennsylvania Hospital, uses the 40th, 75th, and 95th percentile tracks to distinguish bilirubin values in low-, low intermediate-, high intermediate-, and high-risk zones. In the subset of Pennsylvania Hospital patients who had both pre- and postdischarge TSBs measured, the predischarge bilirubin risk zone was a strong predictor of whether an infant developed a postdischarge TSB value greater than the 95th percentile on the nomogram, with 0.6%, 3%, 21%, and 54% of infants who had predischarge TSBs in the low-, low intermediate-, high intermediate-, and high-risk zones, respectively, developing a postdischarge TSB greater than the 95th percentile. (11) This suggested that clinicians could use the predischarge TSB, expressed as a “risk zone” on the bilirubin nomogram, to identify a group of infants who had virtually no risk (low-risk zone predischarge TSB) and a group of infants who had very high risk (high-risk zone predischarge TSB) of developing severe neonatal hyperbilirubinemia. Other investigators have developed and published hour-specific bilirubin nomograms based on data from their hospitals’ patient populations, (12)(13) but none have been formally evaluated for their predictive accuracy.
Clinical Risk Factor Assessment
Epidemiologic studies have identified multiple factors that are associated with an increased or decreased risk of severe neonatal hyperbilirubinemia. The risk factors can be related to maternal or infant characteristics and include items from the medical history, labor and delivery record, physical examination, and blood tests. Used in isolation, such factors have limited predictive ability, but combining multiple factors in clinical prediction rules greatly enhances predictive performance. The AAP’s clinical practice guideline on management of newborn jaundice includes a table of major and minor risk factors as well as protective factors (Table 1), but the guideline does not specify how to use these factors in combination to quantify risk.
|Major Risk Factors||Minor Risk Factors||Decreased Risk|
↵* Race as defined by mother’s description.
TSB=total serum bilirubin, TcB=transcutaneous bilirubin, G6PD=glucose-6-phosphate dehydrogenase, etCOc=end-tidal carbon monoxide
Reprinted with permission from American Academy of Pediatrics Subcommittee on Hyperbilirubinemia. Management of hyperbilirubinemia in the newborn infant 35 or more weeks of gestation. Pediatrics. 2004;11:297–316.
A review of the literature identified several studies in which clinical prediction rules that combine multiple risk factors were developed and evaluated to predict severe neonatal hyperbilirubinemia. One of the earliest was developed by Newman and associates. (14) Their risk index, developed from a nested case-control study of 51,000 newborns in northern California, identified nine predictors that were associated with the development of a TSB of at least 25 mg/dL (427.5 mcmol/L), including exclusive breastfeeding, family history of jaundice in a newborn, neonatal bruising, cephalohematoma, sex, gestational age, and maternal race and age of 25 years or more. The weights for each of the risk factors, which are used in calculating the risk score for an individual infant, are approximately equal to the odds ratio for each of the risk factors in the logistic regression model from which the rule was derived. The test properties of the risk index depend on the positivity criterion used to define a high score. For example, if a risk score greater than 10 is used to define the cutoff for characterizing an infant at high risk for developing severe neonatal hyperbilirubinemia, then the risk score has a sensitivity of 88% and a specificity of 39%, yielding a likelihood ratio of 2.2. Increasing the score used to designate high risk increases specificity but decreases sensitivity.
Keren and colleagues (11) retrospectively abstracted clinical risk factor information from a subset of patients included in the previously cited bilirubin nomogram study and, using methods similar to those employed by Newman, developed a clinical risk factor score to predict the development of severe hyperbilirubinemia, defined as a postdischarge TSB greater than the 95th percentile on the hour-specific bilirubin nomogram. That prediction rule, like Newman’s, included breastfeeding and gestational age but identified vacuum extraction, oxytocin use, and infant weight as additional factors significantly associated with neonatal hyperbilirubinemia. Others have used multivariate logistic regression to develop prediction models for the development of severe neonatal hyperbilirubinemia but have not attempted to translate such models into scoring systems for use by clinicians. Stevenson and colleagues (15) found end-tidal carbon monoxide (ETCO) values, breastfeeding, and birthweight predictive of TSB of at least the 95th percentile on the hour-specific bilirubin nomogram at 96±12 hours after birth. Chou and associates (16) identified maternal race, breastfeeding, and gestational age less than 38 weeks as consistent predictors of TSB of 20 mg/dL (427.5 mcmol/L) or greater and TSB at or greater than the age-specific AAP criteria for considering phototherapy. It is important to note that breastfeeding and gestational age are factors that consistently emerge in multiple studies as the strongest clinical predictors of neonatal hyperbilirubinemia.
Evaluating Risk Assessment Strategies
Before comparing the predictive performance of the various strategies used to predict the risk of severe neonatal hyperbilirubinemia, it is helpful to review the methodological and practical issues that must be considered in any comparison.
MEASURES OF TEST ACCURACY: DICHOTOMOUS TEST RESULTS.
Using a predischarge bilirubin measurement or a clinical risk factor to characterize a newborn’s risk of developing severe hyperbilirubinemia is analogous to using a diagnostic test to determine whether a patient has a particular disease or condition. When multiple risk factors or test results are combined into a formal risk “score” or “index,” the risk assessment takes the form of a clinical prediction rule. Thus, the statistics used to evaluate risk assessment strategies are the same ones used to measure the accuracy of diagnostic tests and clinical prediction rules. The most common of these are sensitivity, specificity, positive predictive value, and negative predictive value–measures of accuracy used when the output of the predictive test is dichotomous (normal/abnormal, above/below a certain threshold). Table 2 demonstrates the typical 2×2 table used to calculate these measures along with definitions of each one.
Positive Predictive Value=a/(a+b)
Negative Predictive Value=d/(c+d)
Likelihood Ratio Positive=(a/[a+c])/(b/[b+d])
Likelihood Ratio Negative=(c/[a+c])/(d/[b+d])
A common method for summarizing test accuracy in one measure is the likelihood ratio, which expresses the odds that a specific test result would occur in a patient who has the outcome of interest as opposed to one who does not have the outcome. Thus, the likelihood ratio positive is calculated by dividing the proportion of patients who have the outcome of interest and have a positive test result by the proportion of patients who do not have the outcome of interest and have a positive test result. The likelihood ratio negative is calculated by dividing the proportion of patients who have the outcome of interest and have a negative test result by the proportion of patients who do not have the outcome of interest and have a negative test result (Table 2). By applying Bayes theorem, clinicians can use the likelihood ratio positive or negative for a test result and the pretest probability (also known as the prevalence) of a disease or outcome to calculate the posttest probability of that outcome. A discussion of how to perform these calculations by hand is beyond the scope of this article, but clinicians should be aware that Bayesian calculators can be found on the Internet or downloaded onto handheld computers to perform these calculations quickly and easily.
MEASURES OF TEST ACCURACY: CONTINUOUS TEST RESULTS.
Another common method for summarizing test accuracy, particularly for tests or prediction rules that have continuous, rather than dichotomous, test results, is the use of receiver operating characteristic (ROC) curves or the c-statistic. The points that define an ROC curve represent the sensitivity and specificity (actually 1-specificity) for a continuous diagnostic test result as the positivity criterion—the threshold used to define a positive test—is varied (Fig. 2). (17) The area under the ROC curve provides a metric for comparing overall test accuracy, but more importantly, the shape of the ROC curve gives valuable information about the degree to which sensitivity is traded for specificity as the positivity criterion is varied. For diagnostic tests or clinical prediction rules intended to assess risk of severe hyperbilirubinemia, clinicians likely desire a test that has high sensitivity (few false-negatives) so no infants are misclassified as low risk when, in fact, they eventually develop severe hyperbilirubinemia. This tends to compromise specificity, which cannot be ignored because it determines what proportion of truly low-risk infants are characterized as such and spared additional testing and parental and provider concern. The shape of the ROC curve shows how much specificity can be increased without a dramatic reduction in sensitivity as the positivity criterion is made more stringent (increased). Figure 2 compares two ROC curves that have equal areas under the curve, but curve A maintains high sensitivity as the positivity criterion and specificity are increased, while curve B displays a more rapid decline in sensitivity as the positivity criterion is increased.
The c-statistic commonly is reported as a measure of overall accuracy of a prediction model and is mathematically equivalent and interchangeable with the area under the ROC curve. It is defined as the proportion of all pairs of patients, one who has and one who does not have the outcome, in which the patient having the outcome had the higher predicted probability (generated by the risk assessment tool) of having the outcome.
In addition to considering the diagnostic accuracy of a risk assessment strategy, clinicians must consider certain practical issues in deciding whether to adopt a risk assessment strategy. These include the accuracy and reliability with which the risk factors used in performing the risk assessment can be obtained, the timing of risk factor availability, and the convenience and cost of the assessment to the patient and clinician.
ACCURACY AND RELIABILITY OF RISK FACTOR DETERMINATION.
Factors used to assess risk of severe hyperbilirubinemia include bilirubin measurements (both serum and transcutaneous), other laboratory tests (Coombs positivity), demographic factors (age, race, ethnicity), historical factors (previous infant requiring phototherapy), details of the perinatal course (vacuum delivery), and physical findings (cephalohematoma, extent of jaundice). The degree to which these factors, all of which have been shown to be associated with severe hyperbilirubinemia, can be used for risk assessment depends on the accuracy and reliability with which they can be obtained. Factors that are more objective, such as laboratory tests and observed events (eg, vacuum delivery), are more likely to be accurately and reliably determined than subjective factors, such as physical findings, or demographic factors, such as race and ethnicity, which are ill-defined social constructs.
TIMING OF RISK FACTOR AVAILABILITY.
The earlier an infant is recognized as being at risk for severe hyperbilirubinemia, the sooner preventive interventions can be initiated. However, not all risk factors are available for use in risk assessment at the same time. Certain historical and demographic factors, such as gestational age and race, are available in the delivery room, whereas feeding history, extent of jaundice, and predischarge bilirubin values become known only later in the birth hospitalization.
CONVENIENCE AND COST.
Screening in the newborn nursery is generally a high-volume operation and, therefore, convenience and cost cannot be ignored. Minimizing the number of factors needed in a risk assessment strategy and using factors that are easily and inexpensively obtained can increase the efficiency and decrease the cost of screening. Patient preference also must be considered. For example, parents are likely to prefer noninvasive bilirubin testing to serum bilirubin testing that requires a heel stick or venipuncture.
Comparing Risk Assessment Strategies: How Do They Measure Up?
Several factors make it difficult to compare the predictive accuracy of recently proposed risk assessment strategies. First, the predicted outcome –severe neonatal hyperbilirubinemia –has been defined differently in the studies evaluating various risk assessment strategies. Some studies used the development of an hour-specific bilirubin greater than the 95th percentile as the predicted outcome, (10)(11)(15) while others used absolute bilirubin greater than 20 mg/dL (342 mcmol/L) or 25 mg/dL (427.5 mcmol/L) (regardless of postnatal age) (14)(16)(18) or a phototherapy treatment threshold value (16) as the outcome of interest. The definition of severe neonatal hyperbilirubinemia may affect both the factors that are found to be predictive in a model and the predictive accuracy of the model. It also is important to acknowledge that none of the published risk assessment strategies has been validated prospectively on an independent sample of patients. Clinical prediction rules seldom predict outcomes in an independent sample of patients as accurately as they do in the sample of patients from which the rule was derived, so their true accuracy when applied in multiple and diverse patient populations is not known. (19) Nonetheless, important and instructive patterns do emerge in comparing alternative risk assessment strategies.
First, studies consistently demonstrate that determining the hour-specific TSB before discharge is the most accurate method for assessing risk of severe hyperbilirubinemia. When compared in the same population, risk assessment strategies that use a predischarge bilirubin value consistently demonstrate a higher c-statistic (equivalent to the area under the ROC curve) compared with risk assessment models that use clinical risk factors alone (Table 3). For example, in the comparison by Keren and associates (11) of these two risk assessment strategies for predicting a postdischarge hour-specific TSB greater than the 95th percentile, the predischarge bilirubin risk zone had better discrimination (c=0.83) than the clinical risk factor score (c=0.71). Similarly, Newman and colleagues (18) found that the predischarge bilirubin value had better discrimination than the risk index for predicting development of a bilirubin greater than 20 mg/dL (342 mcmol/L) (c=0.79 and 069, respectively). Categorizing the predischarge bilirubin value into four risk zones simplifies the process of risk stratification, but, as pointed out by Newman, wastes important information because infants who have predischarge hour-specific bilirubin values higher than the 99th percentile are much more likely to develop severe hyperbilirubinemia than those whose predischarge hour-specific bilirubin values are exactly at the 95th percentile. By converting the predischarge bilirubin into a z-score—a standardized and continuous measure of predischarge risk—Newman showed that the c-statistic for predicting TSB greater than 20 mg/dL (342 mcmol/L) could be increased from 0.79 to 0.83. (19)
|Prediction Model (author, year)||Variables||Outcome Predicted||c-statistic (95% confidence interval)|
|Keren, 2005 (11)||Predischarge TSB risk zone||Postdischarge TSB >95th percentile||0.83 (0.80–0.86)|
|Newman, 2005 (18)||Predischarge TSB risk zone||TSB >20 mg/dL (354 mcmol/L); TSB >25 mg/dL (427.5 mcmol/L)||0.79 (0.77–0.81); 0.83 (0.77–0.89)|
|Clinical Risk Factors|
|Keren, 2005 (11)||Birthweight, gestational age, oxytocin use during delivery, vacuum extraction, method of feeding||Postdischarge TSB >95th percentile||0.71 (0.66–0.76)|
|Newman, 2005 (18)||Race, cephalhematoma, maternal age, gestational age, infant sex||TSB >20 mg/dL (354 mcmol/L); TSB >25 mg/dL (427.5 mcmol/L)||0.69 (95% CI not reported); 0.83 (0.77–0.89)|
|Chou, 2003 (16)||Race, breastfeeding, gestational age||TSB ≥20 mg/dL (354 mcmol/L); TSB ≥age-specific AAP phototherapy criteria||0.79 (95% CI not reported); 0.69 (95% CI not reported)|
|Stevenson, 2001 (15)||End-tidal carbon monoxide level, method of feeding, birthweight||TSB ≥95th percentile at 96±12 h after birth||Not reported|
|Newman, 2005 (18)||TSB z-score plus partial risk index||TSB ≥20 mg/dL (354 mcmol/L)||0.86 (0.84–0.88)|
TSB=total serum bilirubin, CI=confidence interval, AAP=American Academy of Pediatrics
Second, it is clear that no single cutoff for any of the risk assessment strategies can discriminate adequately which infants eventually develop severe hyperbilirubinemia. Conservative cutoffs (eg, predischarge >40th percentile) are needed to produce close to 100% sensitivity (no false-negatives), but these invariably result in poor specificity. In the comparison of a predischarge hour-specific TSB risk zone and a clinical risk factor-based risk assessment of Keren and associates, neither strategy could predict the outcome with 98% or greater sensitivity without compromising specificity (21% and 13%, respectively). (11) Thus, a multitiered (or perhaps even continuous), rather than dichotomous, risk stratification scheme must be used to assign risk of severe hyperbilirubinemia. Risk strata-specific likelihood ratios or computer software can be used to generate predicted probabilities of severe hyperbilirubinemia, and these probabilities can be used to determine subsequent action.
Finally, it is apparent that combining clinical risk factors and predischarge bilirubin values may offer additional predictive performance above either approach used alone, although the incremental benefit has not been studied extensively, and methods for operationalizing this more complex risk assessment strategy have not been developed. Newman and colleagues (18) showed that the addition of clinical risk factors, especially gestational age, to the predischarge hour-specific bilirubin risk zone improved overall discrimination (increase in c from 0.79 to 0.83) and produced further stratification of risk, particularly for children who had predischarge TSBs in the high- and high intermediate-risk zones. Tools that would allow a clinician to combine risk factors easily using, for example, a risk score, risk table, or computer program, are not currently available but would not be difficult to develop.
Limitations of Current Risk Assessment Strategies and Future Research
None of the risk assessment strategies developed to date has been validated prospectively in a large and diverse newborn population. Most previous studies in which risk assessment strategies were developed relied on retrospective chart review or administrative data and were purely observational, with bilirubin values obtained at the discretion of the physician as part of routine care, rather than on every infant, both before and after discharge, as part of a research protocol. These studies, therefore, have multiple potential limitations, including incomplete and inaccurate assessment of risk factors and follow-up bias that occurs when only infants believed to be at risk have the outcome of interest (in this case, a peak bilirubin measurement) verified. Given these limitations, it is unclear if existing risk assessment strategies will have good predictive performance when prospectively applied in all newborns. Therefore, large prospective cohort studies that involve exhaustive risk factor data collection and complete pre- and postdischarge bilirubin measurements are needed to determine whether existing risk assessment strategies have good predictive performance in newborn populations that are ethnically, racially, and geographically diverse.
Current risk assessment strategies help to predict which infants will develop hyperbilirubinemia, but the risk assessments are not paired with recommended actions, leaving clinicians to decide for themselves the best subsequent course of action. Future studies must link risk assessment to management decisions and test the effectiveness and cost-effectiveness of this decision support. Examples of decisions that should be driven by risk assessment information include the frequency and timing of subsequent bilirubin measurements and the timing of follow-up with a clinician. Ideally, the risk assessment and decision support output should be computerized and made available over the Internet on computers in the newborn nursery or on handheld personal digital assistants. An example of such a computerized decision support tool can be found at www.bilitool.org.™ Although it may be difficult to reach consensus on risk-based rules for subsequent management, because management decisions often are affected by the local availability of health-care resources (eg, visiting nurses, home phototherapy, office hours), even broad parameters for risk-based treatment rules should help to standardize newborn jaundice care and prevent the development of severe neonatal hyperbilirubinemia and kernicterus. Studying the effectiveness and cost-effectiveness of alternative risk-based management algorithms will help to make more explicit the trade-offs that clinicians face in balancing a desire to avoid kernicterus with a desire to avoid unnecessary testing and treatment in newborns.
Finally, in the future we need to determine the utility of new predictors as well as new and existing technologies for identifying infants at risk of severe hyperbilirubinemia. Transcutaneous bilirubin meters already are replacing serum bilirubin measurements for predischarge screening in many nurseries. Although these meters appear accurate at lower bilirubin values, their predictive accuracy has yet to be evaluated. Measurement of ETCO to identify infants who are experiencing red blood cell hemolysis eventually may become a cost-effective addition to the risk assessment process. And as we gain better understanding of the genetic determinants of neonatal hyperbilirubinemia, screening for specific genetic markers, such as G6PD enzyme polymorphisms or mutations in glucuronyl transferase, has the potential to refine our risk assessment strategies further.
Dr Bhutani did not disclose any financial relationships relevant to this article. Dr Keren disclosed that he has received material support for past research from Respironics.
- Copyright © 2007 by the American Academy of Pediatrics
Holt LE. The Diseases of Infancy and Childhood: For the Use of Students and Practitioners of Medicine. New York, NY: D. Appleton; 1897
Maisels MJ, Kring E. Length of stay, jaundice, and hospital readmission. Pediatrics. 1998;101:995–998
Maisels MJ, Newman TB. Kernicterus in otherwise healthy, breast-fed term newborns. Pediatrics. 1995;96:730–733
Subcommittee on Hyperbilirubinemia. Management of hyperbilirubinemia in the newborn infant 35 or more weeks of gestation. Pediatrics. 2004;114:297–316
Bhutani VK, Johnson L, Sivieri EM. Predictive ability of a predischarge hour-specific serum bilirubin for subsequent significant hyperbilirubinemia in healthy term and near-term newborns. Pediatrics. 1999;103:6–14
Keren R, Bhutani VK, Luan X, Nihtianova S, Cnaan A, Schwartz JS. Identifying newborns at risk of significant hyperbilirubinaemia: a comparison of two recommended approaches. Arch Dis Child. 2005;90:415–421
Eggert LD, Wiedmeier SE, Wilson J, Christensen RD. The effect of instituting a prehospital-discharge newborn bilirubin screening program in an 18-hospital health system. Pediatrics. 2006;117:e855–e862. Available at: http://pediatrics.aappublications.org/cgi/content/full/117/5/e855
Maisels MJ, Kring E. Transcutaneous bilirubin levels in the first 96 hours in a normal newborn population ≥35 weeks’ gestation. Pediatrics. 2006;117:1169–1173
Stevenson DK, Fanaroff AA, Maisels MJ, et al. Prediction of hyperbilirubinemia in near-term and term infants. Pediatrics. 2001;108:31–39
Chou SC, Palmer RH, Ezhuthachan S, et al. Management of hyperbilirubinemia in newborns: measuring performance by using a benchmarking model. Pediatrics. 2003;112:1264–1273