Prediction of Hyperbilirubinemia in Near-Term and Term Infants

Participant Criteria

Infants were eligible for the study if their gestational age was ≥35 weeks as determined by best obstetric estimate and if enrollment could be accomplished within the first 36 hours of life during the period of February 20, 1998, through February 22, 1999. Exclusion criteria were neonates having any illness that would require admission to the neonatal intensive care unit (NICU) before 24 hours (a workup to rule out sepsis in a neonate who appeared to be well or maternal treatment with antibiotics were not exclusion criteria) or severe congenital anomalies. Also excluded were infants who were in incubators; who had pulmonary disease requiring oxygen or ventilatory assistance via hood, tent, nasal cannula, continuous positive airway pressure, or ventilator; who had significant nasal obstruction in either naris; who had a birth weight of 850 g or less; or who had respiratory rates ≤10 or ≥100 breaths per minute (operating range for the CO-Stat End-Tidal Breath Analyzer [Natus Medical Inc., San Carlos, CA]).

Study Design

This cohort study included all infants who met the eligibility criteria and for whom informed consent was obtained and had mandatory measurements of ETCOc and STB performed at 30 ± 6 hours of age and STB at 96 ± 12 hours. Additional STB measurements were performed according to a flow diagram (Fig 1) and an hourly bilirubin table (Table 1) based on a racially diverse population of 17 854 term- and near-term infants, with 15 014 neonates contributing STB values for the period of 18 to 24 hours.¹⁰ Enrollment consisted of neonates from 9 clinical sites, 4 domestic and 5 international. Each study site enrolled eligible infants serially on a schedule determined by the circumstances and restraints imposed by clinical operations and personnel limitations at the respective institutions. Each site maintained an enrollment log identifying all infants eligible to be studied and selected as well as the reasons for nonenrollment of any eligible neonate. The CO-Stat End-Tidal Breath Analyzer with single-use disposable nasal sampler was used to analyze the breath of all infants for ETCOc.^13,,14The device was calibrated locally every 30 days. The accuracy of the analyzer is ± 0.3 ppm (or μL/L) or 10% of the reading (whichever is greater) for respiratory rates between 10 and 60 breaths/min and ± 0.3 ppm or 15% of the reading (whichever is greater) for breathing rates >60. In a previous bedside evaluation study of the instrument,¹³ the reproducibility and coefficient of variation of ETCOc measurements when sampling neonates and adults were determined to be better than other methods tested, such as hand sampling and semiautomatic electrochemical detection. In addition to ETCOc, other parameters measured by the device include breath CO concentration (uncorrected for ambient CO), ambient CO concentration (reflecting inhaled air), end-tidal carbon dioxide (ETCO₂), and respiratory rate. The device uses side-stream sampling to draw nasal air continuously through the sampler at 60 mL/min. The sampler is made of a clear polymer with an inner and outer diameter of 0.8 and 1.5 mm, respectively. Adhesive wings (8 × 5 mm) allow a maximum insertion depth of 6.0 mm before coming into contact with the lower edge of the nostril. Because participants are not required to control their breathing patterns, this device can be used with infants and others who are unable to accomplish breath control.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

Study protocol decision-making flow diagram.

Blood was collected and the serum was analyzed for STB in each enrolled infant (Fig 1). Each site used its own clinical laboratory and method for all STB measurements (Hitachi 917 Spectrophotometrix Analyzer [Roche-Diagnostics, Inc, Tokyo, Japan], Unistat Bilirubinometer [Reichert Jung, Vienna, Austria] at 2 sites, and the Hitachi 747 (Roche-Diagnostics, Inc] at all other sites).¹⁵ Infants were defined as hyperbilirubinemic if their hours of age-specific STB at any time during the study was ≥95th percentile, as defined by Table 1. In addition, during the first 24 hours, the attending physician had the option of measuring the STB for an infant who exhibited visible jaundice. For infants whose STB was ≥95th percentile, an ETCOc measurement was performed and the infant was excluded from further study. Infants who were not excluded before 24 hours had the first scheduled STB and ETCOc measurements performed at 30 ± 6 hours in coordination with state-mandated newborn screening for inborn errors of metabolism and additional required tests. Infants with STB ≥95th percentile exited the study at this time. All other infants remained in the study. From 24 to 84 hours, measurements of STB were performed at the discretion of the physician. If the physician elected to perform an STB between 24 and 84 hours, then the infant exited the study if the level was ≥95th percentile. If the STB was not ≥95th percentile or if no STB was obtained between 24 and 84 hours, then the infant had a mandatory STB performed at 96 ± 12 hours. If the STB was ≥95th percentile or <40th percentile, then the infant exited the study. Any STB between the 40th and the 94th percentiles required a repeat STB in 24 to 48 hours based on the result. If the 96 ± 12 hour measurement was ≥75th percentile and <95th percentile, then the STB was repeated within 24 hours. If the 96 ± 12 hour measurement was >40th percentile and <75th percentile, then the STB was repeated within 48 hours. An infant who had an STB between the 40th and the 94th percentiles continued to have STB measured every 24 to 48 hours until the STB was ≥95th percentile or <40th percentile, the most recent STB was less than or equal to the STB result 24 to 48 hours earlier, or the infant attained 168 hours of life. At 168 hours, all infants exited the study with ongoing care provided at the discretion of the physician.

Statistical Methods

Of the 1895 participants who were enrolled in the study, 1370 completed the study. Of the excluded infants, 88 were enrolled in the study but did not participate in the testing, 272 were lost to follow-up, 131 had technical problems with preproduction samplers, 32 could not be tested with the CO-Stat device for other reasons (eg, hydrogen interference, irregular breath patterns), and 175 did not adhere to the study protocol. In addition, another 131 infants were tested at a tenth site but were excluded from analysis because measurements of STB were taken with a transcutaneous bilirubinometer rather than performed with the standard blood test. Some infants were excluded for multiple reasons.

The study was conducted in 2 phases. In phase I, 577 infants were tested with a preproduction CO-Stat nasal sampler; in phase II, 793 infants were tested with the production sampler. The ETCOc results from phase I showed a higher mean ± standard deviation (SD) than those values from phase II (1.70 ± 0.54 vs 4.48 ± 0.49 ppm, respectively). This was attributable to a slight outgassing of CO from a material used in the preproduction sampler. The ETCOc data from the 2 phases of the study were combined and normalized by converting the measurements to Z scores (ie, standardized deviation of the measurement from the mean, in SD units). All analyses were conducted in Z units, and the ETCOc values reported throughout this article have been converted back to ppm with the distribution having the mean and SD of phase II.

Receiver operator characteristic (ROC) analyses^16–18 for the prediction of hyperbilirubinemia were performed for 3 different parameters of predicting hyperbilirubinemia: ETCOc at 30 ± 6 hours, STB percentile (as defined by the table) at 30 ± 6 hours, and a combination of STB and ETCOc test at 30 ± 6 hours. A positive result for the combined test was defined as either an STB between the 75th and 95th percentile or an ETCOc greater than the population mean or both at 30 ± 6 hours. Infants who presented with STB ≥95th percentile at 30 ± 6 hours were excluded from the STB and combined ROC analyses because they had been found to be hyperbilirubinemic at the predicting measurement. These infants were not excluded from the ETCOc ROC analysis, because the predicting and outcome measurements were different although they occurred in the same time window.

Two stepwise logistic regression analysis models for the prediction of hyperbilirubinemia (as defined above) were performed on the data: 1 for noninvasive (ie, ETCOc) measurements and 1 for invasive measurements (ie, STB). The following variables were used in the noninvasive model: ETCOc at 30 ± 6 hours, presence of bruising, feeding type (breast milk, formula, or both), birth weight, race (Asian, white or Hispanic, black, other), maternal diabetes, type of labor (vaginal or cesarean section), gender, infection, cephalhematoma, pregnancy-induced hypertension/preeclampsia, gravidity, parity, maternal blood type, and maternal Rh status.

At each step in building the first model, the variable that reduced the total squared error was added first. Two stopping rules were used: 1) the addition of the next variable reduced the overall squared error by <1%, and 2) the probability of hyperbilirubinemia was not statistically different for the candidate variable to be added to the model.