METHODS

Neonates at the Simpson Memorial Maternity Pavilion, Edinburgh, who required blood sampling for SBR were enrolled in the study with parental consent. All neonates were clinically jaundiced but otherwise well. Three bilirubinometers (see below) were applied to the infant’s forehead within 30 minutes of SBR sampling by venepuncture or heel lance. All transcutaneous bilirubinometry (TcB) measurements were performed by one author (PvD) with the infant lying supine. The forehead was not exposed to direct sunlight and care was taken to avoid skin areas that were bruised, excessively hairy, or hypermelanotic. Gestation, birth weight, postnatal age, and ethnicity were recorded. Infants who had received phototherapy or exchange transfusion were excluded.

SBR samples were sent immediately for analysis by an automated Hitachi 911 multichannel analyser. The method used involves oxidation of bilirubin to biliverdin by potassium ferricyanide in the presence of caffeine.¹⁰ This technique is reliable even in the presence of haemolysis up to 5.0 g/l and is therefore particularly suitable for neonatal samples. Interbatch quality control coefficients of variation are 3.0% at 53 μmol/l, 1.4% at 124 μmol/l, and 1.6% at 231 μmol/l. The laboratory participates in the United Kingdom External Quality Assessment Scheme, and shows a mean + 2.4% bias for SBR analysis.

Bilirubinometer specifications and method of use

The devices compared were two SpectRx BiliCheck (SpectRx, USA), designated A and B, and one Minolta AirShields JM102 (now Hill-Rom AirShields International). Both systems rely on spectrophotometric methods.

Detailed information on the SpectRx device has been given elsewhere.⁷ The SpectRx instruction manual and website claim the device to be accurate at gestational age 27–42 weeks, postnatal age 0–20 days, and weight 950–5200 g. It is also claimed to be accurate regardless of skin colour by compensating for the confounding effects of dermal maturity, melanin, and haemoglobin on spectral reflectance. The SpectRx uses disposable probe tips which require calibration before each measurement. The tip is then applied with light pressure to the skin. The SpectRx device is designed to be triggered for five consecutive spectral analyses before it provides one displayed averaged reading in either μmol/l or mg/dl. In other words, it has an internal averaging system in an attempt to improve accuracy. For the study, three of these readings were obtained per device to determine intradevice precision. The average of these readings was used as the operative result in all comparisons with the Minolta, and for calculating interdevice precision.

The Minolta AirShields also measures the spectral reflectance of bilirubin but does not claim to correct for confounding effects. Detail on operation principles is described elsewhere.⁶ It has been extensively studied and validated for use as a screening tool for hyperbilirubinaemia in term infants.^4,11–13 There have also been reports in the literature on its use in ethnic minorities with various skin colouring.^11,12,14,15 The Minolta device does not use disposable parts or require repeated calibration. The probe tip is applied to the skin with increasing pressure until a flash is activated and a reading obtained. Readings are produced as a numeric index. The Minolta was used according to the protocol of the study centre, recording three consecutive readings, and choosing the maximum value for clinical decision making.

Data and statistical analysis

For statistical accuracy, each patient was assessed only once. All statistical calculations were performed using Microsoft Excel 97. Pearson correlation coefficients were calculated using original Minolta index readings. For measurement of agreement by the Bland-Altman method,¹⁶ Minolta index readings were transformed into μmol/l using linear regression of SBR on TcB readings. The transformed values were used for calculation of the differences between SBR and TcB values, the mean difference and 95% confidence intervals of the difference.

RESULTS

From June to September 2000, 64 neonates were enrolled, 19 of whom were preterm (31–35 weeks). Table 1 gives patient details. All were less than 20 days old. Six infants were non-white. None had a haemoglobinopathy. SBR levels ranged from 50 to 349 μmol/l.

Agreement between SpectRx and Minolta devices and SBR

Table 2 shows agreement results for TcB readings against SBR. The 95% confidence intervals of a device reading corresponding to SBR were ± 66.7, ± 67.9, and ± 66.4 μmol/l for the Minolta, SpectRx A, and SpectRx B respectively. The Bland-Altman scatterplots for the devices show a tendency for both Minolta and SpectRx devices to underestimate SBR at higher levels (figs 1–3).

For preterm infants, the Minolta device showed a tendency to overestimate SBR but the device has not been validated for such use. A good correlation with SBR was seen in non-white infants for the Minolta, SpectRx A, and SpectRx B (r = 0.94, 0.95, and 0.99 respectively), but because of small numbers (n = 6), it was deemed inappropriate to calculate statistical agreement.

Accuracy can be determined by agreement, but for practical purposes (such as defining thresholds for performing SBR), the maximum differences between device readings and SBR should be considered. Consequently 100% of infants with hyperbilirubinaemia requiring treatment would be identified. Table 2 also displays the maximum range of differences between devices and SBR. For example, to identify 100% of infants with SBR ≥ 250 μmol/l, blood samples need to be obtained from all infants whose Minolta reading is ≥ 170 μmol/l (index reading 15.5) or whose SpectRx reading is ≥ 150 μmol/l.

Positive predictive value and number of SBR samples required to identify all SBR ≥ 250 μmol/l

When the threshold for each device is defined to achieve 100% sensitivity for SBR ≥ 250 μmol/l, the positive predictive values are similar (table 3). Using the devices to screen for this threshold would reduce SBR sampling by 23% for the Minolta device, 16% for SpectRx A, and 20% for SpectRx B. Thus the SpectRx does not result in fewer SBR samples being taken to identify all infants with significant hyperbilirubinaemia.

Intradevice precision

The median coefficients of variation for the Minolta, SpectRx A, and SpectRx B were 1.49%, 8.94%, and 8.51% respectively. Maximum coefficients of variation were 6.74%, 18.67%, and 27.64% respectively.

Interdevice precision

The 95% confidence interval of difference in readings for the two SpectRx devices was ± 37.4 μmol/l.

DISCUSSION

This study is the first to compare the performance of the new SpectRx BiliCheck with the Minolta AirShields. Given that SBR levels range from 50 to 400 μmol/l in jaundiced infants, both instruments have wide 95% confidence intervals at ± 67 μmol/l. A TcB reading that is different from SBR by up to 100 μmol/l (the maximum difference in this study) can have implications for starting phototherapy or exchange transfusion.⁹ Such a difference could possibly also have an effect on whether an infant develops kernicterus or not. These devices are useful as screening tools for hyperbilirubinaemia, but not as replacements for measuring SBR.

Comparison of SpectRx and Minolta devices

Our study has shown that, as a screening tool, there is no advantage in using the SpectRx device over the Minolta. Although the SpectRx provides readings in μmol/l for easier interpretation, and claims to be accurate regardless of skin colour, we have found that it is neither more accurate than the Minolta device nor does it reduce the number of samples needed to identify significant hyperbilirubinaemia. Agreement is a better measure of accuracy than correlation, and the agreement between the device readings and SBR is poor.

Positive predictive values are acknowledged to vary with disease prevalence, in this case the number of jaundiced infants deemed to require SBR sampling. Hence our positive predictive values should not be extrapolated to a different population of infants, but nevertheless provide useful data for cross comparisons between the devices. The positive predictive values of the devices were similar for all three devices.

Clinical usefulness of the devices as screening tools was assessed by the reduction in SBR sampling while ensuring that all samples with SBR ≥ 250 μmol/l were identified. Thus for the Minolta, SpectRx A, and SpectRx B devices respectively, 15 (23%), 10 (16%), and 13 (20%) SBR samples could have been avoided. The Minolta, in fact, performed slightly better with respect to reduction of SBR sampling than the SpectRx devices. Again, these data are valuable primarily for comparing the devices with each other.

Whichever screening tool is used, the Bland-Altman scatterplots show a tendency for the devices to underestimate SBR at higher levels. This implies that the devices are less accurate when the risk for kernicterus is higher.

Accuracy of SpectRx device

Although the aim of the study was to compare the SpectRx with the Minolta, we simultaneously evaluated accuracy for validation purposes. At that time, no published data on the SpectRx device were yet available in the medical literature. Since then, three papers have been published on the accuracy of the SpectRx device.^7,8,17 The correlation coefficient for the SpectRx in this study (r = 0.87) is comparable to those obtained by Bhutani et al⁷ (r = 0.91), Rubaltelli et al⁸ (r = 0.89) and Knupfer et al¹⁷ (r = 0.73). The 95% confidence intervals of the difference are also similar at ± 67 μmol/l in this study, ± 55 μmol/l in Bhutani et al,⁷ and ± 76 μmol/l in Rubaltelli et al.⁸ Our data are thus analogous to studies carried out elsewhere and add to the literature about the validity of the SpectRx device.

TcB as a replacement for SBR sampling

Perhaps one limitation of this study is that SBR was analysed by a clinical laboratory method rather than the ideal method of high performance liquid chromatography (HPLC). It has been argued that, when compared with HPLC, some clinical laboratory analyses of SBR use automated methods that are subject to errors as wide as those obtained with TcB.⁸ Our study used the Hitachi 911 multichannel analyser with the potassium ferricyanide method, which has been shown to be an accurate and precise assay for total SBR.¹⁰ Furthermore, our laboratory participates in rigorous internal and external quality control procedures. Each institution should be aware of the accuracy of their local clinical laboratory method for measuring SBR before deciding whether transcutaneous devices are suitable replacements.

Rubaltelli et al⁸ propose that transcutaneous measurements could be used to replace invasive SBR sampling altogether. They reported correlation coefficients for SpectRx compared with HPLC (r = 0.89), and laboratory SBR with HPLC (r = 0.93). The corresponding 95% confidence intervals of the difference were ± 76 μmol/l and ± 63 μmol/l (non-significant at p = 0.53). Despite not reaching statistical significance, the 95% confidence intervals were superior for laboratory SBR. The authors concluded that the SpectRx performed as well as laboratory SBR. However, the overall accuracy of their laboratory SBR results was compromised by the use of nine different automated laboratory methods. It is known that certain methods perform more accurately than others,¹⁸ but no information was provided about the individual performance of each of the nine methods.

In addition, no studies have addressed the question of whether skin or serum levels of bilirubin more accurately reflect its deposition in the central nervous system. Some suggest that transcutaneous readings are more representative of tissue deposition of bilirubin,^19,20 whereas others propose that circulating serum bilirubin is more available to move across the blood-brain barrier.⁸ Factors that affect blood-tissue bilirubin distribution include exchange transfusion, phototherapy, albumin levels, pH, and gestational and postnatal age. This debate on the use of TcB as a replacement for SBR sampling would be ideally resolved by conducting a randomised controlled trial of the two diagnostic methods with long term outcome measurements and cost-benefit analyses. This would require a large, multicentre setting with years of multidisciplinary follow up, and the cost would be prohibitive. The current evidence also shows that the transcutaneous methods at present available are not accurate enough to proceed with such a trial.

Summary

The purpose of measuring bilirubin levels, by whatever method, is to identify infants who require treatment. The inaccuracy of existing transcutaneous bilirubinometers means that they should only be used as screening tools for SBR sampling. The newer SpectRx BiliCheck is no more accurate than the Minolta AirShields, and is more complicated to use.