Total bilirubin assay differences may cause inconsistent treatment decisions in neonatal hyperbilirubinaemia

Introduction

The majority of newborns develop some degree of jaundice within the first few days of birth. It is usually harmless, resolves by 2 weeks of age and is due to a high unconjugated bilirubin burden caused by the normal transitional increase in erythrocyte turnover after birth combined with immaturity of hepatic uridine 5′-diphosphoglucuronosyltransferase and increased enterohepatic circulation.

In some cases, hyperbilirubinemia is exacerbated by erythrocyte destruction due to isoimmune or other haemolytic disease, enhanced uptake of bilirubin from the gut, e.g. due to poor feeding or gastrointestinal obstruction, and other genetic or environmental factors [1, 2]. For these infants, treatment to reduce bilirubin may be required as high levels of free (non-albumin bound) unconjugated bilirubin are neurotoxic and may cause kernicterus spectrum disorder (KSD). There are a number of susceptibility factors for neuronal injury due to bilirubin including prematurity, inflammation, haemolysis, low serum albumin, intestinal obstruction and acidosis. The presence of these are taken into account in the decision to treat [3], [4], [5]. Phototherapy is an effective treatment for hyperbilirubinemia in most cases while exchange transfusion is initiated if the infant shows signs of acute bilirubin encephalopathy, total serum bilirubin (TSB) exceeds defined concentrations, phototherapy fails or if other risk factors are present [1].

Irrespective of the cause of hyperbilirubinaemia or the presence of risk factors, all jaundiced infants must be monitored for the development of severe hyperbilirubinaemia and given treatment, if indicated, to prevent KSD.

TSB is currently accepted as the definitive biomarker for the diagnosis of neonatal hyperbilirubinemia and TSB (direct or conjugated bilirubin must not be subtracted) concentration is combined with the presence or absence of defined neurotoxicity risk factors to determine management of hyperbilirubinemia according to published algorithms [1, 6, 7]. However, there is no definition of what constitutes a safe bilirubin concentration and there is concern that vulnerable neonates may acquire KSD if their jaundice is not recognised as severe enough for treatment or if it is prolonged but slightly below treatment thresholds based on TSB concentration [8]. On the other hand, unnecessary use of phototherapy, sometimes requiring hospitalisation, wastes healthcare resources and isolates infants from caregivers at a crucial time for infant bonding [1].

The 2004 American Academy of Pediatrics clinical practice guideline for the management of hyperbilirubinemia in the newborn infant ≥35 weeks of gestation [1] is used as the basis for management of neonatal jaundice in many countries, including Australia [7], [8], [9], [10]. The guideline is consensus-based and uses postnatal hour-specific TSB to designate risk of significant hyperbilirubinemia according to the Bhutani nomogram [1]. If risk is above a certain threshold on the Bhutani nomogram, the hour-specific TSB is then interpreted in light of other risk factors for neurotoxicity, including low serum albumin, to guide the use of phototherapy and/or exchange transfusion, if required, in these infants [1].

The Bhutani nomogram was derived in 1999 using TSB concentrations from a 2,5-dichlorophenyldiazonium tetrafluoroborate (DPD) diazo assay on the Hitachi 747 platform under National Institute of Standards and Technology (NIST) guidelines for accuracy and precision [11]. However, it was noted in the 2004 American Academy of Pediatrics clinical practice guideline that there was significant interlaboratory variability in TSB results and that this should be rectified [1, 12, 13]. In 2008 the College of American Pathologists identified a lack of reliable instrument calibrators as being responsible for differences observed in neonatal bilirubin measurements across a variety of instruments [14]. This variability between laboratories was confirmed in a Dutch study [15] and more recently by a French national working group [16]. A standard reference material for bilirubin (SRM 916) was first certified in 1971 [17]. New material, SRM 916a, was characterised in 1988 [18] and available from the NIST until 2017. The latest NIST SRM 916b was certified and made available for use in 2021 [19].

This study was initiated following a complaint from a clinician who had transferred a newborn to a tertiary hospital for treatment of severe hyperbilirubinemia but, on arrival, the baby was recategorized into a lower risk category due to a 20% difference in TSB between laboratories. Therefore we investigated whether there are clinically significant method-dependent differences in TSB results from neonatal samples.

Materials and methods

Fresh residual plasma samples from hospital-born infants were deidentified and pooled in the index laboratory to obtain 11 samples with a range of total bilirubin concentrations in keeping with neonatal jaundice (75–360 μmol/L). Multiple aliquots of each sample were protected from light with aluminium foil, frozen to −20 °C and transported on dry ice to 4 other laboratories in the Sydney (Australia) region. Table 1 shows the analysers, their manufacturers and methods used. The cobas c702 and ABL90 FLEX were located at the index laboratory; the cobas c501, Alinity c and Architect ci16200 were located in another. The remaining platforms were located in one, each, of the remaining 2 laboratories. Human research ethics approval was not required as the samples were left over and not identifiable. Every participating laboratory was accredited to ISO 15189:2012.

Samples were thawed and kept in foil until analysis in duplicate for total bilirubin, direct bilirubin and the hemolysis, icterus and lipemia indices on the Beckman, cobas, Alinity and Architect instruments. On the VITROS instrument, the BuBc slide assay, that measures and adds conjugated and unconjugated bilirubin to obtain neonatal bilirubin was used as it is the recommended method for neonatal bilirubin on VITROS. ABL90 FLEX does not measure conjugated bilirubin or serum indices. Singlicate measurements were made on the VITROS instrument.

The Beckman AU5822 bilirubin assay was chosen as the comparator assay as the manufacturer claims that it is traceable to SRM 916a in the product information. Radiometer also makes this claim for the ABL90 FLEX. The manufacturers of the other instruments in this study do not claim traceability to SRM 916a in their product information.

Paired samples were used to construct scattergrams with linear trendlines through the origin in Excel. Analyse-it software was used for Passing-Bablok analysis and to construct Bland–Altman difference plots using the same paired samples. Means, standard deviations and percentage coefficients of variation were calculated for duplicate results using Excel.

Data from the Royal College of Pathologists of Australasia Quality Assurance Programs’ (RCPAQAP) Neonatal Bilirubin program from 2021 was also analysed. In this program human serum is supplemented with unconjugated and conjugated (ditaurobilirubin) bilirubin to provide total concentrations between 80 and 320 μmol/L. Most samples contain about 35% conjugated bilirubin. An additional high bilirubin sample (also 320 μmol/L) without added conjugated bilirubin is included to mimic common neonatal samples. The 2 high-concentration samples were value assigned by the Reference Standards Laboratory, Children’s Wisconsin, Milwaukee, WI, USA with a value of 319 μmol/L for both the sample including conjugated bilirubin and the sample with no conjugated bilirubin. The Beckman-Coulter average for these 2 samples was also 319 μmol/L. During 2021 all RCPAQAP Neonatal Bilirubin samples were measured on 4 occasions during the year. The average percentage bias for these 2 high-concentration samples for each method group was assessed using data where the laboratory had measured the sample at least twice. Outliers were excluded using Tukey’s method.

The RCPAQAP analytical performance specifications for the Neonatal Bilirubin program (±8 μmol/L for results up to 80 μmol/L and ±10% for results greater than 80 μmol/L) are based on allowable imprecision of assays and have been used to comment on the performance of individual platforms [20].

Results

No samples were unsuitable for analysis on the basis of hemolysis, icterus or lipemia. Apart from the VITROS laboratory in which samples were measured in singlicate, 2 other laboratories were not able to produce duplicate results for 2 samples due to insufficient sample volume. Conjugated bilirubin (16–49% of total bilirubin, 58–139 μmol/L, depending on method) was present in 3 samples and was <10% (0–14 μmol/L) in all other samples. This did not provide enough data for any analysis other than general observations.

Figure 1 shows the comparison of total bilirubin results from each platform as a scatterplot with RCPAQAP analytical performance specifications superimposed. Using AU5822 as the comparator, results obtained from Architect ci16200 and cobas c702 exceeded the RCPAQAP analytical performance specifications with the Architect ci16200 producing higher results and the cobas c702 lower results.

Figure 1:

Scattergram of results from each platform with linear trendlines set to pass through the origin. Solid black lines indicate RCPAQAP high and low analytical performance specifications (+8 μmol/L for results up to 80 μmol/L and +10% for results greater than 80 μmol/L) for the Neonatal Bilirubin program.

Figure 2 presents the data using Passing-Bablok regression, confirming the linear trends for most method comparisons. However the data suggest that the VITROS BuBc slide has different analytical specificity to the AU5822 as the scatter around the line of best fit is not explained by analytical imprecision of the assay. Additionally there may be a modest effect of the presence of conjugated bilirubin on the Alinity system compared with the AU5822 as the three highest samples, which contain significant fractions of conjugated bilirubin, are non-linear with the lower concentration samples, without conjugated bilirubin.

Figure 2:

Passing-Bablok regression of total bilirubin (µmol/L) results comparing other platforms with the AU5822.

(A) Alinity c, (B) Architect ci16200, (C) cobas c702, (D) cobas c501, (E) ABL90 FLEX, (F) VITROS 5600. The vertical lines represent the bias range of the methods relative to the Beckman-Coulter method in the RCPAQAP neonatal bilirubin program based on the 319 μmol/L target samples with (blue) and without (orange) the presence of conjugated bilirubin. The three highest concentrations as measured by the AU5822 contained conjugated bilirubin (16–49%).

Bland–Altman difference plots (Figure 3) demonstrate a range of positive and negative biases with results distributed as much as 23% higher and 22% lower than AU5822 data. Imprecision estimations, summarised in Table 2, indicate that repeatability was acceptable with a range of coefficients of variation for duplicate analyses of 0–2.4%. The VITROS results, in which total bilirubin is calculated from the sum of unconjugated and conjugated fractions, show the most scatter. However, only the 3 samples indicated by ‘x’ on Figure 3 contained measurable conjugated bilirubin while the other 8 samples had conjugated bilirubin results of 0 μmol/L by this method. The 2 Roche (cobas c702 and cobas c501) instruments were in close agreement whereas the difference between results from Abbott (Architect ci16200 and Alinity c) instruments averaged 9%. Although instrument manufacturers claim that ABL90 and AU5822 are calibrated to SRM 916a, a mean bias of 8% was observed between the 2 methods. This may be explained by an accuracy drift over time due to unavailability of SRM 916a.

Figure 3:

Difference plots of total bilirubin (µmol/L) results from each platform compared with the AU5822. Dashed lines represent 95% confidence limits. (A) Alinity c, (B) Architect ci16200, (C) cobas c702, (D) cobas c501, (E) ABL90 FLEX, (F) VITROS 5600. Samples with high conjugated bilirubin are labelled ‘x’.

Figure 4 presents the results as they would be interpreted according to the Bhutani nomogram (Figure 4A) and phototherapy recommendations for infants at 72 h postnatal age (Figure 4B) [1]. Results show that different treatment decisions, either phototherapy and/or repeated bilirubin measurements, would arise from use of different assays for bilirubin concentrations close to clinical decision points. This applied to some samples analysed with assays performing within the designated RCPAQAP analytical performance specifications for Neonatal Bilirubin (±8 μmol/L for results up to 80 μmol/L and ±10% for results greater than 80 μmol/L).

Figure 4:

Total bilirubin (µmol/L) results colour coded according to risk zone on Bhutani nomogram at 72 h postnatal age (A) and phototherapy initiation for medium-risk infants (B).

RCPAQAP data

The range of biases seen within each method group in the RCPAQAP Neonatal Bilirubin program are shown in Figure 5. Several conclusions can be drawn from this data. Firstly, the QAP material, despite being spiked human serum, cannot be considered commutable, as the biases seen with patient samples are not the same as the biases seen with the QAP materials. Secondly, there appears to be a difference in bias depending on the presence or absence of conjugated bilirubin. Again, this may be a product of these materials rather than what may be expected in patient samples. In addition, there are 8 instrument group – sample type combinations with a within-method bias range (lowest to highest) of 15% or greater. For any individual result this will be combined with assay imprecision as well as any method bias.

Figure 5:

Ranges of biases within and between bilirubin method groups relative to the Beckman-Coulter AU5200 in RCPAQAP Neonatal Bilirubin program 2021: Dark bars – added conjugated bilirubin; pale bars – no added conjugated bilirubin. Symbols (+ and −) represent average bias relative to AU5822 for patient samples in this study where + indicates samples in which conjugated bilirubin was present (3 samples, 16–49% conjugated bilirubin) and – indicates samples with conjugated bilirubin <10% (8 samples).

Discussion

These results demonstrate interlaboratory differences among commercial total bilirubin assays that are likely to impact the management of neonatal jaundice under current guidelines. This has implications, possibly for the neurodevelopment of infants if severe hyperbilirubinemia is undertreated but also for hospital resourcing if it is incorrectly diagnosed and overtreated.

The data do not suggest that analytical imprecision is a significant factor in the differences, rather it is a matter that calibrator values and, perhaps, matrix require closer alignment. This observation has been made previously, in the early 2000s [14] and again in 2020 [16]. The use of bovine serum rather than human serum-based calibrators has been identified as a possible cause of variability among diazo methods. This may be due to the presence of, as yet uncharacterised, inhibitors of the diazo reaction in commercial bovine serum [21]. To the authors’ knowledge there is currently no formal local or global harmonisation activity for TSB. The International Consortium for Harmonization of Clinical Laboratory Results considers total bilirubin harmonisation to be adequate at present [22]. With the release of SRM 916b for pure bilirubin [19] there is opportunity for manufacturers to assess instrument, including point of care, calibrators and align them to the SRM if required. A matrix-matched reference material is also in development [16].

With respect to quality assurance, the ditaurobilirubin-containing RCPAQAP material appears to react differently in some assays to native bilirubin mono- and di-glucuronide in patient samples. Ditaurobilirubin has the same molar absorptivity as unconjugated bilirubin and has been recognised as a suitable surrogate calibrator for conjugated bilirubin [23]. The effect of using ditaurobilirubin in some assays has been described before, with VITROS analysers (Bu + Bc assays) over-recovering and Architect assays under-recovering [21], as seen here. The use of bovine serum as a matrix has also been described to affect quality assurance results [13, 21] however, as the RCPAQAP material is human serum-based, interference by bovine serum is excluded. Elimination of this issue and ensuring commutability of all samples in the Neonatal Bilirubin program are important and require further investigation.

Another consideration is whether the current guidelines are sufficiently robust. A recent metanalysis found that there is insufficient high-quality evidence to support the use of TSB as a prognostic marker for the development of KSD in most neonates, although the authors found limited data to show, with moderate-certainty, a possible association between TSB and KSD in infants with certain risk factors [5]. While interlaboratory differences in TSB may be a factor in this lack of clarity, within-individual biological variation of bilirubin in newborns may also play a role in the interpretation of a measurement at a single timepoint in jaundiced neonates with neonatal jaundice [24]. However, better harmonisation of bilirubin assays is needed in order to progress research on the effects of bilirubin on the neonatal brain, the relationship between TSB levels and KSD and the clinical utility of current guidelines.

Optimisation of neonatal bilirubin testing and hyperbilirubinaemia management will need intervention at a number levels. These include clinical reporting of incidents arising from discordant results for inclusion in post-market surveillance by in vitro diagnostics manufacturers and commitment by manufacturers to ensure metrological traceability of their methods with sufficiently low uncertainty in the final measurements. Commutable quality assurance material with analytical performance specifications that are fit for the purpose of assessing and confirming assay performance, especially at the clinical decision points for neonatal jaundice, is also needed.