Comparison of Five Common Analyzers in the Measurement of Chemistry Analytes in an Authentic Cohort of Body Fluid Specimens

Abstract

Objectives

Interpretation of body fluid (BF) results is based on published studies and clinical guidelines. The aim of this study is to determine whether the assays from five common commercial vendors produce similar results in BFs for 12 analytes in a BF cohort.

Methods

BFs (n = 25) and serum (n = 5) were analyzed on five instruments (Roche cobas c501, Ortho 5600, Beckman AU5800 and DXI800, Siemens Vista 1500, and Abbott Architect c8000) to measure albumin, amylase, total bilirubin, cholesterol, creatinine, glucose, lactate dehydrogenase (LDH), lipase, total protein, triglycerides, urea nitrogen, and carcinoembryonic antigen. Deming regression and Bland-Altman analysis were used for method comparison to Roche.

Results

Results were significantly different from Roche for LDH and lipase on Ortho and lipase on Siemens but similar for both BFs and serum. BF differences were larger than serum differences when measuring creatinine, glucose, and urea nitrogen on Ortho and glucose on Siemens.

Conclusions

Five instruments used to perform BF testing produce results that are not significantly different except for lipase and LDH measurements. Bias of similar magnitude observed in both BF and serum should not affect interpretation. Further investigations into Ortho and Siemens measuring glucose and Ortho measuring creatinine and urea nitrogen are warranted.

Introduction

Clinical laboratories provide a valuable service when performing testing on alternate specimen types such as pleural, synovial, peritoneal, and drain fluids. These specimens are submitted to establish, narrow, and confirm a diagnosis in patients with pathologic effusions. The interpretation of results is often made in comparison to measurement in serum, compared with expected concentrations of the analyte in serum, or using an absolute concentration as a decision limit above or below, which is considered positive or negative for a particular condition. Laboratories offering this testing on a routine basis are challenged to provide interpretations in the information system to assist with result interpretation and clinical decision making. Numerous studies have been published dating back to the 1970s to establish the usefulness of measuring analytes in body fluids (BFs), but based on the paucity of method description and age of some of these studies, laboratories are challenged to confidently transfer the decision limits into use.¹ The aim of this study is to determine whether the assays from five common commercial vendors produce similar results in a cohort of BFs and serum for 11 common chemistry analytes (albumin, amylase, total bilirubin, cholesterol, creatinine, glucose, lactate dehydrogenase [LDH], lipase, total protein, triglycerides, urea nitrogen) and carcinoembryonic antigen (CEA).

In total, 117 remnant BFs after clinical testing was completed were collected over 30 days from Mayo Clinic Central Clinical Laboratory in Rochester, MN (IRB 18-007287). Fluid types consisted of amniotic (n = 2), drain (n = 18), pericardial (n = 3), peritoneal (n = 54), peritoneal dialysate (n = 3), pleural (n = 33), and synovial (n = 4). A cohort of serum samples (n = 31) was also analyzed as a normative analytical control between instruments and methods for an approved sample type. Specimens were aliquoted into four tubes and shipped frozen overnight to each study site except for specimens requiring analysis of LDH, which were gathered over 2 days and shipped at room temperature. Specimens tested for bilirubin were protected from light. Viscous synovial fluids were pretreated with hyaluronidase according to Block et al.²

The multisite study consisted of four tertiary care health systems and five diagnostic instruments. Measurement of BF chemistries was performed at Mayo Clinic Rochester by use of the Roche cobas c501, the University of North Carolina at Chapel Hill by use of the Ortho 5600 (Ortho Clinical Diagnostics), Atrium Health in Charlotte North Carolina by use of the Beckman AU5800, and the Minneapolis Veterans Affairs Hospital by use of the Siemens Vista 1500 (Siemens Medical Solutions) and Abbott Architect c8000. CEA was measured at Mayo Clinic Rochester and Atrium Health on the Beckman Coulter DXI and compared with the other listed instruments Table 1.

Specimens were frozen until shipped to each site on the same day. Upon receipt, LDH was measured, and all frozen samples were stored 2 weeks until testing could be completed at the same time at each site. Frozen samples were prepared for testing first by equilibration to room temperature on a sample rocker for 30 minutes, followed by centrifugation at 3,500 rpm for 5 minutes prior to testing. All measurements for an analyte were tested on the same day by four methods and compared with Roche results obtained prior to freezing.

MATERIALS AND METHODS

Cohort of BF Specimens

Table 1 summarizes the methodology, traceability, measuring range, and representative reference interval for serum for each of the five instruments. Specimens were analyzed according to the manufacturer’s standard protocols for each instrument. Each analyte had 25 BFs (except LDH had 22) and five serum samples measured (Supplemental Table 1 specifies BFs tested for each analyte; all supplemental materials can be found at American Journal of Clinical Pathology online). All instruments used a valid serum calibration and acceptable quality control according to standard laboratory practices. Results below the manufacturer’s analytical measuring range were excluded for analysis.

Diagnostic Instruments and Assays

Results were compiled in Microsoft Excel. Deming regression analysis was performed for each instrument using Roche as the arbitrary reference value using EP Evaluator (Data Innovations) to calculate slope, intercept, r, and bias. Bland-Altman plots were created in OriginPro (OriginLab). Statistical analysis was accomplished using one-way analysis of variance (ANOVA) pairwise comparison of instruments, which used a Holm correction to control the type I error rate for the multiple comparisons with P < .05 considered statistically significant. The intraclass correlation coefficients (ICCs) were also calculated to quantify how consistent the values were between Roche and each of the other four methods using R software (R Foundation for Statistical Computing). Mean difference (method – Roche) and percent difference (method – Roche) / Roche × 100% and standard deviations were calculated to compare disproportionate differences between specimen types for creatinine, glucose, and urea nitrogen.

Overall, there were no statistical differences between the five instruments measuring chemistry analytes (eg, P > .05) except for lipase and LDH Table 2. No distinct bias was observed between serum and BFs or different BF types when measuring albumin, total protein, amylase, LDH, lipase, bilirubin, and triglyceride. Further instrument- and method-specific details are provided for each analyte.

Preanalytical Specimen Handling

RESULTS

Analytical Methods and Calibration Traceability

LDH measurement revealed there was a statistical difference between the five instruments (P < .05). The pairwise comparison and ICC demonstrated Ortho was the outlier Table 2. Figure 1D demonstrates the largest bias (157%) between Roche and Ortho methods, followed by Roche and Beckman (–15%).

Lipase measurement also revealed there was a statistical difference between the five instruments (P < .05). The pairwise comparison and ICC demonstrated significant differences between Roche and both Ortho and Siemens. The pairwise comparison of Ortho to Siemens did not reveal a significant difference, suggesting they produce comparable results. Figure 1E demonstrates Abbott and Beckman methods produced results with the smallest bias (–3.5% and –3.0%, respectively) compared with Roche while the Ortho and Siemens methods had the greatest bias (250% and 227%, respectively).

Figure 2A demonstrates minimal bias with all methods with differences less than +1 mg/dL or 10%. The largest bias overall (0.3 mg/dL or 3%) was between Abbott and Roche.

Bias calculated between Roche and all platforms was within ±5% Table 2. Figure 2B demonstrates most negative bias occurred at low concentrations less than 100 mg/dL.

Overall, the greatest overall bias (–5%) was observed between Roche and Ortho Table 2. Figure 2C demonstrates BFs with a concentration more than 3 mg/dL had a greater mean absolute (SD) and percent difference (SD) (–0.5 [0.3] mg/dL or –6.6% [4.0%]) using Ortho compared with serum (<0.01 [0.1] mg/dL or –0.6% [2.6%]). Peritoneal dialysis fluid (n = 3) had the largest mean (SD) difference (–0.7 [0.3] mg/dL) and percent difference (–8.2% [1.0%]), a trend also observed in drain and peritoneal fluids. The absolute differences remained within ±0.1 mg/dL at concentrations less than 2 mg/dL for all methods. The other methods did not demonstrate such BF to serum bias compared with Roche Figure 2C and Table 3.

The largest overall bias was observed between Roche and Ortho methods (–8.7%) Table 2. Figure 2D demonstrates there were differences observed between BFs and serum for Ortho and Siemens. The peritoneal fluid tested by Ortho had the largest mean (SD) percent difference at –13.5% (2.1%) (n = 5) relative to serum (–1.6% [2.7%], n = 5). The BFs tested by Siemens had a mean –6.7% difference (n = 19 with concentration >50 mg/dL) compared with serum (4.6%, n = 5) with no obvious specimen type differences Table 3.

The greatest amount of overall bias was observed between Ortho and Roche (–4.6%) while all other methods had a bias less than +2% Table 2. Figure 2E demonstrates the absolute differences are less than ±10 mg/dL or ±20% for 90% of samples tested by Ortho, 93% by Abbott and Beckman, and 100% by Siemens.

Data Analysis

The regression analysis demonstrated the greatest amount of bias between Ortho and Roche at –6%, and all other methods had bias within 3% Table 2. The absolute differences were less than ±1.6 mg/dL at concentrations less than 24 mg/dL, with Ortho’s proportional bias (slope = 0.92) giving rise to larger negative percent differences at concentrations more than 30 mg/dL Figure 2F, demonstrating mean difference (SD) and percent difference (SD) of –6.4 (3.2) mg/dL and –9.9% (3.8%), with peritoneal and peritoneal dialysis fluid contributing most significantly to the observed bias Table 3. There was an observed trend for all methods in which serum appeared to have the most positive bias compared with Roche, with no BF-specific trends Table 3.

Albumin

Figure 2G demonstrates the poorest agreement between Siemens and Beckman instruments with slope = 0.72 and bias = –32% Table 2. The best agreement was with Beckman, as expected, when comparing two similar instruments. Overall agreement between all non-Siemens methods demonstrated bias within +5%. The absolute differences below 10 ng/mL were less than ±5 ng/mL, but there were slightly lower correlations (ICCs ranged from 0.92 to 0.95) and percent differences ranging from 50% to –100%.

Bland-Altman plot showing absolute difference for albumin (A) and total protein (B) and percent difference for amylase (C), lactate dehydrogenase (LDH) (D), and lipase (E). The following colors apply to Ortho (black), Abbott (red), Siemens (blue), and Beckman (green). The following symbols apply to peritoneal fluid (■), pleural fluid (●), drain fluid (▲), pericardial fluid (►), synovial fluid (⋆), and serum (▼).

The primary utility in comparing results of BF testing across multiple instruments is to first put the magnitude of bias in perspective of other studies, identify whether it is BF specific, and relate these findings to how BF results are interpreted. This information is valuable to laboratories to gauge the transferability of literature findings from one method to another in the provision of interpretive guidance. These comparisons are also useful when replacing instruments in the laboratory, particularly when the vendor changes. The following is a systematic evaluation of each analyte as it relates to the significance of the differences observed from this study.

For albumin, the magnitude of bias observed between methods in this study is comparable to the bias reported between methods in serum and plasma pools using several instruments with bromocresol green (BCG) and bromocresol purple (BCP) assays compared with a reference nephelometric method.³ They noted greater bias and a larger difference in albumin between serum and plasma using the BCG assays, leading to the conclusion that laboratories should standardize to the BCP methods. One laboratory using an Abbott instrument reported its conversion experience. It demonstrated the expected –0.35 g/dL difference in serum samples but not peritoneal fluid samples (–0.03 g/dL).⁴ The comparison presented herein did not appear to demonstrate a disproportionate bias in serum compared with BFs but rather a difference that depended on albumin concentration. Ultimately serum-ascites albumin gradient (SAAG) (the difference between serum albumin and ascites albumin used to differentiate causes of ascites) was reclassified in 4 of the 22 paired samples using the BCP reagent, suggesting further investigation is warranted.⁴ In conclusion for albumin, while the differences were small, there could be issues with interpreting SAAG values if there is a disproportionate bias in serum and BF matrix that we did not specifically test in this study. Laboratories should be mindful of how well the SAAG decision limit of more than 1.1 g/dL applies to their patient population to properly classify the presence or absence of portal hypertension.⁵

Discussion

Total Protein

For amylase, the magnitude of bias observed between methods in this study is comparable to the bias reported in commutable trueness verification pools⁹ and is commensurate with the observed differences in reference intervals Table 1. These differences have been attributed to the variation in substrates and reaction conditions that is known to contribute to nonharmonized enzyme results. The interpretation of BF amylase involves comparing the concentration in a BF to either the measurement of amylase in serum collected near the time the BF is collected or the comparison of BF amylase to the expected concentration for serum, such as the upper limit of normal for the assay, in which a several-fold elevation above serum suggests the presence of pancreatic fluid.^7,10 In conclusion for amylase, while there is a notable bias observed between methods, it is similar in magnitude between BF and serum, thereby negating the risk for misinterpreting results using a given assay. In the absence of amylase assay harmonization, users must consider the serum concentration or the manufacturer’s suggested reference interval for serum when interpreting results.

Amylase

For LDH, the methods quantify pyruvate except for Ortho, which quantifies LDH activity by the reverse reaction. Overall, the bias observed in BFs is comparable to serum in accordance with the method-specific serum reference intervals Table 1. LDH is measured in BFs to differentiate transudative from exudative or other inflammatory causes of effusion by comparison to serum or expected values in serum,^11-13 although in peritoneal fluid a decision limit above 220 U/L is proposed to differentiate secondary from spontaneous bacterial peritonitis.⁸ We can conclude that while there are statistical differences, the bias appears to be intrinsic to the method and is not specific to specimen type; therefore, the recommendation to compare BF to serum concentration or expected normal concentration should not present any obvious concerns for interpretation. In the case of peritoneal fluid, use of a decision limit should be done with caution using method-specific concentrations.

For lipase, all the assays use different substrate dye combinations to quantify the enzyme activity, and the differences observed correlate with the observed reference values proposed for the assays Table 1. BF lipase interpretation involves comparing the concentration to either the measurement of lipase in serum collected near the time the BF is collected or the comparison to the expected concentration for serum, such as the upper limit of normal for the assay, similar to amylase.^14,15 There is a notable bias observed between lipase methods, and it is similar in magnitude between BF and serum, thereby negating the risk for misinterpreting results using a given assay. Similar to amylase, in the absence of lipase assay harmonization, users must consider the serum concentration or the manufacturer’s suggested reference interval for serum when interpreting results.

For bilirubin, all methods used the diazo dye Table 1. The observed differences were relatively minimal (within 1 mg/dL or 10%) between methods, which is consistent with the magnitude of differences observed in other method comparison studies performed with serum.¹⁶ BF bilirubin is interpreted either using a decision limit of more than 5 mg/dL or in comparison to measurement in serum.^17,18 Differences of 1 mg/dL when used as a decision limit could impart some risk for misinterpretation, but this can be mitigated by interpreting compared with serum concentration.

LDH

For cholesterol, the comparison showed poorer agreement by ANOVA comparison with Ortho compared with other methods, likely due to the significant number of excluded data points below the low limit of quantitation. The methods are expected to be standardized given the National Cholesterol Education Program (NCEP) guidelines used for interpretation, but manufacturers use different standards for traceability Table 1. Prior analysis in serum pools compared several instruments to a reference method, which demonstrated there is negative bias for most methods and that Roche and Abbott (who happen to also use the same dye) have the least bias, which corresponds to our serum comparison findings. Meta-analysis and reviews of studies using cholesterol to detect malignant ascites have proposed a variety of decision limits ranging between more than 40 and 70 mg/dL,^1,19 which likely corresponds to the amount of variation observed between methods at this concentration range. It is worth noting that concentrations of cholesterol are higher in serum compared with BFs, and the Ortho and Siemens manufacturer-defined measuring ranges do not extend below 50 mg/dL, which may limit the sensitivity of these assays for detecting malignant ascites. Further studies defining method-specific decision limits are warranted.

For creatinine, all but Beckman in this study use enzymatic methods and are traceable (through primary or secondary standards) to isotope dilution mass spectrometry, whose desirable total allowable error goals and performance are within 7%.²⁰ The Ortho method demonstrates a larger negative bias in drain, peritoneal fluid, and peritoneal dialysate compared with serum when creatinine concentrations are above 3 mg/dL. BF creatinine is interpreted in comparison to serum,²¹ which could present an issue when using the Ortho method as the bias observed only in BFs may lead to underestimating the BF to serum ratio, particularly in patients with renal impairment whose background creatinine concentration may be elevated. Additional studies are warranted to verify the extent of this risk, as it is dependent on the magnitude of the bias.

For glucose, all methods used hexokinase, except for Ortho using glucose oxidase, which manufacturers have demonstrated achieves desirable total allowable error goals of less than 7%.²² BF glucose is typically measured to identify infection with low concentrations, using decision limits ranging from ~15 to 50 mg/dL, as well as comparison to serum.^8,23-26 Since there appears to be a disproportionate bias between BFs (particularly peritoneal fluid) and serum for the Ortho method and Siemens, it is possible that both decision limits and a BF to serum ratio could be underestimated, and those users should exercise caution.

CONCLUSION

For triglycerides, most assays use a similar design, except for Ortho Table 1, although studies have demonstrated they produce rather similar results within ±15% of the reference method.²⁷ Triglycerides are most often interpreted using single decision limits of less than 50 mg/dL, more than 110 mg/dL for pleural fluid, or more than 187 mg/dL for peritoneal fluid.^28,29 Overall, the differences were quite small in the BFs tested and not anticipated to affect interpretation.

For urea nitrogen, all assays used a common design, except for Ortho Table 1. Method comparison studies in a serum pool (~12 mg/dL) demonstrated bias less than 1 mg/dL,²⁷ which is comparable to the observed results in both serum and BFs in that concentration range. However, at elevated concentrations, there are more marked differences particularly affecting BFs tested on Ortho. Urea nitrogen concentrations in BFs are often compared with serum concentrations, with elevations indicating presence of urine.²¹ This could present an issue when using the Ortho method, as the bias is observed only in BFs and may lead to underestimating the BF to serum ratio, particularly in patients with renal impairment whose background blood urea nitrogen concentration may be elevated. Additional studies are warranted to verify the extent of this risk, as it is dependent on the magnitude of the bias.

Acknowledgments

For urea nitrogen, all assays used a common design, except for Ortho Table 1. Method comparison studies in a serum pool (~12 mg/dL) demonstrated bias less than 1 mg/dL,²⁷ which is comparable to the observed results in both serum and BFs in that concentration range. However, at elevated concentrations, there are more marked differences particularly affecting BFs tested on Ortho. Urea nitrogen concentrations in BFs are often compared with serum concentrations, with elevations indicating presence of urine.²¹ This could present an issue when using the Ortho method, as the bias is observed only in BFs and may lead to underestimating the BF to serum ratio, particularly in patients with renal impairment whose background blood urea nitrogen concentration may be elevated. Additional studies are warranted to verify the extent of this risk, as it is dependent on the magnitude of the bias.