Analytical comparison of two commercially available feline serum amyloid A assays using residual clinical plasma samples

Abstract

Serum amyloid A (SAA) is a major acute-phase protein in cats; however, the agreement between commercially available feline SAA assays has not been fully evaluated. This study compared two analytical systems: the FUJI DRI-CHEM vf-SAA (sandwich fluorescence immunoassay) and VET-SAA Eiken (immunoturbidimetry). Eighty residual plasma samples from client-owned cats were analyzed. Methods were compared using Spearman’s correlation, Passing–Bablok regression, and Bland–Altman percentage difference analysis. These assays revealed a strong association (Spearman’s ρ=0.948). Passing–Bablok regression indicated proportional and constant bias (y=0.3325x + 1.155), and Bland–Altman analysis demonstrated a large negative mean percent difference (−86.2%). These findings suggest that the results are not interchangeable and method-specific reference intervals may be required.

Serum amyloid A (SAA) is a major positive acute-phase protein that increases rapidly after inflammation or tissue injury and decreases quickly during clinical recovery, making it a sensitive biomarker in veterinary medicine [1, 3, 14]. SAA levels in cats increase within hours after experimentally induced inflammation and surgery and are markedly elevated in a broad spectrum of naturally occurring diseases [10, 15]. Accordingly, feline SAA measurements have been used to support diagnosis, treatment monitoring, and prognostic assessment in primary care and referral settings [14, 16, 19].

Multiple analytical platforms are available for SAA quantification, including latex-enhanced immunoturbidimetric assays using clinical chemistry analyzers and cartridge-based sandwich fluorescence immunoassays [6, 11, 17]. However, as these systems differ in terms of antibody reagents, detection principles, calibration, and analytical measurement ranges, direct comparability cannot be assumed. Therefore, agreement between the evaluation methods is crucial to ensure consistent interpretation across different laboratories [6, 11, 17]. Although individual feline SAA assays have been analytically validated [6, 17], direct head-to-head agreement between FUJI DRI-CHEM vf-SAA and VET-SAA Eiken, two platforms commonly used in Japan, has not been evaluated using the same set of residual clinical feline plasma samples.

The aim of this study was to evaluate the correlation and agreement between two commercially available feline SAA assays currently used for feline serum/plasma testing: FUJI DRI-CHEM vf-SAA (measured using FUJI DRI-CHEM IMMUNO AU10V; FUJIFILM Medical Co., Ltd., Tokyo, Japan) and VET-SAA Eiken (measured using an automatic biochemical analyzer, Model 3100; Hitachi High-Tech Corporation, Tokyo, Japan; reagent from Eiken Chemical Co., Ltd., Tokyo, Japan).

This study was reviewed by the Ethics Review Committee of the Veterinary Medical Education Hospital, Faculty of Veterinary Medicine, Okayama University of Science (review number: 2020-0007). The committee determined that performance testing of clinical laboratory reagents using residual samples does not fall within the scope of ethical review; therefore, formal ethics approval was not required. A total of 80 residual lithium-heparinized plasma samples were used. All samples were obtained from client-owned cats that were presented to a veterinary secondary care hospital for various clinical reasons. According to the manufacturer’s instructions, both assays use serum/plasma matrices for veterinary use; therefore, residual lithium-heparinized plasma was used for this analytical comparison [4, 8, 13]. Samples were stored at −30°C for up to approximately 6 months and were thawed only once for analysis.

Precision assessment and method comparison analyses were guided by the American Society for Veterinary Clinical Pathology (ASVCP) quality assurance guidelines for veterinary clinical pathology [7]. The within-run precision of each analytical system was assessed using 10 repeated measurements of three lithium-heparin plasma specimens prepared as a dilution series. Specifically, a residual stored lithium-heparin plasma sample with a high SAA concentration was diluted with residual lithium-heparin plasma from another cat in which SAA was below the assay limit of detection to obtain three concentration levels (levels 1–3). The approximate SAA concentrations of the three dilution series specimens were 7, 97, and 172 mg/L as measured using FUJI DRI-CHEM vf-SAA, and 3, 50, and 85 mg/L as measured using VET-SAA Eiken. Comparative analyses included Spearman’s correlation, Passing–Bablok regression, and Bland–Altman analysis. Because no established reference method exists for feline SAA measurements, we used Passing–Bablok regression, which does not assume either method as a reference. Passing–Bablok regression was performed using Validation Support V64 (Japanese Society of Clinical Chemistry). For presentation, FUJI DRI-CHEM vf-SAA was assigned to the x-axis and VET-SAA Eiken to the y-axis.

Additionally, proportional bias in the Bland–Altman plot analysis was statistically assessed using linear regression of the differences against the means, performed in Microsoft Excel using the Data Analysis ToolPak (Microsoft Corp., Redmond, WA, USA), with a significant nonzero slope (P<0.05) indicating proportional bias.

Outlier candidates were identified based on regression residuals. For each comparison, the residuals were calculated as the difference between the observed and predicted values from the Passing–Bablok regression and standardized using the residual standard deviation. Samples with absolute standardized residuals >3 were considered discrepant. These samples were retained in the analysis but are discussed separately to illustrate the extent of disagreement between the systems.

The within-run imprecision (CVa) assessed using three dilution series specimens (FUJI, approximately 7, 97, and 172 mg/L; Eiken, approximately 3, 50, and 85 mg/L) was 2.2%, 6.1%, and 3.0% for FUJI DRI-CHEM vf-SAA and 6.6%, 0.8%, and 0.3% for VET-SAA Eiken, respectively. Because feline biological variation data are unavailable, the desirable imprecision goal was provisionally set at 12.4% (0.5 ×CVi) using published within-subject biological variation (CVi) for human SAA [2], which has also been applied in a previous feline validation study [17]. All the observed CVa values were below this provisional imprecision goal (12.4%).

The clinical interpretation of SAA differs depending on whether the results are within or above clinically relevant thresholds. Therefore, in addition to the overall method-comparison analyses, results were stratified using assay-specific thresholds. For FUJI DRI-CHEM vf-SAA, the manufacturer-provided upper reference limit (5.49 mg/L) was used. For VET-SAA Eiken, a published threshold (4.0 mg/L) reported for cats was adopted for stratification [20].

Scatter plots illustrating the relationships between the measurement methods are shown in Fig. 1. A strong positive association was observed (Spearman’s ρ=0.948).

Fig. 1.

Scatter plot showing the relationship between the two feline serum amyloid A assays. The solid red line indicates the Passing–Bablok regression line, and the dashed blue line indicates the identity line (y=x). FUJI DRI-CHEM vf-SAA (FUJIFILM Medical Co., Ltd., Tokyo, Japan) values are plotted on the x-axis, whereas VET-SAA Eiken (Eiken Chemical Co., Ltd., Tokyo, Japan) values are plotted on the y-axis.

The Passing–Bablok regression equation for FUJI DRI-CHEM vf-SAA (x) and VET-SAA Eiken (y) was y=0.3325x + 1.155 (slope 95% confidence interval [CI], 0.3047–0.3586; intercept 95% CI, 0.440–1.626). Because the 95% CIs for the slope and the intercept excluded 1 and 0, respectively, these results indicate the presence of proportional and constant systematic biases between the two methods. Consistently, the Bland–Altman plot showed a substantial negative mean percent difference. In the Bland–Altman analysis, linear regression of the percent differences against the means indicated a small but statistically significant proportional component (slope=−0.292% per mg/L, P=0.011; R²=0.079), consistent with a weak proportional bias. The mean percent difference [(Eiken−FUJI)/mean ×100] was −86.2%, and the 95% limits of agreement (mean ± 1.96 SD) were −149.4% to −23.0% (Fig. 2).

Fig. 2.

Bland–Altman plot depicting agreement between the two feline serum amyloid A assays. The solid red line indicates the mean percent difference, and the dashed blue lines indicate the 95% limits of agreement (mean ± 1.96 SD). The comparison was performed between FUJI DRI-CHEM vf-SAA (FUJIFILM Medical Co., Ltd., Tokyo, Japan) and VET-SAA Eiken (Eiken Chemical Co., Ltd., Tokyo, Japan).

In stratified analysis using the FUJI manufacturer-provided reference upper limit (5.49 mg/L), the dispersion of percent differences was greater in samples below this limit (n=7; mean percent difference, −52.2%; 95% limits of agreement, −202.0% to 97.6%) than in samples above this limit (n=73; mean percent difference, −89.5%; 95% limits of agreement, −135.0% to −44.1%). This is consistent with the wider scatter observed in the low-concentration region of the percent-difference Bland–Altman plot. In percent-difference Bland–Altman plots, variability can appear amplified when the mean of paired values is small. Therefore, dispersion in the low-concentration region should be interpreted in light of the threshold-based stratification and the limited number of samples below the FUJI upper reference limit.

Using the thresholds applied for stratification (FUJI upper reference limit, 5.49 mg/L; VET-SAA Eiken threshold, 4.0 mg/L [20]), 6/80 samples were below both thresholds, and 62/80 were above both thresholds. Discordant classification was observed in 12/80 samples: 11 samples were above the FUJI threshold but below the VET-SAA Eiken threshold, whereas 1 sample was below the FUJI threshold but above the VET-SAA Eiken threshold (FUJI 4.1 mg/L vs. Eiken 13.8 mg/L).

Overall, substantial method-dependent systematic differences were observed between FUJI DRI-CHEM vf-SAA and VET-SAA Eiken, as indicated using the Passing–Bablok slope (0.3325) and the large negative mean percentage difference in the Bland–Altman analysis. These findings indicate that the results generated by the two systems should not be considered interchangeable. Accordingly, the reference intervals and threshold values should be interpreted as assay-specific, and the cutoffs established for one assay should not be directly applied to another without appropriate validation.

Although the overall disagreement between the methods was dominated by systematic bias, regression residual analysis identified three specimens showing additional specimen-specific discrepancies beyond the general bias pattern (absolute standardized Passing–Bablok residuals >3). These samples were retained in the analysis but are described separately to illustrate the extent of disagreement. Specifically, SAA concentrations for these specimens were 90.0 versus 48.8 mg/L, 159.8 versus 36.8 mg/L, and 222.8 versus 50.8 mg/L for FUJI DRI-CHEM vf-SAA vs. VET-SAA Eiken, respectively. Clinical information was not linked to the individual residual samples in this study. However, no common gross sample characteristics typically associated with analytical interference (e.g., hemolysis, icterus, or lipemia) were observed among the three specimens. Given the magnitude of these discrepancies relative to the analytical imprecision, they are unlikely to be explained solely by random analytical errors. They may instead reflect specimen-specific matrix effects and/or assay-principle-dependent differences in reactivity.

The repeatability materials were prepared as a dilution series from two residual plasma samples, which limits specimen heterogeneity compared with the residual clinical sample set. Therefore, the apparent magnitude of disagreement observed in the dilution-series materials may differ from that observed across heterogeneous clinical specimens.

The exact cause of the observed disagreement could not be determined because the details of the proprietary assay were not fully disclosed. Nevertheless, the magnitude and direction of the bias suggest that the differences are largely systematic and may, at least in part, reflect differences in the calibration hierarchy and assay design. The VET-SAA Eiken calibrator set (V-SZ90) is described by the manufacturer as standardized against the WHO International Standard for SAA (NIBSC code 92/680) [5, 12] and is not described as cat-specific.

For FUJI DRI-CHEM vf-SAA, publicly available documentation indicates that fluorescence signals are converted to SAA concentrations using calibration information encoded in the cartridge (e.g., via a QR code) [9, 13]. However, explicit information on the traceability of external reference materials is not readily available. Accordingly, differences in the calibration approach could be one of several contributors to the observed systematic bias.

The two methods are based on different immunoassay principles: VET-SAA Eiken is a latex agglutination turbidimetric immunoassay for clinical chemistry analyzers [4], whereas FUJI DRI-CHEM vf-SAA is a cartridge-based sandwich fluorescence immunoassay that uses surface plasmon–enhanced fluorescence technology [9]. Moreover, antibody reactivity may differ across assays, and differences in antibody epitope recognition and sample pretreatment may contribute to systematic bias and occasional specimen-specific discrepancies. Since SAA is associated with high-density lipoprotein in the circulation [18], variability in lipoprotein composition and other matrix factors may influence antigen exposure, thereby affecting immunoreactivity in a method-dependent manner.

This study demonstrated that although two commercially available feline SAA assays showed a high correlation, they lacked agreement and were therefore unsuitable for direct interchangeability.

The substantial magnitude of disagreement observed in the present results, together with the Passing–Bablok regression findings, indicates that the differences are dominated by systematic bias (constant and proportional components) rather than random analytical error. In principle, such systematic differences could be reduced through calibration harmonization; constant bias may be mitigated by aligning the zero or reference point and proportional bias may be mitigated by sensitivity adjustment using reference materials across multiple concentrations. However, establishing a common commutable calibrator for veterinary SAA testing is challenging because calibration is manufacturer-specific. If switching measurement systems is unavoidable (e.g., during longitudinal monitoring), method-comparison data may provide a basis for approximate translation between systems; however, any such use should be interpreted cautiously because specimen-specific discrepancies may occur.

CONFLICT OF INTEREST

The authors declare no conflict of interest.