Analytical verification of the Atellica VTLi point of care high sensitivity troponin I assay

Introduction

Cardiac troponin (cTn) assays have evolved through several decades to the most recent generation coined ‘high sensitivity’ (hsTn) assays. In order to be assigned as ‘high sensitivity’, the required attributes are an imprecision of ≤10 % CV (coefficient of variation) at the 99th percentile upper reference limit (URL) and the ability to measure ≥50 % of concentrations greater than or equal to the assay’s limit of detection (LoD) [1]. Several laboratory troponin assays that fulfil these criteria are well established [2]. Point-of-care testing (POCT) devices for troponin measurement that meet these criteria have been more recently developed [2]. These include the Siemens POC Atellica^® VTLi hsTnI immunoassay [3].

Although accelerated diagnostic pathways in the Emergency Department (ED) have evolved considerably since the emergence of high-sensitivity troponins, the rate limiting factor for decision making is now the turnaround time (TAT) for the troponin assay within the clinical laboratory, considered to be unnecessarily protracted. As a consequence, this adds to ED overcrowding, which is associated with poor patient outcomes. The advent of high sensitivity POCT offers the potential for faster results, circumventing the longer laboratory TATs and facilitating more immediate clinical decision making in the ED [4, 5]. The Siemens POC Atellica^® VTLi hsTnI immunoassay (Forchheim, Germany) is able to produce a result in less than ten minutes and may assist to reduce ED overcrowding.

Initial validation studies have been undertaken on the Siemens POC Atellica^® VTLi hsTnI immunoassay [3, 6]. Using heparinized plasma from the American Association of Clinical Chemists (AACC) now known as the Association for Diagnostics and Laboratory Medicine (ADLM) universal sample bank, key performance characteristics have been defined for this device [3]. In apparently healthy subjects (overall 693, males 363, and females 330 and following exclusion of those with abnormal HbA_1c, NT-proBNP, eGFR, and those on statin medication) hsTnI was measured on multiple POC Atellica^® VTLi immunoassay analyzers. Both male and female subjects measured >50 % of subjects with >80 % measurable above the assay’s LoD, consistent with ‘high sensitivity’ performance [3].

Subsequently, comprehensive validation studies were published for the same device applying Clinical Laboratory Standards Institute (CLSI) protocols [6]. These included limit of blank (LoB), limit of detection (LoD), limit of quantitation (LoQ), applying CLSI-EP17-A2 [7]. Also precision, applying CLSI EP05-A3 [8], linearity applying CLSI EP06-A [9] and analytic specificity with several reagent lots, applying CLSI EP07-3, 3rd Ed [10] and CLSI EP37, 1st Ed [11]. Bland-Altman, Passing-Bablok (r), and hematocrit bias plots compared hsTnI measurement in lithium-heparin plasma (PL) and whole blood (WB) matrices applying CLSI EP09c [12]. Linearity spanned from LoQ to 1,250 ng/L. Specificity was >90 % for 40 potential interferences [6]; no hook effect was detected for cTnI concentrations ranging from LoQ to 1,000,000 ng/L [6]. Positive bias was noted for Li-hep plasma relative to whole blood of 6.3 and 3.8 % in the lower range of ≤50 ng/L and higher range >50 ng/L, respectively [6].

As a prelude to undertaking clinical outcomes studies [13], we undertook verification of the data presented in the above validation studies and made comparisons with laboratory methods to establish degrees of concordance relative to key cut-offs for each assay including LOQ and whole blood URLs.

Materials and methods

Results

Part 2: Comparison with laboratory methods

Laboratory fresh plasma vs. VTLi whole blood

VTLi (whole blood) vs. Beckman hsTnI (plasma)

There were 143 samples with concentrations measured by both assays in Christchurch Hospital Laboratory. The correlation, was 0.98 (95%CI: 0.97–0.99) over the range >20%CV to <maximum concentration (Figure 1A). The regression line slope was 0.53 (95%CI: 0.52–0.53) and intercept was 3.5 ng/L (95%CI: 3.4–3.5). The adjusted kappa statistics were good with the smallest (at the URL: 0.70) (Table 3). At the URL there were more Beckman>URL than VTLi.

Figure 1:

Correlation between assays. Red point is a double-discordant. Blue were outliers identified for the blue regression line (only). (A) Regression line slope 0.53 (95%CI: 0.52–0.53) and intercept 3.5 (95%CI: 3.4–3.5), r=0.98 (95%CI: 0.97–0.99). (B) Regression line slope 0.8 (95%CI: 0.74–0.84) and intercept −2.6 (95%CI: −3.8–−1.9), r=0.90 (95%CI: 0.86–0.93). (C) Regression line slope 0.35 (95%CI: 0.29–0.52) and intercept 4.1 (1.8–5.3), r=0.97 (95%CI: 0.95–0.99).

Table 3:

Summary of agreement between the VTLi and laboratory assays. Adjusted kappa statistics (95 %CI).

	VTLi whole blood vs. laboratory plasma			Frozen plasma concordance
	LoD (95 %CI)	LoQ/20 %CV (95 %CI)	URL/99th percentile (95 %CI)	LoD (95 %CI)	LoQ/20 %CV (95 %CI)	URL/99th percentile (95 %CI)
VTLi vs. Beckman hsTnI	0.90 (0.80–0.96)	0.80 (0.68–0.89)	0.70 (0.58–0.82)	0.83 (0.71–0.92)	0.82 (0.69–0.90)	0.90 (0.82–0.95)
VTLi vs. Roche hsTnT	0.75 (0.62–0.84)	0.73 (0.61–0.83)	0.44 (0.31–0.57)	0.82 (0.73–0.89)	0.75 (0.64–0.83)	−0.1 (−0.24 to 0.04)
VTLi vs. Siemens hsTnI	0.77 (0.56–0.91)	0.84 (0.64–0.95)	0.65 (0.46–0.84)	0.60 (0.55–0.78)	0.68 (0.55–0.78)	0.95 (0.89–0.98)
VTLi vs. Abbott hsTnI	NA	NA	NA	0.83 (0.73–0.91)	0.74 (0.62–0.83)	0.93 (0.87–0.97)

1. NA, not available; LoD, limit of detection; LoQ, limit of quantitation; CV, coefficient of variation; URL, upper reference limit.

There was one (0.7 %) double-discordant with a VTLi <20%CV and Beckman>Female URL. This was an 85 year old Female where the troponin result was rate related. The Beckman hsTnI of 15 ng/L was her baseline (it was also 15 on two previous measures three weeks earlier). In 2020 an Abbott hsTnI was 10 ng/L. Arguably, having a cTn done was not indicated. There was no MI.

In addition, the two measurements at ∼40 ng/L on the VTLi and 5 ng/L on the Beckman were from the same person. This person was pre-syncope, did not collapse, but had pressed a St John’s alarm. They had tachycardia on ECG and no MI was diagnosed.

VTLi (whole blood) vs. Roche hsTnT (plasma)

There were 158 samples with concentrations measured by both assays in Waikato Hospital Laboratory. The correlation was 0.90 (95%CI: 0.86 to 0.93) over the range >20%CV to <maximum concentration (Figure 1B). The regression line slope was 0.35 (95%CI: 0.29 to 0.52) and intercept 4.1 ng/L (95%CI: 1.8 to 5.3). There was one (0.6 %) double-discordant (Unique ID ROC269) with a VTLi <20%CV and Roche>URL. The adjusted kappa statistics were good with the exception of at the URL. Here the adjusted kappa was 0.44 (Table 3). At the URL there were more Roche>URL than VTLi. This is typical for comparisons between hsTnI assays and the Roche hsTnT assay.

VTLi (whole blood) vs. Siemens hsTnI (plasma)

There were 62 samples with concentrations measured by both assays in North Shore Hospital Laboratory. The smaller numbers than the other comparisons increases the confidence intervals. The correlation, was 0.97 (95%CI: 0.95 to 0.99) over the range >20%CV to <maximum concentration (Figure 1C). There were no double-discordant results. The adjusted kappa statistics were good (Table 3).

Additional Bland-Altman plots are provided in the Supplement.

Identical frozen plasma samples vs. VTLi whole blood

There were 98 samples that were measured on whole blood with the VTLi and on plasma on all four laboratory assays. There were up to 204 samples between the VTLi and each assay (Supplementary Table 3). Details are given in the supplement. Correlations were high between the VTLi and laboratory TnI assays and moderate (0.67) with the Roche hsTnT assay. The adjusted kappa statistics were very good or reasonable except for at the URL with Roche hsTnT (Table 3).

In the additional analysis between laboratory assays (Supplement section 2.3), there was very good correlation and concordance at the URL between the hsTnI assays. The correlation was lower between the hsTnT assay and the hsTnI assays. At the URL the concordance between the hsTnT assay and several of the hsTnI assays was poor (Supplementary Table 4).

Discussion

POCT devices for troponin measurement that meet the criteria for ‘high sensitivity’ (imprecision ≤10 % CV at the 99th percentile and able to measure ≥50 % of concentrations greater than or equal to the assay’s LOD) have been recently developed [2]. These include the Siemens POC Atellica^® VTLi hsTnI immunoassay [3] device which has the capacity to deliver results within ten minutes and to facilitate more rapid decision making in critical environments such as the ED.

For intra-assay imprecision, QC level 1 had CV <10 % around the plasma 99th percentile for females, consistent with a ‘high sensitivity’ troponin assay and similar to manufacturer claims. The CV for QC level 2 was slightly over 10 % and slightly higher than manufacturer claims. At a whole blood mean troponin concentration 3.27 ng/L, CV was 28.2 %, thus slightly higher than the expected 20 % CV at the LOQ for whole blood of 3.7 ng/L [6].

Pearson correlation coefficients supported a linear response.

Hemolysis resulted in a consistently positive interference for plasma samples, in a concentration dependent manner, varying by TnI concentration. Although not seen at lower degrees of hemolysis, the low sample (+3 ng/L; +60 %) and high sample (+37 ng/L; +24 %) began to show significant hemolysis interference that appeared to plateau at 250 mg/dL for reasons that are unclear, while the 99th percentile sample began to have significant hemolysis interference at 500 mg/dL (+3.6 ng/L; +24 %). This is a level significantly lower than manufacturer claims of robustness (±10 %) up to hemoglobin 1,000 mg/dL.

This indicates that there is some potential for interference with hemolysis for which one would not necessarily be aware of when analysing whole blood samples. There are newer technological advances such as trans-membrane electrochemistry that may offer the possibility of detecting hemolysis in whole blood samples [16], although unlikely to be applicable soon in the POCT field. It reinforces that care should be taken to avoid hemolysis in taking samples as this may otherwise confound interpretation and in particular, if results from serial samples are being compared where there are differing degrees of hemolysis.

The comparisons with the laboratory methods showed generally good concordance at the LOD for Beckman, Roche and Siemens laboratory assays respectively. Concordances were less good for whole blood URL. Encouragingly, there were only two doubly discordant data points (troponin <LOQ for Atellica and >URL for laboratory assay), one for the Beckman assay (in a patient judged unlikely to have had an MI) and one for the Roche assay under investigation. Note, the VTLi LOQ, 4 ng/L, was also the derived stand-alone single-sample low-risk threshold from a clinical outcome study [17, 18]. The single-sample low-risk threshold derived from our own clinical outcome study for use within an accelerated diagnostic pathway that included electrocardiogram and risk scoring was determined to be 7 ng/L [19].

The comparisons across assays using frozen samples, including the Abbott hsTnI assay, showed good concordance with the VTLi assay. Concordances between the VTLi assay and the laboratory assays were at least non-inferior to the comparisons of laboratory assays with each other. The correlation of the VTLi with the Roche hsTnT and the concordance at the URL was comparatively poor compared to with the other hsTnI assays. However, this was no worse than for any of the other hsTnI assays with the Roche hsTnT assay.

Conclusions

Our verification studies support the performance characteristics of the Atellica^® VTLi device as reported from more rigorous validation studies [3, 6]. The assay characteristics offer acceptable performance for use within the intended medical settings.