Could the Intra-Laboratory Inter-Identical-Instrument Bias Compromise the Interpretation of the Absolute High-Sensitive Troponin Delta Around the 99th Percentile Upper Reference Limit?

Aim of the study: To measure high-sensitive troponin (Access hsTnI) interidentical-instrument bias (DxI800 Beckman Coulter) in order to understand if it can compromise the interpretation of absolute delta value for rapid algorithms 0/1-0/3 hours. Materials and methods: One hundred fifty-nine lithium/heparin plasma samples were processed sequentially on three DxI800 (DxI1, DxI2, DxI3). The results given by the three instruments were analyzed as followed: DxI1 vs. DxI2, DxI1 vs. DxI3, DxI2 vs. DxI3. Statistical analysis was done using the Passing-Bablok regression, Bland-Altman test, and Cohen’s Kappa statistic. Results: PB regression did not show any significant deviation from linearity and no proportional nor constant differences were observed among instruments. Moreover, the mean absolute bias, even though among the three instruments the lowest 95%CI lower limit was -3.75 and the highest 95%CI upper limit was 3.92 ng/L, was within the acceptance limits (all results<reference change value). The concordance between each couple of instruments was mostly strong. Conclusion: Our data suggest that inter-identical-instrument bias needs to be considered before evaluating the clinical diagnostic accuracy of one absolute delta with respect to another, in order to define the minimum absolute delta that the laboratory can guarantee to the clinicians.


Introduction
Measurement of troponins is mandatory for all patients with suspected NSTE-MI [1]. High sensitive troponin (hs-cTn) assays (which are able to measure the troponin in at least 50% of a healthy population, with an analytical coefficient of variation (%CV) lower than 10% at the 99 th percentile upper reference limit (URL)), have been recently recommended in preference to less sensitive ones [1]. The use of hs-cTn assays have basically reduced the troponin-blind interval due to their ability to detect very small concentrations [2]. As result, rule-in and rule-out algorithms, based on the absolute hs-cTn value at presentation (above or below the 99 th percentile URL) and the changing pattern observed 1-3 hours later, are currently recommended by the European Society of Cardiology (ESC) with a Class I recommendation and are considered safe [1][2][3][4][5]. To define whether or not a changing pattern is present, one must consider the blood flow and the time between the onset of symptoms and the obtaining of samples [6]. In addition, previous studies gave warning of possible quantitative analytical shifts at these low concentration ranges that may exceed some of the deltas reported in clinical studies [7][8][9] and it has R statistical package version 3.4.0, library "BlandAltmanLeh") was performed using Passing-Bablok regression to test the linear relationship between the measurements, and Bland-Altman technique to estimate the consistency of the methods [13, 15,16]. The results given by the three instruments were analyzed as followed: DxI1 vs. DxI2, DxI1 vs. DxI3, DxI2 vs. DxI3. The statistical analysis was inclusive of all the results, but this study focuses on the subgroup characterized by values below 50 ng/L. The acceptance of inter-identical-instrument bias calculated with the Bland-Altman test (bias ± 95%CI limits of agreement) was based on the reference change values (RCV= 2.77 (CVa 2 + CVi 2 ) 1/2 where CVi= intra individual variation, taken as 10% [17] and CVa=total imprecision). Finally, method comparison was evaluated by means of the weighted Cohen's Kappa statistic (with linear weights) [18]

Results
Our range of Access TnI+3, measured on 159 pre-existing heparinplasma samples, was <0.01 to 22.4 µg/L and it corresponds to hsTnI ranges 3 to 21839, 3 to 22382 and 5 to 21106 ng/L using DxI1, DxI2 and DxI3, respectively. Four samples were below the LoD and two samples were outliers, as a consequence they were excluded from the statistical analysis. Total imprecision within the laboratory was consistent with claims made by the manufacturer and in accordance with previous studies [19,20] on all instruments.
Due to the fact that we performed the assays using identical instruments, PB regression did not show any significant deviation from linearity (all p values from the linearity test: >0.10) and no proportional nor constant differences were observed among instruments, as expected, considering both the whole range of results Table 1 and those of subgroups below the 50 ng/L Table  1 and Figure 1. Moreover, the mean absolute bias, even though among the three instruments the lowest 95%CI lower limit was -3.75 and the highest 95%CI upper limit was 3.92 ng/L, was virtually zero and within the acceptance limits (all results <RCV; RCV DxI-1: 6.2, RCV DxI-2: 6.4, RCV DxI-3: 6.6 ng/L).
The concordance between each couple of instruments was mostly strong [18]; however, the lower 95%CI limits of agreement sometimes revolved around 0.8, namely moderate agreement (11.6 ng/mL women cutoff:

Discussion
Our data confirmed that analytical performance of each instrument recently been suggested to consider also the biological variation if hs-cTn assay is available, especially around 99 th percentile. As a consequence, most studies conclude that a 50-60% change, if the initial value is less than 99 th percentile, and a 20% change, if the initial value is greater than 99 th percentile, are to be considered significant deltas [6]. However, absolute delta (assay dependent), appears superior to relative per cent changes with hs-cTn assay [10] and in some studies this is particularly so when the initial value is increased [11]. The critical point is that due to the fact that each patient should have more than one sample tested in order to evaluate the hs-cTn changes [2], in a laboratory equipped with more than one instrument, these samples may randomly be assayed by different instruments. This may lead to possible misclassification, as previously demonstrated by a contemporary assay [12]. The aim of this study is to measure inter-identicalinstrument bias in our laboratory in order to understand if it can compromise the interpretation of absolute delta value for rapid algorithm 0-3 hours.

Study design
A training program was scheduled before the beginning of the experiment to ensure that the operators would understand all procedures and could execute them properly [13].
High sensitive troponin I assays (hsTnI) (Access hsTnI, Beckman Coulter, Brea, CA, USA) were performed on sera from anonymized pre-existing heparin-plasma samples with a TnI (Access AccuTnI+3, Beckman Coulter, Brea, CA, USA) result. Tests were processed sequentially on the two DxI800 (Beckman Coulter), DxI-1 and DxI-2, found in the stat laboratory, and on the other, DxXI-3, found in the central routine laboratory.

Assay methods
We used the hsTnI (Access hsTnI, Beckman Coulter, Brea, CA, USA) assay, which is characterized by a 2.3 ng/L LoD, 11.6 ng/L 99 th percentile for women (CV%: 4.2), 19.8 ng/L 99 th percentile for men (CV%: 3.6), 17.5 ng/L 99 th percentile for overall (CV%: 3.7) and the 10%CV at 5.6 ng/L. Evaluation of total imprecision. . These analyses were performed over a period of three weeks. Grubbs' test was used to evaluate the presence of outliers [14]. In addition, daily IQC were evaluated before starting all analytical sessions (IQC-LC, IQC1, IQC2). Same lot of reagents were used over the experimental period (of about two months) and were calibrated using two different calibrator lots.

Statistical analyses
Data analysis (MedCalc® Statistical Software version 18.2.1 and Passing-Bablok regression and Bland-Altman plot comparing DxI-1 vs. DxI-2, DxI-2 vs. DxI-3, DxI-1 vs. DxI-3 hsTnI values in the subgroup <50 ng/L. Figure 1 around the cutoff values lies well within the accepted quality requirements [1][2][3]. Nevertheless, considering the combined effect of the presence of the three instruments, few patient results (k test results described above) were in disagreement between instruments. This was not unexpected and it is, at least in part, also the same if the evaluation is performed using only one instrument and considering its measurement uncertainty (MU) [21]. For example, if the 99th percentile 17.5 ng/L of all the population is supposed to be the real value of a patient then, according to our MU (imprecision and bias) of about 6.3%, the measured value will be, with a 95%CI, within the range (16.4 -18.6 ng/L). Moreover, the measured inter-identical-instrument absolute bias Table 1 and Figure 1 of each couple of instruments, basically from -4 to 4 ng/L in the worst situation, allows us to conclude that the 3 and 5 ng/L absolute delta for short 0/1-0/3 hour algorithms, suggested by the manufacturer, cannot be used in our laboratory because the inter-instrument absolute delta is large enough to significantly interfere with the interpretation of these clinical absolute deltas. Probably this is especially the case when 0/1 hour is evaluated. Absolute deltas of 11 and 22 ng/L could be an analytically applicable choice, even if clinicians should think about these cutoffs as range (11 or 22 ± inter-identicalinstrument absolute bias) instead of dichotomous cutoff. For example, in our laboratory, if a patient has a real absolute delta of 11 ng/L, the measured delta could be 11 ± 4 ng/L, depending on the sequence of instruments used (DxI1-DxI1, DxI2-DxI2, DxI3-DxI3, DxI1-DxI2, DxI2-DxI1, DxXI1-DxI3, DxI3-DxI1, DxI2-DxI3, DxI3-DxI1). Interestingly, even if we used only one instrument, but considering the RCV (for all instrument <7 ng/L) to correctly interpret the absolute delta, the choice to not use 3 and 5 ng/L absolute value would be the same. However, due to the fact that 11 ± 4 ng/L absolute delta is really close to the observed RCV value, our analytical choice was to suggest to the clinicians the 22 ± 4 ng/L absolute delta. In addition, to underline the concept of a delta cutoff as a range, we also added to the final patient report, in the reference interval camp, the indication for absolute delta as 22 ± 4 ng/L.

Conclusion
Our data suggest that inter-identical-instrument bias needs to be considered before evaluating the clinical diagnostic accuracy of one absolute delta with respect to another, in order to define the minimum absolute delta that the laboratory can guarantee to the clinicians (Figure 2). Nevertheless, even if the laboratory has only one instrument, MU and RCV should be evaluated in interpreting values around cutoff and absolute delta respectively. Finally, it is well established that errors in cardiac troponin testing (i.e. heterophile antibodies, non-reproducible false elevations, etc) [22] can occur, so results should always be interpreted alongside the clinical context.