- Email: [email protected]
Serial cardiac troponin differences measured on four contemporary analyzers: Relative differences, actual differences and reference change values compared
Clinica Chimica Acta 413 (2012) 1786–1791
Contents lists available at SciVerse ScienceDirect
Clinica Chimica Acta journal homepage: www.elsevier.com/locate/clinchim
Serial cardiac troponin differences measured on four contemporary analyzers: Relative differences, actual differences and reference change values compared Carel J. Pretorius ⁎, Urs Wilgen, Jacobus P.J. Ungerer Department of Chemical Pathology, Pathology Queensland, Brisbane, Queensland, Australia
a r t i c l e
i n f o
Article history: Received 18 April 2012 Received in revised form 6 June 2012 Accepted 2 July 2012 Available online 10 July 2012 Keywords: Cardiac troponin Myocardial infarction Acute coronary syndrome Reference change value Serial difference
a b s t r a c t Introduction: The diagnosis of myocardial infarction is in part predicated on a rise and/or fall in cardiac troponin (cTn). z-Values incorporate qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi analytical and biological variation to standardize serial differences: z ¼ Δ=
2SD2Analytical þ 2SD2Biological . We investigated the theoretical distributions of actual differences (Δ),
relative differences (%Δ) and z-values and compared the agreement in classification of differences measured on four contemporary platforms. Methods: cTn was measured in 575 sample pairs on the Abbott Architect cTnI, Beckman Coulter Access2 cTnI, Roche Cobas cTnT and Siemens ADVIA Centaur cTnI methods. Results: Good agreement was obtained amongst all methods with a universal z-value cut-off (κ>0.79) and method specific fixed Δ cut-offs (κ>0.82) while suboptimal agreement was obtained between cTnI and cTnT with fixed %Δ cut-offs (κ b 0.50). Conclusion: Fixed Δ cut-offs will perform well at low cTn concentrations while fixed %Δ cut-off values are predicted to perform poorly. z-Values are independent of the cTn concentration, present differences as a continuum of probability and a universal decision level can be used for all cTnI and cTnT methods. Crown Copyright © 2012 Published by Elsevier B.V. All rights reserved.
1. Introduction Cardiac troponin (cTn) is considered the standard biomarker for myocardial infarction (MI) with the universal definition requiring a rise and/or fall with at least one value exceeding the 99th percentile reference limit [1]. With the recognition of non-ischemic etiologies of cTn elevations it became important to accurately define a significant change of cTn in order to improve the diagnosis and clinical management of patients suspected of suffering from acute coronary syndromes (ACS). Both relative (%Δ) and actual numeric (Δ) differences in cTn have been reported to increase diagnostic specificity for MI while maintaining diagnostic sensitivity [2–6]. A range of %Δ decision levels between 20% and 243% has been reported for cardiac troponin-I (cTnI) and cardiac troponin-T (cTnT) [2–4,7]. Α Δ of 6.9
Abbreviations: MI, myocardial infarction; ACS, Acute coronary syndrome; cTn, cardiac troponin; cTnI, cardiac troponin I; cTnT, cardiac troponin T; Δ, Actual numeric difference; %Δ, Relative difference; RCV, Reference change value; LOQ20%, Limit of quantitation defined by a 20% coefficient of variation; SDbiological, Standard deviation of individual biological variability. ⁎ Corresponding author at: Department of Chemical Pathology, Block 7 Floor 3, Herston Hospitals Campus, Herston, 4029, Queensland, Australia. Tel.: +61 7 3646 0083; fax: +61 7 3646 1392. E-mail address: [email protected] (C.J. Pretorius).
to 9.2 ng/L for cTnT and 20 ng/L for a cTnI assay was reported to be clinically superior to %Δ cut-offs [5,6]. It is thus clear that there is no consensus on how to define a significant cTn difference with contradictory cut-off levels proposed by various authors. In view of the variable precision of cTn assays over the measuring range we also noted that little attention was paid to the relationship between the cTn concentration and the various proposed cut-off levels. Comparing the observed %Δ or Δ to a cut-off derived from a reference change value (%RCV or RCV) was proposed as a mechanism to determine if the magnitude of cTn change is significant [8]. Calculating a RCV requires assay specific information on analytical and biological variation. Information on the analytical variation over the measuring range is readily available from laboratories and although the biological variation of cTnI and cTnT has been reported, it is not clear how to apply this information in clinical practice. Some authors have suggested that biological variation should only be taken into account when interpreting results around the 99th percentile cut offs, as these small physiological fluctuations are overwhelmed at higher levels by the pathological release from necrotic tissue [8,9]. Furthermore the %RCV is often reported with an asymmetrical reference range depending on whether an increase or decrease occurred [7,9–11]. Most papers however appear to calculate the %Δ relative to the lower result, irrespective of whether an increase or decrease occurred and this may potentially lead to confusion when compared to a %RCV derived cut-off calculated differently. We are
0009-8981/$ – see front matter. Crown Copyright © 2012 Published by Elsevier B.V. All rights reserved. doi:10.1016/j.cca.2012.07.001
C.J. Pretorius et al. / Clinica Chimica Acta 413 (2012) 1786–1791
not aware of a systematic comparison between empirically determined fixed %Δ or Δ cut-off values and cut-off values determined with a RCV approach. Comparative information on the differences between serial results measured with commercially available cTnI and cTnT assays is incomplete and we are aware of one paper that compared assays directly [5]. Given the lack of standardization amongst cTnI assays, the large reported variances amongst assays [12] and the apparent biological differences that exist between cTnI and cTnT, it is clinically important to understand how assays relate to each other in this respect. In view of the aspects discussed above we defined the terminology and examined some of the theoretical aspects of how to present differences between serial samples. We were particularly interested in the relationship between the degree of cTn elevation and the cut-off values for %Δ and Δ, which we calculated from information on analytical and biological variation at a defined probability. We investigated the distribution of differences obtained on serial patient samples in relation to arbitrarily defined fixed cut-offs and the method specific, but concentration dependent, variable cut-off values. The concordance between four analytical platforms was then determined for differences measured concurrently on patient samples. 2. Materials and methods 2.1. Glossary of terms and definitions 2.1.1. The actual difference (Δ) is the numerical difference between two results in the original units of measurement. Δ = Result2 − Result1, where the sign (+ or –) indicates if an increase or decrease occurred. 2.1.2. The absolute difference is defined as |Δ| = |Result2 − Result1|, again in the original units, but with no indication of whether an increase or decrease occurred. 2.1.3. The relative difference (%Δ) is a ratio where the Δ is expressed relative to a denominator and reported as a percentage. 2.1.3.1. Relative difference as change (%Δchange). %Δchange =(Result2 − Result1)/Result1 × 100, the sign (+ or –) indicates an increase or decrease. The %Δchange results have a numerical range between − 100% and +∞%. 2.1.3.2. Relative difference with respect to the lower result of the pair (%Δlow). %Δlow = (Resulthigh − Resultlow)/Resultlow × 100. The %Δlow results have a numerical range of 0% to +∞%. 2.1.3.3. Relative difference with respect to the mean of the two values (%Δmean). %Δmean = (Result2 − Result1)/0.5(Result2 + Result1) × 100, the sign (+ or –) indicates that an increase or decrease occurred. The %Δmean results have a numerical range of ±200%. 2.1.3.4. %Δmean and %Δlow can easily be inter-converted. %Δmean = %Δlow/ (100 + %Δlow/2). 2.1.4. The reference change value is a cut-off value with a defined probability. The underlying principle is that the variance of differences is estimated from the variances of the observed values. Analytical and biological variances are the main components contributing to the variance of the observed results and the cut-off values are therefore both method specific and concentration dependant.
1787
2.1.4.1. Actual reference change value (RCV). RCV ¼ z ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi q 2SD2 Analytical þ 2SD2Biological . For a 2-sided 95% probability that the difference is significant, a z-value of 1.96 is used (p b 0.05). 2.1.4.2. Relative reference change value (%RCV). %RCV ¼ z rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi 2 2 2 %CV Analytical þ %CVBiological . Analogously the %Δ is compared to the %RCV and if it exceeds this cut-off, the difference is regarded as significant. 2.1.5. The z-value is the probability associated with an observed difference and is obtained by rearranging the formula for RCV (or %RCV): z ¼ Δ= qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2SD2Analytical þ 2SD2Biological . Simplistically, this standardizes the observed difference to a multiple of the estimated standard deviation of the differences. 2.2. Theoretical considerations 2.2.1. Calculating relative differences We prefer to use %Δmean to express relative differences for a number of reasons. Firstly it is explicitly assumed in the simplified calculation of %RCV that there is negligible covariance between the two results i.e. r ≈ 0 [13]. By using %Δchange and even more so with %Δlow the statistical artifact of auto-correlation is introduced, which violates the above stated assumption [14,15]. In the context of a null hypothesis the two results are random variables in the distribution around the true (unknown) value of which the mean value is the best estimate that we have [14]. Furthermore it avoids asymmetrical results depending on the direction of change and results are in a numeric range of − 200% to + 200% that facilitates graphical representation. The probability of a significant difference will tend to be overestimated if %Δlow is compared to a %RCV cut-off and vice versa (see worked example 1 in the online appendix). Comparison of %Δmean to empirically determined %Δlow cut-offs presents little problem as the values can be easily converted (see Section 2.1.3.4). 2.2.2. Biological variation of cardiac troponin The biological variation of cTn is typically reported as a %CV with the implication that this physiological variation increases proportionally as the set point in an individual increases. The graphical presentation of the within-individual biological variation however does not convincingly demonstrate a proportional relationship to the individual median values [9,10]. In our opinion it does not make biological sense to assume that a proportional physiological variation can be transposed onto a pathological release of cTn due to ischemia or necrosis. Such a pathological release is independent of, and overwhelms the small physiological variation observed in cardio-healthy individuals. Even if short term variation in cTn occurs under pathological– ischaemic conditions, it should not be considered in the calculation of a cut-off value as the objective is to differentiate differences due to analytical and physiological variation from pathological changes. Based on these observations we propose that in the context of ACS a constant actual biological variation component (SDbiological) should be taken into consideration when calculating cut-off values. This will ensure that both physiological and analytical variations are considered at cTn values around the 99th percentile cut-offs while the influence of the physiological variation becomes negligible at grossly elevated levels. The median value reported for the short term biological variation of a novel cTnI assay was 2.2 ng/L [11]. Specific data for the methods used in this study were not available and due to the imprecision of the cTnI assays it cannot be accurately obtained for these assays. For the calculations in this paper, based on the approximate linear relationships between the assays, we
1788
C.J. Pretorius et al. / Clinica Chimica Acta 413 (2012) 1786–1791
assumed a median SDbiological 2.5 ng/L for the Roche, 5 ng/L for the Abbott and Beckman Coulter, and 10 ng/L for the Siemens assays. 2.2.3. Calculating z-values z-Values standardize differences as probabilities that are independent of analytical method or cTn concentrations. This allows the use of a universal cut-off (i.e. 1.96 or 2.56) as opposed to method specific and concentration dependent cut-off values. For similar reasons to those stated in Section 2.2.1 above, the analytical and biological variances at the mean level of the two results are used to estimate the variance of differences. According to the universal definition a rise and/or fall of cTn needs to be considered and a two-sided p value to test for significance is appropriate. In acute coronary syndrome changes occur over a period of hours and it is therefore appropriate to use analytical and intra-individual biological variation determined over an equivalent period. The z-value is partly dependent on the magnitude of the SDbiological and the use of a numerically larger value, such as a 99th percentile in place of a median value, will lead to an increase in clinical specificity at the expense of sensitivity [13,16]. 2.3. Study population and analytical methods The study population, quality control procedures, precision parameters, identification of outliers and method comparisons were described in detail elsewhere [12,17]. Serum cTn was measured in duplicate on four contemporary analytical platforms: 1. Abbott Architect i2000SR analytical system with STAT Troponin-I reagent (Abbott Diagnostics, Lane Cove, NSW, Australia). 2. Beckman Coulter Access2 analyzer with Enhanced AccuTnI reagent (Reagent part number A78803, Beckman Coulter Diagnostics, Brae, CA, USA) analyzer. 3. Roche Cobas e601 with TroponinT hs reagent (Roche Diagnostics, Sydney, Australia). 4. Siemens ADVIA Centaur XP with TnI-Ultra reagent (Siemens Healthcare Diagnostics Inc. Deerfield, IL). For the sake of brevity the methods are referred to as Abbott, Beckman Coulter, Roche and Siemens hereafter. The performance characteristics of the respective assays are summarized in Table 1. After exclusion of outliers the actual (Δ) and relative differences (%Δmean) were calculated from the first of the duplicate results in 575 non-overlapping sample-pairs from 406 patients. Results below the limit of quantitation, as defined by a 20% coefficient of variation (LOQ20%), were deemed to be equivalent to the LOQ20% when calculating differences. z-Values were calculated as described in Sections 2.1.5 and 2.2.3 with a constant biological variation component as discussed in Section 2.2.2. The analytical variation component was calculated from the method specific precision profile data [17]. Statistical testing was performed at the 5% significance level (p b 0.05) unless stated otherwise. The regression slopes of the Δ and %Δmean were subdivided into intervals of b 6, 6 to 12, 12 to 24 and > 24 h and tested for significant differences for each analyzer [18]. Agreement in the binary classification between methods was evaluated with Cohen's κ-statistic at a fixed %Δmean cut-off (20%), fixed method specific Δ cut-off values and according to a z-value of Table 1 Summary of the performance characteristics of the Abbott, Beckman Coulter, Siemens and Roche assays.
Abbott Beckman Coulter Siemens Roche
Measurand
99th percentile cut-offa
10% CVb
20% CVb
cTnI cTnI cTnI cTnT
28 40 40 14
48 64 57 6
24 32 29 3
All values are reported in ng/L. a The 99th percentiles are presented as reported by the respective manufacturers. b The values at which a CV of 10% and 20% respectively were attained, were estimated from the total CV determined according to CLSI EP5-A2 and reported elsewhere [17].
1.96. The Δ cut-off values were based on the published empirical data for the Roche cTnT [5] and estimated for the cTnI assays according to the approximate relationships between the assays: Abbott and Beckman-Coulter 15 ng/L, Siemens 30 ng/L and Roche 7 ng/L. 3. Results The median time interval (interquartile range) between repeat samples was 21.8 h (7.0 to 36.3). The median (interquartile range) of the actual differences in ng/L was 3 (0 to 170) for the Abbott, 4 (0 to 269) for the Beckman Coulter, 8 (0 to 464) for the Siemens and 5 (0 to 38) for the Roche analysers respectively. A subgroup analysis of differences did not demonstrate a time dependant aspect and we therefore pooled the data (p b 0.05). 3.1. Influence cTn concentration on Δ and %Δmean cut-off values In Fig. 1 the theoretical cut-off values are obtained with z = 1.96 (p = 0.05), and a constant method specific SDbiological and the precision data of the respective analyzers are presented. The Δ limits were constant at low mean cardiac troponin levels and increased progressively above a mean value of approximately 200 ng/L. In contrast a large and variable %Δ defined the cut-off at low values with relatively constant limits at higher values above a mean of approximately 300 ng/L. Plotting the %Δ against Δ presents the same information independent of the mean value and confirmed a threshold effect for %Δ and Δ. Serial results with a z-value >1.96 generally corresponded to both a Δ > 15 ng/L and a %Δ > 10% on the Abbott and Beckman platforms. The corresponding thresholds for the Siemens were 20 ng/L and 7%, and for Roche 7 ng/L and 5%. z-Values Table 2 Agreement in the binary classification of serial cardiac troponin differences measured with four contemporary assays. ABd % Δmeana Concordant + 210 Relative difference Concordant − 318 Test 1 +, 23 (%) Test 2 − Test 2 +, Test 24 1− κ 0.83 Δb Concordant + 222 Actual difference Concordant − 324 Test 1 +, 15 (ng/L) Test 2 − Test 2 +, 14 Test 1 − κ 0.90 Z > ±1.96 (SDbiological)c z-Value Concordant + 207 Concordant − 335 Test 1 +, 15 Test 2 − Test 2 +, Test 18 1− κ 0.88 a
ASe
BSf
ARg
BRh
SRi
212 302 21
219 308 15
170 255 63
168 252 66
179 245 73
40
33
87
89
78
0.78
0.83
0.47
0.45
0.47
224 324 13
226 327 10
222 314 15
216 309 20
223 314 15
14
12
24
30
23
0.90
0.92
0.86
0.82
0.86
208 330 14
217 336 8
195 324 27
197 323 28
204 324 27
23
14
29
27
20
0.87
0.92
0.79
0.80
0.83
% Δmean; ± 20% cut-off used for all assays (equivalent to a % Δlow ± 22.2%). b Δ: Cut-offs for Abbott and Beckman-Coulter ± 15 ng/L/Siemens ± 30 ng/L/Roche ± 7 ng/L. c z-Values calculated with SDbiological: Abbott and Beckman-Coulter 5 ng/L, Siemens 10 ng/L and Roche 2.5 ng/L. d AB: Abbott vs. Beckman Coulter. e AS: Abbott vs. Siemens. f BS: Beckman Coulter vs. Siemens. g AR: Abbott vs. Roche. h BR: Beckman Coulter vs. Roche. i SR: Siemens vs. Roche.
C.J. Pretorius et al. / Clinica Chimica Acta 413 (2012) 1786–1791
A
3.2. The distribution of measured Δ and %Δmean relative to cut-off values
1000
800
Δ ng/L
1789
The distribution of the serial differences obtained with the Abbott cTnI method across the measuring range is illustrated in Fig. 2 and this was representative for all the methods. The Δ's were mostly distributed in a narrow band relative to the mean cTnI with only 3 (0.5%) results with z > 1.96 below; and 13 results (2.3%) with z b 1.96 above an arbitrary 20 ng/L cut-off. The %Δ's were more uniformly distributed and 21 (3.7%) results with z b 1.96 were above; and 12 (2.1%) with z > 1.96 were below an arbitrary 20% cut-off. Adjusting the Δ cut-off level will mostly affect the classification of cases with a mean value b200 ng/L while adjusting the %Δ cut-off will have an affect across the entire measuring range.
Abbott Beckman Coulter Siemens Roche
600
400
200
3.3. Comparison of differences measured with four analyzers 0 10
100
1000
10000
Mean cTn ng/L
B
200
%Δ mean
150
100
50
4. Discussion
0 10
100
1000
10000
We evaluated relative differences (%Δ), actual differences (Δ) and z-values and directly compared the performance of these approaches on serial samples analyzed concurrently on four commonly used cTnI and cTnT methods. z-Values were not affected by the troponin concentration and were transportable across methods. Both %Δ and Δ cut-offs were method specific and depended on the mean troponin value. Method specific fixed Δ cut-off values were predicted to perform well at low and fixed %Δ cut-off values at high troponin levels. We found good agreement between all methods with method specific fixed Δ cut-offs and z-values, but only moderate agreement between cTnI and cTnT with fixed assay specific %Δ cut-offs.
Mean cTn ng/L
C
100
80
%Δ mean
The Δ and %Δ values obtained are presented in Fig. 3 with the Abbott cTnI method as the arbitrary reference method. The %Δ's demonstrated the following slopes: Beckman Coulter 1.00, Siemens 1.07 and Roche 0.74 with considerable scatter around the regression lines for all the methods. The slopes of Δ's relative to the Abbott were: Beckman Coulter 1.08, Siemens 1.87 and Roche 0.42. The agreement between methods is summarized in Table 2. Good agreement was obtained with fixed method specific Δ cut-off values and with z-values, also between the cTnI and cTnT assays. Good agreement relative to an arbitrary 20% fixed %Δ cut-off was obtained between cTnI methods with κ approximately 0.80 in all instances. In contrast, moderate agreement was obtained between cTnI and cTnT (κ b 0.50) with the cTnI methods classifying more cases as elevated than with cTnT. After adjusting the cut-off of the Roche cTnT assay to 14% according to the slope of the regression equations, the overall agreement did not improve (κ 0.41 to 047).
60
40
4.1. Comparison of z-values, actual- and relative differences
20
0 0
200
400
600
800
1000
|Actual Δ| ng/L Fig. 1. Theoretical distribution of cut-off levels for serial cardiac troponin differences measured with four methods. The absolute actual differences (|Δ|) compared to the mean cardiac troponin level. The absolute relative differences (|%Δ|) compared to the mean cardiac troponin level. Absolute relative differences (|%Δ|) compared to absolute actual differences (|Δ|). Cut-off values for each method were calculated with z= 1.96, a constant biological variation component (Abbot and Beckman 5 ng/L, Siemens 10 ng/L and Roche 2.5 ng/L) and method specific analytical variation data.
calculated with the published proportional individual biological variation (%CVbiological) of 48.2% for cTnT identified less than 10% of the cases with an increased Δ cTnT and were not further evaluated [10].
Both Δ and %Δ cut-off values that corresponded to z = 1.96 were method specific and varied with the mean cardiac troponin level (Fig. 1). The large %Δ cut-off values at low concentrations were a function of decreasing analytical precision and the influence of the relatively large constant biological variation component. At high values the influence of the biological variation component diminished and precision expressed as %CV became constant. At low mean cardiac troponin levels, typically seen in non-ST segment elevation MI, a fixed %Δ cut-off can therefore be predicted to perform poorly. Fixed %Δ cut-offs will however perform well at mean cardiac troponin levels above approximately 300 ng/L (Fig. 1B). The Δ cut-offs were relatively stable at low concentrations, which will facilitate the clinical use of a fixed value in cases with lower elevations of troponin such as non-ST segment elevation MI (Fig. 1A). Since the population studied by Reichlin et al. was confined to non-ST segment elevation MI this may partly explain their observation that a fixed Δ cut-off outperformed a fixed %Δ limit [5]. The
1790
C.J. Pretorius et al. / Clinica Chimica Acta 413 (2012) 1786–1791
A 1000
A
200 Roche cTnT Beckman Coulter cTnI Siemens cTnI
100
%Δ
Δ ng/L
100 0
10 -100
-200 -200
1 10
100
1000
10000
-100
0
100
200
2000
4000
%Δ Abbott
Mean Abbott cTnI ng/L
B 4000
B
Roche cTnT Beckman Coulter cTnI Siemens cTnI
100
Δ ng/L
%Δ mean
2000
10
0
-2000
1 10
100
1000
10000
-2000
0
Δ ng/L Abbott
Mean Abbott cTnI ng/L
Fig. 3. Comparison of actual and relative cardiac troponin differences obtained with four cardiac troponin methods. Relative differences (n = 575). Passing Bablok regression slopes (95% CI) when compared to Abbott were: Beckman Coulter 1.0 (0.98 to 1.01); Siemens 1.07 (1.05 to 1.11) and Roche 0.74 (0.67 to 0.81). All offsets were 0.00 except for the Roche 1.70 (0.86 to 3.09). The dotted lines indicate ±30% decision levels. Actual differences between −4000 and +4000 ng/L are shown (n = 420). Regression analysis was performed with the complete data set (n = 575). Passing Bablok regression slopes (95% CI) compared to Abbott were: Beckman Coulter 1.08 (1.03 to 1.12); Siemens 1.87 (1.77 to 1.97) and Roche 0.24 (0.21 to 0.26). All offsets were 0.00.
C 100
%Δ mean
-4000 -4000
10
1 1
10
100
1000
|Δ| ng/L Fig. 2. Distribution of differences measured on serial samples with the Abbott cTnI method. Absolute actual differences (|Δ|) compared to the mean cardiac troponin level. The dashed line indicates a fixed cut-off value of 20 ng/L. Absolute relative differences (|%Δ|) compared to the mean cardiac troponin level. The dashed line indicates a fixed cut-off value of 20%. Absolute relative differences (|Δ|)compared to absolute actual differences (|%Δ|). Cut-off values for each method were calculated with z = 1.96, a constant biological variation component (5 ng/L) and method specific analytical variation data. Results with a Δ or %Δ of zero are not shown due to the logarithmic axis scaling.
diagnostic performance of fixed Δ cut-offs will deteriorate at higher concentrations, but since significant cardiac pathology due to ischemic and other causes becomes more probable at higher levels, these false positive results may be clinically less relevant. Cardiac troponin levels and serial changes are not critical in the diagnosis of ST segment elevation MI, which is typically associated with gross elevations of cardiac markers. z-Values are unaffected by the distribution of results or choice of analytical method. This will simplify the definition of a significant change in cardiac troponin and it can be used as a universal cut-off. Another potential advantage is that z-values reinforce the concept that the risk of MI is a continuum of probability and not a dichotomous event. While calculating z-values and a variable Δ or %Δ cut-off can be performed with laboratory information systems, it is beyond mental arithmetic which may limit clinical application. A convenient shortcut, that may assist clinical decision making, is to determine a combination of a fixed Δ and %Δ with a given probability for each assay (Fig. 1C). A difference greater than 15 ng/L and 10% with the Abbott cTnI assay will generally correspond to a z-value >1.96, which is expected to occur due to chance in less than 5% of cases (pb 0.05).
C.J. Pretorius et al. / Clinica Chimica Acta 413 (2012) 1786–1791
4.2. Comparison of serial differences measured on four analyzers The comparison of serial differences obtained with three TnI methods and a cTnT method clearly demonstrated that significant systematic variation existed and that a fixed Δ or %Δ cut-offs for all methods would be futile (Fig. 3). The regression slopes of the Δ's corresponded to the regression slopes of the original data and may be largely explained by variability in calibration of the cTnI methods and physiological differences between cTnI and cTnT [12]. The %Δ's of cTnT were significantly lower than those of cTnI and may reflect a physiological difference in their respective clearance, as one would expect the %Δ's of molecules released equimolarly to change in tandem. The wide dispersion of results between the methods was consistent with our previous observation of large variances between methods and between cTnI and cTnT measured on the same samples [12]. The agreement of %Δ between cTnI and cTnT methods was poor with approximately 26% of results classified as discordant even after optimizing the individual cut-offs. The classification of results according to the fixed method specific Δ cut-offs resulted in good agreement amongst all cTnI methods and between cTnI and cTnT and the classification did not alter substantially when small adjustments to the cut-offs were made. This observation can be explained by the distribution of individual Δ's in Fig. 2A where relatively few cases occurred with a Δ between 15 and 25 ng/L and small changes in the fixed cut-off would thus not have a major classification effect. 4.3. Comparison with the findings of others Our results are in general agreement with the finding that Δ's are superior to %Δ in predicting clinical outcome and the threshold-Δ values calculated with our assumptions on biological variation are similar to those established empirically for the Siemens and Roche assays [5,6]. Our prediction that the Δ cut-off will increase with higher cTn concentrations is supported by the observation that the upper range of Δ in coronary care patients without MI also followed this pattern [19]. A direct comparison of cut-off values for relative differences in published reports with our results is not possible as information on the actual cTn values observed in the studies were not readily available. Our data however provides a possible explanation for the wide range of the reported fixed %Δ cut-off values, as the range of cTn concentrations observed in each study will have a direct bearing on the optimal cut-off value. The clinical importance of the discordant classification we observed by the respective methods is uncertain and will be the subject of further study. The good overall agreement in classification of the majority of cases with Δ and z-values is reassuring in view of the described systematic and random variation between cardiac troponin methods and in the classification of results relative to the respective 99th percentiles [12]. This observation supports the use of a universal z-value cut-off; or a fixed assay specific Δ cut-off at low concentrations only, with all the cardiac troponin methods described here. 5. Conclusion We fully support the consensus opinion that the clinical setting is crucial when interpreting cTn results as it does not necessarily follow that analytically and physiologically significant differences always translate into clinically significant findings, nor does the lack thereof always guarantee the absence of pathology [8]. Furthermore non-ischemic causes of elevated cTn can also be dynamic over time. The notion of a single fixed
1791
relative difference (%Δ) threshold over the entire analytical range, across platforms and for both cTnI and cTnT is however fundamentally flawed both from a theoretical and practical perspective, and should be avoided. A fixed method specific actual difference (Δ) cut-off can be used at low cTn values or a universal z-value cut-off across the measuring range. Differences with a z-value above the cut-off cannot be explained by analytical and physiological variation and indicate underlying pathology with a high probability. Finally we recommend that guidelines define a significant change in cTn on the basis of z-values. Supplementary data to this article can be found online at http:// dx.doi.org/10.1016/j.cca.2012.07.001. Acknowledgments We thank Jill Tate for reading the manuscript and for her constructive comments. References [1] Thygesen K, Alpert JS, White HD. On behalf of the Joint ESC/ACCF/AHA/WHF Task Force for the Redefinition of Myocardial Infarction: universal definition of myocardial infarction. J Am Coll Cardiol 2007;50:2173–95. [2] Eggers KM, Jaffe AS, Venge P, Lindahl B. Clinical implications of the change of cardiac troponin I levels in patients with acute chest pain — an evaluation with respect to the Universal Definition of Myocardial Infarction. Clin Chim Acta 2011;412:91–7. [3] Apple FS, Pearce LA, Smith SW, Kaczmarek JM, Murakami MM. Role of monitoring changes in sensitive cardiac troponin I assay results for early diagnosis of myocardial infarction and prediction of risk of adverse events. Clin Chem 2009;55:930–7. [4] Giannitsis E, Becker M, Kurz K, Hess G, Zdunek D, Katus HA. High-sensitivity cardiac troponin T for early prediction of evolving non-ST-segment elevation myocardial infarction in patients with suspected acute coronary syndrome and negative troponin results on admission. Clin Chem 2010;56:642–50. [5] Reichlin T, Irfan A, Twerenbold R, et al. Utility of actual and relative changes in cardiac troponin concentrations in the early diagnosis of acute myocardial infarction. Circulation 2011;124:136–45. [6] Mueller M, Biener M, Vafaie M, et al. Absolute and relative kinetic changes of high-sensitivity cardiac troponin T in acute coronary syndrome and in patients with increased troponin in the absence of acute coronary syndrome. Clin Chem 2012;58:209–18. [7] Apple FS, Collinson PO. Analytical characteristics of high-sensitivity cardiac troponin assays. Clin Chem 2012;58:45–53. [8] Thygesen K, Mair J, Katus H, et al. Recommendations for the use of cardiac troponin measurement in acute cardiac care. Eur Heart J 2010;31:2197–206. [9] Wu AHB, Lu QA, Todd J, Moecks J, Wians F. Short and long term biological variation in cardiac troponin I measured with a high-sensitivity assay: implications for clinical practice. Clin Chem 2009;55:52–8. [10] Vasile VC, Saenger AK, Kroning JM, Jaffe AS. Biological and analytical variability of a novel high-sensitivity cardiac troponin T assay. Clin Chem 2010;56:1086–90. [11] Vasile VC, Saenger AK, Kroning JM, Klee GG, Jaffe AS. Biological variation of a novel cardiac troponin I assay. Clin Chem 2011;57:1080–1. [12] Ungerer JPJ, Wilgen U, Marquart L, O'Rourke PK, Pretorius CJ. Concordance, variance and outliers between 4 contemporary cardiac troponin assays: implications for harmonization. Clin Chem 2012;58:274–83. [13] Harris EK, Yasaka T. On the calculation of a “reference change” for comparing two separate measurements. Clin Chem 1983;29:25–30. [14] Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1986;i:307–10. [15] Gill JS, Zezulka AV, Beevers DG, Davies P. Misleading statistics: Relation between initial blood pressure and its fall with treatment. Lancet 1985;i:567–9. [16] Costongs GMPJ, Janson PCW, Bas BM, et al. Short-term and long-term intra-individual variations and critical differences of chemical laboratory parameters. J Clin Chem Biochem 1985;23:7–16. [17] Pretorius CJ, Dimeski G, O'Rourke PK, et al. Outliers as a cause of false cardiac troponin results: Investigating the robustness of 4 contemporary assays. Clin Chem 2011;57:710–8. [18] Zar JH. Chapter 18 in Biostatistical Analysis. fifth Edition. New Jersey: Prentice Hall; 2010. pp. 372–5. [19] Hammarsten O, Fu MLX, Sigurjonsdottir R, et al. Troponin T percentiles from a random population sample, emergency room patients and patients with myocardial infarction. Clin Chem 2012;58:628–37.
Contents lists available at SciVerse ScienceDirect
Clinica Chimica Acta journal homepage: www.elsevier.com/locate/clinchim
Serial cardiac troponin differences measured on four contemporary analyzers: Relative differences, actual differences and reference change values compared Carel J. Pretorius ⁎, Urs Wilgen, Jacobus P.J. Ungerer Department of Chemical Pathology, Pathology Queensland, Brisbane, Queensland, Australia
a r t i c l e
i n f o
Article history: Received 18 April 2012 Received in revised form 6 June 2012 Accepted 2 July 2012 Available online 10 July 2012 Keywords: Cardiac troponin Myocardial infarction Acute coronary syndrome Reference change value Serial difference
a b s t r a c t Introduction: The diagnosis of myocardial infarction is in part predicated on a rise and/or fall in cardiac troponin (cTn). z-Values incorporate qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi analytical and biological variation to standardize serial differences: z ¼ Δ=
2SD2Analytical þ 2SD2Biological . We investigated the theoretical distributions of actual differences (Δ),
relative differences (%Δ) and z-values and compared the agreement in classification of differences measured on four contemporary platforms. Methods: cTn was measured in 575 sample pairs on the Abbott Architect cTnI, Beckman Coulter Access2 cTnI, Roche Cobas cTnT and Siemens ADVIA Centaur cTnI methods. Results: Good agreement was obtained amongst all methods with a universal z-value cut-off (κ>0.79) and method specific fixed Δ cut-offs (κ>0.82) while suboptimal agreement was obtained between cTnI and cTnT with fixed %Δ cut-offs (κ b 0.50). Conclusion: Fixed Δ cut-offs will perform well at low cTn concentrations while fixed %Δ cut-off values are predicted to perform poorly. z-Values are independent of the cTn concentration, present differences as a continuum of probability and a universal decision level can be used for all cTnI and cTnT methods. Crown Copyright © 2012 Published by Elsevier B.V. All rights reserved.
1. Introduction Cardiac troponin (cTn) is considered the standard biomarker for myocardial infarction (MI) with the universal definition requiring a rise and/or fall with at least one value exceeding the 99th percentile reference limit [1]. With the recognition of non-ischemic etiologies of cTn elevations it became important to accurately define a significant change of cTn in order to improve the diagnosis and clinical management of patients suspected of suffering from acute coronary syndromes (ACS). Both relative (%Δ) and actual numeric (Δ) differences in cTn have been reported to increase diagnostic specificity for MI while maintaining diagnostic sensitivity [2–6]. A range of %Δ decision levels between 20% and 243% has been reported for cardiac troponin-I (cTnI) and cardiac troponin-T (cTnT) [2–4,7]. Α Δ of 6.9
Abbreviations: MI, myocardial infarction; ACS, Acute coronary syndrome; cTn, cardiac troponin; cTnI, cardiac troponin I; cTnT, cardiac troponin T; Δ, Actual numeric difference; %Δ, Relative difference; RCV, Reference change value; LOQ20%, Limit of quantitation defined by a 20% coefficient of variation; SDbiological, Standard deviation of individual biological variability. ⁎ Corresponding author at: Department of Chemical Pathology, Block 7 Floor 3, Herston Hospitals Campus, Herston, 4029, Queensland, Australia. Tel.: +61 7 3646 0083; fax: +61 7 3646 1392. E-mail address: [email protected] (C.J. Pretorius).
to 9.2 ng/L for cTnT and 20 ng/L for a cTnI assay was reported to be clinically superior to %Δ cut-offs [5,6]. It is thus clear that there is no consensus on how to define a significant cTn difference with contradictory cut-off levels proposed by various authors. In view of the variable precision of cTn assays over the measuring range we also noted that little attention was paid to the relationship between the cTn concentration and the various proposed cut-off levels. Comparing the observed %Δ or Δ to a cut-off derived from a reference change value (%RCV or RCV) was proposed as a mechanism to determine if the magnitude of cTn change is significant [8]. Calculating a RCV requires assay specific information on analytical and biological variation. Information on the analytical variation over the measuring range is readily available from laboratories and although the biological variation of cTnI and cTnT has been reported, it is not clear how to apply this information in clinical practice. Some authors have suggested that biological variation should only be taken into account when interpreting results around the 99th percentile cut offs, as these small physiological fluctuations are overwhelmed at higher levels by the pathological release from necrotic tissue [8,9]. Furthermore the %RCV is often reported with an asymmetrical reference range depending on whether an increase or decrease occurred [7,9–11]. Most papers however appear to calculate the %Δ relative to the lower result, irrespective of whether an increase or decrease occurred and this may potentially lead to confusion when compared to a %RCV derived cut-off calculated differently. We are
0009-8981/$ – see front matter. Crown Copyright © 2012 Published by Elsevier B.V. All rights reserved. doi:10.1016/j.cca.2012.07.001
C.J. Pretorius et al. / Clinica Chimica Acta 413 (2012) 1786–1791
not aware of a systematic comparison between empirically determined fixed %Δ or Δ cut-off values and cut-off values determined with a RCV approach. Comparative information on the differences between serial results measured with commercially available cTnI and cTnT assays is incomplete and we are aware of one paper that compared assays directly [5]. Given the lack of standardization amongst cTnI assays, the large reported variances amongst assays [12] and the apparent biological differences that exist between cTnI and cTnT, it is clinically important to understand how assays relate to each other in this respect. In view of the aspects discussed above we defined the terminology and examined some of the theoretical aspects of how to present differences between serial samples. We were particularly interested in the relationship between the degree of cTn elevation and the cut-off values for %Δ and Δ, which we calculated from information on analytical and biological variation at a defined probability. We investigated the distribution of differences obtained on serial patient samples in relation to arbitrarily defined fixed cut-offs and the method specific, but concentration dependent, variable cut-off values. The concordance between four analytical platforms was then determined for differences measured concurrently on patient samples. 2. Materials and methods 2.1. Glossary of terms and definitions 2.1.1. The actual difference (Δ) is the numerical difference between two results in the original units of measurement. Δ = Result2 − Result1, where the sign (+ or –) indicates if an increase or decrease occurred. 2.1.2. The absolute difference is defined as |Δ| = |Result2 − Result1|, again in the original units, but with no indication of whether an increase or decrease occurred. 2.1.3. The relative difference (%Δ) is a ratio where the Δ is expressed relative to a denominator and reported as a percentage. 2.1.3.1. Relative difference as change (%Δchange). %Δchange =(Result2 − Result1)/Result1 × 100, the sign (+ or –) indicates an increase or decrease. The %Δchange results have a numerical range between − 100% and +∞%. 2.1.3.2. Relative difference with respect to the lower result of the pair (%Δlow). %Δlow = (Resulthigh − Resultlow)/Resultlow × 100. The %Δlow results have a numerical range of 0% to +∞%. 2.1.3.3. Relative difference with respect to the mean of the two values (%Δmean). %Δmean = (Result2 − Result1)/0.5(Result2 + Result1) × 100, the sign (+ or –) indicates that an increase or decrease occurred. The %Δmean results have a numerical range of ±200%. 2.1.3.4. %Δmean and %Δlow can easily be inter-converted. %Δmean = %Δlow/ (100 + %Δlow/2). 2.1.4. The reference change value is a cut-off value with a defined probability. The underlying principle is that the variance of differences is estimated from the variances of the observed values. Analytical and biological variances are the main components contributing to the variance of the observed results and the cut-off values are therefore both method specific and concentration dependant.
1787
2.1.4.1. Actual reference change value (RCV). RCV ¼ z ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi q 2SD2 Analytical þ 2SD2Biological . For a 2-sided 95% probability that the difference is significant, a z-value of 1.96 is used (p b 0.05). 2.1.4.2. Relative reference change value (%RCV). %RCV ¼ z rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi 2 2 2 %CV Analytical þ %CVBiological . Analogously the %Δ is compared to the %RCV and if it exceeds this cut-off, the difference is regarded as significant. 2.1.5. The z-value is the probability associated with an observed difference and is obtained by rearranging the formula for RCV (or %RCV): z ¼ Δ= qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2SD2Analytical þ 2SD2Biological . Simplistically, this standardizes the observed difference to a multiple of the estimated standard deviation of the differences. 2.2. Theoretical considerations 2.2.1. Calculating relative differences We prefer to use %Δmean to express relative differences for a number of reasons. Firstly it is explicitly assumed in the simplified calculation of %RCV that there is negligible covariance between the two results i.e. r ≈ 0 [13]. By using %Δchange and even more so with %Δlow the statistical artifact of auto-correlation is introduced, which violates the above stated assumption [14,15]. In the context of a null hypothesis the two results are random variables in the distribution around the true (unknown) value of which the mean value is the best estimate that we have [14]. Furthermore it avoids asymmetrical results depending on the direction of change and results are in a numeric range of − 200% to + 200% that facilitates graphical representation. The probability of a significant difference will tend to be overestimated if %Δlow is compared to a %RCV cut-off and vice versa (see worked example 1 in the online appendix). Comparison of %Δmean to empirically determined %Δlow cut-offs presents little problem as the values can be easily converted (see Section 2.1.3.4). 2.2.2. Biological variation of cardiac troponin The biological variation of cTn is typically reported as a %CV with the implication that this physiological variation increases proportionally as the set point in an individual increases. The graphical presentation of the within-individual biological variation however does not convincingly demonstrate a proportional relationship to the individual median values [9,10]. In our opinion it does not make biological sense to assume that a proportional physiological variation can be transposed onto a pathological release of cTn due to ischemia or necrosis. Such a pathological release is independent of, and overwhelms the small physiological variation observed in cardio-healthy individuals. Even if short term variation in cTn occurs under pathological– ischaemic conditions, it should not be considered in the calculation of a cut-off value as the objective is to differentiate differences due to analytical and physiological variation from pathological changes. Based on these observations we propose that in the context of ACS a constant actual biological variation component (SDbiological) should be taken into consideration when calculating cut-off values. This will ensure that both physiological and analytical variations are considered at cTn values around the 99th percentile cut-offs while the influence of the physiological variation becomes negligible at grossly elevated levels. The median value reported for the short term biological variation of a novel cTnI assay was 2.2 ng/L [11]. Specific data for the methods used in this study were not available and due to the imprecision of the cTnI assays it cannot be accurately obtained for these assays. For the calculations in this paper, based on the approximate linear relationships between the assays, we
1788
C.J. Pretorius et al. / Clinica Chimica Acta 413 (2012) 1786–1791
assumed a median SDbiological 2.5 ng/L for the Roche, 5 ng/L for the Abbott and Beckman Coulter, and 10 ng/L for the Siemens assays. 2.2.3. Calculating z-values z-Values standardize differences as probabilities that are independent of analytical method or cTn concentrations. This allows the use of a universal cut-off (i.e. 1.96 or 2.56) as opposed to method specific and concentration dependent cut-off values. For similar reasons to those stated in Section 2.2.1 above, the analytical and biological variances at the mean level of the two results are used to estimate the variance of differences. According to the universal definition a rise and/or fall of cTn needs to be considered and a two-sided p value to test for significance is appropriate. In acute coronary syndrome changes occur over a period of hours and it is therefore appropriate to use analytical and intra-individual biological variation determined over an equivalent period. The z-value is partly dependent on the magnitude of the SDbiological and the use of a numerically larger value, such as a 99th percentile in place of a median value, will lead to an increase in clinical specificity at the expense of sensitivity [13,16]. 2.3. Study population and analytical methods The study population, quality control procedures, precision parameters, identification of outliers and method comparisons were described in detail elsewhere [12,17]. Serum cTn was measured in duplicate on four contemporary analytical platforms: 1. Abbott Architect i2000SR analytical system with STAT Troponin-I reagent (Abbott Diagnostics, Lane Cove, NSW, Australia). 2. Beckman Coulter Access2 analyzer with Enhanced AccuTnI reagent (Reagent part number A78803, Beckman Coulter Diagnostics, Brae, CA, USA) analyzer. 3. Roche Cobas e601 with TroponinT hs reagent (Roche Diagnostics, Sydney, Australia). 4. Siemens ADVIA Centaur XP with TnI-Ultra reagent (Siemens Healthcare Diagnostics Inc. Deerfield, IL). For the sake of brevity the methods are referred to as Abbott, Beckman Coulter, Roche and Siemens hereafter. The performance characteristics of the respective assays are summarized in Table 1. After exclusion of outliers the actual (Δ) and relative differences (%Δmean) were calculated from the first of the duplicate results in 575 non-overlapping sample-pairs from 406 patients. Results below the limit of quantitation, as defined by a 20% coefficient of variation (LOQ20%), were deemed to be equivalent to the LOQ20% when calculating differences. z-Values were calculated as described in Sections 2.1.5 and 2.2.3 with a constant biological variation component as discussed in Section 2.2.2. The analytical variation component was calculated from the method specific precision profile data [17]. Statistical testing was performed at the 5% significance level (p b 0.05) unless stated otherwise. The regression slopes of the Δ and %Δmean were subdivided into intervals of b 6, 6 to 12, 12 to 24 and > 24 h and tested for significant differences for each analyzer [18]. Agreement in the binary classification between methods was evaluated with Cohen's κ-statistic at a fixed %Δmean cut-off (20%), fixed method specific Δ cut-off values and according to a z-value of Table 1 Summary of the performance characteristics of the Abbott, Beckman Coulter, Siemens and Roche assays.
Abbott Beckman Coulter Siemens Roche
Measurand
99th percentile cut-offa
10% CVb
20% CVb
cTnI cTnI cTnI cTnT
28 40 40 14
48 64 57 6
24 32 29 3
All values are reported in ng/L. a The 99th percentiles are presented as reported by the respective manufacturers. b The values at which a CV of 10% and 20% respectively were attained, were estimated from the total CV determined according to CLSI EP5-A2 and reported elsewhere [17].
1.96. The Δ cut-off values were based on the published empirical data for the Roche cTnT [5] and estimated for the cTnI assays according to the approximate relationships between the assays: Abbott and Beckman-Coulter 15 ng/L, Siemens 30 ng/L and Roche 7 ng/L. 3. Results The median time interval (interquartile range) between repeat samples was 21.8 h (7.0 to 36.3). The median (interquartile range) of the actual differences in ng/L was 3 (0 to 170) for the Abbott, 4 (0 to 269) for the Beckman Coulter, 8 (0 to 464) for the Siemens and 5 (0 to 38) for the Roche analysers respectively. A subgroup analysis of differences did not demonstrate a time dependant aspect and we therefore pooled the data (p b 0.05). 3.1. Influence cTn concentration on Δ and %Δmean cut-off values In Fig. 1 the theoretical cut-off values are obtained with z = 1.96 (p = 0.05), and a constant method specific SDbiological and the precision data of the respective analyzers are presented. The Δ limits were constant at low mean cardiac troponin levels and increased progressively above a mean value of approximately 200 ng/L. In contrast a large and variable %Δ defined the cut-off at low values with relatively constant limits at higher values above a mean of approximately 300 ng/L. Plotting the %Δ against Δ presents the same information independent of the mean value and confirmed a threshold effect for %Δ and Δ. Serial results with a z-value >1.96 generally corresponded to both a Δ > 15 ng/L and a %Δ > 10% on the Abbott and Beckman platforms. The corresponding thresholds for the Siemens were 20 ng/L and 7%, and for Roche 7 ng/L and 5%. z-Values Table 2 Agreement in the binary classification of serial cardiac troponin differences measured with four contemporary assays. ABd % Δmeana Concordant + 210 Relative difference Concordant − 318 Test 1 +, 23 (%) Test 2 − Test 2 +, Test 24 1− κ 0.83 Δb Concordant + 222 Actual difference Concordant − 324 Test 1 +, 15 (ng/L) Test 2 − Test 2 +, 14 Test 1 − κ 0.90 Z > ±1.96 (SDbiological)c z-Value Concordant + 207 Concordant − 335 Test 1 +, 15 Test 2 − Test 2 +, Test 18 1− κ 0.88 a
ASe
BSf
ARg
BRh
SRi
212 302 21
219 308 15
170 255 63
168 252 66
179 245 73
40
33
87
89
78
0.78
0.83
0.47
0.45
0.47
224 324 13
226 327 10
222 314 15
216 309 20
223 314 15
14
12
24
30
23
0.90
0.92
0.86
0.82
0.86
208 330 14
217 336 8
195 324 27
197 323 28
204 324 27
23
14
29
27
20
0.87
0.92
0.79
0.80
0.83
% Δmean; ± 20% cut-off used for all assays (equivalent to a % Δlow ± 22.2%). b Δ: Cut-offs for Abbott and Beckman-Coulter ± 15 ng/L/Siemens ± 30 ng/L/Roche ± 7 ng/L. c z-Values calculated with SDbiological: Abbott and Beckman-Coulter 5 ng/L, Siemens 10 ng/L and Roche 2.5 ng/L. d AB: Abbott vs. Beckman Coulter. e AS: Abbott vs. Siemens. f BS: Beckman Coulter vs. Siemens. g AR: Abbott vs. Roche. h BR: Beckman Coulter vs. Roche. i SR: Siemens vs. Roche.
C.J. Pretorius et al. / Clinica Chimica Acta 413 (2012) 1786–1791
A
3.2. The distribution of measured Δ and %Δmean relative to cut-off values
1000
800
Δ ng/L
1789
The distribution of the serial differences obtained with the Abbott cTnI method across the measuring range is illustrated in Fig. 2 and this was representative for all the methods. The Δ's were mostly distributed in a narrow band relative to the mean cTnI with only 3 (0.5%) results with z > 1.96 below; and 13 results (2.3%) with z b 1.96 above an arbitrary 20 ng/L cut-off. The %Δ's were more uniformly distributed and 21 (3.7%) results with z b 1.96 were above; and 12 (2.1%) with z > 1.96 were below an arbitrary 20% cut-off. Adjusting the Δ cut-off level will mostly affect the classification of cases with a mean value b200 ng/L while adjusting the %Δ cut-off will have an affect across the entire measuring range.
Abbott Beckman Coulter Siemens Roche
600
400
200
3.3. Comparison of differences measured with four analyzers 0 10
100
1000
10000
Mean cTn ng/L
B
200
%Δ mean
150
100
50
4. Discussion
0 10
100
1000
10000
We evaluated relative differences (%Δ), actual differences (Δ) and z-values and directly compared the performance of these approaches on serial samples analyzed concurrently on four commonly used cTnI and cTnT methods. z-Values were not affected by the troponin concentration and were transportable across methods. Both %Δ and Δ cut-offs were method specific and depended on the mean troponin value. Method specific fixed Δ cut-off values were predicted to perform well at low and fixed %Δ cut-off values at high troponin levels. We found good agreement between all methods with method specific fixed Δ cut-offs and z-values, but only moderate agreement between cTnI and cTnT with fixed assay specific %Δ cut-offs.
Mean cTn ng/L
C
100
80
%Δ mean
The Δ and %Δ values obtained are presented in Fig. 3 with the Abbott cTnI method as the arbitrary reference method. The %Δ's demonstrated the following slopes: Beckman Coulter 1.00, Siemens 1.07 and Roche 0.74 with considerable scatter around the regression lines for all the methods. The slopes of Δ's relative to the Abbott were: Beckman Coulter 1.08, Siemens 1.87 and Roche 0.42. The agreement between methods is summarized in Table 2. Good agreement was obtained with fixed method specific Δ cut-off values and with z-values, also between the cTnI and cTnT assays. Good agreement relative to an arbitrary 20% fixed %Δ cut-off was obtained between cTnI methods with κ approximately 0.80 in all instances. In contrast, moderate agreement was obtained between cTnI and cTnT (κ b 0.50) with the cTnI methods classifying more cases as elevated than with cTnT. After adjusting the cut-off of the Roche cTnT assay to 14% according to the slope of the regression equations, the overall agreement did not improve (κ 0.41 to 047).
60
40
4.1. Comparison of z-values, actual- and relative differences
20
0 0
200
400
600
800
1000
|Actual Δ| ng/L Fig. 1. Theoretical distribution of cut-off levels for serial cardiac troponin differences measured with four methods. The absolute actual differences (|Δ|) compared to the mean cardiac troponin level. The absolute relative differences (|%Δ|) compared to the mean cardiac troponin level. Absolute relative differences (|%Δ|) compared to absolute actual differences (|Δ|). Cut-off values for each method were calculated with z= 1.96, a constant biological variation component (Abbot and Beckman 5 ng/L, Siemens 10 ng/L and Roche 2.5 ng/L) and method specific analytical variation data.
calculated with the published proportional individual biological variation (%CVbiological) of 48.2% for cTnT identified less than 10% of the cases with an increased Δ cTnT and were not further evaluated [10].
Both Δ and %Δ cut-off values that corresponded to z = 1.96 were method specific and varied with the mean cardiac troponin level (Fig. 1). The large %Δ cut-off values at low concentrations were a function of decreasing analytical precision and the influence of the relatively large constant biological variation component. At high values the influence of the biological variation component diminished and precision expressed as %CV became constant. At low mean cardiac troponin levels, typically seen in non-ST segment elevation MI, a fixed %Δ cut-off can therefore be predicted to perform poorly. Fixed %Δ cut-offs will however perform well at mean cardiac troponin levels above approximately 300 ng/L (Fig. 1B). The Δ cut-offs were relatively stable at low concentrations, which will facilitate the clinical use of a fixed value in cases with lower elevations of troponin such as non-ST segment elevation MI (Fig. 1A). Since the population studied by Reichlin et al. was confined to non-ST segment elevation MI this may partly explain their observation that a fixed Δ cut-off outperformed a fixed %Δ limit [5]. The
1790
C.J. Pretorius et al. / Clinica Chimica Acta 413 (2012) 1786–1791
A 1000
A
200 Roche cTnT Beckman Coulter cTnI Siemens cTnI
100
%Δ
Δ ng/L
100 0
10 -100
-200 -200
1 10
100
1000
10000
-100
0
100
200
2000
4000
%Δ Abbott
Mean Abbott cTnI ng/L
B 4000
B
Roche cTnT Beckman Coulter cTnI Siemens cTnI
100
Δ ng/L
%Δ mean
2000
10
0
-2000
1 10
100
1000
10000
-2000
0
Δ ng/L Abbott
Mean Abbott cTnI ng/L
Fig. 3. Comparison of actual and relative cardiac troponin differences obtained with four cardiac troponin methods. Relative differences (n = 575). Passing Bablok regression slopes (95% CI) when compared to Abbott were: Beckman Coulter 1.0 (0.98 to 1.01); Siemens 1.07 (1.05 to 1.11) and Roche 0.74 (0.67 to 0.81). All offsets were 0.00 except for the Roche 1.70 (0.86 to 3.09). The dotted lines indicate ±30% decision levels. Actual differences between −4000 and +4000 ng/L are shown (n = 420). Regression analysis was performed with the complete data set (n = 575). Passing Bablok regression slopes (95% CI) compared to Abbott were: Beckman Coulter 1.08 (1.03 to 1.12); Siemens 1.87 (1.77 to 1.97) and Roche 0.24 (0.21 to 0.26). All offsets were 0.00.
C 100
%Δ mean
-4000 -4000
10
1 1
10
100
1000
|Δ| ng/L Fig. 2. Distribution of differences measured on serial samples with the Abbott cTnI method. Absolute actual differences (|Δ|) compared to the mean cardiac troponin level. The dashed line indicates a fixed cut-off value of 20 ng/L. Absolute relative differences (|%Δ|) compared to the mean cardiac troponin level. The dashed line indicates a fixed cut-off value of 20%. Absolute relative differences (|Δ|)compared to absolute actual differences (|%Δ|). Cut-off values for each method were calculated with z = 1.96, a constant biological variation component (5 ng/L) and method specific analytical variation data. Results with a Δ or %Δ of zero are not shown due to the logarithmic axis scaling.
diagnostic performance of fixed Δ cut-offs will deteriorate at higher concentrations, but since significant cardiac pathology due to ischemic and other causes becomes more probable at higher levels, these false positive results may be clinically less relevant. Cardiac troponin levels and serial changes are not critical in the diagnosis of ST segment elevation MI, which is typically associated with gross elevations of cardiac markers. z-Values are unaffected by the distribution of results or choice of analytical method. This will simplify the definition of a significant change in cardiac troponin and it can be used as a universal cut-off. Another potential advantage is that z-values reinforce the concept that the risk of MI is a continuum of probability and not a dichotomous event. While calculating z-values and a variable Δ or %Δ cut-off can be performed with laboratory information systems, it is beyond mental arithmetic which may limit clinical application. A convenient shortcut, that may assist clinical decision making, is to determine a combination of a fixed Δ and %Δ with a given probability for each assay (Fig. 1C). A difference greater than 15 ng/L and 10% with the Abbott cTnI assay will generally correspond to a z-value >1.96, which is expected to occur due to chance in less than 5% of cases (pb 0.05).
C.J. Pretorius et al. / Clinica Chimica Acta 413 (2012) 1786–1791
4.2. Comparison of serial differences measured on four analyzers The comparison of serial differences obtained with three TnI methods and a cTnT method clearly demonstrated that significant systematic variation existed and that a fixed Δ or %Δ cut-offs for all methods would be futile (Fig. 3). The regression slopes of the Δ's corresponded to the regression slopes of the original data and may be largely explained by variability in calibration of the cTnI methods and physiological differences between cTnI and cTnT [12]. The %Δ's of cTnT were significantly lower than those of cTnI and may reflect a physiological difference in their respective clearance, as one would expect the %Δ's of molecules released equimolarly to change in tandem. The wide dispersion of results between the methods was consistent with our previous observation of large variances between methods and between cTnI and cTnT measured on the same samples [12]. The agreement of %Δ between cTnI and cTnT methods was poor with approximately 26% of results classified as discordant even after optimizing the individual cut-offs. The classification of results according to the fixed method specific Δ cut-offs resulted in good agreement amongst all cTnI methods and between cTnI and cTnT and the classification did not alter substantially when small adjustments to the cut-offs were made. This observation can be explained by the distribution of individual Δ's in Fig. 2A where relatively few cases occurred with a Δ between 15 and 25 ng/L and small changes in the fixed cut-off would thus not have a major classification effect. 4.3. Comparison with the findings of others Our results are in general agreement with the finding that Δ's are superior to %Δ in predicting clinical outcome and the threshold-Δ values calculated with our assumptions on biological variation are similar to those established empirically for the Siemens and Roche assays [5,6]. Our prediction that the Δ cut-off will increase with higher cTn concentrations is supported by the observation that the upper range of Δ in coronary care patients without MI also followed this pattern [19]. A direct comparison of cut-off values for relative differences in published reports with our results is not possible as information on the actual cTn values observed in the studies were not readily available. Our data however provides a possible explanation for the wide range of the reported fixed %Δ cut-off values, as the range of cTn concentrations observed in each study will have a direct bearing on the optimal cut-off value. The clinical importance of the discordant classification we observed by the respective methods is uncertain and will be the subject of further study. The good overall agreement in classification of the majority of cases with Δ and z-values is reassuring in view of the described systematic and random variation between cardiac troponin methods and in the classification of results relative to the respective 99th percentiles [12]. This observation supports the use of a universal z-value cut-off; or a fixed assay specific Δ cut-off at low concentrations only, with all the cardiac troponin methods described here. 5. Conclusion We fully support the consensus opinion that the clinical setting is crucial when interpreting cTn results as it does not necessarily follow that analytically and physiologically significant differences always translate into clinically significant findings, nor does the lack thereof always guarantee the absence of pathology [8]. Furthermore non-ischemic causes of elevated cTn can also be dynamic over time. The notion of a single fixed
1791
relative difference (%Δ) threshold over the entire analytical range, across platforms and for both cTnI and cTnT is however fundamentally flawed both from a theoretical and practical perspective, and should be avoided. A fixed method specific actual difference (Δ) cut-off can be used at low cTn values or a universal z-value cut-off across the measuring range. Differences with a z-value above the cut-off cannot be explained by analytical and physiological variation and indicate underlying pathology with a high probability. Finally we recommend that guidelines define a significant change in cTn on the basis of z-values. Supplementary data to this article can be found online at http:// dx.doi.org/10.1016/j.cca.2012.07.001. Acknowledgments We thank Jill Tate for reading the manuscript and for her constructive comments. References [1] Thygesen K, Alpert JS, White HD. On behalf of the Joint ESC/ACCF/AHA/WHF Task Force for the Redefinition of Myocardial Infarction: universal definition of myocardial infarction. J Am Coll Cardiol 2007;50:2173–95. [2] Eggers KM, Jaffe AS, Venge P, Lindahl B. Clinical implications of the change of cardiac troponin I levels in patients with acute chest pain — an evaluation with respect to the Universal Definition of Myocardial Infarction. Clin Chim Acta 2011;412:91–7. [3] Apple FS, Pearce LA, Smith SW, Kaczmarek JM, Murakami MM. Role of monitoring changes in sensitive cardiac troponin I assay results for early diagnosis of myocardial infarction and prediction of risk of adverse events. Clin Chem 2009;55:930–7. [4] Giannitsis E, Becker M, Kurz K, Hess G, Zdunek D, Katus HA. High-sensitivity cardiac troponin T for early prediction of evolving non-ST-segment elevation myocardial infarction in patients with suspected acute coronary syndrome and negative troponin results on admission. Clin Chem 2010;56:642–50. [5] Reichlin T, Irfan A, Twerenbold R, et al. Utility of actual and relative changes in cardiac troponin concentrations in the early diagnosis of acute myocardial infarction. Circulation 2011;124:136–45. [6] Mueller M, Biener M, Vafaie M, et al. Absolute and relative kinetic changes of high-sensitivity cardiac troponin T in acute coronary syndrome and in patients with increased troponin in the absence of acute coronary syndrome. Clin Chem 2012;58:209–18. [7] Apple FS, Collinson PO. Analytical characteristics of high-sensitivity cardiac troponin assays. Clin Chem 2012;58:45–53. [8] Thygesen K, Mair J, Katus H, et al. Recommendations for the use of cardiac troponin measurement in acute cardiac care. Eur Heart J 2010;31:2197–206. [9] Wu AHB, Lu QA, Todd J, Moecks J, Wians F. Short and long term biological variation in cardiac troponin I measured with a high-sensitivity assay: implications for clinical practice. Clin Chem 2009;55:52–8. [10] Vasile VC, Saenger AK, Kroning JM, Jaffe AS. Biological and analytical variability of a novel high-sensitivity cardiac troponin T assay. Clin Chem 2010;56:1086–90. [11] Vasile VC, Saenger AK, Kroning JM, Klee GG, Jaffe AS. Biological variation of a novel cardiac troponin I assay. Clin Chem 2011;57:1080–1. [12] Ungerer JPJ, Wilgen U, Marquart L, O'Rourke PK, Pretorius CJ. Concordance, variance and outliers between 4 contemporary cardiac troponin assays: implications for harmonization. Clin Chem 2012;58:274–83. [13] Harris EK, Yasaka T. On the calculation of a “reference change” for comparing two separate measurements. Clin Chem 1983;29:25–30. [14] Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1986;i:307–10. [15] Gill JS, Zezulka AV, Beevers DG, Davies P. Misleading statistics: Relation between initial blood pressure and its fall with treatment. Lancet 1985;i:567–9. [16] Costongs GMPJ, Janson PCW, Bas BM, et al. Short-term and long-term intra-individual variations and critical differences of chemical laboratory parameters. J Clin Chem Biochem 1985;23:7–16. [17] Pretorius CJ, Dimeski G, O'Rourke PK, et al. Outliers as a cause of false cardiac troponin results: Investigating the robustness of 4 contemporary assays. Clin Chem 2011;57:710–8. [18] Zar JH. Chapter 18 in Biostatistical Analysis. fifth Edition. New Jersey: Prentice Hall; 2010. pp. 372–5. [19] Hammarsten O, Fu MLX, Sigurjonsdottir R, et al. Troponin T percentiles from a random population sample, emergency room patients and patients with myocardial infarction. Clin Chem 2012;58:628–37.