Analytic result variation for high-sensitivity cardiac troponin: Interpretation and consequences

A multicentre prospective observational study (36 sites in 8 Canadian provinces), imbedded in the CODE-MI trial, had clinical laboratories at these sites test 3 samples (normal, female 99th-percentile, and male 99th-percentile concentrations) monthly for 1 year on 67 instruments, producing 2142 results from 6 distinct troponin assays. Mean differences varied per assay with maximum values being ± 4, ± 8, and ± 9 ng/L for the normal, female 99th-percentile, and male 99th-percentile sample, respectively. Using a pragmatic approach, combining all assays, the maximum analytic result variation was ± 3 ng/L for target concentrations less than 10 ng/L and ± 30% at concentrations slightly above. Every laboratory test produces results with some variability. For high-sensitivity cardiac troponin (hs-cTn) testing, the emphasis has been on assay precision measured by the coefficient of variation (CV; SD/mean).1Wu A.H.B. Christenson R.H. Greene D.N. et al.Clinical laboratory practice recommendations for the use of cardiac troponin in acute coronary syndrome: expert opinion from the Academy of the American Association for Clinical Chemistry and the Task Force on Clinical Applications of Cardiac Bio-Markers of the International Federation of Clinical Chemistry and Laboratory Medicine.Clin Chem. 2018; 64: 645-655PubMed Google Scholar The most recent laboratory guidelines regarding hs-cTn tests stipulate that the CV at the 99th-percentile concentration be ≤ 10%.1Wu A.H.B. Christenson R.H. Greene D.N. et al.Clinical laboratory practice recommendations for the use of cardiac troponin in acute coronary syndrome: expert opinion from the Academy of the American Association for Clinical Chemistry and the Task Force on Clinical Applications of Cardiac Bio-Markers of the International Federation of Clinical Chemistry and Laboratory Medicine.Clin Chem. 2018; 64: 645-655PubMed Google Scholar This metric can be evaluated by laboratories, but clinicians remain unaware if a difference in serial hs-cTn results is outside the range of acceptable analytic variation (ie, analytic noise) to allow for reliable clinical decision making. This is important because changes in serial hs-cTn results are often incorporated into diagnostic algorithms and places patients in different risk categories. There are a few published limits for hs-cTn error estimates.1Wu A.H.B. Christenson R.H. Greene D.N. et al.Clinical laboratory practice recommendations for the use of cardiac troponin in acute coronary syndrome: expert opinion from the Academy of the American Association for Clinical Chemistry and the Task Force on Clinical Applications of Cardiac Bio-Markers of the International Federation of Clinical Chemistry and Laboratory Medicine.Clin Chem. 2018; 64: 645-655PubMed Google Scholar,2Diaz-Garzon J. Fernandez-Calle P. Sandberg S. et al.European Federation of Clinical Chemistry and Laboratory Medicine (EFLM) Working Group on Biological Variation and Task Group for the Biological Variation Database. Biological variation of cardiac troponins in health and disease: a systematic review and meta-analysis.Clin Chem. 2021; 67: 256-264PubMed Google Scholar However, these error estimates have been largely derived from single centres, may not be applicable to all assays, and do not incorporate variation obtained from different analysers and batches of reagents. The aim of this prospective nationwide study was to establish limits of analytic variation (or the maximum difference from the target concentration) for hs-cTn testing with the use of assays from all major diagnostic companies. Clinical laboratories, emergency physicians and cardiologists were identified through outreach via the CODE-MI study (NCT03819894) by the first author.3Zhao Y. Izadnegahdar M. Lee M.K. et al.High-Sensitivity Cardiac Troponin–Optimising the Diagnosis of Acute Myocardial Infarction/Injury in Women (CODE-MI): rationale and design for a multicentre, stepped-wedge, cluster-randomised trial.Am Heart J. 2020; 229: 18-28Crossref PubMed Scopus (9) Google Scholar In this 12-month accuracy substudy, sites that expressed interest in CODE-MI were identified and asked to participate. Monthly testing occurred at 4 sites in British Columbia, 1 site in Alberta, 3 sites in Saskatchewan, 3 sites in Manitoba, 15 sites in Ontario, 5 sites in Québec, 1 site in New Brunswick, and 4 sites in Nova Scotia (the sites are listed in the Acknowledgements). Six hs-cTn assays were assessed. Supplemental Table S1 lists the manufacturers, number of instruments, reagent, and calibrator lots. This substudy (a quality assurance study) was deemed exempt from further local research ethics board approval. The process of making human pools with various hs-cTn concentrations has been previously reported.4Kavsak P.A. Caruso N. Beattie J. Clark L. Centrifugation—an important pre-analytical factor for the Abbott Architect high-sensitivity cardiac troponin I assay.Clin Chim Acta. 2014; 436: 273-275Crossref PubMed Scopus (17) Google Scholar For accuracy testing, 3 frozen samples with different hs-cTn concentrations were sent to the respective laboratories (36 aliquots per instrument) with instructions on handling and testing4Kavsak P.A. Caruso N. Beattie J. Clark L. Centrifugation—an important pre-analytical factor for the Abbott Architect high-sensitivity cardiac troponin I assay.Clin Chim Acta. 2014; 436: 273-275Crossref PubMed Scopus (17) Google Scholar: level 1 (a normal concentration), level 2 (near the female 99th-percentile concentration), and level 3 (near the male 99th-percentile concentration). Briefly, level 3 was a plasma pool generated in 2014 that was frozen and retested in 2021 to confirm stability before sending to all laboratories (Supplemental Table S2 provides concentrations obtained in the reference laboratory, difference from listed 99th percentiles, stability claims, with serum and plasma suitable sample types for the assays). Level 2 was a serum pool generated in 2021 for the purpose of constructing a material at the female 99th-percentile concentration (ie, within 20% of 99th percentile), and level 1 was a plasma pool that yielded normal concentrations for all assays. The median concentration from all results for each hs-cTn assay at each level was calculated and set as the target concentration (Table 1). Two different approaches were used to assess analytic variation. First, the difference (absolute and relative) for each hs-cTn result from the target concentration for the specific assay and level was determined (approach 1). Second, either absolute (for target concentrations < 10 ng/L) and relative (for target concentrations ≥ 10 ng/L) differences were used (approach 2), as endorsed by an international task force for hs-cTn.1Wu A.H.B. Christenson R.H. Greene D.N. et al.Clinical laboratory practice recommendations for the use of cardiac troponin in acute coronary syndrome: expert opinion from the Academy of the American Association for Clinical Chemistry and the Task Force on Clinical Applications of Cardiac Bio-Markers of the International Federation of Clinical Chemistry and Laboratory Medicine.Clin Chem. 2018; 64: 645-655PubMed Google Scholar The Tukey test was then applied to identify outliers which were categorised as “far outside” (ie, smaller than the lower quartile − 3 × interquartile range [IQR] or larger than the upper quartile + 3 × IQR) vs “outside” (ie, ± 1.5 × IQR). The mean difference limits after outlier exclusion with 95% confidence intervals (CIs) were calculated (1000 bootstrapped data sets) for each assay at each level for approach 1 with the maximum lower difference and maximum upper difference limits (after outlier exclusion) for below and above 10 ng/L being identified by approach 2. For approach 2, secondary analyses were performed for Roche, Abbott, and Beckman (largest groups) with Ortho and Siemens differences then evaluated against these limits (Supplemental Table S3). We performed analyses (Tukey, median, IQR, mean, 95% CI) with the use of MedCalc Statistical Software version 20.217 (MedCalc Software, Ostend, Belgium), StatsDirect Statistical Software version 3.3.5 (StatsDirect, Birkenhead, UK), and R version 4.2 (R Foundation for Statistical Computing).Table 1The target concentrations and maximum observed variation from target concentrations for each assayCompanyLevelNo. of results;Sex-specific 99th percentile∗The sex-specific 99th percentiles were obtained from the International Federation of Clinical Chemistry and Laboratory Medicine website on February 28, 2023: https://www.ifcc.org/media/479435/high-sensitivity-cardiac-troponin-i-and-t-assay-analytical-characteristics-designated-by-manufacturer-v052022.pdf.Target, ng/LMean absolute difference limits from target, ng/L (95% CI)Mean relative difference limits from target [%] (95% CI)Abbott1n = 904.5± 1.5 (1.3-1.6)± 34.5 (28.9-35.6)2n = 97;Female: 16 ng/L12.6± 2.7 (1.9-3.0)± 21.1 (15.1-23.8)3n = 98Male: 34 ng/L29.7± 8.8 (7.6-9.2)± 29.6 (25.6-31.0)Beckman1n = 994.0± 2.2 (1.9-2.4)± 56.1 (49.8-60.0)2n = 99;Female: 12 ng/L11.5± 2.2 (2.0-2.2)± 18.7 (17.4-19.1)3n = 101;Male: 20 ng/L21.9± 3.9 (2.8-4.2)± 17.8 (12.7-19.2)Ortho1n = 91.3± 0.5 (0.1-0.8)± 41.0 (7.7-61.5)2n = 12;Female: 9 ng/L7.6± 1.3 (0.6-1.6)± 16.6 (7.9-21.1)3n = 12;Male: 13 ng/L19.7± 2.8 (0.8-3.3)± 13.8 (4.1-16.8)Roche1n = 4386.2± 3.8 (2.8-4.7)± 59.9 (45.6-75.2)2n = 436;Female: 9 ng/L8.5± 3.8 (3.0-4.8)± 44.4 (34.9-65.5)3n = 437;Male: 17 ng/L11.7± 3.6 (3.3-3.7)± 30.9 (28.2-31.4)Siemens Atellica1n = 474.9± 1.3 (1.1-1.4)± 26.3 (22.0-28.6)2n = 47;Female: 39 ng/L34.4± 5.4 (3.6-10.8)± 15.8 (10.4-31.5)3n = 48;Male: 54 ng/L40.9± 8.8 (8.1-8.9)± 21.4 (19.6-21.9)Siemens Vista1n = 248.9± 2.2 (1.5-2.3)± 24.4 (6.8-25.8)2n = 24;Female: 54 ng/L45.3± 8.2 (3.5-9.5)± 18.0 (6.8-21.0)3n = 24;Male: 79 ng/L60.1± 7.7 (4.6-8.3)± 12.8 (7.7-13.8)Roche is a high-sensitivity cardiac troponin (hs-cTn) T assay and the other companies’ assays are hs-cTnI. For level 1 there were 3 results that were undetectable for the Ortho hs-cTnI assay (ie, < 1.0 ng/L), therefore resulting in only 9 values, and 1 result for the Roche hs-cTnT assay that was undetectable (ie, < 3.0 ng/L). These results were not included in the analyses.∗ The sex-specific 99th percentiles were obtained from the International Federation of Clinical Chemistry and Laboratory Medicine website on February 28, 2023: https://www.ifcc.org/media/479435/high-sensitivity-cardiac-troponin-i-and-t-assay-analytical-characteristics-designated-by-manufacturer-v052022.pdf. Open table in a new tab Roche is a high-sensitivity cardiac troponin (hs-cTn) T assay and the other companies’ assays are hs-cTnI. For level 1 there were 3 results that were undetectable for the Ortho hs-cTnI assay (ie, < 1.0 ng/L), therefore resulting in only 9 values, and 1 result for the Roche hs-cTnT assay that was undetectable (ie, < 3.0 ng/L). These results were not included in the analyses. The target concentrations for levels 1, 2, and 3 for each assay with the corresponding number of results are presented in Table 1. At normal hs-cTn concentrations (ie, level 1) the maximum absolute difference limits ranged from ± 0.5 ng/L (Ortho, n = 9) to ± 3.8 ng/L (Roche, n = 438) with the relative limits ranging from ± 24.4% (Siemens Vista, n = 24) to ± 59.9% (Roche, n = 438). Near the female 99th-percentile concentrations (level 2), the absolute difference limits ranged from ± 1.3 (Ortho, n = 12) to ± 8.2 ng/L (Siemens Vista, n = 24) with the relative limits ranging from ± 15.8% (Siemens Atellica, n = 47) to ± 44.4% (Roche, n = 436). For level 3 (near the male 99th-percentile concentrations), the absolute difference limits ranged from ± 2.8 (Ortho, n = 12) to ± 8.8 ng/L (Siemens Atellica, n = 48; Abbott, n = 98) with the relative limits ranging from ± 12.8% (Siemens Vista, n = 24) to ± 30.9% (Roche, n = 437). Abbott yielded tighter absolute and relative difference limits for level 1 compared with Beckman, with the opposite effect being observed for level 3. The second approach yielded a near split in the data set with 1155 values in the < 10 ng/L group (54%) and 987 values (46%) in the ≥ 10 ng/L group. After removing the far outside values for the < 10 ng/L group (0.5% of data were outliers), the maximum limits were −3.2 ng/L to 3.4 ng/L (Figure 1). After removing the far outside percent difference values for the ≥ 10 ng/L group (0.7% were outliers), the maximum limits were −31% to 31%. In total, 13 (5 Abbott and 8 Roche) results from 2142 results (0.6%, 95% CI 0.3%-1.0%) were classified as far outside concentrations. When all outside concentrations were removed, the limits were −2.4 ng/L to 2.4 ng/L for < 10 ng/L (2.7% outliers) and −20% to 20% for ≥ 10 ng/L (3.2% outliers), with 63 results (1 Siemens Vista, 3 Siemens Atellica, 12 Abbott, and 47 Roche) from 2142 (2.9%, 95% CI 2.3%-3.8%) being classified as outliers. The limits in approach 2 were similar if derived by Roche, Abbott, Beckman, with 1 result in the ≥ 10 ng/L category from the Siemens and Ortho group being an outlier (1 in 155) (Supplemental Table S3). The data from this multicentre study indicate that as hs-cTn concentrations increase from normal to the 99th percentiles, so too does the absolute variation in observed results from the target concentration. The results are assay dependent. However, no obvious relative (proportional) limit can be applied across all levels and all assays. Rather, the data suggest that at lower concentrations (below the 99th percentile), the maximum difference in absolute terms is somewhat dependent on the assay and number of results used to generate the target concentration, as evident in the comparison of the Roche data (n = 438) vs Ortho data (n = 9) at level 1. It is a common laboratory practice for the same analyte (ie, cTn) to have the same analytic variation limit for specific concentration ranges, and by dichotomizing hs-cTn concentrations below and above 10 ng/L, practical absolute and relative variation limits can be obtained. The exclusion of all outside values in this study (2.9% or 1 in 34 results being outliers) yielded acceptable limits of ± 2 ng/L for hs-cTn concentrations below 10 ng/L and ± 20% for concentrations above. The exclusion of far outside values (0.6%, or 1 in 167, of the results being outliers) yields acceptable limits of ± 3 ng/L for hs-cTn concentrations below 10 ng/L and ± 30% at higher concentrations up to 60 ng/L (or around the 99th percentiles). The latter approach yields an analytic error prevalence of approximately 1% (with these outliers likely be detected clinically as outlier events if a third hs-cTn measurement was performed) and concurs with other clinical safety estimates used for cTn (ie, ≤ 1% miss rate for major adverse cardiac events). Regardless of what limits are used, it does not obviate the need for clinicians to question hs-cTn results that do not align with the clinical status of the patient or the importance of “outlier” results. The study did not assess accuracy at higher target concentrations (ie, multiples of the 99th percentile).5Devereaux P.J. Lamy A. Chan M.T.V. et al.VISION Cardiac Surgery Investigators. High-sensitivity troponin I after cardiac surgery and 30-day mortality.N Engl J Med. 2022; 386: 827-836Crossref PubMed Scopus (33) Google Scholar Also, despite the same instructions being supplied for each site regarding sample handling, we cannot rule out pre-analytic variables also contributing to the observed variation. However, important for these analyses is that during the 12 months of testing, no laboratory reported any errors in testing that would affect the results. In fact, an argument could be made that by participating in the substudy and receiving notifications (ie, monthly reminders and graphic reports) the analytic variation reported here may be smaller than what would be observed in a routine 24/7 laboratory service. Finally, the majority of the results were with hs-cTnT and the majority of the outliers were on instrumentation with older technology. Caution is warranted in applying assay-specific limits (especially for Ortho and Siemens) and at higher concentrations not assessed for the specific assays in this study. This large prospective study provides robust limits for acceptable variation or the maximum difference in hs-cTn measurements on the same sample, with a pragmatic approach for allowable variation being ± 3 ng/L for normal concentrations and ± 30% at concentrations around the 99th percentile. There is substantial variation across various assays. Tighter goals of ± 2 ng/L or ± 20% may require more intervention and effort by the laboratory community and manufacturers to yield analytic error rates ≤ 1%. These results may guide clinicians in their interpretation of hs-cTn levels around the 99th-percentile concentration.