During the period from August 2005 to January 2007 in three urban teaching hospitals, we prospectively enrolled consecutive hospital outpatients (> 18 years of age) who presented to the ED with chest pain suggestive of ACS with the onset or peak occurring within the previous 6 hours. Patients with acute or chronic kidney failure requiring dialysis were excluded. The study was performed according to the principles of the Declaration of Helsinki and approved by the local ethics committee (Comité de Protection des Personnes Ile-de-France VI, CHU Pitié-Salpétrière Hospital, Paris, France). Because routine medical care was unchanged, waiver of informed consent was authorised. We followed most of the recommendations concerning the reporting of diagnostic studies set forth by the Standards for Reporting of Diagnostic Accuracy initiative 16 .

As part of the routine assessment in our institutions, all patients underwent an initial clinical evaluation that included clinical history, a physical examination, 12-lead electrocardiography (ECG), pulse oximetry, routine blood tests and chest X-rays. After these routine tests were done, and before cardiac biomarker results were available, ED physicians were asked to offer an 'empirical' clinical probability of AMI (low, medium or high PTP) based on cardiovascular risk factors, type of chest pain, physical findings and electrocardiogram abnormalities 17 18 . Conventional cardiac troponin I (cTnI) was measured at presentation and, if needed, was repeated after 3 to 9 hours as long as it was clinically indicated. Thus, according to the diagnosis of non-ST elevation MI (NSTEMI) or ST elevation MI (STEMI), the patients were admitted either to the cardiology unit for further evaluation and treatment or directly to the catheterization laboratory for primary percutaneous coronary intervention. However, the timing and treatment of patients were left to the discretion of the attending physicians according to the suspected diagnosis. ED physicians in charge were blinded to the results of HsTnT, and biologists were blinded to the emergency diagnosis suspected by physicians.

To determine the etiologic diagnosis of chest pain at presentation for each patient, two independent experts (ED physicians) who were blinded to the results of HsTnT reviewed all available medical records (including patient history, physical findings, results of laboratory and radiologic testing, ECG, echocardiography, cardiac exercise test, coronary angiography and summary chart at discharge) pertaining to the patient from the time of ED presentation to 30-day follow-up. In the event of diagnostic disagreement, cases were reviewed and adjudicated in conjunction with a third expert (also an ED physician).

AMI was diagnosed according to the joint European Society of Cardiology/American College of Cardiology/American Heart Association/World Heart Federation Task Force redefinition of MI guidelines 6 . Diagnosis of AMI required a cTnI increase above the 10% coefficient of variation (CV) value associated with at least one of the following: symptoms of ischaemia, new ST-T changes or a new Q wave on an electrocardiogram, imaging of new loss of viable myocardium or normal cTnI on admission. Unstable angina was diagnosed in patients with constant normal cTnI levels and a history or clinical symptoms consistent with ACS. Predefined further diagnostic categories included AMI (STEMI with the presence of ST-segment elevation in at least two continuous leads on ECG, new onset of left bundle branch block or NSTEMI), unstable angina, and a third group including cardiac but not coronary symptoms (for example, stable angina, myocarditis, arrhythmias and heart failure), noncardiac symptoms (for example, pulmonary embolism) and chest pain of unknown origin.

To assess the influence of renal function on cTn measurement accuracy, the creatinine level was measured in each patient and then renal function was estimated using the Modification of Diet in Renal Disease study equation 19 .

In two EDs (Cochin Hospital and La Pitié Salpêtrière Hospital, Paris, France), plasmatic cTnI concentrations were routinely measured on an Xpand HM analyzer using the Cardiac Troponin I one-step enzyme immunoassay system (Siemens Healthcare Diagnostics Inc., Newark, NJ, USA). The measurement range extended from 0.04 to 40.00 μg/L. The threshold for this method (0.14 μg/L) corresponds to the lowest substrate concentration that can be reproducibly measured with a CV ≤ 10%. In the remaining ED (Bicêtre Hospital, Le Kremlin-Bicêtre, France), plasmatic cTnI concentrations were routinely measured on an Access analyser (Beckman Coulter, Inc., Brea, CA, USA). The measurement range of this one-step chemiluminescence immunoassay extends from 0.01 to 100.00 μg/L. The threshold (10% CV) given by the manufacturer is 0.06 μg/L.

Heparinised samples collected upon admission and, if available, samples collected 3 to 9 hours later were analysed. Plasmatic highly sensitive cardiac TnT (HScTnT) concentrations were measured using the HScTnT one-step electrochemiluminescence immunoassay on an Elecsys 2010 analyzer (Roche Diagnostics, Meylan, France). The measuring range extended from 0.003 to 10 μg/L. The threshold for this method is 0.014 μg/L and corresponds to the 99th percentile. The CV was found to be < 10% at 0.014 μg/L. In our laboratory, CVs obtained in Roche Diagnostics quality controls containing 0.027 and 2.360 μg/L of HScTnT were < 4%. These analytical performance levels were in accordance with data provided by the manufacturer.

Continuous variables are presented as means ± SD or medians (25th to 75th percentile), and categorical variables are expressed as numbers and percentages. Continuous variables were compared by using the Mann-Whitney U test, and categorical variables were assessed using Pearson's χ² test. Correlations among continuous variables were assessed using the Spearman's rank correlation coefficient. Receiver operating characteristic (ROC) curves were constructed to assess the sensitivity and specificity, positive predictive value (PPV) and negative predictive value (NPV), positive likelihood ratio (LR⁺) and negative likelihood ratio (LR^-) (all data presented with their 95% confidence intervals (95% CIs)) throughout the concentrations of cTnI and HScTnT to compare the accuracy of these markers in the diagnosis of AMI. Comparison of areas under the ROC curve was performed 20 . As this comparison is recognised as potentially insensitive, the net reclassification index (NRI) method was used as recently described 21 . For tests with binary outcomes (such as cTn for the diagnosis of AMI), NRI is defined as the gain in certainty of the first test (cTnI) minus the gain in certainty of the second test (HScTnT) or, alternatively stated, the difference of the sum of the sensitivity and specificity expressed as follows:

NRI is the combination of four components: the proportion of individuals with events who move up or down in a category and the proportion of individuals with nonevents who move up or down in a category. Table 1 is a contingency table comparing diagnostic classifications according to cTnI and HsTnT, with shifts between the two classifications, to represent the possible benefit of HScTnT in terms of the number of patients correctly reclassified. As stated in the Routine assessment subsection above, we separated the study population into two groups: one included the patients assessed as having low or moderate PTP of AMI and the other assessed as having high PTP of AMI.

All hypothesis testing was two-tailed, and P < 0.05 was considered statistically significant. Statistical analysis was performed using StatView for Windows version 5.0 software (SAS Institute, Cary, NC, USA) and MedCalc software for ROC analysis (MedCalc Software, Mariarkerke, Belgium). Graphs were built with GraphPad Prism 5 software (GraphPad Software Inc., La Jolla, CA, USA).

The area under the ROC curve (AUC) for the diagnosis of AMI was 0.940 (95% Confidence Intervall 0.901 to 0.980) (P < 0.001) for initial cTnI compared to 0.926 (0.881 to 0.971) (P < 0.001) for HsTnT. However, there was no significant difference between AUCs (Figure 1). ROC analysis indicated an optimal threshold of HsTnT for the diagnosis of AMI at 0.014 μg/L, with a high sensitivity of 89% (78 to 98) and a high specificity of 82% (78 to 87). The overall diagnostic accuracy of HsTnT was not significantly different compared to that of cTnI, regardless of PTP. Similar results (data not shown) were observed when we considered only NSTEMI patients (that is, after exclusion of the 13 STEMI patients). For the diagnosis of AMI, the sensitivities of HsTnT were higher and the specificities were lower than those of cTnI, regardless of PTP (Table 3). When we assessed the low and moderate PTP populations, the sensitivity of HsTnT was higher (91% (79 to 100) vs. 77% (60 to 95)) but NPV was not (99% (96 to 100) vs. 98% (95 to 99) for cTnI).

Table 3 shows patient classification on the basis of using cTnI or HsTnT to diagnose AMI and highlights the shifts between the two classifications.

Patients were classified into tertiles: tertile 1 (estimated glomerular filtration rate (eGFR) < 67.2 ml^-1 minute^-1 1.73 m^-2), tertile 2 (eGFR from 67.2 to 86.8 ml^-1 minute^-1 1.73 m^-2) and tertile 3 (eGFR ≥ 86.9 ml^-1 minute^-1 1.73 m^-2). Cardiac TnI levels were not significantly different across tertiles. However, HsTnT increased significantly across tertiles (P < 0.001): the lower the eGFR, the higher the HsTnT value. However, in each eGFR tertile, cTnI and HsTnT levels remained significantly different between AMI and no AMI (P < 0.001 for both) (Figure 2). We found no significant differences in the AUCs of cTnI and HsTnT regarding eGFR tertiles, and the optimal threshold value of cTnI did not change across tertiles. Conversely, the optimal threshold value of HsTnT increased only in tertile 1 (0.036 μg/L compared to 0.014 μg/L).

The main limitation of our study is the small sample of patients, especially patients with AMI. We cannot exclude the possibility that better results might have been found with a larger sample. Our sample is comparable to those used in previous studies, however, and most of all, we believe that the imperfect NPV that we describe herein is the major result of our study, which could not have been corrected by including more patients.

Our study has some other limitations. First, we performed only a single measurement of HsTnT. We did not evaluate its kinetics, which would have been interesting, especially in the 'grey zone' (between 0.014 μg/L and 0.050 μg/L). A second value could have provided more data, as previously described in the Giannitsis et al. study 27 , which reported that a doubling in the HsTnT concentration within 3 hours of chest pain (with first negative HsTnT and no electrocardiogram abnormality) was associated with a 100% PPV of a diagnosis of NSTEMI.

Second, we used empirical PTP and not a standardised, validated one 17 18 . However, outcomes in the low and moderate PTP population (only nine with confirmed NSTEMI), and differences in clinical characteristics at admission suggested that even though empirical, this evaluation by the clinician was accurate. Furthermore, one of the strengths of our study was that it evaluated differences in diagnostic performance for the HsTnT regarding PTP as demonstrated for D-dimers and empirical suspicion of pulmonary embolism 28 . Another limitation of our study is that different conventional Tn assays have been used at the two study sites with different threshold values and CVs. These assays were used because they were both local and well-understood methods at the time of the study.

Third, we used two different assays for the comparator (that is, conventional TnI): a Siemens cTnI assay in two centres (CCH and PSL) and a Beckman Coulter assay in the third centre (BCT). The ROC curve for the cTnI is, then, a combined ROC curve of two different assays, making it imprecise. However, the two different ROC curves (for each assay) have similar AUCs.

Last, this study was underpowered to find any significant change in the detection of AMI in the low to moderate PTP patients. However, as the NPV is not perfect in our patient population, we expect that this would remain the case with a larger sample.