Results
The electronic searches identified 3071 records, of which 141 full text articles were assessed for eligibility. Thirty nine of them met the criteria for inclusion in the review, and one additional paper was included from the hand search. Twenty studies (23 papers) were included in the meta-analyses reported here (Fig 1).Table 1 shows their main characteristics, and additional details are provided in Supplementary Tables A , Table B , Table C .
(Enlarge Image)
Figure 1.
Flow chart of selection process. hs-cTnT=high sensitivity cardiac troponin T; ROC=receiver operating characteristics
Two of the 23 included papers reported results from the ongoing multicentre study Advantageous Predictors of Acute Coronary Syndromes Evaluation (APACE), and four reported results from two other studies. The total number of patients in the included studies was 9428, ranging from 13728 to 2072 (median 350, interquartile range 221-491). The reported mean or median age of the included patients ranged from 5429 to 71 years (with the exception of the study by Bahrmann et al, which included only patients aged 70 years or over), and the proportion of men ranged from 49% to 83%. One study included only patients with coronary artery disease, and another included unselected patients aged 70 or over presenting to the emergency department with a non-surgical condition. Ten studies defined specific time from onset of symptoms to presentation as an inclusion criterion, which ranged from four hours to 24 hours. Most of the patients presented to the emergency department within 12 hours of symptom onset, with study medians ranging from 3.5 hours to 6.3 hours, but the average time was reported inconsistently. In 11 papers, the results for non-ST elevation myocardial infarction were reported separately or patients with ST elevation on the initial electrocardiogram were excluded from the study. The median prevalence of acute myocardial infarction was 21.4% (interquartile range 13.3-34.7%) and ranged from 8.0%29 to 56.2%.
All included studies used a composite reference standard based on the contemporary universal definition of myocardial infarction. In terms of reference assays used to diagnose myocardial necrosis, eight studies used serial high sensitivity troponin T assay; 13 studies used standard troponin T or I assays, or a combination of them; one study used a combination of either standard troponin T or I (local assays) and high sensitivity troponin I assays (central laboratory); one study used a combination of standard and high sensitivity troponin T assays; and the reference assay in one study was unclear (in some studies the accuracies according to different reference assays were reported separately).
Twenty studies reported the performance of high sensitivity troponin T at the manufacturer’s recommended cut-off value of 14 ng/L, which represents the 99th centile of a healthy reference population; four studies reported the performance of the test at 3 ng/L (limit of blank) and four at 5 ng/L (limit of detection); results for receiver operating characteristics optimised or other cut-off values were also reported in some papers ( Table 1 and Supplementary Table C ).
Table 2 and Supplementary Table D show the results from the assessment of the methodological quality of the included studies. In approximately half of the studies, patients with ST elevation myocardial infarction were not excluded. As cardiac markers play no role in the diagnosis of this condition, which is made primarily on the results from the electrocardiogram, including patients with this diagnosis may compromise the applicability of the results. The use of high sensitivity troponin T as part of the reference standard may lead to incorporation bias, thus inflating the accuracy estimates, whereas using a standard troponin assay as a reference test may result in patients with minor myocardial infarctions being misclassified as false positives. We investigated the effect of using different generations of reference assays in the meta-regression.
To obtain clinically relevant estimates of the performance of a single baseline measurement of the high sensitivity troponin T assay, we conducted, as far as the data permitted, separate meta-analyses for the different pre-specified cut-off values reported in the papers. The results from these meta-analyses are presented below.
Performance of Assay At 14 ng/L Cut-Off Value
We pooled the results from 20 studies to obtain summary estimates of the sensitivity and specificity at the 14 ng/L cut-off value. When a study reported separately the results for non-ST elevation myocardial infarction and acute myocardial infarction (that is, both patients with ST and non-ST elevation myocardial infarction were included in the study cohort), we included only those for non-ST elevation myocardial infarction, which are clinically more relevant. In a similar way, when both standard troponin and high sensitivity troponin assays were used as reference tests, we included only the results obtained with high sensitivity assays because, being more sensitive, they are able to identify patients with small myocardial infarctions that would be missed by the standard assays. The target condition was acute myocardial infarction in 10 studies and non-ST elevation myocardial infarction in the remaining 10 studies; the reference test was a standard troponin assay in nine studies, high sensitivity troponin assay in eight, either standard or high sensitivity in one, both standard and high sensitivity in one, and unclear in one study. Figure 2 shows a forest plot of the coupled sensitivity and specificity with 95% confidence intervals for each study included in this meta-analysis.
(Enlarge Image)
Figure 2.
Forest plots of coupled sensitivity and specificity at 14 ng/L (99th centile). FN=false negative; FP=false positive; TN=true negative; TP=true positive. See footnote to Table 1 for other abbreviations
Pooling the results produced the following summary estimates: sensitivity 89.5% (95% confidence interval 86.3% to 92.1%), specificity 77.1% (68.7% to 83.7%), positive likelihood ratio 3.9 (2.8 to 5.4), and negative likelihood ratio 0.14 (0.10 to 0.18). The summary receiver operating characteristics plot (Fig 3) shows the summary sensitivity and specificity (the solid blue spot in the middle) and the 95% confidence and prediction regions (the inner and outer ellipses, respectively).
(Enlarge Image)
Figure 3.
Summary receiver operating characteristics plot of sensitivity and specificity at 14 ng/L cut-off value. Each rectangle represents an individual study; size of symbol reflects inverse standard error of specificity (width) and sensitivity (height) estimates; solid spot in middle is summary sensitivity and specificity; inner ellipse represents 95% confidence region, and outer ellipse represents 95% prediction region
As shown in Figures 2 and 3, a significant level of heterogeneity was apparent in the results, greater in specificity than in sensitivity. We investigated the effect of the target condition (acute myocardial infarction versus non-ST elevation myocardial infarction) and the reference test (standard versus high sensitivity troponin assay) on the summary estimates of sensitivity and specificity by adding them as covariates to a bivariate regression model (one covariate at a time) and used a likelihood ratio test to determine the statistical significance of the results. As in two studies a combination of standard and high sensitivity assays were used as a reference test, and the type of the reference assay was unclear in another study, we excluded those three studies from the meta-regression. Without them, the likelihood ratio test showed that the use of different reference standards accounts for some of the variability in the sensitivity (P=0.008) but not in the specificity (P=0.66) (Fig 4). A model that allowed for sensitivity and its variance to vary between studies using different reference tests (standard versus high sensitivity assays) produced the following summary estimates: sensitivity (standard reference assay) 87.7% (82.6% to 89.9%), sensitivity (high sensitivity reference assay) 93.4% (89.8% to 95.7%), specificity 74.7% (73.6% to 75.8%), positive likelihood ratio (standard reference assay) 3.43 (3.08 to 3.82), positive likelihood ratio (high sensitivity reference assay) 3.69 (3.13 to 4.36), negative likelihood ratio (standard reference assay) 0.18 (0.14 to 0.23), negative likelihood ratio (high sensitivity reference assay) 0.09 (0.06 to 0.14). The target condition, on the other hand, had no effect on the results (P=0.79).
(Enlarge Image)
Figure 4.
Summary receiver operating characteristics plot comparing effect of different reference tests (standard troponin assay v high sensitivity troponin assay) on summary estimates of sensitivity and specificity (Inoue 2011, Lotze 2011, and Collinson 2013 were
On the basis of the Cook's distance, we found the following studies to be the most influential in the meta-analysis (in descending order): Khan et al, Melki et al, Invernizzi et al, and Collinson et al (Fig 5). Of these, only Khan et al was identified as an outlier having the highest standardised residuals for specificity (Fig 6). After refitting the model and leaving this study out, we observed no change in sensitivity (89.5% v 89.7%) but specificity decreased from 77.1% to 74.9%. This could be explained by the fact that this study excluded patients without coronary artery disease, thus reducing the probability of false positive results.
(Enlarge Image)
Figure 5.
Influence analysis
(Enlarge Image)
Figure 6.
Outlier detection
Performance of Assay When Either 3 ng/L (Limit of Blank) or 5 ng/L (Limit of Detection) Was Used as Cut-off Value
Seven papers reported the results for 3 ng/L and/or 5 ng/L cut-off values at presentation. Given the small number of studies, pooling the data for each cut-off value separately would have produced unreliable results. Instead, we decided to obtain more precise and reliable summary estimates by including in the meta-analysis all independent 3 ng/L and 5 ng/L data. Two studies reported the results for both cut-off values. As in this analysis we were interested mainly in the sensitivity of the test (its accuracy for ruling out acute myocardial infarction), we decided to include the results for 5 ng/L as the performance at a higher cut-off value would produce a lower sensitivity estimate thus representing the worse case scenario. Owing to the inverse correlation between sensitivity and specificity, we could assume that using even lower cut-off values would further increase the sensitivity of the assay and its ability to rule out the target condition. Thus, from the APACE trial we excluded the results reported by Meune et al, included those reported by Rubini Gimenez et al, and included only the 5 ng/L data reported by Aldous et al. Also, Christ et al reported two different sets of 3 ng/L results, obtained using standard troponin T and high sensitivity troponin T as refrence assays. As in the previous analysis, we included the results obtained by using high sensitivity troponin T as a reference assay, which is more sensitive and, thefeore, more likely to capture small myocardial infarctions. Figure 7 shows a forest plot of the sensitivities and specificities of the included studies. Pooling the results from the six studies produced the following summary estimates: sensitivity 97.4% (94.9% to 98.7%), specificity 42.4% (31.2% to 54.5%), positive likelihood ratio 1.69 (1.40 to 2.05), and negative likelihood ratio 0.06 (0.04 to 0.10) (Fig 8).
(Enlarge Image)
Figure 7.
Forest plot of studies included in meta-analysis of combined 3 ng/L and 5 ng/L. FN=false negative; FP=false positive; TN=true negative; TP=true positive. See footnote to Table 1 for other abbreviations
(Enlarge Image)
Figure 8
Summary receiver operating characteristics plot of sensitivity and specificity for cut-off value of either 3 ng/L or 5 ng/L