a b s t r a c t

Objectives: Cardiovascular disease is the leading cause of mortality and morbidity. Serial troponin tests have been endorsed as essential diagnostic steps to rule out/-in acute myocardial infarction (AMI), and hs-cTn assays have shown promise in enhancing the accuracy and efficiency of AMI diagnosis in the emergency department (ED). Methods: A systematic review and meta-analysis of diagnostic test accuracy studies were conducted to compare the diagnostic performance of various accelerated diagnostic algorithms of hs-cTn assays for patients with symp- toms of AMI. Random-effects bivariate meta-analysis was conducted to estimate the summary sensitivity, spec- ificity, likelihood ratios, and area under receiver operating characteristic curve.

Results: In the systematic review consisting of 56 studies and 67,945 patients, both hs-cTnT and hs-cTnI-based 0-, 1-, 2- and 0-1 h algorithms showed a pooled sensitivity N90%. The hs-cTnI-based algorithm showed a pooled specificity N80%. The hs-cTnT-based algorithms had a specificity of 68% for the 0-h algorithm and of around 80% for the 1-, 2-, and 0-1 h algorithms. The heterogeneities of all diagnostic algorithms were mild (I² b 50%). Conclusion: Both hs-cTnI- and hs-cTnT-based accelerated diagnostic algorithms have high sensitivities but mod- erate specificities for early diagnosis of AMI. Overall, hs-cTnI-based algorithms have slightly higher specificities in early diagnosis of AMI. For patients presenting ED with Typical symptoms, the use of hs-cTnT or hs-cTnI assays at the 99th percentile may help identify patients with low risk for AMI and promote early discharge from the ED.

(C) 2019

Introduction

Cardiovascular disease is the leading cause of mortality and morbid- ity worldwide. In 2013, an estimated 8.14 million people died from acute myocardial infarction (AMI) globally [1-3]. The emergency de- partment (ED) is the main portal of entry for patients with acute chest pain. Approximately 10% of patient visits to the ED are due to chest pain [4], where b10% of the ED patient with chest pain are eventually di- agnosed with AMI [5,6]. The survival and clinical outcome of patients with ST-elevation myocardial infarction rely on a rapid and ac- curate diagnosis to initiate effective, evidence-based medical manage- ment and revascularization [7,8]. Serial troponin tests have been

* Corresponding author at: Health Data Science Research Group, National Taiwan University Hospital, Taipei, Taiwan.

E-mail address: [email protected] (C.-C. Lee).

¹ Equal contribution to this work.

endorsed by American and European guidelines as essential diagnostic steps to rule-out/-in AMI [9-12]. Nevertheless, the exclusion of MI among suspected patients still requires two negative contemporary tro- ponin tests at a 3-6 h interval. MI rule-out delay may result in ED over- crowding while rule-in delay may affect patients’ outcomes [13,14]. ED crowding has become a critical health quality and patient safety issue globally. Rapid rule-out of AMI may avoid anxiety, unnecessary testing or admissions, and alleviate ED crowding.

The introduction of high-sensitivity cardiac troponin (hs-cTn) assays in 2010 has shown tremendous promise in enhancing the accuracy and efficiency of MI diagnosis in the ED [15]. These assays can detect tropo- nins at a level 10- to 100-fold lower than older-generation assays [16,17]. Several accelerated diagnosis algorithms, such as 0-, 1-, or 2-h algorithms, have been developed and implemented based on the highly sensitive detection of new generation troponins [17]. Diagnostic perfor- mance, however, has varied, and cutoffs and methods are heteroge- neous across the literature [16,18]. Previously, a meta-analysis showed

https://doi.org/10.1016/j.ajem.2019.11.035

0735-6757/(C) 2019

a high sensitivity and negative predictive value in ruling out AMI by using a single hs-cTnT measurement below the limit of detection (LOD) combined with a non-ischemic ECG [19]. Nevertheless, the au- thors did not determine the efficacy of hs-cTnI algorithms and the dis- parities across different accelerated diagnosis algorithms. In this study, we performed a systematic review and meta-analysis to investigate the ability of different hs-cTn accelerated algorithms to quickly rule- out or rule-in MI. We also explored the diagnostic accuracy sub- groups by clinical settings and cutoffs [20-26].

Methods

This meta-analysis was performed in accordance with the Preferred Reporting Items for a Systematic Review and Meta-analysis of Diagnos- tic Test Accuracy Studies (PRISMA-DTA) Statement [27]. The pre- planned protocol of this meta-analysis was registered in PROSPERO (CRD42018090832). This study was waived from ethical committee re- view at the study institution.

Search strategy and selection criteria.

We conducted electronic search in Medline and Embase for relevant studies published from inception to May 2018 using keywords such as high sensitivity troponin, high sensitivity troponin-I, or high sensitivity troponin-T. To enhance the sensitivity of the search, we did not include AMI-related terms. We also manually searched the reference list of in- cluded articles for other relevant studies. There were no restrictions on language. Two reviewers screened the titles and abstracts indepen- dently to identify the studies that examined the diagnostic value of hs-cTn assays for diagnosis of AMI. Eligible studies were those that de- fined acute coronary syndrome or myocardial infarction as the end- points and estimated the sensitivity, specificity, or odds ratio or provided sufficient data to construct a 2 x 2 contingency table. Only the study with the largest sample size was included if there were mul- tiple studies derived from the same cohort with overlapping study pe- riods. We excluded case reports, case series, review articles, editorials, and clinical guidelines. Any discrepancies were resolved by consensus.

Data extraction

Potentially relevant articles were retrieved for full-text review. Two authors independently performed study selection and data extraction. The extracted variables included study size, diagnosis algorithm, diagnosis marker, cutoffs for the hs-cTn, AMI prevalence, sensitivity, specificity, positive predictive value, negative predictive value, study setting, and presenting time. Presenting time was defined as the time from the onset of chest pain to ED presentation.

Quality assessment

The quality of the selected studies was appraised independently by two reviewers using Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) [28]. QUADAS-2 is a standardized tool for risk of Bias assessment with four domains: patient selection, index test, ref- erence standard, and flow and timing. The first three domains are also used for assessment of applicability of clinical practice. Discrepancies between the reviewers were resolved by a consensus discussion with a third author.

Data synthesis and analysis

The summary estimates of sensitivity, specificity, and likelihood ratio for the included studies were derived using the bivariate model for diagnostic meta-analysis. A bivariate model was employed to sum- marize the results from individual diagnostic test studies while keeping the two-dimensionality of the data. This bivariate approach assumes a bivariate distribution for the logit transformed sensitivity and specific- ity. Pairs of sensitivity and specificity were jointly analyzed in the

bivariate model, and the inherent correlation between them was assessed by a random effects approach. The goodness-of fit of model and the bivariate normality assumption were assessed by the quantile plot. We also constructed a hierarchical summary receiver operating characteristic curve (ROC) plotting sensitivity versus specificity and es- timated the area under Receiver operating characteristic curve to summarize the true- and false-positive rates for the hs-cTn ability to predict the outcomes, irrespective of the different cutoff points used in various studies. We plotted the 95% confidence regions for the summary points. We also plotted the 95% prediction region, which represents the 95% confidence for the diagnostic accuracy if a future study assuming that the model is true [29]. For any zero observations in the 2 x 2 con- tingency tables, 0.5 was added to reduce bias. We evaluated the degree of between-study heterogeneity using the quantity I². To search for po- tential sources of heterogeneity, we used the spike plot to identify influ- ential observations using the Cook’s distance [30] and a scatter plot to check the outliers that are greater than two times standardized resid- uals. We tested for publication bias by using the Deek’s test [31]. When the publication bias test was positive, we used the trim-and-fill method to identify potential missing studies and re-compute the effect estimate. A secondary analysis was conducted a priori, restricted to studies evaluating the diagnostic accuracy hs-cTnT 0-h algorithms, to examine the following sources of heterogeneity: research settings, cut- off values, outcome definitions, prevalence of MI, and presenting time. All analyses were performed using STATA version 11.0 (Stata Corp, Col- lege Station, TX, USA). All statistical tests were two-tailed. Statistical sig- nificance was defined as a p-value b0.05.

Results

A flowchart of inclusion and exclusion processes is shown in Fig. 1. A total of 6250 records were identified in Medline, Embase, and by manual search. After initial screening, 146 studies were re- trieved for full text review. Finally, 56 studies were included for data extraction and meta-analysis. As there were six articles investi- gating the diagnostic accuracy of both the hs-cTnT and the hs-cTnI, 40 and 22 articles for hs-cTnT and hs-cTnI data, respectively, were included.

The 56 articles included in this study applied Universal Definition of Myocardial Infarction as the gold standard in diagnosing AMI [11,32- 34]. Of note, the choice of diagnostic marker was heterogeneous. There were 31 studies that used contemporary troponin assays as the gold standard while 25 studies used hs-cTn assays.

Characteristics and quality of the included studies

eTables 1 and 2 summarize the study-level characteristics in reports that assessed the diagnostic accuracy of the hs-cTnT and the hs-cTnI. A total of 28,283 (4615 AMI) patients receiving hs-cTnT assays and 39,662 (5213 AMI) patients receiving hs-cTnI assays were included. Most of the studies (hs-cTnT: 33 of 40; hs-cTnI: 20 of 22) were con- ducted in the ED setting. There were 32, 6, and 8 studies that evaluated the diagnostic value of the hs-cTnT at initial presentation (0 h), 1, and 2 h, respectively, while there were 14, 5, and 5 for hs-cTnI-based accel- erated diagnostic algorithms. Fig. 2 shows the risk of bias and applicabil- ity of included studies. Most studies reported blinding of the standard reference. All studies were recorded as low risk of bias in the flow and timing section. Fifteen studies were recorded as having high risk of bias in the reference test because they did not adopt the current guide- line to assess the outcomes. Upon the evaluation of applicability, most studies were recorded as low risk of bias in adherence to the reference standard and index test. Ten studies were scored as high risk of bias in the patient selection section for not performing patient selection within 24 h.

Fig. 1. PRISMA flowchart of study selection. There are 6 studies including both hs-cTnT and hs-cTnI researches in the same research.

Diagnostic accuracy of hs-cTnT and hs-cTnI in different algorithms Hs- cTnT assay

There were 32 studies that investigated the diagnostic accuracy of hs-cTnT in a 0-h diagnosis algorithm with a total of 20,811 patients with a 17% prevalence of AMI. The pooled sensitivity and specificity of hs-cTnT at initial presentation (0-h algorithm) were 93% (95% CI: 89-95%) and 68% (95% CI: 61-75%) (Table 1, upper panel). The 1- and

2-h algorithm subgroups had similar sensitivity but higher specificity (84% [95% CI: 75-90%] and 79% [95% CI: 77-81%]) compared with the 0-h algorithm (68% [95% CI: 60-75%]). Being paralleled with the in- creased specificity for three 1- and 2-h algorithms, the AUROC increased from 0.89 (95% CI: 0.86-0.92) for the 0-h algorithm to 0.95 (95% CI:

0.92-0.96) for the 1-h algorithm and 0.90 (95% CI: 0.87-0.92) for the 2-h algorithm. However, it is important to note that the 95% confidence intervals of the AUROC for the 2-h algorithm overlap with those of the

Fig. 2. Summary of methodological quality of included studies by QUADAS-2 checklist.

Table 1

Summary of pooled accuracy estimates according to the types of troponin and algorithms

Troponin type by	No. of	Sensitivity	Specificity	Positive likelihood	Negative likelihood	AUROC	I2 (95% CI)	Publication bias
algorithm	studies	(95% CI)	(95% CI)	ratio	ratio	(95% CI)		(Deek’s test P-value)
high sensitivity troponin T
0 h algorithm	32	0.93[0.89-0.95]	0.68[0.60-0.75]	2.9[2.4-3.6]	0.11[0.07-0.15]	0.89[0.86-0.92]	30[16-44]	0.85
1 h algorithm	6	0.92[0.87-0.96]	0.84[0.75-0.90]	5.7[3.7-8.7]	0.09[0.05-0.15]	0.95[0.92-0.96]	13[0-29]	0.33
2 h algorithm	8	0.95[0.92-0.96]	0.79[0.77-0.81]	4.6[4.2-5.1]	0.07[0.05-0.10]	0.90[0.87-0.92]	4[0-13]	0.11
0-1 h delta algorithm	4	0.93[0.86-0.97]	0.81[0.70-0.89]	5.0[3.0-8.3]	0.08[0.04-0.17]	0.94[0.92-0.96]	15[0-37]	0.35
High sensitivity troponin I
0 h algorithm	14	0.93[0.85-0.97]	0.80[0.67-0.89]	4.6[2.8-7.5]	0.08[0.04-0.18]	0.93[0.91-0.95]	40[27-71]	0.45
2 h algorithm	5	0.95[0.91-0.98]	0.86[0.75-0.92]	6.7[3.8-11.9]	0.05[0.03-0.10]	0.97[0.95-0.98]	15[0-35]	0.51
0-1 h delta algorithm	5	0.95[0.86-0.98]	0.82[0.69-0.90]	5.1[3.1-8.4]	0.06[0.03-0.16]	0.94[0.92-0.96]	28[0-59]	0.48

Abbreviation: AUROC, area under receiver operating characteristic curve; CI, confidence interval.

0-h algorithm unlike other non-overlapping confidence intervals. eFigure 1A shows the summary ROCs for the hs-cTnT 0-h algorithm. The 1- and 2-h algorithms had much higher positive likelihood ratios (5.7 [3.7-8.7] and 4.6 [4.2-5.1]) and a lower negative likelihood ratio

(0.09 [0.05-0.15] and 0.07 [0.05-0.10]) compared with the 0-h algo- rithm. There were four studies that used the delta hs-cTnT value be- tween 0 and 1 h to diagnose MI. The pooled sensitivity and specificity of this delta value were 93% (95% CI: 86-97%) and 81% (95% CI: 70-89%), respectively.

Hs-cTnI assay

A total of 14 studies comprised of 20,754 patients with a 14% preva- lence of AMI, for the hs-cTnI assay were included in the 0-h diagnosis al- gorithm subgroup. As shown in Table 1 (lower panel), the pooled sensitivity was 93% (95% CI: 82-98%), comparable to the pooled sensitiv- ity of hs-cTnT assays, while the specificity was 80% (95% CI: 67-89%), higher than the hs-cTnT assay in the 0-h algorithm. The AUROC of the 0-h algorithm was 0.93 (95% CI: 0.91-0.95) (eFigure 1B). The AUROCs for the 2-h (0.97 [95% CI: 0.95-0.98]) and 0-1 h delta (0.94 [95% CI:

0.92-0.96]) algorithms were similar to the 0-h algorithm. There were no significant differences among the three algorithms as the 95% confi- dence intervals of all parameters were broadly overlapped.

Subgroup analyses of hs-cTnT 0-h algorithm

When categorized by clinical setting, the ED setting (0.90 [95% CI: 0.87-0.92]) has a higher AUROC than non-ED settings (0.84 [95% CI: 0.87-0.92]). When examining subgroups using different hs-cTnT cut- offs, studies using a cutoff value of 14 ng/L had the highest positive like- lihood ratio (3.6 [95% CI: 3.0-4.4]) compared with studies using 5 ng/L

(1.7 [95% CI: 1.5-2.0]) and studies using 3 ng/L (1.4 [95% CI: 1.1-1.6]) as cutoff values. However, lowering the cutoff value results in an im- proved negative likelihood (5 ng/L: 0.04 [95% CI: 0.02-0.09]); 3 ng/L

(0.01 [95% CI: 0.00-0.47]). The AUROC was the highest in the 14 ng/L subgroup (0.90 [95% CI: 0.88-0.93]), when compared to the 5 ng/L cut- off (0.85 [95% CI: 0.82-0.88]) and to the 3 ng/L cutoff (0.60 [95% CI: 0.55-0.64]). Studies with a low prevalence of MI (b20%) had a high summary sensitivity (96% [95% CI: 91-98%]) while studies a with high prevalence of MI (>=20%) had lower sensitivity (88% [95% CI: 84-91%]) but a higher specificity (79% [95% CI: 74-83%]). Timing of presentation did not affect the sensitivity of the algorithms by the hs-cTn subtype in our analysis (Table 2). The subgroup analysis for studies evaluating hs-cTnI 0-h algorithm followed a similar pattern although fewer sub- group analyses were feasible due to insufficient data (eTable 2). Lastly, only the Abbott hs-cTnI assay was regulatory approved as the hs-cTnI assay in North America. Accordingly, we performed the analyses restricting on studies using the Abbott assay. Accuracy estimates were similar to those pooling all hs-cTnI assays. Accuracy estimates were summarized in eTable 3.

Sensitivity analysis and publication bias

The goodness of fit and bivariate normality analyses (eFigure 2A, B, both upper and lower panel) suggested that the random effects bivar- iate model was robust. Influence analysis and outlier detection identi- fied two outlier studies for the hs-cTnT assay (eFigure 2C, D, upper panel) and one outlier study for the hs-cTnI assay (eFigure 2C, D, lower panel). When we removed the outlier studies, the I² decreased from 30% to 23% and from 49% to 40%, respectively. Nevertheless, there were only minimal changes in the pooled estimates, which did not significantly affect the overall estimates. Subgroup analyses by

Table 2

Summary of subgroup analysis of the included studies by settings, markers and outcomes.

Variables	No. of	Sensitivity	Specificity	Positive likelihood	Negative likelihood	AUROC	I2 (95% CI)	Publication bias
	studies	(95% CI)	(95% CI)	ratio	ratio	(95% CI)		(Deek’s test P-value)
Settings
ED settings	26	0.93[0.89-0.95]	0.66[0.57-0.75]	2.8[2.2-3.5]	0.11[0.07-0.16]	0.90[0.87-0.92]	28[13-43]	0.89
Non ED settings	7	0.92[0.80-0.97]	0.73[0.65-0.80]	3.4[2.7-4.4]	0.10[0.04-0.29]	0.84[0.81-0.87]	37[2-72]	0.52
Cut-offs
14 ng/L	26	0.90[0.86-0.92]	0.75[0.70-0.80]	3.6[3.0-4.4]	0.14[0.10-0.18]	0.90[0.88-0.92]	14[4-24]	0.41
5 ng/L	7	0.98[0.96-0.99]	0.44[0.35-0.53]	1.7[1.5-2.0]	0.04[0.02-0.09]	0.85[0.82-0.88]	17[0-39]	0.30
3 ng/L	6	1.00[0.85-1.00]	0.27[0.16-0.42]	1.4[1.1-1.6]	0.01[0.00-0.47]	0.60[0.55-0.64]	80[47-100]	0.17
Prevalence of MI
b20%	17	0.96 [0.91, 0.98]	0.56 [0.45, 0.67]	2.2 [1.7, 2.8]	0.07 [0.03, 0.14]	0.86[0.83-0.89]	43[21-66]	0.83
>=20%	16	0.88 [0.84, 0.91]	0.79 [0.74, 0.83]	4.1 [3.3, 5.0]	0.15[0.11, 0.20]	0.91[0.88-0.93]	6[0-15]	0.91
Presenting time
<=4 h	11	0.91[0.80, 0.96]	0.80[0.62, 0.90]	4.5[2.4, 8.6]	0.11 [0.05, 0.24]	0.93[0.90-0.95]	39[10-67]	0.47
N 4 h	10	0.91[0.87, 0.94]	0.75[0.69, 0.80]	3.6[2.9, 4.6]	0.11 [0.07, 0.18]	0.90[0.87-0.92]	11[0-23]	1.00

Abbreviation: NSTEMI, non-ST elevation myocardial infarction; AMI, acute myocardial infarction; ACS, acute coronary syndrome; AUROC, area under receiver operating characteristic curve; CI, confidence interval; ED, emergency department.

different algorithms, settings, cutoff values, prevalence of MI, presenting time, and outcome definitions showed mild heterogeneity (I² b 50%). One exception was the cutoff value of 3 ng/L (I² = 80%), which could be due to a small number of available studies. No publication bias was found in all analyses except for the hs-cTnT 2-h algorithm (Deek’s test p-value = 0.04). We performed a trim-and-fill analysis by incorporating the theoretical missing studies and re-computing the summary effect estimates. The original pooled diagnostic OR for the hs-cTnT 2-h algo- rithm is 66.0 [95% CI: 50.2-87.3]. After incorporating theoretical missing studies, the updated meta-analysis showed a diagnostic OR of 51.7 (95% CI: 11.9-64.9). The filled funnel plot was shown in eFigure 3.

Discussion

In this systematic review consisting of 56 studies and 67,945 pa- tients, we demonstrated both hs-cTnT- and hs-cTnI assays showed a high sensitivity (93%) to rule out AMI on admission. Using a LoD (5 ng/L) or limit of blank (LoB) (3 ng/L) as the cutoff values further in- creased the sensitivity to 98% and 100% at the expense of decreased specificity. We did not find that the 1-h or 2-h algorithms for either hs-cTnT or hs-cTnI assays could further improve the sensitivity, how- ever the 1-h or 2-h algorithms did show an increased specificity for the hs-cTnT assays. The timing of presentation and settings do not seem to affect the accuracy in our analysis. This result was consistent with the meta-analysis by Pickering et al., although the authors adopted LoD combined with a non-ischemic ECG reading [19].

Based on the 2012 expert consensus, high sensitivity assays must have a coefficient of variance of b10% at the 99th percentile concentra- tion of the reference population (URL) and detected the concentration below 99 percentile N50% of normal population to differentiate between Conventional assays and high sensitivity assays, [35]. In a previous meta-analysis, Li et al. including 4 studies and 2863 patients showed that the hs-cTn assay could reach a summary sensitivity of 89% (95% CI: 85-91%) and a summary specificity of 89% (95% CI: 85-92%). How- ever, the authors did not report the accuracy estimates for the hs-cTnT or hs-cTnI assay separately. Furthermore, this meta-analysis did not take account for the different timing of hs-cTn measurements [36]. Kavsak et al. demonstrated that a cutoff of hs-cTnT at 14 ng/L at presen- tation had the highest probability to identify patients at high risk for car- diovascular disease [37]. Westwood et al. later updated the meta- analysis by pooling data from 18 studies and showed that using a LoB threshold in a presentation sample could reach a negative likelihood ratio of 0.10 (95% CI: 0.05 to 0.18) to ruled-out AMI [38]. They also showed that a positive test at 2 h has a positive likelihood ratio of 8.42 (95% CI: 6.11 to 11.60) to rule in an AMI [36,39]. We confirmed Westwood’s finding by showing a negative likelihood ratio of 0.01 (95% CI: 0.00-0.47) for LoB and a negative likelihood ratio of 0.04 (95% CI: 0.02-0.09). By pooling more data for the 2-h algorithms, we showed that the positive likelihood ratio was 4.6 (95% CI: 4.2-5.1) for hs-cTnT assay and 6.7 (95% CI: 3.8-11.9) for the hs-cTnI assay. Zhelev Z et al. also published a meta-analysis addressing the accuracy of the hs-cTnT 0-h algorithm. They showed a summary sensitivity of 0.90 (95% CI: 0.86-0.92) at 14 ng/L and a summary sensitivity of 0.97 (95% CI: 0.95 to 0.99) at 3-5 ng/L [40]. Our results were fairly consistent with their findings, however, through more extensive data collection, our results showed an improved sensitivity. We showed that at the cut- off values of 3, 5, and 14 ng/L, the sensitivity was 100%, 98%, and 90%, re- spectively. Zhelev et al. did not report the effect estimates for hs-cTnI assays, which we showed that they would have a better specificity as compared with the hs-cTnT assay in the 0-h algorithm.

Previous studies demonstrated that the conventional CTnI or cTnT assays had a comparable accuracy in the diagnosis of MI, but the cTnT might carry more prognostic implications than the cTnI [41]. The cTnT assay has been shown to be able to identify a group of patients with worse prognosis among patients that tested positive for cTnI tests [42]. There were only a few studies that made a direct comparison

between the hs-cTnT and the hs-cTnI for the accuracy of early diagnosis of AMI. The largest study (2226 patients) to date by Gimenez et al. re- vealed small but potentially important differences between hs-cTnT and hs-cTnI assays [43]. They showed that the hs-cTnI assay was supe- rior for the early presenters (b3 h since chest pain onset) (AUC: 0.92 [95% CI: 0.89-0.94]) compared with the hs-cTnT (AUC: 0.89 [95% CI: 0.86-0.91]) while the hs-cTnT assay was superior in late presenters (AUC: 0.96 [95% CI: 0.94-0.96]) vs. hs-cTnI (AUC: 0.94 [95% CI:

0.93-0.95]). Our results corroborate this finding. We showed that hs- cTnI assays have superior specificity in the 0-h algorithm but have com- parable sensitivity and specificity in the 2-h and 0/1-h delta algorithms. Results of our meta-analysis are concordant with the findings of the largest study that made a direct comparison between hs-cTnI and hs- cTnT [43].

To our knowledge, this study is by far the most comprehensive meta-analysis on the diagnostic performance of high-sensitivity cardiac troponins for early diagnosis of AMI. The large number of studies affords us the ability to report the accuracy of hs-cTnT and hs-cTnI assays using different algorithms. In addition, the heterogeneities of the overall and almost all subgroup analyses were mild statistically. Reporting positive and negative likelihood ratios is another strength of this study. Likeli- hood ratios are less likely affected by the change of disease prevalence. This enhances the generalizability of the findings. Findings of this meta- analysis may have major clinical implications for clinicians, laboratory experts, and hospital administrators in decision-making guiding them which rapid AMI diagnosis algorithms to use at their institutions.

There are several limitations of the meta-analysis that should be noted. First, various cutoff values were used in individual studies. We adopted two approaches to solve the cutoff heterogeneity problems. The first approach applied the statistical approach (bivariate model) to model the distribution of sensitivities and specificities in each study given the inherent negative correlation between sensitivities and spec- ificities between different studies that use different cutoff values. The second approach provides clinically relevant accuracy estimates for sub- groups using the same cutoff value analysis in the hs-cTnT and hs-cTnI 0-h algorithms. Second, the inclusion criteria for patients in each in- cluded study, to some extent, differ. The cutoff for ED-presenting time since chest pain onset, definition of typical chest pain, and timing for routine ECG varied in each study, which may increase the heterogene- ity. Finally, like other studies that estimated the diagnostic accuracy of hs-cTn, potential verification bias is prevalent in the included studies. Not all enrolled patients received the same outcome verification process because invasive Diagnostic procedures, such as coronary angiography or other advanced imaging exams could not be performed for all en- rolled patients.

Conclusion

Our meta-analysis showed that for patients presenting ED with typ- ical symptoms, the use of hs-cTnT or hs-cTnI assays at the 99th percen- tile may help identify patients with low risk for AMI and promote early discharge from the ED. Using LoD or LoB cut off values for the hs-cTnT test can identify patients with even lower risk of AMI at the expense of increased false positive results. Further evaluation is warranted for clinical decision guidelines incorporating hs-cTn assays or combining multiple biomarker assays for triaging patients with symptoms sugges- tive of AMI.

Funding source

This work is supported by research grant from Taiwan National Min- istry of Science and Technology Grants MOST 104-2314-B-002-039- MY3, and MOST 105-2811-B-002-031. No funding bodies had any role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Declaration of competing interest

The authors have no conflicts of interest to disclose.

Acknowledgement

We would like to thank Kia Byrd, MD, MPH at Harvard Medical School for her critical review of this manuscript.

Contributors

CCL conceived, designed the study, supervised data collection, de- signed the search strategy, and critically reviewed the manuscript for important intellectual content. SSH coordinated the systematic review, acquired the data, carried out the initial analyses, wrote the first draft of the manuscript, and revised the manuscript. YHY wrote the first draft of the manuscript. YTH involved in the data collection and extraction. WTH and JYP interpreted the results, helped with language editing and critically revised the manuscript. KI conceive the idea, reviewed the manuscript and provided critical insights.

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi. org/10.1016/j.ajem.2019.11.035.

References

1. Reed GW, Rossi JE, Cannon CP. Acute myocardial infarction. Lancet 2017;389: 197-210.
2. Global Burden of Disease Study C. Global, regional, and national incidence, preva- lence, and years lived with disability for 301 acute and chronic diseases and injuries in 188 countries, 1990-2013: a systematic analysis for the Global Burden of Disease Study 2013. Lancet 2015;386:743-800.
3. Roth GA, Huffman MD, Moran AE, et al. Global and regional patterns in cardiovascu- lar mortality from 1990 to 2013. Circulation 2015;132:1667-78.
4. Bhuiya FA, Pitts SR, McCaig LF. Emergency department visits for chest pain and ab- dominal pain: United States, 1999-2008. NCHS Data Brief 2010:1-8.
5. Roger VL, Go AS, Lloyd-Jones DM, et al. Heart disease and stroke statistics-2012 update: a report from the American Heart Association. Circulation 2012;125: e2-220.
6. Pines JM, Isserman JA, Szyld D, Dean AJ, McCusker CM, Hollander JE. The effect of physician risk tolerance and the presence of an observation unit on decision making for ED patients with chest pain. Am J Emerg Med 2010;28:771-9.
7. Mehta SR, Granger CB, Boden WE, et al. Early versus delayed invasive intervention in acute coronary syndromes. N Engl J Med 2009;360:2165-75.
8. Antman EM, Cohen M, Bernink PM, et al. The TIMI risk score for unstable angina/ non-ST elevation MI: a method for prognostication and therapeutic decision mak- ing. JAMA 2000;284:835-42.
9. Dolci A, Panteghini M. The exciting story of cardiac biomarkers: from retrospective detection to gold diagnostic standard for acute myocardial infarction and more. Clin Chim Acta 2006;369:179-87.
10. Danese E, Montagnana M. An historical approach to the Diagnostic biomarkers of acute coronary syndrome. Ann Transl Med 2016;4:194.
11. Anderson JL, Adams CD, Antman EM, et al. ACC/AHA 2007 guidelines for the man- agement of patients with unstable angina/non-ST-elevation myocardial infarction: a report of the American College of Cardiology/American Heart Association Task Force on Practice Guidelines (writing committee to revise the 2002 guidelines for the Management of Patients with Unstable Angina/non-ST-elevation myocardial in- farction) developed in collaboration with the American College of Emergency Physi- cians, the Society for Cardiovascular Angiography and Interventions, and the Society of thoracic surgeons endorsed by the American Association of Cardiovascular and Pulmonary Rehabilitation and the Society for Academic Emergency Medicine. J Am Coll Cardiol 2007;50:e1-157.
12. Myocardial infarction redefined-a consensus document of The Joint European Soci- ety of Cardiology/American College of Cardiology Committee for the redefinition of myocardial infarction, Eur Heart J 2000;21:1502-13.
13. Sun BC, Hsia RY, Weiss RE, et al. Effect of emergency department crowding on out- comes of admitted patients. Ann Emerg Med 2013;61:605-11 e6.
14. Hoot NR, Aronsky D. Systematic review of emergency department crowding: causes, effects, and solutions. Ann Emerg Med 2008;52:126-36.
15. Reichlin T, Hochholzer W, Bassetti S, et al. Early diagnosis of myocardial infarction with sensitive cardiac troponin assays. N Engl J Med 2009;361:858-67.
16. Garg P, Morris P, Fazlanie AL, et al. Cardiac biomarkers of acute coronary syndrome: from history to high-sensitivity cardiac troponin. Intern Emerg Med 2017;12: 147-55.
17. McRae AD, Innes G, Graham M, et al. Comparative evaluation of 2-hour rapid diag- nostic algorithms for acute myocardial infarction using high-sensitivity cardiac tro- ponin T. Can J Cardiol 2017;33:1006-12.
18. Vasile VC, Jaffe AS. High-sensitivity cardiac troponin for the diagnosis of patients with acute coronary syndromes. Curr Cardiol Rep 2017;19:92.
19. Pickering JW, Than MP, Cullen L, et al. Rapid rule-out of acute myocardial infarction with a single high-sensitivity cardiac troponin T measurement below the limit of de- tection: a collaborative meta-analysis. Ann Intern Med 2017;166:715-24.
20. Boeddinghaus J, Nestelberger T, Twerenbold R, et al. Direct comparison of 4 very early rule-out strategies for acute myocardial infarction using high-sensitivity car- diac troponin I. Circulation 2017;135:1597-611.
21. Sorensen NA, Neumann JT, Ojeda F, et al. Challenging the 99th percentile: a lower tro- ponin cutoff leads to low mortality of chest pain patients. Int J Cardiol 2017;232:289-93.
22. Wassef AW, Hiebert B, Saeed MF, Tam JW. Novel High-sensitivity troponin assay re- quires higher cut-off value to separate acute myocardial infarction from non-acute myocardial infarction in a high-risk population. Can J Physiol Pharmacol 2015;93: 873-7.
23. Giannitsis E, Katus HA. Pros and cons of high-sensitivity assays for cardiac troponin. Nat Rev Cardiol 2012;9:616-8.
24. Newby LK. Myocardial infarction rule-out in the emergency department: are high- sensitivity troponins the answer?: comment on “One-hour rule-out and rule-in of acute myocardial infarction using high-sensitivity cardiac troponin T”. Arch Intern Med 2012;172:1218-9.
25. Charpentier S, Chenevier-Gobeaux C. 2015 ESC guidelines: 1-hour rule-out and rule- in of acute myocardial infarction with High-sensitive troponin T. Presse Med 2016; 45:859-64.
26. Eggers KM, Aldous S, Greenslade JH, et al. Two-hour diagnostic algorithms for early assessment of patients with acute chest pain-implications of lowering the cardiac troponin I cut-off to the 97.5th percentile. Clin Chim Acta 2015;445:19-24.
27. McInnes MDF, Moher D, Thombs BD, et al. Preferred reporting items for a systematic review and meta-analysis of diagnostic test accuracy studies: the PRISMA-DTA state- ment. JAMA 2018;319:388-96.
28. Whiting PF, Rutjes AW, Westwood ME, et al. QUADAS-2: a revised tool for the qual- ity assessment of diagnostic accuracy studies. Ann Intern Med 2011;155:529-36.
29. Harbord RM, Deeks JJ, Egger M, Whiting P, Sterne JA. A unification of models for meta-analysis of diagnostic accuracy studies. Biostatistics 2007;8:239-51.
30. Cook RD. Detection of influential observation in linear regression. Technometrics 1977;19:15-8.
31. Deeks JJ, Macaskill P, Irwig L. The performance of tests of publication bias and other sample size effects in systematic reviews of diagnostic test accuracy was assessed. J Clin Epidemiol 2005;58:882-93.
32. Thygesen K, Alpert JS, White HD. Universal definition of myocardial infarction. J Am Coll Cardiol 2007;50:2173-95.
33. Thygesen K, Alpert JS, Jaffe AS, et al. Third universal definition of myocardial infarc- tion. J Am Coll Cardiol 2012;60:1581-98.
34. Hamm CW, Bassand JP, Agewall S, et al. ESC guidelines for the management of acute coronary syndromes in patients presenting without persistent ST-segment eleva- tion: the task force for the management of acute coronary syndromes in pa- tients presenting without persistent ST-segment elevation of the European Society of Cardiology (ESC). Eur Heart J 2011;32:2999-3054.
35. Newby LK, Jesse RL, Babb JD, et al. ACCF 2012 expert consensus document on prac- tical clinical considerations in the interpretation of Troponin elevations: a report of the American College of Cardiology Foundation task force on clinical expert consen- sus documents. J Am Coll Cardiol 2012;60:2427-63.
36. Li WJ, Chen XM, Nie XY, et al. Early diagnostic and prognostic utility of high-sensitive troponin assays in acute myocardial infarction: a meta-analysis. Intern Med J 2015; 45:748-56.
37. Kavsak PA, Worster A, Ma J, et al. High-sensitivity cardiac troponin risk cutoffs for acute cardiac outcomes at emergency department presentation. Can J Cardiol 2017;33:898-903.
38. Westwood M, van Asselt T, Ramaekers B, et al. High-sensitivity troponin assays for the early rule-out or diagnosis of acute myocardial infarction in people with acute chest pain: a systematic review and cost-effectiveness analysis. Health Technol As- sess 2015;19:1-234.
39. Sethi A, Bajaj A, Malhotra G, Arora RR, Khosla S. Diagnostic accuracy of sensitive or high-sensitive troponin on presentation for myocardial infarction: a meta-analysis and systematic review. Vasc Health Risk Manag 2014;10:435-50.
40. Zhelev Z, Hyde C, Youngman E, et al. Diagnostic accuracy of single baseline measure- ment of Elecsys troponin T high-sensitive assay for diagnosis of acute myocardial in- farction in emergency department: systematic review and meta-analysis. BMJ 2015; 350:h15.
41. van der Linden N, Klinkenberg LJ, Bekers O, et al. Prognostic value of basal high- sensitive cardiac troponin levels on mortality in the general population: a meta- analysis. Medicine (Baltimore) 2016;95:e5703.
42. Tiwari RP, Jain A, Khan Z, et al. Cardiac troponins I and T: molecular markers for early diagnosis, prognosis, and accurate triaging of patients with acute myocardial infarc- tion. Mol Diagn Ther 2012;16:371-81.
43. Rubini Gimenez M, Twerenbold R, Reichlin T, et al. Direct comparison of high- sensitivity-cardiac troponin I vs. T for the early diagnosis of acute myocardial infarc- tion. Eur Heart J 2014;35:2303-11.