Troponin Cut-Offs for Acute Myocardial Infarction in Patients with Impaired Renal Function—A Systematic Review and Meta-Analysis

Identifying acute myocardial infarction in patients with renal disease is notoriously difficult, due to atypical presentation and chronically elevated troponin. The aim of this study was to identify a specific troponin T/troponin I cut-off value for diagnosis of acute myocardial infarction in patients with renal impairment via meta-analysis. Two investigators screened 2590 publications from MEDLINE, Embase, PubMed, Web of Science, and the Cochrane library. Only studies that investigated alternative cut-offs according to renal impairment were included. Fifteen articles fulfilled the inclusion criteria. Six studies were combined for meta-analysis. The manufacturer’s upper reference level for troponin T is 14 ng/L. Based on the meta-analyses, cut-off values for troponin in patients with renal impairment with myocardial infarction was 42 ng/L for troponin I and 48 ng/L for troponin T. For patients on dialysis the troponin T cut-off is even higher at 239 ng/L. A troponin I cut-off value for dialysis patients could not be established due to lack of data. The 15 studies analyzed showed considerable diversity in study design, study population, and the definition of myocardial infarction. Further studies are needed to define a reliable troponin cut-off value for patients with kidney disease, especially in dialysis patients, and to allow necessary subanalysis.


Introduction
According to the fourth definition of myocardial infarction, the three diagnostic pillars for acute myocardial infarction (AMI) include typical symptoms, typical ECG changes, and a fall or rise in cardiac enzymes-preferably cardiac troponin (cTn) with one of the cTns being above the 99th percentile of the upper reference limit (URL) [1]. Renal impairment is known to be strongly associated with increased risk of cardiovascular disease [2]. The fourth universal definition of myocardial infarction (UDMI) suggests similar criteria for diagnosing AMI among patients with chronic kidney disease (CKD) as those for the general population [1]. However, both symptoms of AMI and changes in ECGs are different in CKD patients in comparison with the background population. The most common symptom associated with AMI in CKD patients is dyspnea and not chest pain [3]. Also, ECG abnormalities, such as left bundle branch block (LBBB), right bundle branch block (RBBB),

Protocol and Registration
The protocol was registered on 28 April 2020 (ID CRD42020162299).

Eligibility Criteria
We included studies evaluating cTn in adult patients with suspected renal impairment for AMI. Only articles in English were considered eligible. No limitations as to publication date were set.
Inclusion criteria: age above 18 years; only full-text articles in peer-reviewed journals; only studies that investigated alternative cut-off values according to renal impairment.
Exclusion criteria: posters or abstracts; letters to the editors; and other grey literature. The search procedure included the following search engines and websites: Embase, MEDLINE, PubMed, Cochrane Library, and Web of Science. The search strategy was assisted by a specialized librarian and can be seen under Supplementary Materials.

Study Selection
Preliminary selection of articles was performed by titles and abstract, followed by a detailed full text selection based on inclusion and exclusion criteria. Assessment was executed by Jan Dominik Kampmann (JDK) and Jeff Granhøj (JG). In the case of disagreement, a referee Frans Brandt (FB) was consulted. The online systemic review manager program Covidence™ was used during the extraction period.

Data Extraction
Using a standardized data extraction form investigator, JDK extracted relevant details and results. The year of publication, study population, study design, and definition of renal impairment are listed in Table 1, and cut-off-value-related data in Table 2. The extracted data were verified by JG.

Methodological Quality Assessment
Quality and bias were assessed by two independent investigators, JDK and JG, using the Quadas2 score. The Quadas2 score is a validated tool for quality assessment of diagnostic accuracy [11]. The tool content was tailored by exchanging the domain "Index test" with reporting bias defined by how much detail the cTn measurement was described in (detection limit, manufacture, 99th percentile) and how well renal impairment assessment was described (estimated glomerular filtration rate (eGFR) formula, assay used for creatinine measurement). In cases of disagreements regarding the QUADAS-2 results, consensus was reached by discussion between JDK and JG. If no agreement could be reached, FB was consulted as third reviewer.

Statistical Analysis
A bivariate mixed-effect model on the sensitivity and specificity transformed by way of the inverse probit function, similar to the model implemented in the R-package diagmeta was employed in order to calculate the optimal cut-off value in accordance with the area under the curve (AUC) for cardiac troponin T(cTnT) and cardiac troponin I(cTnI) in patients with renal impairment [12]. We chose this model because several studies had multiple and varying numbers of cut-off values with corresponding sensitivity and specificity. The AUC showed that the optimal link function was probit for all studies, and the optimal cut-off value was estimated by way of the summary receiver operating characteristic (SROC) curve. Therefore, the optimal cut-off point maximizes the area under the curve of the SROC curve, which is an estimate of the true underlying ROC curve of the studies in the analysis. We assumed that the covariance structure had no random intercept and a common random slope. The choice for this structure was based upon that different cut-off values that were calculated on different patients and a more complex model was unable to converge. A SROC curve for cTnT and cTnI stratified for dialysis for all studies was utilized to assess bias, because challenges in interpretation of funnel plots arise when each study has multiple cut-off values and corresponding sensitivity and specificity. The optimal cut-off points, their corresponding sensitivity and specificity, and their 95% confidence intervals were calculated as described in [12]. The analysis was done in Rstudio™ with the R-package diagmeta.

Results
As our paper is a systematic review on troponin cut-off values in CKD patients and a meta-analysis, results and discussion are divided into a narrative synthesis and a metaanalysis in order to improve readability.

Narrative Synthesis
A total of 2590 publications were screened. A flowchart according to "Referred Reporting Items for Systematic Reviews and Meta-Analyses: the PRISMA Statement 2009" is presented in Figure 1 [13]. During the title and abstract screening, 2544 studies did not fulfill the inclusion criteria. Publications in other languages than English and publications not dealing with cTn cut-off values in patients with AMI and renal impairment were excluded. The remaining 46 publications were again assessed for eligibility according to inclusion and exclusion criteria. A total of 15 articles published from 2007 [14]-2019 [15], were preliminarily included. A total of six studies, which were considered to be adequate in terms of AMI and renal impairment definition, were selected for the meta-analysis.
One of the four retrospective studies was a multicenter study [18]. Three of the eight prospective studies were multicenter studies [23][24][25]. Two of the post hoc analyses were based on a multicenter study [27,28]. The other studies were single-center studies [14- In the following all 15 studies are presented in combination. Studies featured in the meta-analysis and the meta-analysis itself are discussed in detail at the end of the result section. Study characteristics of the included studies can be seen in Table 1.
Chotivanawan et al. [21] and Canney et al. [18], included asymptomatic patients in order to establish a mean cTn value in asymptomatic renal impairment patients or to establish the long-term risk of cardiovascular events, respectively.

Defining the Endpoint AMI
There was a heterogeneous definition of AMI in the included articles. Four articles used NSTEMI as an endpoint [16,23,24,28], Miller-Hodges et al. included cardiovascular death within 30 days additionally to NSTEMI diagnosis [23].
Two articles chose different endpoints. Canney et al. established a cut-off for CV risk in asymptomatic CKD patients [18]. Chotivanawan et al. suggested a cut-off based on the mean hs-cTnT in asymptomatic patients with CKD stage 3-5 [21].  [17,27]; and Flores-Solis et al. [19] and Twerenbold et al. [24] to the ACS consensus documents by the ESC, ACCF, AHA, and WHF from 2007. Iwasaki et al. used the ESC, American College of Cardiology, American Heart Association and World Heart Federation Task Force consensus document [20].

Confirmation of the Diagnosis Myocardial Infarction
Out of the 13 publication that had ACS or AMI as endpoint, six authors specified by whom the diagnosis of AMI/ACS was suggested [15,17,20,25,27,28].
Out of the 13 publications, 6 used a varying number of cardiologists to confirm the diagnosis AMI/ACS [15,17,20,25,27,28]. In two articles the cardiologists were blinded for the cTn result [20,28]. Kraus et al. used two cohorts in their study. In one of the cohorts, the AMI diagnosis was based on the diagnosis at hospital discharge, in the other cohort the AMI diagnosis was made by two independent cardiologists [28].

Patient Characteristics
The proportion of men with renal impairment in the study population ranged from 34.8% [14] to −68% [19]. The age of patients with renal impairment in the studies was between 60.9 [26] and 79.0 years [24]. One article did not report the age and sex distribution of participants with renal impairment [27]. The size of study population ranged from 46 [14] to 2284 patients [26].

Quality Assessment
In order to determine the risk of bias and applicability of the individual studies, the QUADAS-2 score was applied. Results are shown in Figures 2 and 3. The proportion of men with renal impairment in the study population ranged from 34.8% [14] to −68% [19]. The age of patients with renal impairment in the studies was between 60.9 [26] and 79.0 years [24]. One article did not report the age and sex distribution of participants with renal impairment [27]. The size of study population ranged from 46 [14] to 2284 patients [26].

Quality Assessment
In order to determine the risk of bias and applicability of the individual studies, the QUADAS-2 score was applied. Results are shown in Figures 2 and 3.
The risk for bias was low in the studies in general. Only two studies [14,22] scored "unclear" in terms of reporting bias and none of the studies had a high risk for reporting bias or patient flow according to QUADAS-2 score. Patient selection was more problematic with three studies [14,16,17], having a high risk for bias and four studies [15,20,24,28] with an unclear risk of bias. Three studies [16,18,19] were considered high risk in terms of reference standard bias, ten [14,15,17,20,[22][23][24][25]27,28] low risk, and two studies [21,26] were regarded as unclear.
Chenevier-Gobeaux et al. [27] and Twerenbold et al. [25], were the best studies according to our evaluation in terms of risk of bias and concerns regarding applicability, with low-risk scores in all categories. The study with most concerns regarding applicability was by Canney et al. [18], with high risk scoring throughout. Risk of bias was highest for the study by Sukanthasarn et al. [14], with only one low-risk score in reference standard, high-risk inpatient selection, and unclear results in reporting bias and flow and timing.
Two articles looked exclusively at end stage renal patients (ESRD) patients on dialysis resulting in a higher cut-off value [15,26]. Yang et al. showed a lower cut-off value for cTn in the CKD stage 5 group not on dialysis compared to CKD 4 patients. Soeiro et al. suggest a higher cut-off value in patients without renal failure compared to patients with renal impairment (6.05 ng/L compared to 5.20 ng/L) [16]. Iwasaki et al. used a fourpattern semiquantitative measurement of cTnT, where only changes between 0.1-2.0 ng/mL were displayed numerically [20].

Meta-Analysis
Six papers [15,17,[25][26][27][28] were eligible for meta-analysis defined by relevant study population, AMI definition, and comparable statistical method. The respective studies provided the following amount of cut-off values in total: Gobeaux et al. A total of 15 studies were included for the qualitative synthesis and of these, six studies were included in the meta-analysis. Three articles were excluded as the authors  The risk for bias was low in the studies in general. Only two studies [14,22] scored "unclear" in terms of reporting bias and none of the studies had a high risk for reporting bias or patient flow according to QUADAS-2 score. Patient selection was more problematic with three studies [14,16,17], having a high risk for bias and four studies [15,20,24,28] with an unclear risk of bias. Three studies [16,18,19] were considered high risk in terms of reference standard bias, ten [14,15,17,20,[22][23][24][25]27,28] low risk, and two studies [21,26] were regarded as unclear.
Chenevier-Gobeaux et al. [27] and Twerenbold et al. [25], were the best studies according to our evaluation in terms of risk of bias and concerns regarding applicability, with low-risk scores in all categories. The study with most concerns regarding applicability was by Canney et al. [18], with high risk scoring throughout. Risk of bias was highest for the study by Sukanthasarn et al. [14], with only one low-risk score in reference standard, high-risk inpatient selection, and unclear results in reporting bias and flow and timing.
Two articles looked exclusively at end stage renal patients (ESRD) patients on dialysis resulting in a higher cut-off value [15,26]. Yang et al. showed a lower cut-off value for cTn in the CKD stage 5 group not on dialysis compared to CKD 4 patients. Soeiro et al. suggest a higher cut-off value in patients without renal failure compared to patients with renal impairment (6.05 ng/L compared to 5.20 ng/L) [16]. Iwasaki et al. used a four-pattern semiquantitative measurement of cTnT, where only changes between 0.1-2.0 ng/mL were displayed numerically [20].

Meta-Analysis
Six papers [15,17,[25][26][27][28] were eligible for meta-analysis defined by relevant study population, AMI definition, and comparable statistical method. The respective studies provided the following amount of cut-off values in total: Gobeaux et al. A total of 15 studies were included for the qualitative synthesis and of these, six studies were included in the meta-analysis. Three articles were excluded as the authors only accepted the diagnosis myocardial infarction when a significant coronary artery occlusion was present on angiography [16,20,22]. One study was not eligible for metaanalysis due to the inclusion of unstable angina pectoris as endpoint [19]. Two articles only included asymptomatic patients and were excluded [18,21]. Two other studies were not considered for meta-analysis since both tested a cut-off value from a previous study [23] or an alternative algorithm [24]. The Sukanthasarn et al. [14] article was excluded due to high risk scores in applicability and risk of bias in the QUADAS-2 assessment.
The curve was computed by gathering the provided cut-off values from the different studies together with the respective sensitivity and specificity.
The meta-analysis established that the optimized cut-off for cTnI and cTnT from all the six studies included is at 73 (25.75; 119.45). The sensitivity and specificity of the cut-off values were 0.78 and 0.84, respectively, with an AUC of 0.89 (see Figure 4). A total of 15 studies were included for the qualitative synthesis and of these, six studies were included in the meta-analysis. Three articles were excluded as the authors only accepted the diagnosis myocardial infarction when a significant coronary artery occlusion was present on angiography [16,20,22]. One study was not eligible for metaanalysis due to the inclusion of unstable angina pectoris as endpoint [19]. Two articles only included asymptomatic patients and were excluded [18,21]. Two other studies were not considered for meta-analysis since both tested a cut-off value from a previous study [23] or an alternative algorithm [24]. The Sukanthasarn et al. [14] article was excluded due to high risk scores in applicability and risk of bias in the QUADAS-2 assessment.
The curve was computed by gathering the provided cut-off values from the different studies together with the respective sensitivity and specificity.
The meta-analysis established that the optimized cut-off for cTnI and cTnT from all the six studies included is at 73 (25.75; 119.45). The sensitivity and specificity of the cutoff values were 0.78 and 0.84, respectively, with an AUC of 0.89 (see Figure 4).  For patients with eGFR < 60 mL/min/1.73 m 2 not on dialysis, the optimized cut-off for cTnT according to meta-analysis was at 48 ng/L (23.95; 71.83). Four studies were included, providing 11 cut-offs. Kraus      The cTnI cut-off for patients with eGFR < 60 mL/min/1.73 m 2 not on dialysis was established at 42 ng/L (33.83; 51.08). Kraus et al. provided one cut-off and Twerenbold 30. The cut-off for cTnI for patients on dialysis could not be analyzed as only one study using troponin I included dialysis patients [15]. Kraus et al. used a hs-TnI assay. Twerenbold used three cardiac cTnI and three hs-cTnI assays. The sensitivity and specificity for the cut-off were 0.77 and 0.86 respectively with an AUC of 0.89. See Figure 7. The cTnI cut-off for patients with eGFR < 60 mL/min/1.73 m 2 not on dialysis was established at 42 ng/L (33.83; 51.08). Kraus et al. provided one cut-off and Twerenbold 30. The cut-off for cTnI for patients on dialysis could not be analyzed as only one study using troponin I included dialysis patients [15]. Kraus et al. used a hs-TnI assay. Twerenbold used three cardiac cTnI and three hs-cTnI assays. The sensitivity and specificity for the cut-off were 0.77 and 0.86 respectively with an AUC of 0.89. See Figure 7.  Table 3 shows all the different estimated cut-offs with the respective sensitivity, specificity, AUC, and CI values. Abbreviations: CI (confidence interval), AUC (area under the curve). *: confidence interval calculated for sensitivity given specificity, + : confidence interval calculated for the specificity, given sensitivity.  Table 3 shows all the different estimated cut-offs with the respective sensitivity, specificity, AUC, and CI values. Abbreviations: CI (confidence interval), AUC (area under the curve). *: confidence interval calculated for sensitivity given specificity, + : confidence interval calculated for the specificity, given sensitivity.

Narrative Synthesis
The interpretation of cTn in CKD patient with AMI is challenging. The prevalence of chronically elevated cTn is high and especially the first cTn measurement is difficult to interpret. While waiting for a second cTn to establish dynamic changes, relevant interventions may be delayed.
The studies included in this review were heterogeneous in terms of definition of renal impairment and definition of AMI.
The definition of renal impairment varied due to the different eGFR formulas used. The definition of chronic kidney disease opposed to a random eGFR measurement was vague in many of the included studies. It has not been established whether acute renal failure does have a different impact on cTn compared to chronic kidney disease and might have had an impact on the results.
No study presented information on dialysis details or as to the timing when the cTn sample was obtained. Studies have shown different results on the influence of dialysis on cTn [29,30]. Without those details, the cTn measurements may be unreliable. The cut-off span from 75 ng/L for hemodialysis (HD) patients to 144 ng/L for the seven PD patients included in the study was wide. PD patients tended to have higher baseline cTn [31]; however, a twice-as-high cut-off seems extreme.
Only three studies [17,18,20] provided different cut-off values for cTn according to different stages of CKD. The cut-off values showed considerable diversity. Only Chenevier-Gobaux et al. described that the area under the curve in their study did not vary according to eGFR categories; however, only 75 patients with CKD were involved in the study [27].
The suggested cut-off values were considerably higher in studies with a predominantly Asian study population [17,20]. In contrast to the study by Yang et al. [17] who suggested a cut-off level for hs-TnT of 129 ng/L, Twerenbold [25]  The articles included in the review used different definitions for diagnosing AMI. The definition of AMI has changed significantly from 2000 to 2007. One article combined two cohorts, which used different definition of AMI as endpoint [28]. This might have influenced the final diagnosis.
NSTEMI is more common in CKD patients than STEMI [3,5]. The ECG changes and symptoms can cause problems in CKD patients due to atypical presentation, making NSTEMI hard to diagnose, and leaving the physician dependent on cTn results [4,32,33]. The diagnostic dilemma has been shown in Twerenbold et al.'s study in which the two independent cardiologists disagreed more frequently when it came to diagnosing NSTEMI in patients with renal impairment in comparison to patients with normal eGFR (13.1% vs. 9.1%, p = 0.006) [24]. It is therefore debatable if NSTEMI as endpoint is reliable.
NSTEMI is a clinical diagnosis and coronary angiography is not mandatory [34]. Previous studies have shown that 40-50% of patient have coronary stenosis before starting dialysis [35] causing potential bias when using coronary occlusion as an endpoint as was done in some studies.
Chenevier-Gobeaux et al. underlined the importance of age as influential factor on cTn alteration [27]. The authors propose that an adapted threshold for cTn is required for patients with low eGFR and ≥70 years of age [27]. Kraus et al. presented an algorithm based on admission cTn and dynamic changes [28]. Flores-Solís et al. suggested including CK-MB, which improved the diagnostic accuracy in their study [19]. Twerenbold et al. tested the recently proposed 0/1-h European Society of Cardiology (ESC) algorithm for patients with low eGFR [24]. The 0/1-h algorithm is designed for patients with NSTEMI and is based on cTn concentrations at presentation and their absolute change after 1 h. Using slightly higher cut-offs for ruling out NSTEMI yielded only a small improvement in their study and the authors did not recommend altering the cut-off for the 0/1-h algorithm for patients with low eGFR [24].

Meta-Analysis
According to our meta-analysis, the cTnI cut-off lies at 42 ng/L and at 48 ng/L for cTnT. For patients on dialysis the cTnT cut-off is as high as 240 ng/L.
The calculated cut-off values for cTn in patients with renal impairment and myocardial infarction were in general higher than the URL suggested by the assay manufacturers.
The manufacturer's upper reference level for cTnT is 14 ng/L. The suggested cut-off for cTnT is 3.4 times higher than the URL for non-dialysis patients and 17 times higher for dialysis patients.
In cTnI the URL ranges from 9 (TNI Siemens) to 42 ng/mL (TNI Beckman Coulter), corresponding to a cut-off 4.7 times higher than, or roughly identical to, the manufacturer's URL.
The sensitivity and specificity of the generated cut-offs were generally high, with a sensitivity of 0.76 for TnT in patients without dialysis and 0.88 in dialysis patients respectively. The corresponding specificity was 0.78 and 0.92 respectively. For cTnI cut-off in non-dialysis patients, the sensitivity was 0.77 and specificity 0.86. This gives us some confidence that our cut-offs are clinically useful. However, it should be noted that only the cTnT cut-off value for non-dialysis patients was based exclusively on highly sensitive assays. The other meta-analysis consisted of data were from both highly sensitive and less-sensitive assays.

Limitations
There are many conditions that cause elevated cTn levels [21,36]. Both the sensitivity and specificity of cTn can be affected by various factors, such as the respective assay, manufacturer, sex, and the age of the patient [7,25,27,37,38]. A within-day and between-day reference change from 46% to −32% and from 81% to −45% respectively in cTnI have been described [39]. Due to this, an optimized cut-off should therefore always be interpreted with caution.
Our meta-analysis depended on multiple suggested cut-off values from the different studies. Most of these cut-offs derive from Twerenbold et al. [25]. However, this study scored highly in the QUADAS-2 quality score, emphasizing its quality, and are in line with cut-offs from the other high-scoring study by Chenevier-Gobeaux [27]. The featured TnI assays in Twerebold's work had a wide range of 99th percentile from 9 ng/L to 42 ng/L, yet the optimized cut off-range was much narrower, ranging from 26 ng/L to 46 ng/L [25]. In order to produce meaningful troponin cut-off values, our study includes the current hs-Tn as well as cTn assays. However, although the 99th percentile values may be lower for highly sensitive assays, the cut-off values were similar in high cTn and hs-Tn assays [25]. Therefore, we argue that using both assays does not alter the quality of the cut-off value.
We acknowledge that an assay specific cut-off value and a subanalysis for NSTEMI is crucial. This, however, was not possible with the included studies.
The cTnI cut-off for patients not on dialysis and the cTnT cut-off for patients on dialysis is based on only two studies. The analysis was only possible due to the several cut-offs provided by the respective studies and the respective sensitivity and specificity analysis. Several subanalyses could not be performed due to lack of data, including differences of cut-off values according to cTn versus hs-cTn assays, differences in the respective CKD stages, and differences in cut-off values in HD patients and PD patients.

Conclusions
Our review consisted of highly diverse studies. We suggest cut-offs for cTnI of 42 ng/L, and for cTnT of 48 ng/L for non-dialysis patients with eGFR < 60 mL/min/1.73 m 2 . For patients on dialysis, we suggest a cTnI cut-off of 240 ng/L. However, these cut-offs were only based on two studies, providing a substantial risk of bias. Yet, the high sensitivity and specificity of the cut-offs emphasize the validity of the optimized cut-offs. We suggest further studies with a high number of patients and homogenous definition of renal impairment and AMI to confirm or modify the findings. Further subdivision according to eGFR would be desirable in order to optimize and personalize cTn cut-offs, especially for dialysis patients. A stratification for NSTEMI vs. STEMI would provide important information. The limitations and challenges of our study can be seen as an inspiration to improving future study designs on troponin cut-off values. Trials comparing multiple highly sensitive assays are desirable, since all assays have their own individual cut-of value. In dialysis patients, cTn cut-offs should be subdivided into HD and PD. Optimized cTn cut-off values can never stand alone. Clinical assessment and thorough anamneses will always be pivotal for diagnosing AMI. Therefore, the cut-off values presented in our study should be regarded as a suggestion rather than a final conclusion, and we underline the importance of further studies on the subject.

Informed Consent Statement: Not applicable.
Data Availability Statement: Data sharing not applicable. No new data were created or analyzed in this study. Data sharing is not applicable to this article.