High-Sensitivity Cardiac Troponin Concentration and Risk of First-Ever Cardiovascular Outcomes in 154,052 Participants

1Cardiac troponins are structural proteins in the contractile apparatus of cardiac myocytes that are released into the circulation after cardiac myocyte cell death. Since the introduction of the first troponin assays in the early 1990s, measurement of circulating cardiac troponin concentration has become a cornerstone in the diagnosis and acute management of myocardial infarction (MI). Development of high-sensitivity assays allowed the detection of cardiac troponin I and T (hs-cTnI and hs-cTnT, respectively) at lower concentrations, which enabled earlier identification of myocardial injury and increased diagnostic accuracy in patients with suspected MI.

2At the same time, these methodological advances have led to a fundamental change in the clinical interpretation of cardiac troponin assay results. Although patients traditionally have been classified as being “troponin-positive” or “troponin-negative” in the context of the diagnosis of acute MI, new-generation assays allow a continuous interpretation of findings (i.e., higher vs. lower concentration). Furthermore, high-sensitivity assays can detect circulating cardiac troponin in individuals with no clinically manifest myocardial damage or previous cardiovascular disease (CVD).

3Recent studies in subjects recruited from the general population have suggested that elevations in circulating cardiac troponin are associated with a higher risk of a first-ever CVD event,,,,,,,,,,,,,,,,,,,,,,,,,,. The largest such study, from the BiomarCaRE (Biomarkers for Cardiovascular Risk Assessment in Europe) consortium, reported a hazard ratio (HR) for CVD of 1.92 (95% confidence interval [CI]: 1.76 to 2.10) comparing the highest versus the lowest quintile of hs-cTnI concentration. Although these studies have yielded promising findings, their interpretation and comparison have been hampered by use of different scales and inconsistent adjustment of effect estimates. To clarify associations of cardiac troponins with incident CVD, the present study collated effect estimates from published studies conducted in populations free of previous CVD, supplemented them with de novo data, and combined them in a meta-analysis.

4We systematically searched the electronic databases PubMed, Web of Science, and EMBASE for prospective studies published up to September 2016 that reported on associations of cardiac troponin concentration with incident CVD outcomes. The search strategy is detailed in Online Table 1. Studies were eligible for inclusion if they: 1) had measured hs-cTnI and/or hs-cTnT with a high-sensitivity assay; 2) had recruited participants not based on having a history of CVD at baseline; and 3) had recorded CVD outcomes over more than 1 year of follow-up. We further obtained unpublished tabular data from the Bruneck Study through direct correspondence with study investigators. Using standardized data extraction protocols, 2 independent reviewers (P.W. and J.D.W.E.) extracted, by consensus, information on geographical location, baseline survey dates, study design, exclusion of participants with pre-existing CVD, mean age at baseline, proportion of male subjects, proportion of white participants, sample type, cardiac troponin assay manufacturer, assay detection limit, proportion of participants with detectable cardiac troponin, and number of participants in the study. In relation to incident outcomes, the reviewers extracted data on duration of follow-up, the specific composition of reported endpoints, the number of outcomes accrued, effect sizes, and the degree of statistical adjustment of any reported association. The degree of adjustment was classified as “o” when relative risks (RRs) were adjusted for age and sex only, “+” when RRs were adjusted for conventional risk factors (age, sex, smoking, diabetes, blood pressure, and total and high-density lipoprotein cholesterol [HDL-C]); “++” when RRs were also adjusted for C-reactive protein (CRP) or serum creatinine; and “+++” after further adjustment for B-type natriuretic peptides. If multiple publications on the same study were available, the most up-to-date or comprehensive information was used. Methods and results are reported in accordance with the Preferred Reporting Items for Systematic Review and Meta-Analysis Protocols guidelines.

5PROSPER (Pravastatin in Elderly Individuals at Risk of Vascular Disease Study) was a randomized, double-blind, placebo-controlled pravastatin trial in individuals with pre-existing CVD or risk factors thereof (i.e., smoking, hypertension, diabetes),. The present analysis involved 4,402 participants with no previous CVD and with complete information on hs-cTnT and covariates, recruited at study centers in Scotland, Ireland, and the Netherlands (Online Figure 1). Measurement of hs-cTnT used plasma samples obtained 6 months after randomization using a high-sensitivity electrochemiluminescence immunoassay. The occurrence of incident outcomes was adjudicated by the PROSPER Endpoints Committee during the in-trial phase (mean duration: 3.2 years) and ascertained with routine health data thereafter. The institutional ethics review boards of all centers approved the protocol, and all participants gave written informed consent. The protocol adhered to the principles of the Declaration of Helsinki. Further details are provided in the Online Appendix.

6The statistical analysis was conducted according to a pre-defined statistical analysis plan. The combined CVD endpoint was defined as fatal coronary heart disease (CHD), nonfatal MI, and fatal plus nonfatal stroke (ischemic, hemorrhagic, or unclassified). In PROSPER, we assessed cross-sectional associations between hs-cTnT and other characteristics by using linear and logistic regression models (for continuous and categorical data, respectively), adjusted for age, sex, center, and treatment arm. HRs were calculated for CVD outcomes stratified according to treatment arm and were progressively adjusted for age, sex, and center (model 1), plus smoking, history of diabetes mellitus, history of hypertension, systolic blood pressure, total cholesterol, HDL-C, and body mass index (model 2), plus CRP, estimated glomerular filtration rate (eGFR), and N-terminal pro–B-type natriuretic peptide (NT-proBNP) (model 3).

7To characterize shapes of associations with CVD outcomes, HRs were calculated across quartiles of detectable hs-cTnT concentration by using participants with undetectable hs-cTnT as the reference group, and HRs were plotted against the median hs-cTnT concentration within each category. For the outcomes of CVD and fatal CVD, effect modification was explored across clinically relevant subgroups with formal tests of interaction. We also tested whether assessment of hs-cTnT, in addition to age, sex, smoking status, history of diabetes mellitus, systolic blood pressure, and levels of total cholesterol and HDL-C, could improve the prediction of 10-year CVD risk. C-index changes as well as categorical and continuous net reclassification indexes (NRIs) were calculated by using previously published methods,. Categorical NRIs were calculated across 10-year predicted risk categories of <15%, 15% to <25%, and ≥25%, chosen to reflect the elevated risk in the study population; 95% CIs were estimated for these prediction metrics by using bootstrap resampling with 999 repetitions.

8Study-specific risk estimates were pooled by using random effects meta-analyses. Odds ratios and HRs were assumed to approximate the same underlying measure of RR. When studies reported RRs of various levels of adjustment, the most adjusted estimate was used. To enable a consistent approach to analysis, RRs and 95% CIs in each study were standardized to a common scale (i.e., to reflect a comparison of the top third vs. the bottom third of the population's cardiac troponin distribution, using methods previously described). Consistency of findings across studies was assessed with standard chi-square tests and the I2 statistic. Subgroup analyses were conducted by using meta-regression across pre-specified study-level characteristics. Evidence of publication bias was assessed by using funnel plots and Egger's asymmetry test. Statistical tests were 2-sided and used a significance level of p < 0.05 for principal analyses and p < 0.01 for subgroup analyses.

9In PROSPER, hs-cTnT was detectable in 3,853 (87.5%) of 4,402 participants. The median concentration of hs-cTnT was 7 ng/l (interquartile range: 4 to 11 ng/l), and 85% of hs-cTnT values were within the normal range of ≤14 ng/l. Concentration of hs-cTnT correlated positively with age, male sex, body mass index, blood pressure, NT-proBNP, CRP, history of diabetes mellitus, and history of hypertension (Online Table 2). Concentration of hs-cTnT was lower in current smokers and correlated inversely with renal function assessed by eGFR. There was no difference in hs-cTnT according to trial arm.

10In the overall cohort, 694 fatal CVD events were recorded over a follow-up duration of up to 11.3 years (mean: 8.2 years). At the Scottish center, 519 fatal and nonfatal CVD events, 405 CHD events, and 269 stroke events were recorded. After adjusting for age, sex, smoking, history of diabetes mellitus, history of hypertension, systolic blood pressure, total cholesterol, HDL-C, and body mass index, the shapes of associations between baseline hs-cTnT concentration and CVD outcomes were approximately log linear (Figure 1). The adjusted HRs for a comparison of the top third versus the bottom third of the hs-cTnT concentration were 1.55 (95% CI: 1.23 to 1.96) for CVD, 2.16 (95% CI: 1.74 to 2.67) for fatal CVD, 1.85 (95% CI: 1.42 to 2.42) for CHD, and 1.21 (95% CI: 0.88 to 1.67) for stroke. There was some attenuation of estimates after further adjustment for CRP, eGFR, and NT-proBNP (Online Table 3). There was no evidence for effect modification by a range of participant characteristics, including age, sex, diabetes, eGFR, and NT-proBNP, but there was evidence for a somewhat stronger association in never- or ex-smokers compared with current smokers (Online Figure 2).

11We screened 4,645 records and identified 28 eligible prospective studies,,,,,,,,,,,,,,,,,,,,,,,,,, reporting on a total of 154,052 participants (Table 1, Online Figure 3). Eighteen studies were based in Europe, 7 in North America, 1 in Asia, and 2 were multinational. The mean age was 56.1 years, 52.8% of participants were male, and 88.6% were white. Seventeen studies had measured hs-cTnI and 11 had measured hs-cTnT. The proportion of participants with detectable cardiac troponin concentrations was 80.0% overall (82.6% in hs-cTnI studies and 69.7% in hs-cTnT studies). Multivariable regression analyses weighting studies according to their size identified that the proportion of participants with detectable troponin was higher in hs-cTnI studies (p = 0.007) and in older study populations (8.3% per decade higher mean age; p = 0.009) but was unrelated to sex distributions (p = 0.155). Figure 2 displays the proportion of troponin detection according to study, assay type, and mean age of the study population.

12The combined RR in the top third versus the bottom third of cardiac troponin concentration was 1.43 for CVD, 1.67 for fatal CVD, 1.59 for CHD, and 1.35 for stroke (Figure 3). Forest plots for each outcome are shown in Online Figures 4 to 7. The level of between-study heterogeneity was low for stroke (I2 = 0%; p = 0.764) and high for other outcomes (I2 >75%; all p < 0.001). Magnitudes of associations with CVD and fatal CVD were largely similar across studies grouped according to geographical regions, in studies nested versus those not nested in clinical trials (Online Figure 8), and for different mean ages of study participants, proportions of male participants, and numbers of recorded events (Online Figure 9). Associations for fatal CVD were somewhat stronger in studies based in North America (p = 0.010) (Online Figure 8) and studies measuring hs-cTnT rather than hs-cTnI (p = 0.027) (Online Figure 8). Associations with CVD were stronger in studies with a higher proportion of nonwhite participants (p = 0.004) (Online Figure 9). There was evidence for publication bias in reporting associations with fatal CVD (pEgger = 0.027) but not for the other outcomes assessed (Online Figure 10).

13The addition of hs-cTnT concentration to a prediction model containing information on conventional risk factors improved the prediction of fatal CVD (Online Table 4). It improved the C-index by 0.028 (95% CI: 0.007 to 0.050; p = 0.018), the categorical NRI by 0.123 (95% CI: 0.074 to 0.153; p < 0.001), and the continuous NRI by 0.357 (95% CI: 0.277 to 0.436; p < 0.001). In contrast, for the overall CVD outcome, smaller improvements in the continuous NRI of 0.152 (95% CI: 0.052 to 0.253; p = 0.003) were observed, whereas no improvements were observed in the C-index (p = 0.51) and the categorical NRI (p = 0.25). Detailed reclassification data are provided in Online Table 5.

14In the present report that combined de novo data from the PROSPER study with a comprehensive literature-based meta-analysis, we reliably quantified associations of baseline cardiac troponin measurements with first-ever CVD outcomes (Central Illustration). We further determined the added predictive value of such measurements in the primary prevention of CVD.

15Overall, the present systematic review and meta-analysis involved data from 28 long-term prospective studies with a total of 154,052 participants without previous CVD. All included studies measured cardiac troponins hs-cTnT or hs-cTnI with high-sensitivity assays and reported detectable levels in 80% of people, on average. Pooled estimates from the meta-analysis suggest that individuals with cardiac troponin values in the top third of the population distribution are at 43% increased risk of any CVD, 59% increased risk of CHD, and 67% increased risk of fatal CVD outcomes. A smaller but significant increase in stroke risk (35%) was noted. It is likely that this association was driven by cardiac and heart rhythm abnormalities causing ischemic stroke, including paroxysmal atrial fibrillation. An analysis of the Atherosclerosis Risk in Communities Study according to stroke subtype revealed preferential associations of hs-cTnT with cardioembolic stroke. Furthermore, Wrigley et al. recently reported that, in patients with acute ischemic stroke, an elevated troponin concentration was associated with structural cardiac disease detected by using echocardiography.

16Our findings in the meta-analysis were further substantiated by analyses of PROSPER study data, showing that subjects with higher blood concentrations of hs-cTnT were at higher risk of developing fatal and nonfatal CVD and CHD. Associations were independent of conventional CVD risk factors and persisted even after additional adjustment for NT-proBNP, eGFR, and CRP. We could also confirm a previous suggestion from the BiomarCaRE consortium that the risk of these diseases increased approximately log linearly with higher cardiac troponin concentration, with no evidence for a threshold effect at a certain level. It is striking that we observed significant positive associations with CVD outcomes even though 85% of the study population had hs-cTnT values within the normal range (≤14 ng/l) at baseline.

17We also detected significant between-study heterogeneity for the associations with the outcomes of all CVD and fatal CVD. Studies varied in their age and sex profile, outcome definitions, follow-up durations, and degrees of statistical adjustment, but none of these factors had an influence on the strengths of associations observed. However, our analysis revealed possibly stronger associations in studies based in North America and with a higher proportion of nonwhite participants. Although it is well established that nonwhite participants have higher cardiac troponin values compared with white participants,,,,, single studies have so far not detected that associations of cardiac troponins with CHD or stroke are modified according to ethnicity. Furthermore, we noted a trend toward stronger associations for hs-cTnT than for hs-cTnI. To shed more light on this preliminary finding, future research is needed that evaluates, in a direct comparison, whether the clinical utility of hs-cTnT and hs-cTnI measurements depends on the assay generation or on patient characteristics, including age, sex, and indicators of myocardial stress. Although findings in PROSPER and 3 previous studies,, suggested that the prognostic value of cardiac troponins does not vary according to NT-proBNP levels, analysis of large-scale individual-participant data is required to assess multiplicative interactions adequately.

18The traditional view has been that cardiac troponins are elevated in case of myocardial necrosis only; however, evidence has emerged that other heart diseases such as atrial fibrillation and cardiac structural and functional abnormalities, can also lead to modest increases in cardiac troponin concentrations in circulation. Another relevant factor could be subclinical coronary atherosclerosis. Coronary angiography studies have shown close correlations between the severity of coronary atherosclerosis and cardiac troponin concentration. It is speculated that clinically silent ruptures of coronary plaques trigger microembolizations and, consequently, the shedding of troponin from cardiomyocytes via membranous blebs. A third mechanism could be the link between cardiac troponin concentration and cardiac stress characterized by activation of the adrenergic and renin-angiotensin-aldosterone systems. This relationship has been demonstrated in patients with heart failure.

19Our data also suggested that cardiac troponin assessment could be a useful adjunct to conventional risk factors in predicting CVD risk. In line with a preferential association of cardiac troponins with fatal CVD, assessment of hs-cTnT in the PROSPER study yielded greater improvements for predicting fatal CVD than overall CVD. For fatal CVD, it yielded improvements in the C-index as well as categorical and continuous NRIs. For the overall CVD outcome, hs-cTnT assessment only improved the continuous NRI.

20Another group of biomarkers that has emerged as a potential adjunct to CVD risk prediction are the natriuretic peptides. In the Natriuretic Peptides Studies Collaboration, a large collaborative individual-participant-data meta-analysis, NT-proBNP demonstrated associations with CVD (RR: 1.76; 95% CI: 1.56 to 1.98) and CHD (RR: 1.67; 95% CI: 1.45 to 1.93) that were of similar magnitude to those observed in PROSPER for hs-cTnT. The association between hs-cTnT and CVD risk was weakened in PROSPER after adjustment for NT-proBNP but remained significant. It might be that a combination of these biomarkers or a more extensive panel of complementary biomarkers will provide the greatest improvement in prediction. However, this approach might be limited by cost, which requires further study.

21Furthermore, the data presented from PROSPER and studies in the current meta-analysis all classified participants on the basis of a single measurement of cardiac troponin. Diurnal variability in cardiac troponin has been observed with moderately higher levels during the morning, and short-term variability was present both in those with and without known coronary artery disease. This biological variation will need to be explored further to inform selection of appropriate cutoffs for use in clinical practice. Evaluations of the prognostic value of changes in cardiac troponin over time have suggested that the trajectory of cardiac troponin might provide additional prognostic information,.

22Recent data from the WOSCOPS (West of Scotland Coronary Prevention Study) trial showed that pravastatin significantly lowered hs-cTnI in a manner consistent with lower CVD risk. This recent paper, along with our summation of available prospective data on troponins and CVD risk, suggests that more research on troponins as markers of clinical utility beyond the diagnosis of an acute coronary syndrome is urgently needed.

23This study provided de novo data from the PROSPER study and was strengthened by concurrent reporting of a comprehensive review of the literature regarding the association between cardiac troponin and incident CVD. Transformation of study-specific HRs to reflect RR between those with cardiac troponin values in the upper third compared with the lower third allowed a common scale to be used when pooling studies and has been used in the assessment of a number of biomarkers. It did, however, assume that the relationship between the biomarker of interest and disease risk was linear. In the meta-analysis, there was significant between-study heterogeneity that could have resulted from several differences between the included studies as discussed earlier. Inherent limitations exist in the use of study-level data in this meta-analysis, and the availability of individual-participant data would allow for further standardization of analytical techniques, covariate adjustment, and subgroup analysis.

24Elevated cardiac troponin concentration, well within the normal range, was associated with an increased risk of incident CVD outcomes in the general population. This association was strongest for fatal CVD and persisted after adjustment for conventional CVD risk factors. Further research on cardiac troponin as a useful marker of risk prediction seems warranted.COMPETENCY IN MEDICAL KNOWLEDGE: Meta-analysis of published studies suggests that people with elevated circulating troponin concentrations in absence of clinically manifest myocardial injury are at a relatively high risk of developing cardiovascular events.TRANSLATIONAL OUTLOOK: Future research is needed to clarify the predictive value of cardiac troponin measurements in relation to conventional cardiovascular risk factors.

25COMPETENCY IN MEDICAL KNOWLEDGE: Meta-analysis of published studies suggests that people with elevated circulating troponin concentrations in absence of clinically manifest myocardial injury are at a relatively high risk of developing cardiovascular events.

26TRANSLATIONAL OUTLOOK: Future research is needed to clarify the predictive value of cardiac troponin measurements in relation to conventional cardiovascular risk factors.