Association of Troponin T Detected With a Highly Sensitive Assay and Cardiac Structure and Mortality Risk in the General Population FREE

Cardiac troponins T (cTnT) and I are the preferred biomarkers for the diagnosis of acute myocardial infarction (MI).^1- 2 Increasingly it has been recognized that elevated troponin levels may be detected in other clinical scenarios in which acute myocardial injury may occur^3- 4 as well as in several chronic disease states, including coronary artery disease (CAD), heart failure, and chronic kidney disease (CKD).^5- 7 Troponins T and I are also occasionally detectable in individuals from the general population using standard assays.^8- 11 Although the prevalence in the population is low, detectable troponin associates strongly with structural heart disease⁸ and an increased risk of death and adverse cardiovascular events.^9- 11 These findings suggest that troponin may be useful for detecting subclinical cardiovascular disease and assessing cardiovascular disease risk in the general population; however, the low prevalence of detection with standard assays would limit the utility of troponin measurement for these clinical applications.⁸

Recently, a highly sensitive assay for cTnT has been developed that detects levels approximately 10-fold lower than those detectable with the standard assay. In patients with suspected acute coronary syndromes, this assay improves accuracy for the diagnosis of MI compared with the standard cTnT assay.¹² In patients with chronic heart failure¹³ and chronic CAD,¹⁴ circulating cTnT is detectable in almost all individuals with the highly sensitive assay, and higher levels correlate strongly with increased cardiovascular mortality. Here, we report an evaluation of the highly sensitive assay in a general population cohort with detailed cardiovascular phenotyping and long-term follow-up for mortality.

The Dallas Heart Study (DHS) is a multiethnic, population-based cohort study of Dallas County residents. The study was approved by the University of Texas Southwestern institutional review board, and all participants provided written informed consent. Details of the study design and participant selection have been described previously.¹⁵

The random probability sample included intentional oversampling of black individuals to comprise 50% of the cohort. Enrollment occurred between July 2000 and September 2002, and data collection was performed in 3 phases, beginning with an initial in-home visit (n = 6101) to collect demographic information, medical history, blood pressure, and anthropometric measurements. Participants aged 30 to 65 years were asked to participate in a second in-home visit (n = 3557) to collect fasting blood and urine samples and then a final visit at University of Texas Southwestern Medical Center (n = 2971) where dual-energy x-ray absorptiometry (DEXA) scanning for body composition, detailed cardiovascular phenotyping by electron-beam computed tomography (EBCT), and cardiac and aortic magnetic resonance imaging (MRI) were performed.

Sampling weights were calculated for each participant to reflect differential probabilities of selection and attrition of participants between visits. These sampling weights permit the generation of unbiased estimates of population frequencies in Dallas County (additional methods presented in the eSupplement).¹⁵ The present study includes all 3546 individuals with cTnT levels measured using the highly sensitive assay, including 2501 with cardiac MRI and 2770 with EBCT.

Blood was collected into EDTA tubes, refrigerated at 4°C for 4 hours or less, centrifuged, and the plasma removed and stored at −70°C. Brain-type natriuretic peptide,¹⁶ N-terminal pro-brain-type natriuretic peptide (NT-proBNP),¹⁶ and high-sensitivity C-reactive protein (CRP)¹⁷ were measured as previously described. Levels of cTnT were measured previously using a standard assay with a lower limit of detection of 0.01 ng/mL (Elecysys-2010 Troponin T; Roche Diagnostics, Indianapolis, Indiana).⁸ For the present study, cTnT levels were measured using a highly sensitive assay on an automated platform (Elecsys-2010 Troponin T hs STAT, Roche Diagnostics), with a lower detection limit of 0.003 ng/mL and a reported 99th percentile value in apparently healthy individuals of 0.014 ng/mL.¹⁸

Detailed methods for MRI, EBCT, and DEXA scans in the DHS have been reported.^19- 21 EBCT was performed in duplicate on a single scanner (Imatron Inc, San Bruno, California) to assess coronary artery calcium (CAC), with results averaged. Scans were scored by the Agatston method and classified as 0 to ≤10, >10 to ≤100, >100 to ≤400, and >400 Agatston units.¹⁹ DEXA was used to estimate body composition, dividing total body mass into bone, fat, and lean mass components.¹⁹ Cardiac and aortic MRI was performed using a 1.5-Tesla system (Intera; Philips Medical Systems, Best, the Netherlands). Left ventricular mass, wall thickness, end diastolic volume (LVEDV), and ejection fraction (LVEF) were calculated from short-axis sequences. Left ventricular hypertrophy (LVH) was defined as left ventricular mass greater than 89 g/m² in women and greater than 112 g/m² in men, based on a phenotypically normal subpopulation of the DHS cohort.²⁰ Details of aortic MRI and compliance methods are presented in the eSupplement.

Race/ethnicity was self-reported. Coronary heart disease (CHD) was defined as self-reported prior MI or revascularization. Cardiovascular disease was defined as CHD or self-reported history of heart failure or stroke. For purposes of categorizing baseline risk, Framingham Risk Score (FRS) estimates of 10-year CHD events were determined using published algorithms²² with a correction for statin treatment,²³ and participants were categorized into low-risk (<10%), intermediate-risk (10%-20%), and high-risk (>20%) categories; participants with diabetes or CHD were categorized into the high-risk group. Other variable definitions are provided in the eSupplement.

The National Death Index was queried to determine participant mortality through December 31, 2007. Deaths were classified as cardiovascular if they included International Statistical Classification of Diseases, 10th Revision codes I00-I99.²⁴

Continuous variables are reported as median (interquartile range) and categorical variables as proportions. The prespecified primary analyses assessed cTnT level as an ordered categorical variable with secondary analyses as a dichotomous variable (detectable vs undetectable levels). For the analyses of cTnT as a categorical variable, participants were divided into 5 a priori–determined categories based on cTnT levels determined using the highly sensitive assay: those with undetectable cTnT were placed in the first category; those with cTnT levels greater than or equal to the previously reported 99th percentile value (≥0.014 ng/mL) were placed in the fifth category, and those with cTnT levels between 0.003 and 0.014 ng/mL were divided into tertiles for categories 2 through 4. Demographic and clinical variables and cardiovascular phenotypes were compared across cTnT categories using the Jonckheere-Terpstra trend test, which is a nonparametric trend test for ordered classes.²⁵ Sensitivity analyses were performed restricted to participants without prevalent cardiovascular disease and those in the lowest FRS category.

Logistic regression was used to identify variables independently associated with detectable cTnT. Candidate variables tested included age, sex, race/ethnicity, diabetes, hypertension, prior MI, heart failure or angina, hospitalization within the past year, estimated glomerular filtration rate, CAC, LVEF, left ventricular mass, lean mass by DEXA, and body surface area. Left ventricular mass was replaced with LVEDV and left ventricular wall thickness in a second model.

All-cause and cardiovascular disease mortality were estimated using the Nelson-Aalen estimator,²⁶ and cumulative incidence curves were compared across cTnT categories using the log-rank test. Multivariable Cox proportional hazards models were used to determine associations of cTnT categories with mortality after serially adjusting for traditional risk factors plus high-sensitivity CRP level, estimated glomerular filtration rate, and NT-proBNP level. Sensitivity analyses were performed replacing cTnT categories with cTnT as a log-transformed continuous variable, with undetectable values assigned a value just below the lower detection limit of the assay (2.99 ng/mL). Time-dependent C statistics²⁷ were calculated for models with and without cTnT and compared using bootstrap resampling, with model fit assessed by the likelihood ratio test and calibration by the Hosmer-Lemeshow statistic. The integrated discrimination index (IDI) was calculated for the addition of cTnT to each model.²⁸ The IDI represents the improvement in average sensitivity minus any increase in 1 − specificity when a variable is added to a model.

Sample weighting was used for determining the prevalence of detectable cTnT in Dallas County and in subsets of the population (eSupplement). For all other analyses evaluating associations within the DHS cohort, no sample weighting was used. All P values are 2-sided; P < .05 was considered statistically significant. No adjustment was made for multiple comparisons. Statistical analyses were performed using SAS version 9.2 (SAS Institute Inc, Cary, North Carolina).

The sample-weight–adjusted prevalence of detectable cTnT in Dallas County adults was 25.0% (95% confidence interval [CI], 22.7%-27.4%) using the highly sensitive cTnT assay and 0.7% (95% CI, 0.3%-1.1%) using the standard assay. Among individuals without cardiovascular disease, CKD, subclinical heart disease, diabetes, or hypertension, the prevalence of detectable cTnT with the highly sensitive assay was 16.2% (95% CI, 13.3%-19.1%) (Table 1). The prevalence of a cTnT concentration of 0.014 ng/mL or greater (the 99th percentile cutpoint for diagnosis of MI) was 2.0% (95% CI, 1.5%-2.6%) in Dallas County adults and 1.9% (95% CI, 1.0%-2.0%) in those without prior cardiovascular disease (Table 1).

The prevalence of detectable cTnT varied markedly with age, ranging from 14.0% (95% CI, 11.2%-16.9%) in participants aged 40 to 50 years to 57.6% (95% CI, 47.0%-68.2%) in those aged 60 to 65 years (P < .001) (Table 1). Large differences in prevalence were also seen according to sex and race/ethnicity: men were 3-fold more likely to have detectable levels than women (37.1% [95% CI, 33.3%-41.0%] vs 12.9% [95% CI, 10.6%-15.2%], P < .001), and black participants had a significantly higher prevalence of detectable cTnT than Hispanic or white participants (P < .001) (Table 1). The prevalence of detectable cTnT stratified by sex and race/ethnicity is shown in eFigure 1.

Two-thirds of participants in the highest cTnT category had undetectable cTnT levels with the standard assay (Table 2). The prevalence of hypertension increased from 27.2% (95% CI, 25.5%-28.9%) to 70.9% (95% CI, 61.8%-79.0%) and prevalence of diabetes from 7.7% (95% CI, 6.7%-8.8%) to 41.0% (95% CI, 32.2%-50.3%) across categories of increasing cTnT levels. Estimated glomerular filtration rate decreased from 99.5 mL/min per 1.73 m² (95% CI, 87.1-114.4) to 88.0 (95% CI, 62.7-107.9) mL/min per 1.73 m² (P < .001 for trend) across categories of higher cTnT levels, but smoking status and levels of low-density lipoprotein cholesterol did not change. Although higher body mass index was modestly associated with increasing cTnT levels, DEXA measurements of body composition demonstrated discordant associations with lean and fat mass: cTnT was positively associated with lean mass but inversely (and weakly) associated with fat mass (Table 2).

Across cTnT categories, left ventricular mass increased markedly, as did left ventricular wall thickness, and the proportion of individuals classified as having LVH increased from 7.5% (95% CI, 6.4%-8.8%) to 48.1% (95% CI, 36.7%-59.6%) (P < .001 for trend). Modest increases in LVEDV and decreases in LVEF were also seen with higher cTnT levels. CAC score and abdominal aortic wall thickness increased while aortic compliance decreased across higher cTnT categories (Table 2). Strong graded associations were also evident between increasing severity of selected pathologic phenotypes and the prevalence of detectable cTnT, including indexed left ventricular mass, LVEF, CAC score, and CKD (eFigure 2).

Self-reported heart failure, CHD, and cardiovascular disease were more frequent with higher cTnT levels (Table 2). In sensitivity analyses excluding participants with cardiovascular disease (eTable 1), similar associations were observed between cTnT categories and demographic variables, risk factors, and cardiac phenotypes. Additionally, when these analyses were restricted to individuals predicted by the FRS to be at low risk (estimated 10-year risk of CHD <10% [n = 2680]), cTnT remained associated with multiple pathological cardiac phenotypes (eTable 2).

In a multivariable logistic regression model with detectable cTnT as the dependent variable, male sex, older age, black race, history of heart failure, lower estimated glomerular filtration rate, and higher left ventricular mass independently associated with detectable cTnT, whereas prior MI or angina, LVEF, CAC score, lean mass, and body surface area did not (C statistic of model, 0.81) (eTable 3). When left ventricular mass was replaced with left ventricular wall thickness and LVEDV, the model performed similarly and both wall thickness and LVEDV remained associated with detectable cTnT.

Over a median follow-up period of 6.4 years (interquartile range, 6.0-6.8 years), 151 total deaths and 62 cardiovascular disease deaths occurred. Unadjusted all-cause mortality occurred in 1.9% (95% CI, 1.5%-2.6%) of participants with undetectable cTnT vs 9.1% (95% CI, 7.4%-11.1%) of those with detectable cTnT (P < .001) and increased from 1.9% (95% CI, 1.5%-2.6%) to 28.4% (95% CI, 21.0%-37.8%) across higher cTnT categories (P < .001) (Figure 1). Similar trends were observed for cardiovascular disease mortality (Figure 1). Detectable cTnT and increasing cTnT categories remained associated with mortality after excluding participants with cTnT detected by the standard assay, as well as those with prevalent CHD or cardiovascular disease. Associations of cTnT level with all-cause and cardiovascular disease mortality were consistent in subgroups defined by sex, race/ethnicity, age, hypertension, and diabetes (Figure 2, eFigure 3, eTable 4, and eTable 5).

In a series of Cox proportional hazards models adjusting for traditional risk factors (model 1) and further adjusting for high-sensitivity CRP level (model 2), higher cTnT categories demonstrated a graded association with all-cause and cardiovascular disease mortality. The addition of cTnT to the baseline models significantly improved the C statistic (0.818 vs 0.793; P = .001), IDI (0.044 [95% CI, 0.025-0.063]; P < .001), and model fit (P < .001) (Table 3). Only modest attenuation of the hazards was seen with further adjustment for CKD categories (model 3). Although more significant attenuation was seen with additional adjustment for NT-proBNP level (model 4), cTnT in the fourth category (hazard ratio [HR], 2.0 [95% CI, 1.2-3.4]) and fifth category (HR, 2.8 [95% CI, 1.4-5.2]) remained independently associated with all-cause mortality in the fully adjusted models, and the addition of cTnT improved model fit (P = .02) and the IDI (0.010 [95% CI, 0.002-0.018]; P = .01) but not the C statistic (0.827 [95% CI, 0.789-0.864] vs 0.821 [95% CI, 0.783-0.859]); P = .12) (Table 3). In contrast, in the same fully adjusted model, cTnT level measured with the standard assay did not associate with all-cause mortality and did not improve model fit, the IDI, or the C statistic (P > .15 for each). NT-proBNP level was associated with all-cause (P < .001) and cardiovascular disease (P < .001) mortality in the fully adjusted model, but high-sensitivity CRP level was not (P = .44 for all-cause mortality and P = .28 for cardiovascular disease mortality). Sensitivity analyses using log-transformed cTnT as a continuous variable yielded qualitatively similar results, including in the fully adjusted all-cause mortality model (HR per log unit change, 1.5 [95% CI, 1.1-1.9]; P = .006).

Circulating cardiac troponin T is detectable in approximately 25% of adults aged 30 to 65 years in the general population using a novel highly sensitive assay, with wide variation in its prevalence according to age, sex, and race/ethnicity. Higher levels of cTnT measured with the highly sensitive assay, well below the detection range of standard assays, are associated with cardiac structural abnormalities including LVH (both left ventricular wall thickening and dilation) and left ventricular systolic dysfunction. These associations were consistent in lower-risk subgroups defined by the absence of known cardiovascular disease or a low FRS category. Moreover, higher cTnT levels demonstrate a graded association with all-cause and cardiovascular disease mortality independent of traditional risk factors, renal function, and levels of other biomarkers such as high-sensitivity CRP and NT-proBNP.

The important contributions of chronic sources of myocardial injury to troponin release have become increasingly evident as troponin measurements have been applied to ambulatory patient populations, particularly with highly sensitive assays. For example, nearly 100% of clinical trial participants with stable chronic heart failure and CAD had detectable cTnT levels using a highly sensitive assay.^13- 14

We applied the highly sensitive assay to a randomly sampled general population cohort with a large proportion of women and ethnic minorities and found that cTnT was commonly detectable in adults at levels well below the detection limit of a standard assay. Although several prior studies have measured levels of troponins in general population cohorts, they have used standard troponin assays that are much less sensitive than the assay used in the present study.^8- 11,29 For instance, the prevalence of detectable cTnT with a standard assay in this same young DHS cohort was only 0.7%.^8,29 The finding that male sex, older age, and black race are associated with higher troponin levels has been suggested previously, but prior studies used less-sensitive assays and had a low rate of detectable troponins, precluding adequately powered multivariable analyses.^{8- 9,29- 30}

In univariable analyses, cTnT was associated with multiple cardiac risk factors, measures of atherosclerosis burden, and cardiac structural phenotypes, most notably left ventricular mass and wall thickness. In multivariable analyses, diabetes, hypertension, worse renal function, and increased left ventricular mass, wall thickness and chamber dilation, but not CAC score, remained independent determinants of detectable cTnT. These findings should be interpreted in the context of prior studies investigating standard troponin assays in elderly individuals from the general population²⁹ as well as in patients with chronic CAD,³¹ which suggest that chronic elevation of troponin levels is mediated to a greater extent by indices of heart failure (such as higher left ventricular mass, lower LVEF, or increased NT-proBNP levels) than indices of atherosclerosis or ischemia. In the Prevention of Events with Angiotensin Converting Enzyme Inhibition (PEACE) trial, which applied a highly sensitive assay to patients with chronic CAD, cTnT was associated with death and heart failure but not MI.¹⁴ In contrast, in patients with acute coronary syndromes, low levels of troponin correlate with angiographic evidence of greater lesion severity and complexity³² and identify patients at greater risk for ischemic complications than for death or heart failure.^33- 34 Taken together, these findings suggest important differences in the pathophysiology of troponin release in the chronic compared with the acute setting.

Prior studies have described associations between increased troponin levels detected with standard assays and future risk for mortality.^9- 11 Here, we report that these associations extend to much lower troponin levels not detected with assays in current clinical use. Indeed, only one-third of the participants in the highest cTnT category had cTnT levels measurable with the standard-generation cTnT assay. Higher cTnT levels associated strongly with future risk for total and cardiovascular disease mortality, and consistent associations were observed in individuals without prior cardiovascular disease and those who would be deemed at low risk based on the FRS. Associations with mortality were only modestly attenuated after adjustment for traditional risk factors, high-sensitivity CRP level, and CKD. Further adjustment for levels of NT-proBNP, which was also associated with total and cardiovascular disease death in this cohort, resulted in more substantial attenuation of the HRs, suggesting that NT-proBNP and cTnT provide partly overlapping information concerning cardiac structural and functional abnormalities. However, even after full adjustment, cTnT in the highest 2 categories remained independently associated with all-cause mortality, with adjusted HRs in a range that would suggest potential clinical utility.³⁵ The addition of cTnT to a variety of risk prediction models increased model fit and the IDI, a metric of discrimination.

The present findings suggest that future studies should be performed to assess whether measurement of cTnT levels with a highly sensitive assay adds value to traditional cardiovascular risk factors as well as measures of renal function and NT-proBNP levels for risk assessment in the general population. In the fully adjusted model, which included NT-proBNP levels and renal function, improvements in the C statistic and the IDI with cTnT were small, and the clinical significance of changes of this magnitude are not clear. Future studies will need to evaluate clinical performance metrics, such as net reclassification improvement, in data sets with long-term follow-up for nonfatal end points, to fully evaluate the incremental value of measurement of cTnT levels. In addition, appropriate diagnostic and therapeutic responses would need to be defined before population-based screening could be recommended. However, our data showing associations with LVH and other cardiac structural abnormalities as well as with mortality among individuals classified at low risk using the FRS suggest that low levels of cTnT may identify subclinical structural heart disease and contributors to cardiovascular disease risk not fully captured by current risk-assessment tools. The independent and additive associations of cTnT and NT-proBNP suggest that combinations of these 2 markers may perform better than either marker alone.

Our finding of a high prevalence of detectable cTnT in the general population has important, and complex, implications for the use of highly sensitive troponin assays for diagnosing acute MI in the hospital setting. Among patients with clinical presentations suspicious for MI, higher-sensitivity assays improve diagnostic sensitivity, particularly early after presentation, but reduce specificity.^12,36 When applied to patients with a high clinical suspicion for MI, the net result is improved accuracy. However, if applied to individuals with a lower likelihood of MI, but with factors associated with higher cTnT levels such as older age, male sex, black race, hypertension, diabetes, CKD, and LVH, the results of the highly sensitive troponin assays will have lower specificity, and false-positive diagnoses of MI will be more common. Because current guidelines recommend more intensive treatment for patients with suspected acute coronary syndromes who have elevated troponin levels,^1- 2 it is possible that widespread application of highly sensitive assays without integrating new approaches to discriminate acute injury³⁷ will expose some patients to unnecessary risk and expense. Additional investigation is needed before the full clinical implications of highly sensitivity troponin assays can be determined.

This study has several strengths. The sampling strategy used for the DHS permits generation of unbiased population estimates of the prevalence of detectable cTnT among Dallas County adults. The availability of cardiac MRI and EBCT data in this large population-based cohort allows detailed cardiac phenotypes to be assessed for their associations with cTnT. The measurement of cTnT by standard as well as highly sensitive assays allows demonstration of the incremental value of the novel assay.

This study also has several limitations that merit comment. First, not all participants completed cardiac imaging studies. Second, adjudicated nonfatal outcome data are not yet available for the DHS; thus, full evaluation of the incremental value of measuring cTnT levels for risk prediction over currently available risk assessment algorithms such as the FRS could not be performed. Third, only a single baseline measurement of cTnT level and other covariates was performed. Finally, the number of cardiovascular deaths was relatively small, limiting statistical power for this outcome.

Using a highly sensitivity assay, cTnT was detectable in approximately 25% of adults in the general population and was associated with structural heart disease and risk of subsequent all-cause mortality. Higher cTnT levels, below the detection range of currently available assays, may be considered a marker of “end organ” cardiovascular damage from a variety of risk factors and pathological cardiac and vascular processes.