Association of Repeatedly Measured High-Sensitivity–Assayed Troponin I with Cardiovascular Disease Events in a General Population from the MORGAM/BiomarCaRE Study

Abstract

BACKGROUND

High-sensitivity troponin I (hs-cTnI) concentrations reflect myocardial stress. The role of hs-cTnI in predicting long-term changes in the risk of cardiovascular disease (CVD) in general populations is not clearly defined.

METHODS

We investigated whether the change in 3 repeated measures of hs-cTnI collected 5 years apart in a prospective Danish study (3875 participants, initially aged 30–60 years, 51% female, disease free at baseline) improves 10-year prediction of incident CVD compared to using a single most recent hs-cTnI measurement. The change process was modelled using a joint (longitudinal and survival) model and compared to a Cox model using a single hs-cTnI measure adjusted for classic CVD risk factors, and evaluated using discrimination statistics.

RESULTS

Median hs-cTnI concentrations changed from 2.6 ng/L to 3.4 ng/L over 10 years. The change in hs-cTnI predicts 10-year risk of CVD (581 events); the joint model gave a hazard ratio of 1.31 per interquartile difference in hs-cTnI (95% CI 1.15–1.48) after adjustment for CVD risk factors. However, the joint model performed only marginally better (c-index improvement 0.0041, P = 0.03) than using a single hs-cTnI measure (c-index improvement 0.0052, P = 0.04) for prediction of CVD, compared to a model incorporating CVD risk factors without hs-cTnI (c-index 0.744).

CONCLUSIONS

The change in hs-cTnI in 5-year intervals better predicts risk of CVD in the general population, but the most recent measure of hs-cTnI, (at 10 years) is as effective in predicting CVD risk. This simplifies the use of hs-cTnI as a prognostic marker for primary prevention of CVD in the general population.

Cardiac troponin is a marker of necrosis and of myocardial infarction in emergency situations (1) but can also act as a long-term biomarker in predicting risk of cardiovascular disease (CVD),¹⁰ heart failure, and death in the general population (2, 3). Cardiac troponin I measured with a high-sensitivity assay (hs-cTnI) can be detected in 80%–90% of the general population (3, 4) and adds predictive information beyond established risk factors for fatal and nonfatal CVD in men and women across a wide age-range (2, 3, 5, 6). As troponin concentrations below the limits of conventional troponin assays can be prognostically important (3), biological distributions and long-term variation of troponin concentrations and their relationship with other risk factors across general populations need to be characterized. Longitudinal studies repeatedly measuring troponin can address this and estimate the prognostic importance of measuring changes for predicting risk of disease. The determinants of temporal increases in troponin (hs-cTnT) over 6 years in a general population (median 57 years) have recently been shown to include increasing age, male sex, hypertension, diabetes, and obesity (7). Changes in troponin over 1 year were associated with increased risk of myocardial infarction in stable coronary heart disease patients (8). In healthy elderly populations (>65 years) changes in troponin over 3–5 years were associated with increased risk of heart failure, cardiovascular and all-cause mortality, and atrial fibrillation (9, 10, 11, 12). However, incorporating this change led to minimal improvement in risk prediction (9, 10, 11). In a healthy general population (mean age 56 years) changes in troponin over 6 years failed to improve prediction of coronary heart disease but improved prediction of heart failure risk (13). No studies have assessed the prognostic implications of changes in troponin in general populations over longer time scales.

Using a prospective population cohort initially aged 30–60 years, which collected hs-cTnI and other risk factors at 3 time points over 10 years with a further 16 years of follow-up for incident CVD events, we determined (a) if change in hs-cTnI is associated with increased risk of CVD, and (b) whether prediction of CVD can be improved by modelling change in hs-cTnI. We developed prognostic models comparing the trend in hs-cTnI over 10 years to a single most recent hs-cTnI measure.

Methods

STUDY DESIGN

The MONICA 1 population cohort at the RCPH (Research Centre for Prevention and Health) represents 11 municipalities in Copenhagen, Denmark. Men and women aged 30, 40, 50, and 60 years were randomly sampled from the national population register (14). They were examined in 1982–1984 [Round 1 (R1)], then reexamined in 1987–1988 [Round 2 (R2)] and again in 1993–1994 [Round 3 (R3)] (14) (Fig. 1). Participants received a physical examination, a self-administered questionnaire, and a blood sample was drawn at each exam. Smoking status, blood pressure (BP), body mass index (BMI), and blood lipids were measured in a standardized way. Prevalent diabetes was defined as self-reported doctor-diagnosed diabetes, use of diabetes medication or diabetes history recorded in registry data at each round. The study was approved by ethics committees; participants consented to all examinations and follow-up of their medical records. At recruitment, preexisting cardiovascular disease (myocardial infarction or stroke) was self-reported. The outcome was the first major cardiovascular event (including first fatal or nonfatal definite or possible myocardial infarction, coronary death or unclassifiable death, unstable angina, cardiac revascularization, and probable ischemic stroke). Follow-up was achieved through linkage to the National Cause of Death Register and National Hospital Discharge Register until December 2009 with only 40 participants (1.05%) lost to follow-up (15, 16). Registry diagnoses have been validated against the MONICA criteria (17).

Study design of the prospective Danish cohort.

BIOMARKER MEASUREMENTS AND ADJUSTMENTS

Serum was separated from blood, then stored at 4 °C (1–3 days) during transfer to the laboratory for storage at −20 °C and subsequently at −80 °C. Storage times are summarized in Table 1 in the Data Supplement that accompanies the online version of this article at http://www.clinchem.org/content/vol63/issue1. Biomarkers were measured in the MORGAM/BiomarCaRE laboratory. hs-cTnI concentrations were determined by the ARCHITECT STAT high sensitivity Troponin I immunoassay (Abbott Diagnostics, ARCHITECT i2000SR). The limit of detection (LOD) was 1.9 ng/L, with values below this imputed (see below). The assay supported a 10% coefficient of variation at a concentration of 5.2 ng/L. Intraassay and interassay CVs at this concentration were 4.3% and 6.3%. Because of skewed distributions, a cubic root transformation was applied to hs-cTnI for consistency with previous research (2, 3).

STATISTICAL ANALYSIS

Missing data occurred across the 3 rounds either due to participants failing to attend examination, insufficient information on risk factors, or insufficient serum for a biomarker test. However, their vital status was followed up. Participation rates across the rounds are given in online Supplemental Table 2. Missing data for risk factors were minimal (<0.1%) for those attending exams. The missing data rate for biomarkers ranged from 0.15% to 8.8% and for hs-cTnI was 4.8% at R1, 8.8% at R 2 and 0.51% at R3. Missing data were addressed through a multiple imputation (MI) model which captured the longitudinal trajectory of hs-cTnI and other variables. Imputed values were applied to the risk prediction analysis models. Individuals recruited at R1 were included in the imputation model. The MI model for men and women combined included classic CVD risk factors, biomarkers, and case status at the start and end of the follow-up. Twenty imputed datasets were created using chained equations (18). Predictive mean matching was used for all variables, except hs-cTnI, where a normal linear regression model was used. When the missing values corresponded to values below the LOD the imputed values were drawn from a truncated normal distribution, otherwise the imputed values were drawn from a normal distribution. Time-to-event information was included in the imputation model (19) with further details in the online Supplemental Material. A sensitivity analysis was performed on a restricted dataset of those study participants with hs-cTnI and risk factors measured at R3, who may have had some missing measures before that round, which were imputed (see online Supplemental Methods).

PREDICTION MODELS

Prediction modelling is outlined in Fig. 1. We developed prognostic models comparing the trend in hs-cTnI over 10 years to a single most recent hs-cTnI measure. We needed to keep the measurement period (10 years of change) separate from the follow-up period because prognostic models cannot use “future measures” (i.e., R2 or R3 in this case) as predictors; therefore a new baseline was set at 10 years that became the starting point for predictions using a further 16 years of follow-up after this point.

A Cox proportional hazards model was constructed containing cardiovascular risk factors: sex, systolic BP (SBP), HDL-cholesterol, non-HDL cholesterol (difference between total and HDL cholesterol), prevalent diabetes, smoking status, and BMI measured at R3 (Model 1). The R3 measurements were the baseline for the follow-up and age was used as the time-scale in the analysis. The follow-up extended up to 16 years later. To this model, hs-cTnI measured at R3 was added (Model 2). Hazard ratios (HRs) are reported per 0.4 times the cubic root of the hs-cTnI concentration in ng/L which corresponds roughly to the interquartile range in the cohort.

Model 2 was compared to models 3–5, which incorporated the history of hs-cTnI and the cardiovascular risk factors measured in rounds 1–3. Model 3 was similar to model 2 but used the change in hs-cTnI from rounds 1–3 as the predictor instead of the concentration of hs-cTnI at R3. Those developing CVD before R3 were excluded from analysis.

Model 4 combined analysis of the repeated measurements and time to cardiovascular event. It was a joint model (JM) of the association between the repeated measurements at rounds 1–3 and the risk of CVD (20). The model formed 2 parts, a multilevel model for longitudinal trend in hs-cTnI and a proportional hazards survival model for cardiovascular events (see online Supplemental Equation). The multilevel model included a random intercept reflecting a different starting value of hs-cTnI for each participant and its relationship with other covariates, and a random slope for time from baseline. Sex, age at the first examination (measuring the cross sectional effect of age), and time from baseline (measuring the longitudinal effect of age) were fixed effects in this model. The survival model incorporated: sex, SBP, HDL cholesterol, non-HDL cholesterol, prevalent diabetes, smoking status, and BMI, updated at each round. hs-cTnI was included in the survival model via the estimate from the multilevel model. The JM was fitted to all follow-up data but the validation step involving Model 4 used only the follow-up data after R3 (see below).

We also fitted a Cox model with time-dependent covariates using the same covariates as Model 4 and using the same timespan and number of events (Model 5). All survival models used age as the time scale (21, 22). For the Cox models, those developing CVD before R3 were excluded. For the JMs information available after an individual developed CVD was not used. Prevalent cases of CVD at the first examination (2.18%, n = 83) were excluded from the analyses.

FITTING AND VALIDATION OF THE PREDICTION MODELS

Prediction models 1–5 were evaluated with the c-index (23) and net reclassification index (NRI) (24) using the 10 years probabilities of CVD derived from the models using R3 as the baseline. 10-fold cross validation was used to adjust the risk estimates for over optimism in assessing model performance on the same dataset where it was developed (25, 26). The analyses were carried out in both sexes combined, and performed using R v3.1.1 [R Core Team (2013)] (27). Our study conforms to TRIPOD (Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis) guidelines for reporting prediction models (28).

Results

PARTICIPANT CHARACTERISTICS

Participant characteristics for each examination round are given in Table 1 (imputed data) and online Supplemental Table 1 (complete case data). From R1 to R3, the prevalence of diabetes increased from 2.2% to 4.3%, smoking decreased from 46.7% to 39.5%, BP treatment increased from 5.8% to 12.5%, and systolic BP increased from 123.3–129.9 mmHg.

hs-cTnI ranged from 0 to173.0 ng/L at R1 to 0.3 to 164.6 ng/L at R3. Median hs-cTnI concentrations changed from 2.6 ng/L to 3.6 ng/L to 3.4 ng/L across the 3 rounds, with overall changes illustrated in Fig. 2. hs-cTnI was above the LOD (>1.9 ng/L) in 68.2% of participants at R1 (61.8% men, 38.2% women), 90.4% at R2 (50.8% men, 49.2% women), and 84.9% at R3 (55.1% men, 44.9% women). Ninety-five percent of the changes in hs-cTnI from R1 to R3 were relatively small, between −3.5 and 6.48 ng/L (see online Supplemental Table 3). Higher concentrations of hs-cTnI were observed in men, but over time hs-cTnI increased in both sexes. hs-cTnI values increased over time from R1 to R2 in the full cohort and continued to increase in CVD cases, reaching a plateau in noncases from R2 to R3 (see online Supplemental Fig. 1).

Spaghetti plot visually illustrating the change in hs-cTnI over time.

Spearman correlation coefficients for hs-cTnI from R1 to R3 ranged from 0.59–0.69. (R1 to R2 P = 0.60, R2 to R3 P = 0.69, R1 to R3 P = 0.59). This range was similar although lower than correlations observed for SBP across 3 rounds (R1 to R2 P = 0.76, R2 to R3 P = 0.75, R1 to R3 P = 0.67) or HDL cholesterol (range 0.75–0.81). Correlations for BMI across rounds were higher (range 0.85–0.91). For the prediction models, the number of participants for each risk set after exclusions is given in online Supplemental Table 4.

ASSOCIATION OF HISTORY OF CHANGE IN hs-cTnI CONCENTRATIONS AND CVD

During the follow-up period after R3 (1993–1994, median follow-up 16.6 years), 444 participants had a CVD event. hs-cTnI at R3 was a strong predictor of risk of CVD (HR 1.18) per interquartile difference in the cubic root of hs-cTnI (95% CI 1.08–1.30; P <0.001) after adjustment for cardiovascular risk factors (Model 2 in Table 2 and online Supplemental Table 5). The 10-year change in hs-cTnI (difference R3-R1), after adjustment for risk factors was also positively associated with CVD (HR 1.16; 95% CI 1.02–1.31; P = 0.023) (Model 3 in Table 2 and online Supplemental Table 5).

ASSOCIATION OF LONGITUDINAL TREND IN hs-cTnI CONCENTRATIONS WITH CVD

During the follow-up after R1 (median follow-up 27.5 years) 581 incident CVD cases were observed. After risk factor adjustment, hs-cTnI was positively associated with CVD in the JM, with an HR of 1.31 (95% CI 1.15–1.48), per interquartile difference in the cubic root; P <0.001 (Model 4 in Table 2 and Table 3). The Cox model with time dependent covariates yielded an HR of 1.22 (95% CI 1.12–1.32) P <0.001 (Model 5 in Table 2 and online Supplemental Table 5). Both these models account for changing levels of cardiovascular risk factors and hs-cTnI at each round.

According to the longitudinal part of the JM, hs-cTnI increased with age at R1 by 0.007 (95% CI 0.006–0.008, P <0.001) cube units and with time from baseline (coefficient 0.016, 95% CI 0.014–0.017, P <0.001). hs-cTnI was higher in men than women by 0.226 (CI 0.205–0.247, P <0.001) cube units (Table 3).

PREDICTION MODELLING FOR TROPONIN

The probabilities of 10-year risk of CVD were estimated from the Cox models based on 444 CVD events. The c-index for model 1 with cardiovascular risk factors was 0.744 (95% CI 0.717–0.772). Adding a single most recent hs-cTnI measurement at R3 (i.e., Model 2) to this, marginally improved prediction (c-index 0.750, improvement 0.0052 P = 0.043) (Table 4). Model 3, replacing the last hs-cTnI measurement with the 10-year change in hs-cTnI, did not improve prediction upon the single measure of hs-cTnI (Model 2). Model 4, the JM incorporating the longitudinal trend in hs-cTnI and risk of CVD, improved prediction but only marginally when compared to Model 2. The c-index for Model 4 was 0.754 (95% CI 0.726–0.782), improvement 0.004 P = 0.03) compared to Model 2 (Table 4). The continuous NRI measure for Model 4 was 0.23 (P <0.001) when compared to Model 2.

SENSITIVITY ANALYSIS

Sensitivity analysis based on those with hs-cTnI and risk factors measured at R3 (n = 2339) are presented in online Supplemental Tables 6–7. After risk factor adjustment, hs-cTnI was positively associated with CVD in the JM with a HR of 1.30 (95% CI 1.15–1.47) per interquartile difference in cubic root of hs-cTnI (see online Supplemental Table 6). The addition of hs-cTnI measured at R3 to cardiovascular risk factors (Model 2) failed to improve prediction, based on 304 CVD events, (c-index improvement 0.004, P = 0.089), when compared to Model 1, whose c-index was 0.755 (see online Supplemental Table 7). None of the models incorporating change in hs-cTnI improved prediction when compared to the model with a single measure of hs-cTnI at R3. (see online Supplemental Table 7). See online Supplemental methods for further details.

Discussion

We questioned whether monitoring changes in troponin over time in general populations could lead to better prediction of future CVD events compared to a single measure of hs-cTnI. We focused on 10-year risk of CVD, an established metric in CVD risk scores, so we could make a realistic comparison between a single measure to longitudinal measures. We tested this in a longitudinal cohort with clear separation between measurement period (10 years of change in hs-cTnI) and follow-up period used for prediction (16 years) as prognostic models cannot use “future” measures as predictors. Therefore a new “baseline” at 10 years became the starting point for predictions. Our study had longer intervals of measurement than comparable studies (9, 10) and a larger number of events (9); we used a joint modelling approach which more sensitively monitored the change in hs-cTnI (Model 4) than incorporating change as a predictor in the models (Model 3). We found that hs-cTnI increases over time in the general population and the change in hs-cTnI was associated with increased risk of fatal and nonfatal CVD. Our findings applied across a wide age range including a younger group, whereas previous findings for association of change in hs-cTnI or hs-cTnT with increased risk of heart failure, cardiovascular death and all-cause mortality were evident in older groups (>65 years) (9, 10) and a middle-aged group for coronary heart disease (13). We found 3 measures of hs-cTnI can be better than 1 in characterizing risk of CVD and improving prediction. In terms of risk prediction in the general population, however, the magnitude of the gain was minimal and offered no significant practical advantage over a single most recent measure of hs-cTnI for long-term prediction of CVD risk.

The associations noted show that change in hs-cTnI is significantly linked with cardiovascular disease, confirming previous reports of association with cardiovascular mortality (10) and coronary heart disease (13). Rather than categorizing troponin as these studies did, which may reduce precision and power (29), we used a continuous measure. Our associations were significant and had tighter confidence intervals based on a sample of n = 3178 compared to n = 1797 (10) or n = 3448 (13). Moreover the HR of continuous hs-cTnI was not directly comparable to the HRs of categorical variables. The associations showed that the change in hs-cTnI incorporated to the JM had a highly significant independent effect if added after other updated risk factors, highlighting its prognostic value independently of other risk factors. We quantified the impact of change in hsTnI compared to change in systolic BP on prediction of CVD (see online Supplemental Methods). The change in SBP contributed twice as much to prediction of CVD as change in hsTnI (measured in relative contributions to the JM), which is unsurprising because SBP is a key modifiable risk factor for CVD; however, hsTnI eclipsed some established risk factors such as HDL cholesterol and BMI in its impact. hs-cTnI and change in hs-cTnI offer the potential for identifying individuals at higher risk of CVD events but not currently identified by established risk factors.

Our statistical approach was sensitive enough to detect significant improvements in risk prediction by monitoring changes in hs-cTnI compared to a single measure. JM sensitivity may be a result of including the CVD outcome when estimating the longitudinal hs-cTnI trend, which avoids diluting the association between hs-cTnI and CVD, whereas Cox time-dependent models tend to underestimate the true risk for hs-cTnI (30, 31). However, median hs-cTnI concentrations were relatively low from 2.6 ng/L to 3.4 ng/L over 10 years, but followed predictable trends. Concentrations frequently changed (7, 9, 10, 32), although most patients had concentrations below diagnostic cutoffs for myocardial infarction (4, 7, 33). hs-cTnI concentrations across rounds were correlated and similar to other cardiovascular risk factors such as systolic BP but lower than BMI suggesting strong tracking over time. Detectable hs-cTnI concentrations increased from 68% at R1–84.9% at R3, consistent with levels in other general populations (2, 3, 4) and in a previous longitudinal study (7). hs-cTnI concentrations increased with age and it may be that the single last measure of hs-cTnI was sufficient for optimal prediction because this represented a time when hs-cTnI could become prognostically more relevant as the population aged. Other studies have found minimal benefit of incorporating change in hs-cTnI/hs-cTnT to prediction but have taken different statistical approaches and outcomes precluding a direct comparison. McEvoy et al. (13) compared a model with CVD risk factors incorporating the δ change in TnT from 2 measurements, 6 years apart, n = 3448, including 622 CHD events. They found the addition of change in hsTnT offered no significant gain in discrimination (c-index difference 0.0004, P value 0.1). This is consistent with our adjusted model including δ change in hs-cTnI (Table 4, c-index difference −0.003, P = 0.11) although the models are not directly comparable given the different troponin measurements and statistical approaches. However, they did find that incorporating the change in TnT improved prediction of heart failure and all-cause mortality (13). In older cohorts minimal predictive benefits have also been found. deFilippi et al. incorporated relative change in hsTnT over 2 years as a covariate in a Cox model marginally improved classification [IDI (integrated discrimination improvement)] for heart failure and cardiovascular mortality but not c-index in an elderly cohort with approximately 8 years follow-up (10). Eggers et al. incorporated relative change in hs-cTnI over 5 years as a time-dependent covariate in a Cox regression with approximately 4 years follow-up, finding that this failed to improve discrimination of cardiovascular mortality (n = 32 events) in an elderly cohort (>65 years) (9). Taking a similar approach in this cohort, the change in hs-cTnI was found to be less strongly associated with cardiovascular events (n = 163) than the change in GDF15 (growth differentiation factor 15) or NTProBNP (N-terminal pro brain natriuretic peptide) (11). The change in hs-cTnI was compared against baseline hs-cTnI or not formally compared prognostically against a single last measure (9, 10, 11). Our investigation, by comparison, has unequivocally tested whether the change in troponin adds prognostic information to 10-year risk models compared to a single hs-cTnI measure and has laid a framework for testing in other cohorts.

Missing data are a study limitation for longitudinal studies and it is difficult to distinguish data missing at random from missing not at random. MI maximized the inclusion of relevant information for the prediction models. Omitting information from study participants with missing data can attenuate associations and cause bias (34), as we observed in a sensitivity analysis when we restricted the dataset to those available at R3. We cannot exclude the possibility that sample degradation may have affected the biomarker measurements leading to overestimation of the change in hs-cTnI. However, previous studies have found that hs-cTnI assessment is robust to long-term storage (3) and if samples deteriorated at the same rate it should not affect the predictive value of hs-cTnI in the models even if absolute concentrations are affected. Changes in medication use over 26 years may have affected the predictive value of hs-cTnI over time but little empirical data on such effects exist.

In summary, although a change in hs-cTnI improved prediction, it did not substantially improve estimates beyond a single most recent measure of hs-cTnI. A single measure of hs-cTnI is sufficient for 10-year prediction of CVD risk, simplifying the use of hs-cTnI as a stable prognostic marker for primary prevention of CVD and endorses prediction models of hs-cTnI for 10-year risk of CVD derived from other larger general population studies (2, 3). However, other factors such as physical activity (35), hyperglycemia (7, 36) and renal function (37) can influence changes in hs-cTnI. Further work will examine the potential for hs-cTnI to monitor changes in risk for the selection of higher risk patients and targeting of therapy (6, 38).

10 Nonstandard abbreviations

 
• CVD

  cardiovascular disease

 
• hs-cTnI

  high sensitivity cardiac troponin I

 
• cTnT

  cardiac troponin T

 
• BMI

  body mass index

 
• R1

  round 1

 
• R2

  round 2

 
• R3

  round 3

 
• BP

  blood pressure

 
• LOD

  limit of detection

 
• MI

  Multiple Imputation

 
• SBP

  systolic BP

 
• HR

  hazard ratio

 
• JM

  joint model

 
• NRI

  net reclassification index.

Acknowledgments

Annex: Sites and Key Personnel of the Contributing MORGAM Centres: Denmark: T. Jørgensen (head), J. Vishram, A. Borglykke, C. Agger. Finland: MORGAM Data Centre–National Institute for Health and Welfare, Helsinki: K. Kuulasmaa (head), Z. Cepaitis, A. Haukijärvi, M. Niemelä, L. Paalanen, O. Saarela, T. Palosaari. Germany: BiomarCaRE Biomarker Laboratory–University Heart Centre Hamburg: S. Blankenberg, T. Zeller. MORGAM Management Group: K. Kuulasmaa (chair Helsinki, Finland), S. Blankenberg (Hamburg, Germany), M. Perola (Helsinki, Finland), M. Ferrario (Varese, Italy), A. Evans (former chair, Belfast, UK), F Kee (Belfast, UK), A. Peters (Munich, Germany), V. Salomaa (Helsinki, Finland), D.-A Trègöuet (Paris, France), H. Tunstall-Pedoe (Scotland, UK).

References

Author notes