Abstract
The relation of high-sensitivity cardiac troponin I (hs-cTnI) concentration and presence of obstructive coronary artery disease (CAD) in patients without myocardial infarction (MI) is unclear. Study participants selected from individuals free of MI undergoing coronary angiography with or without intervention were enrolled and hs-cTnI measured. A gradient boosting model (GBM) was implemented to build a model for detection of CAD. Cox proportional hazard regression was used to assess the association of hs-cTnI and adverse cardiovascular outcome. Among 978 study participants, 607 (62%) patients had CAD. Higher concentrations of hs-cTnI were associated with chronic kidney disease, heart failure, CAD, male sex, current tobacco use, anemia, age and LDL-cholesterol. Prior history of CAD, male sex, type 2 diabetes mellitus, hs-cTnI, anemia, age and HDL- cholesterol were the most influential factors for detection of CAD. The GBM had an area under the curve of 0.82, accuracy of 75%, sensitivity of 88%, specificity of 52%, positive predictive value of 76% and negative predictive value of 72% for detection of CAD. Increase in one log unit of hs-cTnI was significantly associated with increased risk of incident MI (HR =1.34, 95% CI =1.22–1.47, P <.001), CV mortality (HR = 1. 24, 95% CI = 1.11–1.39, P <.001), and composite of incident MI or CV mortality (HR=1.29, 95% CI = 1.20–1.40, P <.001). In conclusion, among individuals without acute MI and CAD, higher concentrations of hs-cTnI was associated with the presence of CAD and linked to increased risk of future cardiovascular events.
Keywords: Myocardial injury, High-sensitivity troponin I, machine learning, model
Measurement of cardiac troponin has transformed the evaluation and management of patients with suspected or proven acute myocardial infarction (MI) 1. However, cardiac injury, identified by high sensitivity cardiac troponin (hs-cTn) above the 99th percentile of a normal population may be present among those without acute MI 2; in this setting, concentrations of cTn reflect cardiomyocyte necrosis from a wide array of non-ischemic cardiac diseases 3. Notably, an hs-cTn above the 99th percentile in this context may indicate an increased risk of future cardiac events, including incident MI 4–8. This implies potential links between abnormal hs-cTn to presence and severity of obstructive coronary artery disease (CAD) in those without prevalent MI 9, however more information regarding the optimal interpretation of myocardial injury is needed 10. One major limitation to studies of hs-cTn interpretation is the lack of an angiographic gold-standard against which an abnormal result may be evaluated, particularly when an abnormal result in an individual without an acute coronary syndrome is encountered. Such information would help clinicians view hs-cTn assay results not only in the context of MI, but as reflecting multiple pathophysiologic pathways of myocardial injury 11. In the present analysis, we evaluated a widely-utilized, validated hs-cTnI relative to its associations with CAD in those without acute MI.
Methods
The design of the Catheter Sampled Blood Archive in Cardiovascular Diseases (CASABLANCA) study has been described previously 12. Briefly, 1251 patients undergoing coronary and/or peripheral angiography with or without intervention between 2008 and 2011 were prospectively enrolled at the Massachusetts General Hospital in Boston, MA. Patients were referred for angiography for various acute and non-acute indications including acute coronary syndromes, HF, abnormal stress tests, stable chest pain, claudication, and routine pre-operative evaluation. After excluding patients who underwent peripheral angiography only and those with acute MI (including those previously diagnosed with unstable angina but who had an abnormal hs-cTn), our final study cohort for this analysis consisted of 978 patients (Figure 1).
Figure 1:
CONSORT Diagram for the present analysis.
Medical record review from time of enrolment to end of follow-up was performed. Median follow-up was 3.67 years with a maximum follow-up of 8 years. For identification of clinical endpoints, review of medical records as well as phone follow-up with patients and/or managing physicians was performed. The Social Security Death Index and/or postings of death announcements were used to confirm vital status. A detailed definition of endpoints for CASABLANCA was previously published 12. Specific to this analysis, end-point adjudication was performed using the guidance of the Universal Definition of Myocardial Infarction.
A total of 15 mL of blood was obtained immediately before the angiographic procedure through a centrally placed vascular access sheath. The blood was immediately centrifuged for 15 min, serum and plasma aliquoted on ice, and frozen at −80 °C until biomarker measurement. The samples for this study were analyzed after the first freeze–thaw cycle for baseline biomarker values only. Specific to this study, we measured concentrations of hs-cTnI (Abbott Diagnostics, Abbott Park, IL); we used sex-specific cut-offs of 16 ng/L for females and 34 ng/L for males to define ≥ 99th percentile hs-cTnI concentration13. Furthermore, we categorized participants based on hs-cTnI concentration into three Low/Intermediate/High risk category based on sex-specific cut-offs (For males: Low < 6 ng/L, Medium 6–12 ng/L, High > 12 ng/L. For females: Low < 4 ng/L, Medium 4–10 ng/L, High > 10 ng/L). Other markers considered in this analysis include kidney injury marker-1 (KIM-1; Singulex Inc, Alameda CA), high sensitivity c-reactive protein (hs-CRP; Siemens, Newark, DE), N-terminal prohormone of B type natriuretic peptide (NT-proBNP; Siemens Inc, Newark DE), low density lipoprotein-cholesterol (LDL-C), high density lipoprotein-cholesterol (HDL-C), and soluble ST2 (sST2; Critical Diagnostics, San Diego, CA).
The CASABLANCA participants selected for this analysis consisted of 978 individuals. Baseline characteristics across CAD categories (none to minimal CAD 0–29%, mild CAD 30–49%, moderate CAD 50–69%, and obstructive CAD ≥ 70%) were compared. For continuous variables, an ANOVA test was used if approximately normally distributed and a Kruskal-Wallis test was used if non-normally distributed. For categorical variables, Pearson’s chi-squared test was used if all expected cell counts were >5 and Fisher’s exact test was used otherwise (Table 1 and Table 2). The concentration values for all biomarkers were log-transformed. Overall, 2.62% of the data were missing; we used Multivariate Imputation via Chained Equations (MICE) package in R for data imputation. To assess association with CAD ≥70%, first, we conducted logistic regression with the least absolute shrinkage and selection operator (LASSO) penalization to perform predictor selection, which can help reduce the dimensions of a prediction model. To determine the penalty factor (lambda), a tenfold cross-validated error plot of the LASSO model was constructed. The optimal lambda was determined by choosing the most regularized and parsimonious model within 1 standard error from the minimum. In addition to the traditional multivariable logistic regression model, various machine learning approaches to detect CAD >70% were explored, including random forest, gradient boosting model (GBM), adaptive index modeling (AIM), conditional forest, and elastic net modeling. The data was first split into a train set (80%) for model building and a test set (20%) for evaluating model performance. To assess the performance of these models, area under the curve (AUC), accuracy, sensitivity, specificity, PPV, and NPV were calculated.
Table 1:
Patient characteristics by level of obstructive CAD
| Variable | Non-Obstructive CAD 217 (22.2%) |
CAD (30–49%) 90 (9.2%) |
CAD (50–69%) 64 (6.5%) |
CAD (≥70%) 607 (62.1%) |
Overall 978 (100%) |
P value* |
|---|---|---|---|---|---|---|
|
| ||||||
| Age, year | <0.001 | |||||
| Mean (SD) | 62.1 (11.3) | 66.1 (11.2) | 68.2 (10.7) | 67.1 (11.1) | 65.9 (11.3) | |
| Sex | <0.001 | |||||
| Female | 96 (44.2%) | 34 (37.8%) | 26 (40.6%) | 117 (19.3%) | 273 (27.9%) | |
| Male | 121 (55.8%) | 56 (62.2%) | 38 (59.4%) | 490 (80.7%) | 705 (72.1%) | |
| Hypertension | 131 (60.4%) | 64 (71.1%) | 47 (73.4%) | 476 (78.4%) | 718 (73.4%) | <0.001 |
| Type II Diabetes Mellitus | 29 (13.4%) | 14 (15.6%) | 11 (17.2%) | 178 (29.3%) | 232 (23.7%) | <0.001 |
| Current Smoker | 27 (12.4%) | 13 (14.4%) | 8 (12.5%) | 79 (13.0%) | 127 (13.0%) | 0.97 |
| Missing | 3 (1.4%) | 1 (1.1%) | 0 (0%) | 6 (1.0%) | 10 (1.0%) | |
| History of CAD | 36 (16.6%) | 31 (34.4%) | 33 (51.6%) | 399 (65.7%) | 499 (51.0%) | <0.001 |
| CKD | 13 (6.0%) | 5 (5.6%) | 7 (10.9%) | 88 (14.5%) | 113 (11.6%) | 0.002 |
| Prior Angioplasty | 3 (1.4%) | 8 (8.9%) | 5 (7.8%) | 87 (14.3%) | 103 (10.5%) | <0.001 |
| Anemia | 44 (20.3%) | 14 (15.6%) | 18 (28.1%) | 189 (31.1%) | 265 (27.1%) | 0.002 |
| Missing | 51 (23.5%) | 14 (15.6%) | 11 (17.2%) | 102 (16.8%) | 178 (18.2%) | |
| Race | 0.03 | |||||
| White | 194 (89.4%) | 85 (94.4%) | 62 (96.9%) | 576 (94.9%) | 917 (93.8%) | |
| Black | 14 (6.5%) | 2 (2.2%) | 0 (0%) | 8 (1.3%) | 24 (2.5%) | |
| Asian/Pacific | 2 (0.9%) | 0 (0%) | 0 (0%) | 7 (1.2%) | 9 (0.9%) | |
| Hispanic | 4 (1.8%) | 2 (2.2%) | 1 (1.6%) | 12 (2.0%) | 19 (1.9%) | |
| Native American | 0 (0%) | 0 (0%) | 1 (1.6%) | 1 (0.2%) | 2 (0.2%) | |
| Other/Unknown | 3 (1.4%) | 1 (1.1%) | 0 (0%) | 3 (0.5%) | 7 (0.7%) | |
| hs-cTnI (ng/L) | <0.001 | |||||
| median (Q1-Q3) | 3.2 (1.5 – 6.0) | 3.9 (2.1 – 7.6) | 3.4 (2.0 – 8.8) | 4.7 (2.3 – 11) | 4.1 (2.1 – 9.3) | |
| Missing | 0 (0%) | 2 (2.2%) | 0 (0%) | 1 (0.2%) | 3 (0.3%) | |
| KIM-1 (ng/ml) | <0.001 | |||||
| median (Q1-Q3) | 120 (80 – 180) | 120 (91 – 180) | 180 (120 – 260) | 160 (100 – 260) | 150 (97 – 230) | |
| Missing | 0 (0%) | 1 (1.1%) | 0 (0%) | 2 (0.3%) | 3 (0.3%) | |
| Cystatin C (mg/dl) | 0.05 | |||||
| median (Q1-Q3) | 0.75 (0.66 – 0.91) | 0.76 (0.68 – 0.89) | 0.80 (0.67 – 0.92) | 0.79 (0.69 – 0.99) | 0.78 (0.68 – 0.96) | |
| Missing | 55 (25.3%) | 19 (21.1%) | 17 (26.6%) | 122 (20.1%) | 213 (21.8%) | |
| hsCRP (mg/L) | 0.93 | |||||
| median (Q1-Q3) | 2.4 (1.0 – 5.2) | 2.8 (1.1 – 4.9) | 2.3 (1.0 – 5.7) | 2.6 (1.0 – 5.6) | 2.5 (1.0 – 5.3) | |
| Missing | 13 (6.0%) | 3 (3.3%) | 1 (1.6%) | 29 (4.8%) | 46 (4.7%) | |
| NT-proBNP (pg/ml) | 0.28 | |||||
| median (Q1-Q3) | 1400 (440 – 3500) | 1300 (500 – 4100) | 2200 (620 – 5600) | 1400 (510 – 3700) | 1400 (510 – 3800) | |
| Missing | 1 (0.5%) | 1 (1.1%) | 0 (0%) | 1 (0.2%) | 3 (0.3%) | |
| LDL Cholesterol (mg/dl) | <0.001 | |||||
| median (Q1-Q3) | 89 (73 – 110) | 78 (64 – 98) | 75 (61 – 100) | 76 (60 – 96) | 79 (63 – 100) | |
| Missing | 15 (6.9%) | 3 (3.3%) | 1 (1.6%) | 32 (5.3%) | 51 (5.2%) | |
| HDL Cholesterol (mg/dl) | <0.001 | |||||
| median (Q1-Q3) | 49 (38 – 58) | 47 (38 – 59) | 49 (36 – 61) | 41 (33 – 50) | 43 (34 – 53) | |
| Missing | 13 (6.0%) | 3 (3.3%) | 1 (1.6%) | 29 (4.8%) | 46 (4.7%) | |
| sST2 (ng/mL) | 0.86 | |||||
| median (Q1-Q3) | 36 (25 – 50) | 36 (28 – 48) | 36 (28 – 53) | 36 (27 – 47) | 36 (27 – 48) | |
| Missing | 17 (7.8%) | 7 (7.8%) | 10 (15.6%) | 56 (9.2%) | 90 (9.2%) | |
For continuous variables, an ANOVA test was used to compare groups if approximately normally distributed and a Kruskal-Wallis test was used if non-normally distributed. For categorical variables, Pearson’s chi-squared test was used if all expected cell counts were >5 and Fisher’s exact test was used otherwise. Anemia was defined as hemoglobin <12 mg/dL. History of coronary artery disease was self-reported.
CAD: obstructive coronary artery disease, CKD: chronic kidney disease, PCI: percutaneous coronary intervention, hs-cTnI: high sensitive cardiac troponin I, Kim-1: kidney injury marker 1, hsCRP: high sensitive c-reactive protein, NT-proBNP: N-terminal prohormone of brain natriuretic peptide, LDL-C: low density lipoprotein-cholesterol, HDL-C: high density lipoprotein-cholesterol, sST2: soluble ST2
Table 2.
Presence and Severity of CAD as a Function of hs-cTnI Concentrations Dichotomized Around the 99th Percentile Value (34 ng/L in male and 16 ng/L in female)
| hs-cTnI <99th percentile | hs-cTnI ≥99th percentile | P value | |
|---|---|---|---|
|
| |||
| ≥30% Coronary stenosis in ≥2 arteries | 545/878 (62.1%) | 68/92 (73.9%) | 0.03 |
| ≥30% Coronary stenosis in ≥3 arteries | 421/878 (47.9%) | 55/92 (59.8%) | 0.03 |
| ≥50% Coronary stenosis in ≥2 arteries | 438/878 (49.9%) | 59/92 (64.1%) | 0.009 |
| ≥50% Coronary stenosis in ≥3 arteries | 296/878 (33.7%) | 41/92 (44.6%) | 0.04 |
| ≥70% Coronary stenosis in ≥2 arteries | 333/878 (37.9%) | 44/92 (47.8%) | 0.06 |
| ≥70% Coronary stenosis in ≥3 arteries | 186/878 (21.2%) | 22/92 (23.9%) | 0.54 |
For comparison, Pearson’s chi-squared test was used if all expected cell counts were >5 and Fisher’s exact test was used otherwise.
CAD: coronary artery disease, hs-cTnI; high sensitive-cardiac troponin I
After comparing the performance of the various machine learning algorithms, GBM was selected as the best model based on AUC in the test set 14. The GBM package in R was used 15, with number of trees set to 2000, shrinkage set to 1%, interaction depth set to 3, and the minimum number of observations in the terminal nodes of the trees set to 20. To find the optimal number of trees to use for prediction, 10-fold cross validation was applied. Relative influence was also calculated for the variables that contributed to the model.
To further understand risk associated with higher concentration of hs-cTnI, Cox proportional hazards models assessed biomarker concentrations relative to primary clinical outcome (the combined outcomes of incident MI and cardiovascular death). The model was adjusted for age, sex, hypertension, diabetes mellitus, smoker, dyslipidemia, history of CAD. To assess the performance of the models, Harrell’s C statistics, Akaike information criterion (AIC) and the Bayesian information criterion (BIC), Hosmer-Lemeshow were calculated. The added value of hs-cTnI was calculated by both cutpoint-free and cutpoint-based Net reclassification index (NRI) (10, 30) and integrated discrimination index (IDI). All p-values reported were 2-sided. A P value <0.05 was considered statistically significant. All statistical analyses were performed using the R version 3.6.2 (R Foundation for Statistical Computing, Vienna, Austria. URL: https://www.R-project.org/).
Results
This study consisted of 978 patients without AMI who underwent coronary angiography. A CONSORT Diagram is depicted in Figure 1 to clarify study flow. Table 1 shows the distribution of baseline characteristics of study population across CAD categories. 607 (62.1%) patients had ≥70% CAD, 64 (6.5%) had 50–69% CAD, 90 (9.2%) had 30–49% CAD, and 217 (22.2%) had non-obstructive CAD. Across CAD categories, there were significant differences for age, sex, history of hypertension, history of type II diabetes mellitus, CKD, CAD, and prior angioplasty. No significant differences were observed in distribution of cystatin C, hsCRP, NT-proBNP and sST2 across CAD categories.
Table 2 demonstrates the presence and severity of CAD as a function of hs-cTnI concentration, which shows modest association between hs-cTnI above the 99th percentile and presence of obstructive CAD of various severity.
Figure 2 demonstrates the influential variables for hs-cTnI concentrations as determined by Lasso regression. Prevalent CKD, heart failure, CAD, male sex, current smoking status, anemia, greater age, and LDL-c were the most influential factors respectively.
Figure 2:
Influential factors for concentrations of hs-cTnI entered as a log-transformed continuous variable.
hs-cTnI: high sensitivity cardiac troponin I, hxckd: history of chronic kidney disease, hxhf: history of heart failure, obstructive: obstructive coronary artery disease, ldlc: low density lipoprotein.
Table 3 depicts the independent predictors of CAD ≥70% using multivariable logistic regression using the variables determined to be important predictors from Lasso regression. These included an hs-cTnI ≥ 99th percentile, history of CAD including those with revascularization, male sex, age per year, type II diabetes mellitus and HDL-cholesterol concentraitons. The final model had a good calibration (Harrell’s C statistic=0.77), goodness of fit (Hosmer Lemeshow p-value =0.39). Addition of hs-cTnI to the model decreased both AIC and BIC and improved the predictive ability as measured by IDI (P = 0.02) and NRI (P <0.001).
Table 3.
Independent predictors of obstructive CAD >70% using logistic regression
| Odd ratio (95% Confidence interval) | P value | |
|---|---|---|
|
| ||
| (Intercept) | 0.21 (0.07–0.58) | 0.003 |
| Age, (per year) | 1.01 (1.00–1.03) | 0.04 |
| Men | 2.63 (1.88–3.69) | <0.001 |
| HDL-C, (per 1 mg/dL) | 0.98 (0.97–0.99) | <0.001 |
| Type II Diabetes Mellitus | 2.02 (1.38–2.98) | <0.001 |
| History of coronary artery disease | 4.62 (3.41–6.28) | <0.001 |
| Anemia | 1.30 (0.93–1.88) | 0.13 |
| Log hs-cTnI | 1.25 (1.11–1.40) | <0.001 |
HDL-C: high density lipoprotein-cholesterol, hs-cTnI; high sensitivity-cardiac troponin I
Table 4 details result from a GBM for independent predictors of CAD, including age, sex, HDL cholesterol, history of CAD, type II diabetes mellitus, anemia, and log hs-cTnI. The best iteration of number of trees was found using 10-fold cross validation to be 336. Predictions using these trees on the train dataset (80% of observations) resulted in an AUC of 0.82 (95% CI 0.79–0.85), accuracy of 75%, sensitivity of 88%, and specificity of 52%. Applying this model to the validation dataset (20% of observations) resulted in an AUC of 0.75 (95% CI 0.68–0.81), accuracy of 68 %, sensitivity of 82 %, and specificity of 38 %. Supplemental Figure 1 shows the relative influence of variables in the GBM, with the most influential variables being history of CAD, HDL cholesterol, and log hs-cTnI.
Table 4.
Performance of GBM for detection of obstructive CAD (≥70%)
| Dataset | N | AUC (95% CI) | Accuracy | Sensitivity | Specificity | PPV | NPV |
|---|---|---|---|---|---|---|---|
|
| |||||||
| Train (80%) | 782 | 0.82 (0.79, 0.85) | 75% | 88% | 52% | 76% | 72% |
| Test (20%) | 196 | 0.75 (0.68, 0.81) | 68% | 82% | 38% | 65% | 60% |
AUC: area under curve, PPV: positive predictive value, NVP: negative predictive value, GBM: gradient boosting model, CAD: coronary artery disease, N: number, CI: confidence interval
Time to first CV death/acute MI and the individual outcomes were shorter in those with hs-cTnI above the 99th percentile concentration (Figure 3). We then evaluated whether higher concentrations beyond the 99th percentile concentration were associated with future risk. Table 5 demonstrate the hazard ratios of log hs-cTnI for development of incident MI, CV death and the composite of the two. Higher hs-cTnI was associated with increased risk of incident MI (HR = 1.34 per log-unit increase, 95% CI =1.22–1.47, P <.001), CV mortality (HR = 1. 24 per log unit increase, 95% CI = 1.11–1.39, P <.001). Notably, risk for each outcome appeared consistent regardless of baseline degree of CAD. Furthermore, among hs-cTnI risk categories, higher hs-cTnI was associated with each outcome examined (Figure 4). Notably, considering adjudicated type 1 and type 2 MI as separate outcomes, one log-unit increase in hs-cTnI was associated with similarly increased risk for incident type 1 MI (HR = 1. 38, 95% CI = 1.15–1.68, P=.007) and type 2 MI (HR = 1. 31, 95% CI = 1.17–1.47, P <.001). Cumulative hazard curves detailing time to incident type 1 and type 2 MI as a function of hs-cTnI concentration (99th percentile as well as risk categories) are shown in Supplementary Figure 2 and 3.
Figure 3:
Cumulative hazard curves detailing time to incident A) acute MI/CV death, B) acute MI, and C) CV death as a function of hs-cTnI among those without acute MI at baseline. Those with concentrations ≥99th percentile value had higher risk for each outcome measure. MI: myocardial infarction, CV: cardiovascular, hs-cTnI: high sensitive cardiac troponin I, ≥99th percentile hs-cTnI concentration: male > 34 ng/L, female > 16 ng/L
Table 5.
Association of one log unit increase in hs-cTnI with adverse cardiovascular outcome
| Variable | Number of events (%) | Hazard ratio | 95% confidence interval | P value |
|---|---|---|---|---|
|
| ||||
| Incident MI | ||||
| All participants (N=978) | 141 (14.4%) | 1.35 | 1.23–1.48 | <0.001 |
| Obstructive CAD >70% (n=607) | 110 (18.1%) | 1.29 | 1.16–1.43 | <0.001 |
| Nonobstructive CAD (N=217) | 13 (6.0%) | 1.49 | 1.03–2.17 | 0.04 |
| CV death | ||||
| All participants (N=978) | 105 (10.7%) | 1.25 | 1.12–1.40 | <0.001 |
| Obstructive CAD >70% (n=607) | 79 (13.0%) | 1.25 | 1.10–1.42 | <0.001 |
| Nonobstructive CAD (N=217) | 14 (6.5%) | NA | NA | NA |
| MI/CV death | ||||
| All participants (N=978) | 206 (21.1%) | 1.30 | 1.20–1.41 | <0.001 |
| Obstructive CAD >70% (n=607) | 153 (25.2%) | 1.27 | 1.15–1.39 | <.001 |
| Nonobstructive CAD (N=217) | 25 (11.5%) | 1.37 | 1.00–1.86 | 0.05 |
MI: Myocardial infarction, CV: cardiovascular, CAD: coronary artery disease,
Model adjusted for age, sex, hypertension, type II diabetes, current smoking status, dyslipidemia, history of CAD.
NA: not applicable due to low statistical power
Figure 4.
Cumulative survival curves detailing time to incident A) acute MI/CV death, B) acute MI, and C) CV death as a function of risk category among those without acute MI at baseline. Those with highest risk had lower survival probability. MI: myocardial infarction, CV: cardiovascular. Low Risk: males < 6 ng/L, females < 4 ng/L. Medium Risk: male 6–12 ng/L, female 4–10 ng/L. High Risk: male> 12 ng/L, female > 10 ng/L.
Discussion
Understanding the meaning of troponin results in those without acute MI may be challenging without a clinical and angiographic standard. In this large cohort study of patients who were free of acute MI at baseline undergoing angiographic procedures, we examined the meaning of hs-cTnI concentrations. We identified numerous variables predictive of presence and severity of CAD including hs-cTnI. In turn, among other variables, presence of CAD (defined for the analysis as at least one angiographic stenosis ≥70% in severity) was influential in determining hs-cTnI concentrations. In a machine learning model, higher concentrations of hs-cTnI were linked to CAD in a derivation model and validated in 20% of the samples in the analysis. Lastly, elevated hs-cTnI was associated with increased risk of future CV events in both obstructive and non-obstructive CAD patients, predicting incident MI (both type 1 and 2) as well as death.
Advances in troponin assay technology have led to the development of highly sensitive measurement of the biomarker. Such hs-cTn assays have transformed the evaluation of patients with suspected or proven acute MI. However, it is increasingly clear that higher concentrations of hs-cTnI may be observed in circumstances other than acute MI 16. Myocardial injury beyond that of an acute MI may arise secondary to many cardiac and non-cardiac conditions 17–19; studies have linked concentrations of hs-cTn to presence and severity of underlying CAD, even in the absence of clinical MI 20, 21. This suggests in certain circumstances sub-clinical myocardial ischemia may play a role, linking higher hs-cTnI to CAD. Studies have proposed that in patients without myocardial infarction, injury triggered by displacement of thrombi in small and/or more heavily disease coronary vessels may be a potential cause for associations between CAD and hs-cTnI 22, 23.
In this analysis, we examined concentrations of hs-cTnI both as a dichotomous variable (using the 99th percentile value) as well as a continuous variable (either in log-transformed outcomes models or in a categorical manner in time-to-event curves). In each case, higher values were linked to presence of CAD. While it is logical to utilize the 99th percentile concentration as it identifies presence of myocardial injury by the Universal Definition of Myocardial Infarction 17, dichotomization by nature reduces the power of extremes seen when considering values in a ranked or continuous manner. The results of this study suggest presence of injury is important to identify presence of CAD, however, values above the 99th percentile would be expected to have stronger association with presence and severity of CAD as well as worse prognosis.
Despite significant association with presence and severity of CAD, hs-cTnI alone is not sufficiently able to predict presence of coronary obstruction in those without a clinical picture of ischemia. In this regard, together with hs-cTn, other variables including biomarkers may provide additional information to clarify mechanism of underlying myocardial injury 24, 25. Recently, Neumann and colleagues in a cohort of 748 patients of whom 138 had MI and 221 had myocardial injury evaluated the association of 29 biomarkers with diagnosis of MI or myocardial injury 26. They identified 5 biomarkers (adiponectin, NT-proBNP, pulmonary and activation-regulated chemokine, transthyretin, copeptin) in addition to hs-cTnI, to build a prognostic model to discriminate between MI and non-ischemic myocardial injury. The model had a promising discrimination ability with an area under curve of 0.84. However, this study lacks angiographic data to confirm the diagnosis of MI, relying only on clinical adjudication. Among patients free of AMI referred for coronary angiography, McCarthy and colleagues 25 developed a model inclusive of 3 clinical variables (male, age, previous PCI) and 3 biomarkers (hs-cTnI [Siemens Atellica immunoassay], adiponectin and kidney injury molecule-1) to detect the presence of CAD as evident by more than 70% obstructive disease in at least one vessel; the model had excellent performance in both internal and external cohorts. The present analysis extends previous observations, now including the Abbott hs-cTnI in the mix of biomarkers liked to CAD and leverage a machine learning strategy to optimize the accuracy of CAD detection.
The links between hs-cTnI concentrations and major cardiovascular events such as incident MI and death provides supportive evidence for the associations between the biomarker and prevalent CAD. It is noteworthy that in CASABLANCA, type of MI was adjudicated with both type 1 and 2 considered in this analysis. Results indicate that baseline hs-cTnI not only predicted incident type 1 MI but also type 2 events. Clinical importance of type 2 MI is increasingly recognized 27, yet treatment strategies are not yet developed for this highly prevalent form of MI. The results from this study provide further support for the importance of biomarkers to identify presence and predict risk for complications from CAD and support ongoing efforts to better define coronary anatomy and biomarker patterns among those with type 2 MI in an effort to develop prevention and treatment strategies.
The empirical results reported herein should be considered considering its limitations. First, electrocardiographic data were unavailable in CASABLANCA. Hence, we could not assess the significance of ischemic changes on ECG besides troponin concentration for the detection of CAD in patients without myocardial infarction. Second, most of the study participants in CASABLANCA were White individuals, which might limit the generalizability of study results given the influence of race on hs-cTn concentrations 28. Third, our results were based on one measurement of hs-cTnI at a single point in time, so we could not assess the effect of the possible change in concentrations over time relative to risk for CAD. Lastly, although our machine learning model had a satisfactory performance in the validation cohort, the model still should be externally validated in other cohorts.
In conclusion, this study demonstrated the prognostic role of hs-cTnI for detection of CAD in patients free of MI at the time of blood sampling and predict complications from its presence. Nevertheless, it is important to understand that hs-cTnI concentration can rise due to many factors other than CAD. Using a machine learning algorithm, combination of hs-cTnI and clinical history offer a promising tool for detection of CAD.
Supplementary Material
Source of funding:
Supported by Abbott Diagnostics. Dr. Mohebi is supported by the Barry Fellowship. Dr. McCarthy is supported by the NHLBI T32 postdoctoral training grant (5T32HL094301–12). Dr. Januzzi is supported by the Hutter Family Professorship.
Footnotes
Disclosures
Dr. Jackson and Dr. Murtagh are full-time employees and shareholders of Abbott. Dr. Gaggin has received research grant support from Roche Diagnostics, Jana Care, Ortho Clinical, Novartis, Pfizer, Alnylam, Akcea; consulting income from Amgen, Eko, Merck, Roche Diagnostics, Pfizer; Stock ownership for Eko; Research payments for clinical endpoint committees from Radiometer. She has also received research payment for clinical endpoint committees from Baim Institute for Clinical Research for Abbott, Siemens and Beckman Coulter. Dr. Januzzi is supported by the Hutter Family Professorship; is a Trustee of the American College of Cardiology; is a board member of Imbria Pharmaceuticals; has received grant support from Abbott Diagnostics, Applied Therapeutics, Innolife and Novartis; has received consulting income from Abbott Diagnostics, Boehringer-Ingelheim, Janssen, Novartis, Roche Diagnostics; and participates in clinical endpoint committees/data safety monitoring boards for AbbVie, Siemens, Takeda and Vifor. The remaining authors have nothing to disclose.
ClinicalTrials.gov identifier: NCT00842868
References
- 1.McCarthy CP, Raber I, Chapman AR, Sandoval Y, Apple FS, Mills NL and Januzzi JL Jr, Myocardial Injury in the Era of High-Sensitivity Cardiac Troponin Assays: A Practical Approach for Clinicians. JAMA Cardiol. 2019;4:1034–1042. [DOI] [PubMed] [Google Scholar]
- 2.Love SA, Sandoval Y, Smith SW, Nicholson J, Cao J, Ler R, Schulz K and Apple FS. Incidence of Undetectable, Measurable, and Increased Cardiac Troponin I Concentrations Above the 99th Percentile Using a High-Sensitivity vs a Contemporary Assay in Patients Presenting to the Emergency Department. Clin Chem. 2016;62:1115–1119. [DOI] [PubMed] [Google Scholar]
- 3.Kelley WE, Januzzi JL and Christenson RH. Increases of cardiac troponin in conditions other than acute coronary syndrome and heart failure. Clin Chem. 2009;55:2098–2112. [DOI] [PubMed] [Google Scholar]
- 4.Samman Tahhan A, Sandesara P, Hayek SS, Hammadah M, Alkhoder A, Kelli HM, Topel M, O’Neal WT, Ghasemzadeh N, Ko YA, Gafeer MM, Abdelhadi N, Choudhary F, Patel K, Beshiri A, Murtagh G, Kim J, Wilson P, Shaw L, Vaccarino V, Epstein SE, Sperling L and Quyyumi AA. High-Sensitivity Troponin I Levels and Coronary Artery Disease Severity, Progression, and Long-Term Outcomes. J Am Heart Assoc. 2018;7:e007914. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Eggers KM, Venge P, Lindahl B and Lind L. Cardiac troponin I levels measured with a high-sensitive assay increase over time and are strong predictors of mortality in an elderly population. J Am Coll Cardiol. 2013;61:1906–1913. [DOI] [PubMed] [Google Scholar]
- 6.Everett BM, Brooks MM, Vlachos HE, Chaitman BR, Frye RL, Bhatt DL and Group BDS. Troponin and Cardiac Events in Stable Ischemic Heart Disease and Diabetes. N Engl J Med. 2015;373:610–620. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Sinning C, Keller T, Zeller T, Ojeda F, Schluter M, Schnabel R, Lubos E, Bickel C, Lackner KJ, Diemert P, Munzel T, Blankenberg S, Wild PS and Gutenberg Health S. Association of high-sensitivity assayed troponin I with cardiovascular phenotypes in the general population: the population-based Gutenberg health study. Clin Res Cardiol. 2014;103:211–222. [DOI] [PubMed] [Google Scholar]
- 8.Ndrepepa G, Braun S, Schulz S, Mehilli J, Schomig A and Kastrati A. High-sensitivity troponin T level and angiographic severity of coronary artery disease. Am J Cardiol. 2011;108:639–643. [DOI] [PubMed] [Google Scholar]
- 9.Olson F, Engborg J, Gronhoj MH, Sand NP, Lambrechtsen J, Steffensen FH, Nybo M, Gerke O, Mickley H and Diederichsen AC. Association between high-sensitive troponin I and coronary artery calcification in a Danish general population. Atherosclerosis. 2016;245:88–93. [DOI] [PubMed] [Google Scholar]
- 10.Januzzi JL Jr., and McCarthy CP. Trivializing an Elevated Troponin: Adding Insult to Injury? J Am Coll Cardiol. 2019;73:10–12. [DOI] [PubMed] [Google Scholar]
- 11.Januzzi JL Jr., and McCarthy CP. Untangling Myocardial Injury: Moving From Consequence to Cause. J Am Coll Cardiol. 2021;78:791–793. [DOI] [PubMed] [Google Scholar]
- 12.Gaggin HK, Liu Y, Lyass A, van Kimmenade RR, Motiwala SR, Kelly NP, Mallick A, Gandhi PU, Ibrahim NE, Simon ML, Bhardwaj A, Belcher AM, Harisiades JE, Massaro JM, D’Agostino RB, Sr. and Januzzi JL, Jr. Incident Type 2 Myocardial Infarction in a Cohort of Patients Undergoing Coronary or Peripheral Arterial Angiography. Circulation. 2017;135:116–127. [DOI] [PubMed] [Google Scholar]
- 13.Bhatia PM and Daniels LB. Highly Sensitive Cardiac Troponins: The Evidence Behind Sex-Specific Cutoffs. J Am Heart Assoc. 2020;9:e015272. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Trevor Hastie RT, Friedman Jerome. The Elements of Statistical Learning: Data Mining, Inference, and Prediction.; 2009. [Google Scholar]
- 15.Brandon Greenwell BB, Cunningham Jay,. Generalized Boosted Regression Models R package version 2.1.8. 2020. [Google Scholar]
- 16.Meigher S, Thode HC, Peacock WF, Bock JL, Gruberg L and Singer AJ. Causes of Elevated Cardiac Troponins in the Emergency Department and Their Associated Mortality. Acad Emerg Med. 2016;23:1267–1273. [DOI] [PubMed] [Google Scholar]
- 17.Thygesen K, Alpert JS, Jaffe AS, Chaitman BR, Bax JJ, Morrow DA, White HD and Executive Group on behalf of the Joint European Society of Cardiology /American College of Cardiology /American Heart Association /World Heart Federation Task Force for the Universal Definition of Myocardial I. Fourth Universal Definition of Myocardial Infarction (2018). Circulation. 2018;138:e618–e651.30571511 [Google Scholar]
- 18.Weil BR, Suzuki G, Young RF, Iyer V and Canty JM Jr, Troponin Release and Reversible Left Ventricular Dysfunction After Transient Pressure Overload. J Am Coll Cardiol. 2018;71:2906–2916. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Turer AT, Addo TA, Martin JL, Sabatine MS, Lewis GD, Gerszten RE, Keeley EC, Cigarroa JE, Lange RA, Hillis LD and de Lemos JA. Myocardial ischemia induced by rapid atrial pacing causes troponin T release detectable by a highly sensitive assay: insights from a coronary sinus sampling study. J Am Coll Cardiol. 2011;57:2398–2405. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Januzzi JL Jr, Suchindran S, Coles A, Ferencik M, Patel MR, Hoffmann U, Ginsburg GS, Douglas PS and Investigators P. High-Sensitivity Troponin I and Coronary Computed Tomography in Symptomatic Outpatients With Suspected CAD: Insights From the PROMISE Trial. JACC Cardiovasc Imaging. 2019;12:1047–1055. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.McCarthy CP, Ibrahim NE, Lyass A, Li Y, Gaggin HK, Simon ML, Mukai R, Gandhi P, Kelly N, Motiwala SR, van Kimmenade RRJ, Massaro JM, D’Agostino RB, Sr. and Januzzi JL, Jr. Single-Molecule Counting of High-Sensitivity Troponin I in Patients Referred for Diagnostic Angiography: Results From the CASABLANCA (Catheter Sampled Blood Archive in Cardiovascular Diseases) Study. J Am Heart Assoc. 2018;7:e007975. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Korosoglou G, Lehrke S, Mueller D, Hosch W, Kauczor HU, Humpert PM, Giannitsis E and Katus HA. Determinants of troponin release in patients with stable coronary artery disease: insights from CT angiography characteristics of atherosclerotic plaque. Heart. 2011;97:823–831. [DOI] [PubMed] [Google Scholar]
- 23.Rittersma SZ, van der Wal AC, Koch KT, Piek JJ, Henriques JP, Mulder KJ, Ploegmakers JP, Meesterman M and de Winter RJ. Plaque instability frequently occurs days or weeks before occlusive coronary thrombosis: a pathological thrombectomy study in primary percutaneous coronary intervention. Circulation. 2005;111:1160–1165. [DOI] [PubMed] [Google Scholar]
- 24.Ibrahim NE, Januzzi JL Jr, Magaret CA, Gaggin HK, Rhyne RF, Gandhi PU, Kelly N, Simon ML, Motiwala SR, Belcher AM and van Kimmenade RRJ. A Clinical and Biomarker Scoring System to Predict the Presence of Obstructive Coronary Artery Disease. J Am Coll Cardiol. 2017;69:1147–1156. [DOI] [PubMed] [Google Scholar]
- 25.McCarthy CP, Neumann JT, Michelhaugh SA, Ibrahim NE, Gaggin HK, Sorensen NA, Schaefer S, Zeller T, Magaret CA, Barnes G, Rhyne RF, Westermann D and Januzzi JL Jr, Derivation and External Validation of a High-Sensitivity Cardiac Troponin-Based Proteomic Model to Predict the Presence of Obstructive Coronary Artery Disease. J Am Heart Assoc. 2020;9:e017221. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Neumann JT, Weimann J, Sorensen NA, Hartikainen TS, Haller PM, Lehmacher J, Brocks C, Tenhaeff S, Karakas M, Renne T, Blankenberg S, Zeller T and Westermann D. A Biomarker Model to Distinguish Types of Myocardial Infarction and Injury. J Am Coll Cardiol. 2021;78:781–790. [DOI] [PubMed] [Google Scholar]
- 27.McCarthy CP, Kolte D, Kennedy KF, Vaduganathan M, Wasfy JH and Januzzi JL Jr,. Patient Characteristics and Clinical Outcomes of Type 1 Versus Type 2 Myocardial Infarction. J Am Coll Cardiol. 2021;77:848–857. [DOI] [PubMed] [Google Scholar]
- 28.Winchester DE, Kline K, Estel C, Mahtta D, Taasan S and Peacock FW. Associations between cardiac troponin, mortality and subsequent use of cardiovascular services: differences in sex and ethnicity. Open Heart. 2018;5:e000713. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.