Author + information
- Received November 16, 2016
- Revision received November 30, 2016
- Accepted December 1, 2016
- Published online February 27, 2017.
- Nasrien E. Ibrahim, MDa,
- James L. Januzzi Jr., MDa,b,∗ (, )
- Craig A. Magaret, MSc,
- Hanna K. Gaggin, MD, MPHa,b,
- Rhonda F. Rhyne, BPharm, MBAc,
- Parul U. Gandhi, MDd,
- Noreen Kelly, MDe,
- Mandy L. Simon, DNP, FNP-BCa,
- Shweta R. Motiwala, MDe,
- Arianna M. Belcher, MSa and
- Roland R.J. van Kimmenade, MD, PhDf
- aMassachusetts General Hospital, Division of Cardiology, Boston, Massachusetts
- bHarvard Clinical Research Institute, Cardiometabolic Trials, Boston, Massachusetts
- cPrevencio, Inc., Kirkland, Washington
- dYale University, Cardiology, New Haven, Connecticut
- eBrigham and Women's Hospital, Cardiology, Boston, Massachusetts
- fDepartment of Cardiology, Radboud University Medical Centre, Nijmegen, the Netherlands
- ↵∗Address for correspondence:
Dr. James L. Januzzi Jr., Division of Cardiology, Massachusetts General Hospital, 32 Fruit Street, Yawkey 5984, Boston, Massachusetts 02114.
Background Noninvasive models to predict the presence of coronary artery disease (CAD) may help reduce the societal burden of CAD.
Objectives From a prospective registry of patients referred for coronary angiography, the goal of this study was to develop a clinical and biomarker score to predict the presence of significant CAD.
Methods In a training cohort of 649 subjects, predictors of ≥70% stenosis in at least 1 major coronary vessel were identified from >200 candidate variables, including 109 biomarkers. The final model was then validated in a separate cohort (n = 278).
Results The scoring system consisted of clinical variables (male sex and previous percutaneous coronary intervention) and 4 biomarkers (midkine, adiponectin, apolipoprotein C-I, and kidney injury molecule–1). In the training cohort, elevated scores were predictive of ≥70% stenosis in all subjects (odds ratio [OR]: 9.74; p < 0.001), men (OR: 7.88; p <0.001), women (OR: 24.8; p < 0.001), and those with no previous CAD (OR: 8.67; p < 0.001). In the validation cohort, the score had an area under the receiver-operating characteristic curve of 0.87 (p < 0.001) for coronary stenosis ≥70%. Higher scores were associated with greater severity of angiographic stenosis. At optimal cutoff, the score had 77% sensitivity, 84% specificity, and a positive predictive value of 90% for ≥70% stenosis. Partitioning the score into 5 levels allowed for identifying or excluding CAD with >90% predictive value in 42% of subjects. An elevated score predicted incident acute myocardial infarction during 3.6 years of follow up (hazard ratio: 2.39; p < 0.001).
Conclusions We described a clinical and biomarker score with high accuracy for predicting the presence of anatomically significant CAD. (The CASABLANCA Study: Catheter Sampled Blood Archive in Cardiovascular Diseases; NCT00842868)
Despite efforts toward better recognition of risk factors and preventive treatments, the prevalence of coronary artery disease (CAD) in the general population remains high, with nearly 1 in 5 people >65 years of age affected by the diagnosis. Indeed, heart disease is the leading cause of death for both men and women, with CAD the most common affliction, killing >370,000 people annually (1). As such, CAD is a public health concern, and an efficient manner for its noninvasive detection could potentially result in reduction of morbidity, mortality, and cost of this disease process.
In the context of a patient with risk factors for CAD, clinicians often use stress testing with or without adjunctive imaging to assist their evaluation of obstructive CAD. Drawbacks to stress testing include variable sensitivity and specificity, limitations with respect to accuracy in certain types of body habitus (including overweight/obese patients and women), as well as the need for ionizing radiation. More recently, coronary calcium assessment by computed tomography (CT) scanning, as well as CT angiography, has been used to identify severe CAD; both techniques identify CAD presence and severity independent of and substantially incremental to clinical risk scores (2). However, CT angiography has similar drawbacks to stress testing. In addition, the application of imaging to large numbers of patients suspected of having CAD would not be practical. Lastly, both stress testing and CT imaging come with the challenge of high costs.
A relatively unexplored approach for identifying significant CAD is the use of clinical and biomarker scoring systems. Wilson et al. (2) developed one of the earliest CAD risk-prediction models in the Framingham Heart Study based on traditional risk factors. Such models predict risk for events but lack discrimination for anatomically significant CAD (3). More recently, Bolton et al. (4) showed that addition of genetic testing to conventional risk factors improved prediction of CAD. Lastly, in testing 359 patients referred for coronary angiography, LaFramboise et al. (5) found significant differences in several circulating proteins in those patients with significantly obstructive CAD versus those without. These latter results provided proof of concept but were in a much smaller population of patients with low pre-test probability for disease presence.
Accordingly, the goal of the present study was to identify clinical and biomarker predictors of clinically significant CAD in an at-risk population of subjects enrolled in the CASABLANCA (Catheter Sampled Blood Archive in Cardiovascular Diseases) study undergoing coronary angiography for numerous indications (6). We hypothesized that the addition of plasma biomarkers to known clinical risk factors might increase the accuracy of predicting clinically significant CAD.
Patients and Methods
All study procedures were approved by the Partners HealthCare Institutional Review Board and conducted in accordance with the Declaration of Helsinki.
The design of the CASABLANCA study has been described previously (6). Briefly, a convenience sample of 1,251 patients undergoing coronary and peripheral angiography with or without intervention between 2008 and 2011 were prospectively enrolled at the Massachusetts General Hospital in Boston, Massachusetts. Patients were referred for these procedures for numerous reasons, including angiography after acute processes such as myocardial infarction (MI), unstable angina pectoris, and heart failure, as well as for nonacute indications, such as the diagnostic evaluation of stable chest pain and failed stress testing or pre-operatively before heart valve surgery.
After obtaining informed consent, detailed clinical and historical variables and reason for referral for angiography were recorded at the time of the procedure. Results of coronary angiography (based on visual estimation at the time of the procedure) were recorded; the left main, left anterior descending, left circumflex, and right coronary artery were each considered major coronary arteries, and the highest percent stenosis within each major coronary artery or their branches was recorded. For the purposes of this analysis, we characterized “significant” coronary stenosis as ≥70% luminal obstruction. Although less severe stenoses might be associated with risk for cardiovascular events, we elected to use a widely accepted standard for defining angiographic “significance.”
Medical record review from time of enrollment to end of follow-up was undertaken. For identification of clinical endpoints, review of medical records as well as telephone follow-up were performed with patients and/or their managing physicians. The Social Security Death Index and/or postings of death announcements were used to confirm vital status. A detailed definition of endpoints for CASABLANCA has been published previously (6). The following clinical end events were identified, adjudicated, and recorded by study investigators: death, new nonfatal MI (starting at 3 days after the procedure), heart failure, stroke, transient ischemic attack, peripheral arterial complication, and cardiac arrhythmia. For any recurring events, each discrete event was recorded. In addition, deaths were adjudicated for the presence/absence of a cardiovascular cause. Study investigators judging angiographic severity of CAD or events during follow-up were blinded to results of all biomarker testing.
We obtained 15 ml of blood immediately before and immediately after 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.
After a single freeze–thaw cycle, 200 μl of plasma was analyzed for a panel of 109 biomarkers (Online Table 1) on a Luminex xMAP technology platform (Luminex Corporation, Austin, Texas) that uses multiplexed, microsphere-based assays in a single reaction vessel. Multiplexing was accomplished by assigning each protein-specific assay a microsphere set labeled with a unique fluorescence signature. An assay-specific capture antibody was conjugated covalently to each unique set of microspheres and bound to the protein of interest. Assay-specific, biotinylated detecting antibodies were added, followed by a streptavidin-labeled fluorescent “reporter” molecule; the bound amount of fluorescence generated is proportionate to the protein level. A minimum of 100 individual microspheres from each unique set were analyzed, and the median value of the protein-specific fluorescence was logged.
Patients from the CASABLANCA study selected for the present analysis consisted of the initial 927 patients chronologically who received a coronary angiogram. These include patients who might also have received a peripheral angiogram concomitantly. All patients who received a coronary angiogram chronologically subsequent to this set were isolated for potential future analytic use and were not available to the practitioners of this analysis for the entire scope of work in this study.
The 927 patients selected for analysis were randomly split into a training set (n = 649 [70%]) and a holdout validation set (n = 278 [30%]). Given this sample size of 278 patients in the validation set, and a positive outcome rate of 64% with an assumed 5% type I error rate, we had 95% power to detect a risk ratio of at least 1.4, comparing the event rate above versus below the diagnostic threshold. This threshold was evaluated using a power calculation for a difference in proportions.
Baseline characteristics were compared between those with and without ≥70% coronary stenosis in at least 1 major coronary artery (Table 1); dichotomous variables were compared by using 2-sided Fisher exact tests, and continuous variables were compared by using 2-sided 2-sample Student t tests. The biomarkers compared were tested with the Wilcoxon rank sum test because their concentrations were not normally distributed. For any biomarker result that was unmeasurable, a standard approach of imputing concentrations 50% below the limit of detection was used.
All studies for biomarker selection and the development of a diagnostic model were conducted exclusively on the training set. To facilitate the predictive analysis, the concentration values for all proteins were transformed as follows: 1) they were log-transformed to achieve a normal distribution; 2) outliers were clipped at the value of 3 times the median absolute deviation; and 3) the values were rescaled to a distribution with a zero mean and unit variance. The starting sets of variables consisted of all 109 proteins as well as clinical factors in the CASABLANCA dataset. In addition to the biomarkers, we began with a broad selection of variables that were chosen for their possible clinical relevance (age; body mass index; race; sex; smoking status; statin use; and history of CAD, chronic obstructive pulmonary disease, cerebrovascular disease, diabetes, hyperlipidemia, heart failure, hypertension, MI, peripheral arterial disease, percutaneous coronary intervention [PCI], cardiac surgery, and arrhythmias).
Candidate panels of proteins and clinical features were selected via least angle regression. In this method, factors are selected one at a time and evaluated for predictive performance and goodness of fit at each step in that if the new variable improves the score performance, that variable is retained in the panel, and another variable is added. If the new variable does not improve the performance, it is removed from the panel, and the next-best variable is selected. This method is repeated until there are no options left that satisfy the algorithm's goodness-of-fit requirements. With this panel of interest, predictive analyses were run on the training set by using least absolute shrinkage and selection operator with logistic regression, predicting the outcome of the presence of ≥70% obstruction in at least 1 coronary artery, using only the variables in the panel of interest. From the analysis results, the least absolute shrinkage and selection operator’s shrinkage performance was used to determine when a given variable was not contributing significantly to the model; in those cases, we removed the variable and repeated the analysis.
When we reached the point where we had a final panel, a final model was built by using the entire training set, validating it on the validation set. Candidates were subjected to further analysis of discrimination via iterative model building, assessing change in area under the curve (AUC) with the addition of biomarkers to the base model, along with assessment of improvement in calibration from their addition through minimization of the Akaike or Bayesian information criteria and goodness of fit in Hosmer-Lemeshow testing. The final model included the clinical variables male sex and previous PCI and 4 biomarkers (midkine, adiponectin, apolipoprotein C-I [apo C-I], and kidney injury molecule [KIM]–1).
Multivariable logistic regression was used to evaluate the performance of the model in the training set as a whole as well as in several relevant subgroups; diagnostic odds ratios (ORs) with 95% confidence intervals (CIs) were generated. Subsequently, the final model was evaluated with the validation set: to do so, we generated the score distribution within the validation cohort, followed by receiver-operating characteristic (ROC) testing with valor of the score as a function of the AUC. Operating characteristics of the score were calculated, with sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) generated. The range of the diagnostic model was then partitioned into 5 different risk levels, corresponding to multiple levels of CAD risk. The partitions were determined according to specific PPV and NPV thresholds in the training set, and the validation set was evaluated against these partitions.
Lastly, to evaluate prognostic meaning of the CAD score, age- and CAD score–adjusted Cox proportional hazards analyses were performed to evaluate whether a score above the optimal threshold for CAD diagnosis also predicted future acute MI; hazard ratios (HRs) were estimated for an elevated CAD score as well as per-unit score increases with 95% CIs. Lastly, time to first acute MI event as a function of elevated CAD score was calculated, displayed as Kaplan-Meier survival curves, and compared by using log-rank testing.
All statistics were performed by using R software, version 3.3 (R Foundation for Statistical Computing, Vienna, Austria). P values are 2-sided, with a value <0.05 considered significant.
Baseline characteristics of study subjects, dichotomized as a function of presence or absence of significant CAD, are detailed in Table 1. There were numerous baseline characteristics that differed between those in the training set who had at least 1 coronary stenosis ≥70% (n = 428) and those who did not. Notably, of all the biomarkers measured, those with severe CAD had lower concentrations of adiponectin and apo C-I and higher concentrations of KIM-1 and midkine at baseline. In the validation set, baseline characteristics were similar (Online Table 2). Online Figure 1 shows the distribution of maximum stenoses observed in the training cohort.
CAD scoring system
In the training cohort, independent predictors of CAD ≥70% in any 1 vessel included clinical variables (male sex and previous PCI) and 4 biomarkers (midkine, adiponectin, apo C-I, and KIM-1). Model fitting performed on the validation cohort (Table 2) shows that the addition of each individual biomarker to clinical variables improved discrimination, while simultaneously improving calibration for coronary stenosis of ≥70%, as evidenced by minimization of the Akaike or Bayesian information criteria and with concomitant goodness of fit through Hosmer-Lemeshow testing. With respect to the biomarkers, candidates were retained if they strengthened the model and/or improved calibration.
In multivariable logistic regression, among those in the training cohort, our score strongly predicted severe CAD in all subjects (OR: 9.74; 95% CI: 6.05 to 16.1; p < 0.001). To better understand performance of the score in various subgroups, we then examined score performance in men (OR: 7.88; 95% CI: 4.31 to 14.9; p < 0.001) and women (OR: 24.8; 95% CI: 7.11 to 111.6; p < 0.001), as well as in those with no history of CAD (OR: 8.67; 95% CI: 4.38 to 17.9; p < 0.001).
Individual scores were then calculated for patients in the validation cohort, and results were expressed as a function of CAD presence. In doing so, a bimodal score distribution was revealed (Figure 1), with higher prevalence of severe CAD in those with higher scores and lower prevalence among those with lower scores. When the diagnostic CAD score was evaluated as a continuous measure, higher raw CAD scores were directly correlated with a higher percent coronary stenosis (r2 = 0.32; p < 0.001) (Online Figure 2).
In ROC testing, for the gold standard diagnosis of ≥70% stenosis of any major epicardial coronary artery, the scores generated had an AUC of 0.87 (p < 0.001) (Figure 2). The AUC of the score for predicting severe CAD in subjects presenting without an acute MI was 0.87 (p < 0.001).
With regard to the operating characteristics of the CAD algorithm across various scores (Online Table 3), we found 77% sensitivity, 84% specificity, PPV of 90%, and NPV of 67% for severe CAD at the optimal score cutpoint. In subjects with a history of CAD, the score had a sensitivity of 84%, specificity of 66%, PPV of 90%, and NPV of 53% for prediction of CAD. In subjects with no history of CAD, the score had a sensitivity of 78%, specificity of 80%, PPV of 80%, and NPV of 78% for prediction of CAD. The CAD score was also tested for performance in patients presenting with and without an MI with comparable performance.
When the score was divided into 5 categories of predicted risk (Table 3), 42% of subjects could be “ruled in” or “ruled out” for severe CAD with an NPV of 91% and a PPV of 93%, respectively. Consistent with the linear correlation between raw score and median stenosis severity, scores in the lowest category (ruling out a significant stenosis with 91% NPV) were associated with lowest median angiographic stenosis, whereas intermediate scores were associated with intermediate severity maximal stenoses; the highest scores (ruling in significant coronary artery stenoses with 93% PPV) were associated with the highest median angiographic stenosis severity (Figure 3).
Of the 649 subjects in the training set, 154 had exercise stress tests without imaging and 174 had nuclear stress tests; of the 278 in the validation set, 47 had exercise stress tests without imaging and 61 had nuclear stress tests. Among these patients undergoing cardiac stress testing per standard of care either before or after their diagnostic coronary angiogram, the CAD score was substantially more accurate for predicting angiographically severe CAD (again, 0.87 vs. 0.52; p < 0.001 for difference in AUC), albeit in a diagnostically ambiguous, nonblinded, and nonrandomized comparison.
During a mean follow-up of 3.6 years, in the entire cohort of subjects, the CAD scoring system independently predicted subsequent incident acute MI in age- and score-adjusted models (HR: 2.23; 95% CI: 1.53 to 3.25; p < 0.001). When modeled as a continuous variable, the score was similarly predictive, with higher scores predictive of higher risk for incident acute MI (HR: 1.2 per unit score increase; 95% CI: 1.1 to 1.32; p < 0.001). Those with a dichotomously elevated score had a shorter time to first event than those with a lower CAD score, as evidenced by rapid and sustained divergence of the Kaplan-Meier survival curves (log rank p value < 0.001) (Figure 4).
Using patients from the CASABLANCA study cohort of patients referred to coronary angiography for a broad range of indications, this study describes a novel scoring system to predict the presence of severe epicardial CAD (≥70% stenosis in at least 1 major vessel). This score combined clinical variables and concentrations of 4 relevant biomarkers (Central Illustration). For the diagnosis of ≥70% stenosis of any major epicardial coronary artery, the score generated had an area under the ROC curve of 0.87 in the validation set and, at the optimal cutpoint, the score was both highly sensitive (77%) and specific (84%) for the diagnosis of CAD, with a PPV of 90%. Importantly, the CAD score performed particularly well in women. Although 1 element of the score was previous PCI, the score performance was similar in subjects with no history of CAD. Also, it was accurate for predicting significant epicardial stenoses in those without prevalent MI. As would be expected with an accurate model to diagnose severe CAD, our scoring system also reassuringly provided prognostic information regarding risk for incident acute MI.
The importance of biomarkers to supplement clinical variables in predictive modeling has become more appreciated, but such analyses are usually to predict prognosis for events (7). For noninvasive diagnosis of CAD, to our knowledge, such efforts are less widely explored. After derivation of a candidate list of assays with biological plausibility for the detection of underlying CAD, we performed a broad search of >100 biomarkers; in doing so, 4 biomarkers (midkine, adiponectin, apo C-I, and KIM-1) were found, each of which either added to model discrimination and/or calibration for predicting epicardial stenoses ≥70%.
Midkine is a heparin-binding cytokine/growth factor with a molecular weight of 13 kDa. Found in multiple organs, midkine promotes migration of leukocytes, induction of chemokines, and suppression of regulatory T cells in damaged tissues (8). In the heart, preclinical data support a potential role of midkine in the pathophysiology of CAD, where it promotes endothelial cell proliferation and enhances plaque infiltration of inflammatory cells (9,10). Notably, midkine-deficient mice exhibited significantly lower neointimal formation (9). To our knowledge, this study is the first description of midkine as a diagnostic tool for coronary atherosclerosis in humans. The role of adiponectin, a 244–amino acid peptide secreted by adipose tissue, includes regulation of glucose and fatty acid metabolism. Low concentrations of adiponectin have been previously linked to the presence of CAD (11), with low adiponectin concentrations also predicting vascular disease progression (12). Apo C-I plays a pivotal role for regulation of triglycerides in fasting and postprandial conditions (13). Previously identified in an unbiased proteomics screen as a predictor of significant CAD (14), clinical studies have provided conflicting information regarding apo C-I and risk for atherosclerosis (15). Our data suggest that this biomarker might be useful for predicting the presence of angiographically significant CAD. Lastly, KIM-1 is a proximal renal tubular marker whose concentrations have been linked to acute kidney injury (16); although modestly linked to other cardiovascular outcomes such as heart failure (17), to our knowledge this report is the first associating concentrations of KIM-1 with the presence of angiographically significant CAD.
Taken together, the biomarkers in our score represent a unique pathophysiological mix of vascular injury and plaque inflammation (midkine), abnormal glucose and fatty acid metabolism (adiponectin), hyperlipidemia (apo C-I), and renal dysfunction/injury (KIM-1), explaining why the “orthogonal” information provided by these biomarkers added independent discrimination and calibration to the predictive models (Central Illustration).
We selected coronary stenosis severity ≥70% as a widely accepted threshold for determination of angiographic significance. Although our strategy was to develop a scoring system to predict presence of such coronary stenoses, we do not in any way ignore the fact that CAD stenoses of lesser severity may be biologically and clinically important; patients with such lesions should receive secondary prevention. Notably, although we focused on development of a strategy to identify or exclude clinically significant lesions, the linear association between our score and median angiographic stenosis severity suggest that a lower score would be more likely to be associated with lesser severe angiographic CAD, intermediate scores with intermediate severity, and highest scores with lesions of greatest angiographic severity.
Clearly, the score we have developed deserves validation in other unique cohorts, including potentially in patients with risk factors for CAD for whom a physician is considering the risk/benefit ratio of sending the patient for coronary angiography. A score like this might aid in such a decision. This scoring has the potential to avoid the high cost of stress tests or coronary angiography, because in comparison, the cost of biomarker testing is generally considerably lower. Of course, the challenges to acquire detailed invasive angiographic information to validate such novel approaches make initial use of cohort studies such as CASABLANCA more logical. Further exploration of the score in patients with different pre-test probabilities for significant CAD will be a next step.
Although supported by a careful methodologic approach, limitations to our study exist. For example, the CASABLANCA cohort was predominantly male, white, and representative of patients in a tertiary care referral center, although a major advantage of our cohort was its detailed characterization. In addition, the training and validation cohorts were taken from the same population, although they were randomly selected; however, there are plans to validate this score in an external cohort in the future. Also, patients referred to angiography enrolled in our study had an a priori reason for the procedure; accordingly, due to Bayesian considerations, the pre-test probability for significant CAD was higher than if a community-based cohort without indications for invasive angiography was studied. On the other hand, the score performed very well across multiple groups, including those with no prior known CAD, those presenting without acute MI, and notably in women, who have a poorer prognosis than men with CAD and who represent a diagnostic challenge (18). In addition, many subjects did not have a stress test performed before referral for angiography; as such, we were unable to correlate presence of ischemia with the degree of coronary stenosis.
We have developed a clinical and biomarker scoring strategy to reliably diagnose severe epicardial CAD. Advantages of such a reliable clinical and biomarker score include the fact such a technology can be widely disseminated in a cost-effective manner, is easily interpreted, and might be associated with a well-defined sequence of therapeutic steps to reduce risk for CAD-related complications, such as antiplatelet or lipid-lowering therapy. Further studies using our scoring system are planned.
COMPETENCY IN MEDICAL KNOWLEDGE: A risk score for coronary atherosclerosis that combines 2 clinical features (male sex and previous PCI) with the results of 4 biomarker assays (adiponectin, apolipoprotein C-I, KIM-1, and midkine) can predict the presence of ≥70% angiographic coronary stenosis with high sensitivity and specificity.
TRANSLATIONAL OUTLOOK: Further studies are needed to validate this risk assessment tool in a broader population of patients and to define its role in the selection of patients for stress testing and/or coronary angiography.
The authors thank Dr. Amalia Magaret for her role in the statistical review of the paper.
For supplemental figures and tables, please see the online version of this article.
This work was supported by a grant from Prevencio, Inc. Dr. Ibrahim is supported by the Dennis and Marilyn Barry Fellowship in Cardiology. Dr. Januzzi is supported in part by the Hutter Family Professorship in Cardiology; has received grant support from Siemens, Singulex, and Prevencio; consulting income from Roche Diagnostics, Critical Diagnostics, Sphingotec, Phillips, and Novartis; and participates in clinical endpoint committees/data safety monitoring boards for Novartis, Amgen, Janssen, and Boehringer Ingelheim. Dr. Gaggin is supported in part by the Ruth and James Clark Fund for Cardiac Research Innovation; and has received grant support from Roche and Portola; consulting income from Roche Diagnostics, American Regent, Amgen, Boston Heart Diagnostics, and Critical Diagnostics; and research payments for clinical endpoint committees for EchoSense and RadioMeter. Mr. Magaret is a consultant to Prevencio, Inc. Ms. Rhyne is an employee of Prevencio, Inc. All other authors have reported that they have no relationships relevant to the contents of this paper to disclose. Drs. Ibrahim and Januzzi contributed equally to this work. Sanjay Kaul, MD, served as Guest Editor for this paper.
- Abbreviations and Acronyms
- Apo C-I
- apolipoprotein C-I
- area under the curve
- coronary artery disease
- confidence interval
- computed tomography
- hazard ratio
- kidney injury molecule–1
- myocardial infarction
- negative predictive value
- odds ratio
- percutaneous coronary intervention
- positive predictive value
- receiver-operating characteristic
- Received November 16, 2016.
- Revision received November 30, 2016.
- Accepted December 1, 2016.
- American College of Cardiology Foundation
- ↵(2015) Centers for Disease Control and Prevention. Heart Disease Facts, https://www.cdc.gov/heartdisease/facts.htm. Accessed September 1, 2016.
- Wilson P.W.,
- D’Agostino R.B.,
- Levy D.,
- Belanger A.M.,
- Silbershatz H.,
- Kannel W.B.
- Pen A.,
- Yam Y.,
- Chen L.,
- Dennie C.,
- McPherson R.,
- Chow B.J.
- Bolton J.L.,
- Stewart M.C.,
- Wilson J.F.,
- Anderson N.,
- Price J.F.
- Gaggin H.K.,
- Bhardwaj A.,
- Belcher A.,
- et al.
- Goliasch G.,
- Kleber M.E.,
- Richter B.,
- et al.
- Sumi Y.,
- Muramatsu H.,
- Takei Y.,
- Hata K.I.,
- Ueda M.,
- Muramatsu T.
- Sato W.,
- Kadomatsu K.,
- Yuzawa Y.,
- et al.
- Rizza S.,
- Gigli F.,
- Galli A.,
- et al.
- Persson J.,
- Strawbridge R.J.,
- McLeod O.,
- et al.
- Jong M.C.,
- Dahlmans V.E.,
- van Gorp P.J.,
- et al.
- Bonventre J.V.
- Kohli P.,
- Gulati M.