Canadian Journal of Cardiology 31 (2015) 709e716

Clinical Research

A New Integrated Clinical-Biohumoral Model to Predict Functionally Significant Coronary Artery Disease in Patients With Chronic Chest Pain Chiara Caselli, PhD,a Daniele Rovai, MD,a Valentina Lorenzoni, MSc,b Clara Carpeggiani, MD,a Anna Teresinska, PhD,c Santiago Aguade, MD,d Giancarlo Todiere, MD,e Alessia Gimelli, MD,e Stephen Schroeder, MD,f Giancarlo Casolo, MD,g Rosa Poddighe, MD, PhD,g Francesca Pugliese, MD,h Dominique Le Guludec, MD,i Serafina Valente, MD,j Gianmario Sambuceti, MD,k Pasquale Perrone-Filardi, MD,l Silvia Del Ry, MSc,a Martina Marinelli, PhD,a Stephan Nekolla, PhD,m Mikko Pietila, MD,n Massimo Lombardi, MD,e Rosa Sicari, MD,a Arthur Scholte, MD, PhD,o Jose Zamorano, MD,p Philipp A. Kaufmann, MD,q S. Richard Underwood, MD,r Juhani Knuuti, MD,n Daniela Giannessi, MSc,a and Danilo Neglia, MD, PhD;a,e The EVINCI Study Investigators a

Institute of Clinical Physiology, National Research Council, Pisa, Italy; b Scuola Superiore Sant’Anna, Pisa, Italy; c Institute of Cardiology, Warsaw, Poland; d Institut Catala de la Salut, Barcelona, Spain; e Fondazione Toscana G. Monasterio, Pisa, Italy; f Kliniken des Landkreises Göppingen, Göppingen, Germany; g Ospedale della Versilia, Lido di Camaiore, Italy; h Centre for Advanced Cardiovascular Imaging, NIHR Cardiovascular Biomedical Research Unit at Barts, William Harvey Research Institute, Barts and The London School of Medicine, Queen Mary University of London, London, United Kingdom; i APHP, Groupe Hospitalier Bichat-Claude Bernard, De partement Hospitalo-Universitaire FIRE and Universite Paris Diderot, Paris, France; j Azienda Ospedaliero Universitaria Careggi, Firenze, Italy; k Università di Genova, Genoa, Italy; l Università di Napoli Federico II, Naples, Italy; m Klinikum rechts der Isar der Technischen Universitat Munchen, Munchen, Germany; n University of Turku and Turku University Hospital, Turku, Finland; o Leiden University Medical Center, Leiden, The Netherlands; p University Hospital Clinico San Carlos, Madrid, Spain; q University Hospital Zurich, Zurich, Switzerland; r Imperial College London, United Kingdom

See editorial by Azzalini and Jolicoeur, pages 696-698 of this issue. ABSTRACT

  RESUM E

Background: In patients with chronic angina-like chest pain, the probability of coronary artery disease (CAD) is estimated by symptoms, age, and sex according to the Genders clinical model. We investigated the incremental value of circulating biomarkers over the Genders model to predict functionally significant CAD in patients with chronic chest pain.

Introduction : Chez les patients souffrant de douleurs thoraciques pseudo de maladie coronarienne (MC) est estime e par angineuses, la probabilite les symptômes, l’âge et le sexe selon le modèle clinique Genders. Nous  la valeur incre mentale des biomarqueurs circulants par avons examine dire la MC fonctionnellement signifirapport au modèle Genders pour pre cative des patients souffrant de douleurs thoraciques chroniques.

Received for publication December 11, 2014. Accepted January 29, 2015. Corresponding author: Dr Chiara Caselli, CNR Institute of Clinical Physiology, Area della RicercadVia Moruzzi, 1 56100 Pisa, Italy. Tel.: 39050-3152019; fax: 39-050-3152166. E-mail: [email protected] See page 716 for disclosure information.

In patients with chest pain symptoms suggestive of stable coronary artery disease (CAD), the goal of noninvasive diagnostic screening is to identify individuals with functionally significant CADdie, obstructive coronary lesions causing myocardial ischemiadwho should undergo invasive coronary angiography and eventually coronary revascularization to

http://dx.doi.org/10.1016/j.cjca.2015.01.035 0828-282X/Ó 2015 Canadian Cardiovascular Society. Published by Elsevier Inc. All rights reserved.

710

Canadian Journal of Cardiology Volume 31 2015

Methods: In 527 patients (60.4 years, standard deviation, 8.9 years; 61.3% male participants) enrolled in the European Evaluation of Integrated Cardiac Imaging (EVINCI) study, clinical and biohumoral data were collected. Results: Functionally significant CADdie, obstructive coronary disease seen at invasive angiography causing myocardial ischemia at stress imaging or associated with reduced fractional flow reserve (FFR < 0.8), or bothdwas present in 15.2% of patients. High-density lipoprotein (HDL) cholesterol, aspartate aminotransferase (AST) levels, and highsensitivity C-reactive protein (hs-CRP) were the only independent predictors of disease among 31 biomarkers analyzed. The model integrating these biohumoral markers with clinical variables outperformed the Genders model by receiver operating characteristic curve (ROC) (area under the curve [AUC], 0.70 [standard error (SE), 0.03] vs 0.58 [SE, 0.03], respectively, P < 0.001) and reclassification analysis (net reclassification improvement, 0.15 [SE, 0.07]; P ¼ 0.04). Cross-validation of the ROC analysis confirmed the discrimination ability of the new model (AUC, 0.66). As many as 56% of patients who were assigned to a higher pretest probability by the Genders model were correctly reassigned to a low probability class (< 15%) by the new integrated model. Conclusions: The Genders model has a low accuracy for predicting functionally significant CAD. A new model integrating HDL cholesterol, AST, and hs-CRP levels with common clinical variables has a higher predictive accuracy for functionally significant CAD and allows the reclassification of patients from an intermediate/high to a low pretest likelihood of CAD.

thodes : Nous avons recueilli les donne es cliniques et bioMe carts types, 8,9 ans; 61,3 % humorales chez 527 patients (60,4 ans, e tude europe enne EVINCI de participants de sexe masculin) inscrits à l’e (Evaluatin of Integranted Cardiac Imaging). sultats : La MC fonctionnellement significative, c’est-à-dire la MC Re e à l’angiographie invasive causant l’ische mie obstructive observe e à la re duction de la myocardique à l’imagerie de stress ou associe serve de de bit fractionnaire (RDF < 0,8), ou les deux, e tait pre sente re rol à lichez 15,2 % des patients. Les concentrations de choleste ines de haute densite  (HDL), d’aspartate-aminotransfe rase poprote ine C re active haute sensibilite  (CRP-hs) e taient les (AST) et de prote dicteurs inde pendants de la maladie parmi les 31 bioseuls pre s. Le modèle inte grant ces marqueurs biohumorals marqueurs analyse aux variables cliniques surpassait le modèle Genders par la courbe ristique d’efficacite  du re cepteur (ROC; surface sous la courbe caracte [SSC], 0,70 [erreur type (ET) 0,03] vs 0,58 [ET, 0,03], respectivement, lioration nette de la P < 0,001) et l’analyse de reclassification (ame e de reclassification, 0,15 [ET, 0,07]; P ¼ 0,04). La validation croise  de discrimination du nouveau l’analyse ROC confirmait la capacite taient re partis modèle (SSC, 0,66). Jusqu’à 56 % des patients qui e  avant test plus e leve e par le modèle Genders selon une probabilite taient correctement re partis dans une classe de faible probabilite  e gre . (< 15 %) par le nouveau modèle inte cision pour pre dire Conclusions : Le modèle Genders a une faible pre grant la MC fonctionnellement significative. Un nouveau modèle inte rol à HDL, d’AST et de CRP-hs aux vales concentrations de choleste cision plus e leve e pour la MC riables cliniques communes a une pre fonctionnellement significative et permet la reclassification des pa interme diaire/e leve e avant test à une probatients d’une probabilite  faible de MC. bilite

improve outcome. The recent European guidelines on stable CAD suggest a specific diagnostic pathway in patients with chest pain of uncertain origin.1 Patients with a low pretest likelihood of CAD (< 15%), estimated by the Genders clinical predictive model, do not need further testing. In patients with an intermediate or high pretest likelihood (
15%), a stress test to detect ischemia is recommended.1 In the presence of extensive myocardial ischemia, patients are identified as being at high risk of future cardiac events (annual mortality > 3%), and coronary angiography with possible revascularization is indicated. Thus, the whole diagnostic process in these patients is conditioned by the accurate estimation of the pretest likelihood of CAD. Current models estimate the pretest likelihood of CAD on the basis of symptom characteristics, age, and sex, with or without inclusion of major risk factors, and these models have been revised to adapt to contemporary patient populations.2,3 However, the available models have been tested only for their ability to predict the presence of anatomic CAD (coronary stenosis > 50%) at invasive coronary angiography and not to predict functionally significant CAD, which is known to be associated with high prognostic risk. Accordingly, there is a need to improve actual models to better select patients at higher risk and ultimately avoid a suboptimal diagnostic and therapeutic yield with invasive coronary angiography.4 Circulating biomarkers are able to stratify the risk of cardiovascular events in the general population and in patients with stable CAD, acute coronary syndromes, or heart failure.5-7 Recently, a model based on multiple biomarkers was tested to

predict CAD in patients referred for coronary angiography.8 No clinical or biohumoral model has ever been tested for the ability to predict a functionally significant disease. With this in mind, we undertook the present study to comparatively evaluate the accuracy of an integrated model, which combines clinical data with circulating biomarkers, and of a current clinical model to predict functionally significant CAD in a contemporary European population of patients with chronic chest pain of uncertain origin.9 Methods Study design, population, and diagnostic protocol This investigation is part of the Evaluation of Integrated Cardiac Imaging (EVINCI) trial (http://www.clinicaltrials. gov, NCT00979199).9 Patients with stable chest pain or equivalent symptoms and an intermediate probability of CAD were prospectively enrolled at 14 European centres. Patients with acute coronary syndrome, known CAD, left ventricular ejection fraction < 35%, more than moderate valve disease, cardiomyopathy, or contraindications to stress imaging were excluded. In patients fulfilling inclusion and exclusion criteria (Supplemental Table S1), blood samples were collected before noninvasive imaging. According to the protocol, patients underwent a study of coronary anatomy by coronary computed tomography angiography and at least 1 coronary artery functional imaging test,9 including myocardial perfusion imaging (MPI) and ventricular wall motion imaging

Caselli et al. Results From the EVINCI Study

(WMI). Standard acquisition and analysis protocols were based on available international guidelines.10-12 If at least 1 noninvasive anatomic or functional imaging study was abnormal, as judged by the recruiting centre, patients underwent invasive coronary angiography (ICA) and measurement of fractional flow reserve (FFR) when indicated. Further clinical management was at the discretion of the local supervising clinician. Revascularization procedures within 30 days of ICA were recorded. Ethical approval was provided by each participating centre, and all participants gave written informed consent. Blood collection and analysis Blood withdrawal and initial processing of blood samples were performed at the 14 centres according to a standardized protocol. Specifically, blood specimens were centrifuged, plasma and serum aliquots were provisionally stored in a local refrigerator at 80 C and shipped to the coordinating centre (CNR, Clinical Physiology Institute, Pisa, Italy). The final collection and cryoconservation of samples were performed at a biological bank at the core laboratory. The patients who completed the protocol and whose plasma samples were available at the central biohumoral core laboratory were eligible for inclusion in this biohumoral substudy. The selected biohumoral markers and their relative analytical methods are reported in the Supplemental Table S2. The methods used were previously standardized in the core laboratory regarding sensitivity, accuracy, reproducibility, and working range (determination of analytes with an imprecision < 10%). The homeostatic model assessment index was calculated as fasting glucose (mg/ dL)  fasting insulin (pmol/L)/8.66.13 All determinations were made in duplicate to reduce the analytical variability. The operators who analyzed the blood samples were blinded to the clinical data. Study end point The diagnostic end point was functionally significant CAD, which was defined, in agreement with the European Society of Cardiology 2013 CAD guidelines,1 as the presence of at least 1 of the 3 following findings at stress imaging and invasive coronary angiography: 1. > 50% stenosis of the left main coronary artery or the proximal left anterior descending (LAD) artery, left circumflex (LCx) artery, or right coronary artery (RCA) associated with severe ischemia on stress imaging. Myocardial ischemia was considered severe if it involved > 10% of the left ventricular myocardium, as documented by a summed difference score
5 at stress MPI or by a segmental difference score
3 at stress WMI. 2. > 50% stenosis of the left main coronary artery or proximal LAD artery (or both), LCx artery, or RCA, associated with a FFR < 0.80. 3. > 90% stenosis of the left main coronary artery or proximal LAD artery, or both. Data analysis Differences according to the predefined diagnostic end point were assessed by use of an independent t test or

711

Mann-Whitney test in the case of continuous variables according to their distribution and by means of a c2 or Fisher exact test in case of categorical data. Univariate and multivariate logistic regression analyses were used to estimate the association between biohumoral variables and the end point. To correctly estimate the variance in the data, given the multicentre nature of the study, all analyses were performed accounting for clustering of patients within each centre. All regression models were stratified over centres using pooling estimates. Covariate selection in multivariate analysis was done using clinical and statistical criteria. Specifically, all variables with P < 0.10 at univariate analysis were considered for multivariable models. Backward and forward stepwise selections were used to assess model stability using P < 0.10 as a threshold to include or exclude a variable. The clinical model, including clinical data (age, sex, and type of angina) according to the Genders model,2 was tested in the EVINCI population for the ability to predict functionally significant CAD. A multivariate model including only laboratory variables was developed to identify the biohumoral independent predictors of the study end point. Clinical data (age, sex, and type of angina) were added to obtain an integrated clinical and biohumoral model (EVINCI model). A limit of 10 events (positive diagnosis) for each dependent variable was considered. The discrimination and risk reclassification ability of the integrated model were evaluated and compared with the Genders clinical model. Discrimination ability was assessed by calculating the area under the receiver operating characteristic curve (AUC ROC), and C-statistic was used to make comparisons. Improvement in reclassification was evaluated by examining the proportion of individuals reclassified correctly by the EVINCI model using the categorical net reclassification improvement metric, with 15% as a cutoff value. Patients with a low probability were differentiated from those with an intermediate or high probability (
15%) of significant disease, the latter of whom should undergo further testing according to the 2013 European Society of Cardiology (ESC) stable CAD guidelines.1 Cross-validation of the ROC analysis and cross-validated Hosmer-Lemeshow test results were obtained to assess the internal validity of the new model. Moreover, an indication of external validity of the model was assessed by performing ROC analysis in an independent population of patients with suspected or known CAD hospitalized at the National Research Council Institute of Clinical Physiology/Fondazione Toscana G. Monasterio of Pisa and registered in a dedicated database (IMAGE). Details of this database have been previously described.14 Specifically, 186 consecutive patients with suspected CAD who were hospitalized between January 2000 and October 2005 and fulfilled the inclusion and exclusion criteria of the EVINCI study were selected from the whole population for model validation. To evaluate the clinical utility of the integrated EVINCI model, predicted probability estimates from that model (referred to as the EVINCI score) were obtained and used to categorize patients according to quartiles of this multivariable score. The proportions of patients falling below or above the 15% probability cutoff using the Genders score and the EVINCI score were compared by c2 test.

712

Canadian Journal of Cardiology Volume 31 2015

Table 1. Patient characteristics of the study population according to the presence/absence of functionally significant CAD Variables Demographics Age, y Male sex (%) Cardiovascular risk factors (%) Family history Diabetes mellitus Hypertension Hypercholesterolemia Obesity Smoking within the past year Symptoms (%) Typical angina Atypical angina Nonanginal chest pain Anatomic CAD (%) > 50% stenosis Revascularization within 30 d (%) Percutaneous coronary intervention Coronary artery bypass grafting

Study population (n ¼ 527)

No significant CAD (n ¼ 447)

Significant CAD (n ¼ 80)

60.4 (8.9) 323 (61.3)

60.2 (9.0) 263 (58.8)

61.3 (8.4) 60 (75.0)

186 138 332 313 123 128

164 116 278 261 106 101

(35.3) (26.2) (63.0) (59.4) (23.3) (24.3)

(36.7) (26.0) (62.2) (58.4) (23.7) (22.6)

22 22 54 52 17 27

P value 0.300 0.006

(27.5) (27.5) (67.5) (65.0) (21.3) (33.8)

0.113 0.772 0.365 0.267 0.632 0.032

134 (25.4) 318 (60.4) 75 (14.2)

112 (25.1) 276 (61.7) 59 (13.2)

22 (27.5) 42 (52.5) 16 (20.0)

0.190

166 (32.7)

88 (20.6)

80 (100)

< 0.001

90 (17.1) 13 (2.5)

50 (11.2) 6 (1.3)

40 (50) 7 (8.8)

< 0.001 < 0.001

CAD, coronary artery disease.

All analyses were performed using Stata, version 10 (StataCorp LP, College Station, TX). A 2-sided value of P < 0.05 was considered statistically significant. Results Plasma samples were collected at a central core laboratory in 619 of the 697 (89%) patients enrolled in the EVINCI study. Of these 619 patients, 50 were excluded from the analysis because of inadequate quality of the biological samples, and 42 patients were excluded because they did not complete the study protocol. Thus, 527 participants were eligible for the present investigation. In the whole population, the pretest likelihood of anatomic CAD (> 50% coronary stenosis) according to the Genders model was 49% (interquartile range, 34%-59%). Mean age was 60.4 years, the standard deviation was 8.9 years, and 61.3% of patients were men. The majority of participants (60.4%) had atypical chest pain, whereas 25.4% had typical chest pain, and 14.2% had nonanginal chest pain. Anatomic CAD was diagnosed in 167 (33%) patients. Eighty patients had functionally significant CAD (15.2%) and differed from the remaining population because of a higher frequency of male sex, smoking habit, anatomic CAD, and revascularization procedures (Table 1). In detail, 18 patients had a > 90% stenosis of the left main coronary artery or the proximal LAD artery, 46 patients had at least 1 50%-90% stenosis of major coronary vessels associated with severe ischemia at stress imaging, and 16 patients had at least 1 50%-90% stenosis of major coronary vessels associated with an FFR < 0.8. Patients with functionally significant CAD presented with a worse lipid profile (lower high-density lipoprotein [HDL] cholesterol), lower apolipoprotein A1 levels, an increased total-to-HDL cholesterol ratio, higher levels of aspartate aminotransferase (AST), and high sensitive C-reactive protein (hs-CRP) levels than patients without significant CAD (Supplemental Table S3). Prediction of functionally significant CAD The Genders clinical model had a limited ability to predict functionally significant CAD in the EVINCI population, showing an area under the ROC curve of 0.58 (standard error

[SE], 0.04). Extension of the clinical model including risk factorsdie, diabetes, hypertension, dyslipidemia, and smokingddid not significantly increase the predictive accuracy (P ¼ 0.176). When tested for its ability to detect a softer diagnostic end point (anatomic CAD), the Genders model had an area under the ROC curve of 0.63 (SE, 0.03). Again, extension of the clinical model including risk factors did not significantly increase the predictive accuracy (P ¼ 0.161). From the several biohumoral variables individually associated with the study end point at univariate analysis, the multivariate model identified AST and hs-CRP levels and HDL cholesterol as the only independent predictors of functionally significant CAD (Table 2). At ROC curve analysis, the biohumoral model built using these 3 variables showed an AUC of 0.68 (SE, 0.03). The integration of AST and hs-CRP levels and HDL cholesterol with clinical variables into a new model (Table 3) increased the area under the ROC curve up to 0.70 (SE, 0.03) (P ¼ 0.002 vs Genders model) (Fig. 1). Reclassification analysis The same cutoff value of 15% used in the 2013 ESC stable CAD guidelines to differentiate patients with a low probability from those with an intermediate or high probability of anatomic CAD was chosen to identify patients with a low probability from those with an intermediate or high probability of functionally significant CAD.1 The integrated model showed a net reclassification improvement value of 0.15 (SE, 0.07) in comparison with the Genders model (P ¼ 0.041) for the study end point. This improvement was driven largely by the reclassification of 109 of 447 participants without significant CAD from a higher to a low probability class (Table 4). Model validation The internal validation confirmed the discriminatory ability of the EVINCI model. The AUC from the crossvalidated ROC curve was 0.66, whereas computation of the cross-validated Hosmer-Lemeshow test indicated a c2 value equal to 4.6 and P ¼ 0.799. The clinical and biohumoral

Caselli et al. Results From the EVINCI Study

713

Table 2. Univariate and multivariate analysis of the association between biohumoral variables and functionally significant CAD Biomarkers Univariate analysis Apolipoprotein A1 HDL cholesterol Aspartate aminotransferase High-sensitivity C-reactive protein Total/HDL cholesterol Total cholesterol Leptin Uric acid Heme oxigenase-1 Interleukin-6 LDL cholesterol Insulin Creatinine Matrix metalloproteinase-9 Urea HOMA index Matrix metalloproteinase-2 Osteopontin Thyroid-stimulating hormone Free thyroxine Apolipoprotein B Lipoprotein (a) Triglycerides Adiponectin Alanine aminotransferase Glycemia Free tri-iodothyronine Alkaline phosphatase Gamma-glutamyl transferase Total protein Multivariate analysis HDL cholesterol Aspartate aminotransferase High-sensitivity C-reactive protein

Odds ratio (95% CI)

P value

c2 test

0.99 (0.99-0.99)

< 0.001

50.96

0.96 (0.95-0.98) 1.02 (1.01-1.036) 1.62 (1.10-2.40)

< 0.001 0.002 0.02

16.26 9.59 5.93

1.23 0.99 0.97 1.18 1.05 1.10 0.99 1.01 2.22 1.00 1.01 1.03 1.00 1.00 0.95 0.98 1.00 1.00 1.00 0.99 1.00 1.00 0.88 1.00 1.00 0.96

(1.03-1.47) (0.98-1.00) (0.93-1.01) (0.97-1.43) (0.989-1.12) (0.97-1.25) (0.99-1.00) (1.00-1.03) (0.70-7.06) (1.00-1.00) (0.99-1.04) (0.98-1.09) (1.00-1.01) (1.00-1.01) (0.83-1.09) (0.94-1.03) (0.99-1.01) (0.99-1.01) (1.00-1.00) (0.97-1.02) (0.99-1.02) (0.99-1.01) (0.51-1.55) (0.99-1.02) (0.99-1.01) (0.61-1.52)

0.96 (0.94-0.98) 1.02 (1.01-1.03) 1.57 (1.10-2.23)

0.02 0.03 0.10 0.11 0.11 0.14 0.16 0.17 0.18 0.23 0.28 0.30 0.36 0.43 0.45 0.46 0.38 0.51 0.49 0.62 0.63 0.65 0.66 0.80 0.85 0.87

5.21 4.68 2.73 2.58 2.56 2.22 2.01 1.86 1.82 1.47 1.18 1.08 0.83 0.63 0.57 0.56 0.48 0.43 0.43 0.25 0.24 0.20 0.19 0.07 0.04 0.03

< 0.001 < 0.001 0.01

CAD, coronary artery disease; CI, confidence interval; HDL, high-density lipoprotein; HOMA, homeostatic model; LDL, low-density lipoprotein.

characteristics of the 186 patients from the IMAGE database selected for external validation are summarized in Supplemental Table S4. The EVINCI model also showed good performance in this population with a higher prevalence of functionally significant CAD (40%). The area under the ROC curve was equal to 0.72 (SE, 0.04), similar to that observed for the EVINCI population.

Figure 1. Comparison of the accuracy to predict functionally significant coronary artery disease by receiver operating characteristic curves obtained for the clinical model (Genders) and the integrated clinical-biohumoral model. AUC, area under the curve.

EVINCI score Figure 2 illustrates the actual prevalence of functionally significant CAD according to quartiles of the EVINCI score obtained by integrating clinical and biohumoral data. The prevalence of significant disease increased progressively (P < 0.001) according to the EVINCI score quartiles, and the patients in the highest quartile had about a 5-fold greater prevalence of functionally significant CAD compared with individuals in the lowest quartile. Similarly, the frequency of revascularization increased from the first to the fourth quartiles. Interestingly, the frequency of revascularization exceeded that of functionally significant CAD in all the probability quartiles. Of the 527 patients in the study population, 301 (57%) patients with a
15% probability by the Genders model shifted to the low-probability class when using the EVINCI

Table 3. New integrated clinical and biohumoral model to predict functionally significant CAD Significant CAD Male sex Age Chest pain (vs atypical chest pain) Nonanginal chest pain Typical chest pain Biohumoral variables HDL cholesterol, mg/dL (per unit of change) High-sensitivity C-reactive protein, mg/dL (per unit of change) Aspartate aminotransferase, IU/L (per unit of change)

Coefficient Odds ratio

95% CI

0.49 0.03

1.64 1.03

1.49-2.78 1.02-1.09

0.40 0.20

1.50 1.22

1.41-1.57 0.96-1.87

0.04

0.97

0.95-0.99

0.46

1.60

1.01-1.05

0.02

1.02

1.02-0.95

Table 4. Reclassification table for the integrated EVINCI model vs Genders probability using 15% probability of functionally significant CAD as cutoff

Variable Significant CAD No significant CAD

CAD, coronary artery disease; CI, confidence interval; HDL, high-density lipoprotein.

Probability categories EVINCI model

Probability categories Genders model

< 15%


15%

Total

< 15%
15% Total < 15%
15% Total

14 14 28 164 109 273

15 37 52 48 126 174

29 51 80 212 235 447

CAD, coronary artery disease; EVINCI, Evaluation of Integrated Cardiac Imaging.

714

Figure 2. Prevalence of functionally significant coronary artery disease and frequency of revascularization in the study population according to quartiles of the Evaluation of Integrated Cardiac Imaging (EVINCI) score.

score (Fig. 3). The prevalence of functionally significant CAD was 24.3% and 8.3% in the higher and lower EVINCI probability classes (P < 0.001), respectively. Discussion This study demonstrates that the currently used Genders clinical model has a low performance in discriminating patients with functionally significant CAD among those with stable chest pain of uncertain origin. In these patients, the new EVINCI model, which integrates clinical data with a few biohumoral markers, outperforms the Genders model in predicting significant disease. This approach allows correct classification of a low pretest probability class in as many as 57% of patients with an intermediate/high probability estimated by the current models. Comparison of predictive models of CAD The model developed by Diamond and Forrester in the 1970s and largely used worldwide,15 has been updated by Genders et al.2 in a contemporary population of patients referred for coronary angiography. The same authors extended their model to a population at lower risk, ie, patients referred for coronary computed tomography angiography, and included cardiovascular risk factors among the predictive variables.3 This extended model allowed for more accurate estimation of the pretest probability of anatomic CAD compared with the initial model. In the EVINCI population, both the original and extended Genders models showed a moderate value in predicting anatomic CAD. None of these models had been previously tested for their ability to predict the presence of functionally significant CAD, ie, obstructive coronary artery stenosis causing myocardial ischemia at stress imaging or associated with reduced FFR, or both. In the present study, only 80 of the 167 patients with anatomic CAD had evidence of functionally significant disease. The performance of the Genders model to predict this harder diagnostic end point was suboptimal. We developed a new integrated clinical-biohumoral model that was demonstrated to be accurate in recognizing patients known to be at

Canadian Journal of Cardiology Volume 31 2015

Figure 3. Number of patients shifting from a higher to a lower probability of coronary artery disease (from
15% to < 15%) applying the Evaluation of Integrated Cardiac Imaging (EVINCI) score instead of the Genders score.

high risk of events at follow-up. There are several factors that may explain the different performance of predictive models in our population. The Genders models were specifically designed for predicting anatomic CAD; thus, it is not surprising that these models had a lower predictive value with respect to functionally significant disease. Genders initially studied candidates for invasive coronary angiography, in whom the prevalence of anatomic CAD was high (58.3% of patients). In contrast, we enrolled candidates for noninvasive screening of stable chest pain or equivalent symptoms. These patients had a lower prevalence of anatomic CAD at invasive angiography (33%) and a lower prevalence of typical angina (25% vs 53%), younger age (60 years [SE, 9 years] vs 62 years [SE, 10 years]) and there was a higher prevalence of female patients (39% vs 23%). In the retrospective analysis performed by Genders, referral bias and differences in local practice could have caused the selection of patients at the opposite spectrum of disease probability. Thus, in Genders’ population, the correlation of clinical variables with the presence/absence of CAD is expected to be higher than in the EVINCI population, which included patients with a more homogeneous probability of disease. Circulating biomarkers as predictors of CAD Circulating biomarkers putatively associated with mechanisms underlying coronary atherosclerosisdincluding vascular inflammation, aberrant lipid regulation, metabolic hormones, and extracellular matrix remodellingdwere selected for evaluation in this study. Among the 36 preliminary targets, only HDL cholesterol, hs-CRP, and AST were independent predictors of functionally significant CAD. These biomarkers largely accounted for the better predictive accuracy of the EVINCI model expressing the residual risk over more common clinical determinants. The present results support the relevance of inflammation, lipid regulation, and metabolism as main determinants of

Caselli et al. Results From the EVINCI Study

ischemic coronary disease.8 Specifically, among biomarkers, low HDL cholesterol was the most powerful independent predictor of functionally significant CAD in the EVINCI population. This is not surprising in view of the existing compelling evidence of low HDL cholesterol as an independent predictor of coronary disease and major cardiovascular events.16,17 Together with its ability to promote cholesterol efflux from macrophages in the arterial wall, HDL cholesterol has several additional vascular protective actions, including anti-inflammatory properties and favourable effects on endothelial function.18,19 Consistent with the hypothesis that inflammation plays a role in the pathogenesis of CAD and its more severe manifestations, hs-CRP was positively associated with functionally significant CAD in the present study. CRP is a well-known nonspecific indicator of inflammatory status,20 and for more than a decade, data from large-scale prospective cohorts in the United States and Europe have indicated its predictive value for the estimation of cardiovascular risk. In a recent metaanalysis of 52 prospective studies that included almost 250,000 participants without a history of cardiovascular disease, HDL cholesterol and CRP were independent and additive predictors of future cardiovascular events over clinical and demographic variables.21 An increased circulating level of AST was also an independent predictor of functionally significant CAD in our population. There is increasing interest in liver enzymes as novel markers of cardiovascular risk related to multiple cardiometabolic risk factors,22 in particular diabetes and insulin resistance syndrome.22-24 Also in the present study, increased levels of liver enzymes (AST and alanine transaminase) were correlated with the presence of hyperglycemia or insulin resistance (data not shown). However, an association between serum liver enzyme activity and the presence and severity of angiographically documented premature CAD has also been documented independently of other confounding factors such as components of metabolic syndrome and hs-CRP concentration,25 as in the present study. Whether AST could represent metabolic alterations underlying coronary artery disease, an association with inflammatory parameters,24 or even subtle myocardial damage consequent to coronary artery disease and ischemia could be a matter of future investigations.26 Clinical implications According to the 2013 ESC Guidelines for the management of stable CAD,1 the estimation of pretest likelihood of disease is a key point to address further diagnostic assessment and ultimately appropriate treatment in patients with chronic chest pain. Moreover, the ultimate goal of the whole diagnostic screening is not only the recognition of patients with anatomic CAD but also, more specifically, patients with functionally significant coronary lesions able to cause myocardial ischemia. These patients have a higher risk of future coronary events and as such may benefit from invasive procedures and eventual revascularization. In current practice, patients without evidence of ischemia often undergo revascularization based only on the anatomic detection of coronary artery lesions, as is also suggested in the present population by an excess of revascularization procedures compared with the

715

presence of documented functionally significant CAD (Fig. 2). Thus, the availability of predictive models able to more accurately estimate the pretest likelihood of functionally significant coronary artery disease could increase the diagnostic and therapeutic yield of invasive procedures.4 This need is even more relevant in populations currently referred for noninvasive screening of CAD, which are characterized, as in the EVINCI study, by a high prevalence of atypical symptoms and a low prevalence of disease. The current clinical predictive models of CAD have not been tested against such a hard diagnostic end point. In the present study, we could demonstrate that current clinical predictive models are not effective for this purpose, whereas an integrated model combining common clinical variables with a few circulating biomarkers has high predictive accuracy. In patients with known CAD, circulating biomarkers are used to identify cardiovascular risk factors, monitor their trends, assess the effectiveness and safety of therapy, and predict cardiovascular risk.1 Biomarkers are less frequently used for screening purposes in patients with suspected CAD. In our study, for the first time to our knowledge, biohumoral markers were used to integrate clinical variables to predict functionally significant CAD. The cost of dosing the circulating biomarkers included in the EVINCI score is low, roughly V10/patient. In addition, based on the present results, an easy-to-use online probability calculator could be developed to be used as a point-of-care screening tool. After validation in a larger population and in a randomized controlled trial, the EVINCI score might represent a user-friendly low-cost screening strategy that is potentially useful for better discrimination of higher-risk patients and to exclude low-risk patients from further procedures in the context of a personalized medicine approach. Study limitations The low prevalence of the diagnostic end point (functionally significant CAD) in the present population posed some limitations in the statistical analysis. No more than 8 variables could be included in predictive models to avoid overfitting, and cutoff values of pretest probability higher than 15% could not be used. Moreover, it was unfeasible to select from the population both a sample group for model development and a new sample to be used as a validation group. We selected biomarkers on the basis of previous experimental and clinical studies that reported their association with atherosclerosis, vascular remodelling, inflammation, and metabolism, all processes underlying CAD. Other biomarkers expressing myocardial damage and associated with prognosis in stable CAD, such as high-sensitivity cardiac troponins, might have provided additional information but were not available in the present study. Conclusions A new integrated model including clinical and biohumoral variables allowed the prediction of functionally significant CAD in a population of patients with chronic chest pain and a low prevalence of disease, outperforming current clinical models. The present results could have relevant implications

716

for cost containment and effectiveness of the diagnostic process in patients with suspected CAD. The integrated model allows reclassification to lower probability in more than 50% of patients who are candidates for further testing by current clinical models and accurate identification of patients with an increased likelihood of high-risk disease who would benefit from noninvasive imaging and eventual invasive procedures. Funding Sources This work was supported by a grant from the European Union FP7-CP-FP506 2007 project (grant agreement no. 222915, EVINCI). Disclosures The authors have no conflicts of interest to disclose. References

Canadian Journal of Cardiology Volume 31 2015 12. Wahl A, Paetsch I, Gollesch A, et al. Safety and feasibility of high-dose dobutamine-atropine stress cardiovascular magnetic resonance for diagnosis of myocardial ischaemia: experience in 1000 consecutive cases. Eur Heart J 2004;25:1230-6. 13. Matthews DR, Hosker JP, Rudenski AS, et al. Homeostasis model assessment: insulin resistance and beta-cell function from fasting plasma glucose and insulin concentrations in man. Diabetologia 1985;28:412-9. 14. Iozzo P, Rossi G, Michelassi C, et al. Interpretation of the “obesity paradox”: a 30-year study in patients with cardiovascular disease. Int J Cardiol 2013;168:112-6. 15. Diamond GA, Forrester JS. Analysis of probability as an aid in the clinical diagnosis of coronary-artery disease. N Engl J Med 1979;300:1350-8. 16. Mangalmurti SS, Davidson MH. The incremental value of lipids and inflammatory biomarkers in determining residual cardiovascular risk. Curr Atheroscler Rep 2011;13:373-80. 17. Arsenault BJ, Boekholdt SM, Kastelein JJ. Lipid parameters for measuring risk of cardiovascular disease. Nat Rev Cardiol 2011;8: 197-206.

1. Task Force Members, Montalescot G, Sechtem U, Achenbach S, et al. 2013 ESC guidelines on the management of stable coronary artery disease: the Task Force on the management of stable coronary artery disease of the European Society of Cardiology. Eur Heart J 2013;34: 2949-3003.

18. Barter PJ, Nicholls S, Rye KR, et al. Antiinflammatory properties of HDL. Circ Res 2004;95:764-72.

2. Genders TS, Steyerberg EW, Alkadhi H, et al. A clinical prediction rule for the diagnosis of coronary artery disease: validation, updating, and extension. Eur Heart J 2011;32:1316-30.

20. Libby P, Crea F. Clinical implications of inflammation for cardiovascular primary prevention. Eur Heart J 2010;31:777-83.

3. Genders TS, Steyerberg EW, Hunink MG, et al. Prediction model to estimate presence of coronary artery disease: retrospective pooled analysis of existing cohorts. Eur Heart J 2012;12:e3485. 4. Patel MR, Peterson ED, Dai D, et al. Low diagnostic yield of elective coronary angiography. N Engl J Med 2010;362:886-95. 5. Siasos G, Tousoulis D, Athanasiou D, et al. Novel risk factors related to stable angina. Curr Pharm Des 2013;19:1550-61. 6. Tousoulis D, Hatzis G, Papageorgiou N, et al. Assessment of acute coronary syndromes: focus on novel biomarkers. Curr Med Chem 2012;19:2572-87. 7. Giannessi D. Multimarker approach for heart failure management: perspectives and limitations. Pharmacol Res 2011;64:11-24.

19. Riwanto M, Landmesser U. High density lipoproteins and endothelial functions: mechanistic insights and alterations in cardiovascular disease. J Lipid Res 2013;54:3227-43.

21. Kaptoge S, Di Angelantonio E, Pennells L, et al. C-reactive protein, fibrinogen, and cardiovascular disease prediction. N Engl J Med 2012;4: 1310-20. 22. Porter SA, Pedley A, Massaro JM, et al. Aminotransferase levels are associated with cardiometabolic risk above and beyond visceral fat and insulin resistance: the Framingham Heart Study. Arterioscler Thromb Vasc Biol 2013;33:139-46. 23. Steinvil A, Shapira I, Ben-Bassat OK, et al. The association of higher levels of within-normal-limits liver enzymes and the prevalence of the metabolic syndrome. Cardiovasc Diabetol 2010;9:30. 24. Hanley AJ, Williams K, Festa A, et al. Insulin resistance atherosclerosis study. Elevations in markers of liver injury and risk of type 2 diabetes: the insulin resistance atherosclerosis study. Diabetes 2004;53:2623-32.

8. LaFramboise WA, Dhir R, Kelly LA, et al. Serum protein profiles predict coronary artery disease in symptomatic patients referred for coronary angiography. BMC Med 2012;10:157.

25. Masoudkabir F, Karbalai S, Vasheghani-Farahani A, et al. The association of liver aminotransferase activity with presence and severity of premature coronary artery disease. Angiology 2011;62:614.

9. Neglia D, Rovai D, Caselli C, et al. Detection of significant coronary artery disease by noninvasive anatomic and functional imaging. Circ Cardiovasc Imaging 2015;8.

26. Lott JA, Stang JM. Serum enzymes and isoenzymes in the diagnosis and differential diagnosis of myocardial ischemia and necrosis. Clin Chem 1980;26:1241-50.

10. Sicari R, Nihoyannopoulos P, Evangelista A, et al. Stress echocardiography expert consensus statement: European Association of Echocardiography (EAE) (a registered branch of the ESC). Eur J Echocardiogr 2008;9:415-37. 11. Hesse B, Tägil K, Cuocolo A, et al. EANM/ESC procedural guidelines for myocardial perfusion imaging in nuclear cardiology. Eur J Nucl Med Mol Imaging 2005;32:855-97.

Supplementary Material To access the supplementary material accompanying this article, visit the online version of the Canadian Journal of Cardiology at www.onlinecjc.ca and at http://dx.doi.org/10. 1016/j.cjca.2015.01.035.