Author + information
- Received November 26, 2017
- Revision received March 6, 2018
- Accepted March 20, 2018
- Published online January 7, 2019.
- Pierre Lantelme, MD, PhDa,b,∗ (, )
- Hélène Eltchaninoff, MD, PhDc,
- Muriel Rabilloud, MD, PhDd,e,f,g,
- Géraud Souteyrand, MD, PhDh,
- Marion Dupré, MDc,
- Marco Spaziano, MDi,j,
- Marc Bonnet, MDa,
- Clément Becle, MDa,b,
- Benjamin Riche, PhDd,e,f,g,
- Eric Durand, MD, PhDc,
- Erik Bouvier, MDi,
- Jean-Nicolas Dacher, MD, PhDk,
- Pierre-Yves Courand, MD, PhDa,b,
- Lucie Cassagnes, MD, PhDl,
- Eduardo E. Dávila Serrano, PhDb,
- Pascal Motreff, MD, PhDh,
- Loic Boussel, MD, PhDb,m,
- Thierry Lefèvre, MDi and
- Brahim Harbaoui, MD, PhDa,b
- aCardiology Department, Hôpital Croix-Rousse and Hôpital Lyon Sud, Hospices Civils de Lyon, Lyon, France
- bUniversity of Lyon, CREATIS UMR5220, INSERM U1044, INSA-15 Lyon, France
- cCardiology Service, Rouen–Charles-Nicolle University Hospital Center, National Institute of Health and Medical Research U644, Rouen, France
- dHospices Civils de Lyon, Service de Biostatistique et Bioinformatique, F-69003 Lyon, France
- eUniversité de Lyon, F-69000 Lyon, France
- fUniversité Lyon 1, F-69100 Villeurbanne, France
- gCNRS, UMR 5558, Laboratoire de Biométrie et Biologie Evolutive, Equipe Biostatistique-Santé, F-69100 Villeurbanne, France
- hDepartment of Cardiology, Gabriel Montpied University Hospital Center, Image Science for Interventional Techniques, Cardiovascular Interventional Therapy and Imaging, National Scientific Research Center UMR 6284, University of Auvergne, Clermont-Ferrand, France
- iInstitut Cardiovasculaire Paris Sud, Ramsay–Générale de Santé, France
- jDepartment of Cardiology, McGill University Health Center, Montreal, Canada
- kRadiology Department, Rouen–Charles-Nicolle University Hospital Center Rouen, France
- lRadiology Department, Gabriel Montpied University Hospital Center, and Institut Pascal, TGI UMR6602 CNRS UCA SIGMA, Faculté de Médecine, Clermont-Ferrand, France
- mRadiology Department, Hôpital Croix-Rousse, Hospices Civils de Lyon, Lyon, France
- ↵∗Address for correspondence:
Prof. Pierre Lantelme, Cardiology Department, Hôpital Croix-Rousse and Hôpital Lyon Sud, 103 Grande Rue de la Croix-Rousse, F-69004, Lyon, France.
Objectives The aim of this study was to develop a new scoring system based on thoracic aortic calcification (TAC) to predict 1-year cardiovascular and all-cause mortality.
Background A calcified aorta is often associated with poor prognosis after transcatheter aortic valve replacement (TAVR). A risk score encompassing aortic calcification may be valuable in identifying poor TAVR responders.
Methods The C4CAPRI (4 Cities for Assessing CAlcification PRognostic Impact) multicenter study included a training cohort (1,425 patients treated using TAVR between 2010 and 2014) and a contemporary test cohort (311 patients treated in 2015). TAC was measured by computed tomography pre-TAVR. CAPRI risk scores were based on the linear predictors of Cox models including TAC in addition to comorbidities and demographic, atherosclerotic disease and cardiac function factors. CAPRI scores were constructed and tested in 2 independent cohorts.
Results Cardiovascular and all-cause mortality at 1 year was 13.0% and 17.9%, respectively, in the training cohort and 8.2% and 11.8% in the test cohort. The inclusion of TAC in the model improved prediction: 1-cm3 increase in TAC was associated with a 6% increase in cardiovascular mortality and a 4% increase in all-cause mortality. The predicted and observed survival probabilities were highly correlated (slopes >0.9 for both cardiovascular and all-cause mortality). The model’s predictive power was fair (AUC 68% [95% confidence interval [CI]: 64% to 72%]) for both cardiovascular and all-cause mortality. The model performed similarly in the training and test cohorts.
Conclusions The CAPRI score, which combines the TAC variable with classical prognostic factors, is predictive of 1-year cardiovascular and all-cause mortality. Its predictive performance was confirmed in an independent contemporary cohort. CAPRI scores are highly relevant to current practice and strengthen the evidence base for decision making in valvular interventions. Its routine use may help prevent futile procedures.
Transcatheter aortic valve replacement (TAVR) represents the standard of care for relieving aortic stenosis (AS) in high-risk patients; it is also preferred in symptomatic intermediate-risk patients according to the latest European Society guidelines (1). Although TAVR efficiently normalizes the gradient across the aortic valve, approximately 25% of high-risk patients die within the first year following the procedure (2). Because of the competing risk of noncardiovascular mortality or heart failure (HF), the true benefit of TAVR is difficult to estimate. Although perioperative mortality accounts for a minor part of overall deaths and is diminishing with technical refinement (3), most of this residual risk remains high. This points to a need to better identify potentially poor TAVR responders who would not benefit from the procedure. Lack of benefit may be defined as death and/or absence of functional improvement during a relatively short period after the procedure (6 months to 1 year) (4). Within this time frame, it is likely that adverse outcomes are mainly driven by factors present before the TAVR procedure. Among various candidates (4), predictors of improvement of left ventricular (LV) function are probably highly important but remain insufficiently characterized. Vascular after-load deserves scrutiny as it becomes preeminent after relief of AS (5). Our group recently established the prognostic significance of aortic calcifications assessed manually by computed tomography (CT) on the outcomes after TAVR. In particular, ascending aortic calcification was predictive of HF (6), and total aortic calcification burden was predictive of cardiac mortality (7). This prognostic implication may also concern noncardiovascular mortality (8,9).
The C4CAPRI (4 Cities for assessing CAlcification PRognostic Impact; NCT02935491) study aimed at developing a score based on aortic calcification burden combined with classical predictors, to predict 1-year cardiovascular and all-cause mortality after TAVR.
This was a multicenter study performed in 4 high-volume French centers. Two different cohorts were considered: a “training cohort” used to develop the risk scores and a “test cohort” in which the predictive value of the model was tested.
All patients undergoing TAVR for severe AS during the study period were part of the FRANCE 2 (2,10) and of the FRANCE TAVI registries (11). The training cohort encompassed all patients treated by TAVR for severe AS at Clermont-Ferrand University Hospital, Lyon Croix-Rousse University Hospital, Institut Cardiovasculaire Paris Sud and Rouen University Hospital between January 2010, and December 2014. The test cohort comprised all patients implanted from January to December 2015 at the Lyon and Paris centers, reflecting the most recent practices available. Patients were included in the analysis if a pre-operative CT scan was available and thoracic aortic calcification (TAC) was assessed.
The C4CAPRI study was approved by the Ethical Committee (Comité de Protection des Personnes SUD-EST IV, L16-56) and by the Commission Informatique et Liberté (CNIL N° 16-065). All patients provided written informed consent to anonymous processing of their data.
CT acquisition was performed on CT scanners with at least a 4-cm z-coverage: Brilliance 64 and iCT (Philips, Best, the Netherlands) or Discovery CT750 HD (GEMS, Waukesha, Wisconsin). For each examination, the whole thoracic aorta was studied. CT scans were performed after intravenous injections of iodine-based contrast agents with electrocardiogram (ECG) gating, tube voltage ranging from 100 to 140 kV, and adapted mAs. Reconstruction parameters for axial slices ranged from 0.625 to 0.8. All images were anonymized, transferred to a core lab, and analyzed by 3 operators who were blinded to outcomes data.
Calcifications were extracted using a semiautomated dedicated software based on an open source environment available at (12). The rater first delineates the ascending, horizontal, and descending thoracic aorta by placing at least 3 points (Figure 1). The main axis of the aorta was calculated using a third-order b-spline. An adjustable tube embedding the whole aorta was created, and an initial threshold set at 550 UH was applied in the tube to detect calcifications. This threshold could be adapted to improve the performance of the algorithm. The results of the segmentation were visually asserted, and each calcification was manually adjusted (by addition or subtraction) and validated by the user. A connectivity algorithm, based on graph theory, was subsequently applied to segment each calcification. The algorithm is a simple recursive function initialized in the center of each calcification after thresholding. For each pixel, the 28 neighboring pixels were checked, and the algorithm was re-run for each pixel if its value was above the threshold. For each patient, TAC was calculated from the aortic sinus to the aortic hiatus as previously described (7); valvular calcifications were excluded.
Interobserver and intraobserver reproducibility, assessed in a subset of randomly selected patients from the training cohort (N = 50 and 75, respectively), was high with intraclass correlation coefficients estimated, respectively, at 0.997 (95% confidence interval [CI]: 0.994 to 0.998) and 0.997 (95% CI: 0.996 to 0.998).
Two TAVR systems were mainly used: a self-expandable prosthesis (Medtronic CoreValve ReValving System, Medtronic, Minneapolis, Minnesota) and a balloon-expandable prosthesis (Edwards SAPIEN valve, Edwards Lifesciences, Irvine, California); various routes (transfemoral, transapical, subclavian) and types of anesthesia were used in the different centers. The variables collected have been described elsewhere (6,10) and encompass demographics, cardiovascular history, biological variables, and echocardiographic variables.
The primary endpoint was cardiovascular death at 1 year. Secondary endpoint was all-cause mortality at 1 year. Vital status was obtained by telephone contact with patients, their relatives, caregivers, or physicians and by on-site planned visits. Follow-up was censored at 1 year following TAVR. Cardiovascular deaths were adjudicated by 2 experienced cardiologists, blinded to patient characteristics and TAC. Cardiovascular deaths were defined according to the VARC-2 criteria (13). Data collection was performed through dedicated web-based case report forms in each center, which were merged for analysis. Range checks to identify extreme values and assessments of internal consistency were applied during upload.
Construction of CAPRI risk score
The characteristics of the cohorts were described using the absolute and relative frequencies for the qualitative characteristics and the mean and the standard deviation for the quantitative characteristics. TAC was considered as a 3-class variable defined according to the tertiles for the analysis using the Kaplan-Meier method and as a continuous variable in the model used to build the risk score. In addition to the continuous TAC, the risk score was constructed from known prognostic mortality factors according to the literature (2,10,14–16) and expert opinion, split into 3 groups: 1) demographic and comorbidities; 2) atherosclerotic disease; and 3) cardiac function (Supplemental Appendix). Nonredundant factors were selected in each group after estimating correlations among factors by Spearman coefficient, tetrachoric, or polychoric coefficients according to the nature of the factors. Cox regression models were used to quantify the effects of the retained factors on mortality hazard. The prediction improvement linked to the added factors was tested using the likelihood ratio test as recommended (17) and quantified using the integrated discrimination improvement (IDI) index (18). The calibration of the final model was estimated by plotting observed survival probabilities against deciles of prediction and its discriminative performance by using the cumulative/dynamic area under the ROC curve (AUC). The bootstrap method was used to quantify and correct the optimism of the estimated AUCs. The risk scores of mortality were computed using the linear combination of the factors included in the Cox model, weighted by the regression coefficients. The same strategy of analysis was applied to cardiovascular and all-cause mortality. The predictive performance of the risk scores was evaluated on the independent test cohort by assessing its calibration and estimating the AUC.
To illustrate the impact of the risk scores on the identification of potentially futile TAVR, futility thresholds corresponding to the expected 1-year mortality (between 15% and 25% for cardiovascular mortality and between 25% and 35% for all-cause mortality) of a medically treated population (19) were estimated in the training cohort. These thresholds were used in the test cohort to give an estimate of potentially futile TAVR: that is, those patients in whom the mortality probability after TAVR would be equal to or greater than that without intervention (Supplemental Appendix).
The analysis was performed using the statistical software SAS version 9.3 (SAS Institute Inc., Cary, North Carolina), and R version 3.3.1. (R Foundation for Statistical Computing, Vienna, Austria).
Characteristics of patients and procedures
The flow of patients is shown in Figure 2. The characteristics of the training and of the test cohorts were similar concerning risk profile, route for TAVR, type of prosthesis, echocardiographic parameters, and TAC (Table 1, Supplemental Table 1). There were no substantial differences to the excluded patients (Supplemental Table 2). The rate of procedural success was high, and a marked decrease of transaortic gradient accompanied by a small increase of left ventricular ejection fraction (LVEF) was observed (Supplemental Table 1).
In the training cohort, 237 deaths were observed during the first year following TAVR, 170 of which were from cardiovascular causes. The 1-year cardiovascular and all-cause mortalities in the training cohort were estimated at 13.0% (95% CI: 11.2% to 14.8%) and 17.9% (95% CI: 15.9% to 19.1%), respectively. In the test cohort, 30 deaths—21 from cardiovascular causes—were observed. Cardiovascular and all-cause mortalities in this group were thus 8.2% (95% CI: 4.7% to 11.6%) and 11.8% (95% CI: 7.7% to 15.7%).
Correlation between factors belonging to the same group and TAC
For all 3 groups of risk factors, the correlations with TAC were low (Figure 3, Supplemental Figure 1). The highest correlation was between TAC and peripheral vascular disease or coronary disease, with coefficients estimated at 0.17.
Predictive power of TAC for cardiovascular and all-cause mortality
The survival curves of the 3 classes of TAC were significantly different for cardiovascular and all-cause mortality, respectively (Supplemental Figure 2), with a survival much lower for the patients with a TAC above the second tertile in comparison with the 2 other classes of TAC.
Among demographic variables and comorbidities, age, sex, glomerular filtration rate (GFR), and chronic obstructive pulmonary disease (COPD) were included in the Cox model. The inclusion significantly improved the prediction of cardiovascular and all-cause mortality compared with the model without covariates (p = 0.002, p < 0.0001, respectively). Inclusion of the atherosclerotic disease factors (coronary artery disease, history of stroke or transient ischemic attack, or history of peripheral vascular disease) did not significantly improve the predictive power of the model (p = 0.75 for cardiovascular and p = 0.16 for all-cause mortality). Addition of factors linked to cardiac function (LVEF, pulmonary pressure, mean gradient, dyspnea [New York Heart Association functional class], mitral regurgitation) to the 2 previous groups improved the prediction of cardiovascular and all-cause mortality significantly (p = 0.0001, p < 0.00001, respectively).
Adding TAC to the other factors improved the prediction of cardiovascular and all-cause mortality significantly (p < 0.01 and p = 0.04, respectively). An increase of 1 cm3 was associated with a 6% increase in cardiovascular mortality (hazard ratio [HR]: 1.06, 95% CI: 1.01 to 1.10) (Table 2) and a 4% increase in all-cause mortality (HR: 1.04, 95% CI: 1.00 to 1.08) (Table 3).
Calibration and discrimination of the final model
The calibration of the model including the 3 groups of factors and the TAC was good for cardiovascular (Supplemental Figure 3A) and all-cause mortality (Supplemental Figure 3B). Slopes were 0.90 for cardiovascular and 0.95 for all-cause mortality, respectively. The IDI index measuring the improvement of the discrimination ability of the model by adding the TAC was significant: estimated at 0.006 (95% CI: 0.000 to 0.025) for cardiovascular mortality and at 0.004 (95% CI: 0.000 to 0.018) for all-cause mortality (Supplemental Figures 4A and 4B). The AUC for cardiovascular mortality was 68% (95% CI: 64% to 72%); that for all-cause mortality was 68% (95% CI: 64% to 72%). The optimism of the AUC of the final models was low: estimated at 0.04% for cardiovascular mortality and 0.07% for all-cause mortality.
In comparison, the Euroscore performed less well for cardiovascular and all-cause mortality, with AUCs estimated at 56% (95% CI: 52% to 60%) and 57% (95% CI: 53% to 61%), respectively. It is interesting that adding TAC to the Euroscore was also able to improve its discrimination ability with an estimated IDI at 0.007 (95% CI: 0.001 to 0.021) for cardiovascular mortality and at 0.006 (95% CI: 0 to 0.018) for all-cause mortality (Supplemental Figures 4C and 4D). The AUCs of TAC combined with Euroscore were estimated at 59% (95% CI: 55% to 62%) and 58% (95% CI: 54% to 62%), respectively.
Risk scores and thresholds
Tables 2 and 3 present the coefficients of the risk scores constructed on the training cohort for predicting cardiovascular and all-cause mortality, respectively. Figure 4 presents the components of the risk scores and the correspondence between risk scores and predicted mortality in the training cohort. The threshold of risk score corresponding to a predicted cardiovascular mortality of 20% was estimated at 0.7 (Supplemental Table 3). The associated sensitivity and specificity were estimated at 33.4% and 87.5%, respectively. The threshold of risk score corresponding to a predicted all-cause mortality of 30% was estimated at 0.78. The associated sensitivity and specificity were estimated at 27.8% and 90.8%, respectively.
Predictive performance of the risk scores on the test cohort
The calibration of the risk scores on the test cohort was good (Supplemental Figures 3C and 3D), with slopes estimated at 0.92 and 0.95 for cardiovascular and all-cause mortality, respectively. The AUCs were estimated at 66% (95% CI: 65% to 67%) and 67% (95% CI: 65% to 67%) for cardiovascular and all-cause mortality, respectively. For patients with risk scores above the 20% mortality threshold, cardiovascular mortality probability at 1 year was estimated at 29.6%, compared with 6% for patients with risk scores below or equal to the threshold. For all-cause mortality, the 1-year probability of death in patients with risk scores above the 30% threshold was estimated at 50%, versus 8.5% in patients with risk scores below or equal to the threshold. This corresponds to 112 and 77 of 1,000 patients, respectively, in whom TAVR would potentially be futile. The stringency of the criteria may be varied, as illustrated in Figure 5.
The CAPRI score is a newly developed, dedicated score encompassing an original and meaningful variable—TAC—in addition to classical variables representative of comorbidities and atherosclerotic and cardiac factors. A 1-cm3 increase in TAC predicted a 6% increase in cardiovascular mortality and a 4% increase in all-cause mortality. The CAPRI score has a good discriminative ability for 1-year cardiovascular and all-cause mortality after TAVR.
The CAPRI score components and performance
The identification of patients in whom nonmodifiable factors will predispose to adverse outcomes post-TAVR is of utmost importance, both for ethical and economical reasons. The CAPRI score performs well when predicting cardiovascular and all-cause mortality. The validation process strictly followed recommendations (20) and represents the most ambitious attempt so far to develop a risk model for TAVR candidates. As prediction models tend to perform better on the data from which they were constructed than on new data, strategies of bootstrapping were used for internal validation to limit the optimism bias. External validation was also used as a confirmatory step to assess the predictive performance of the scores. The CAPRI score included 3 groups of meaningful factors: demographic and comorbidities, atherosclerotic disease, and cardiac function (Figure 4). The components of these 3 groups have been largely found to have an important prognostic value among patients undergoing TAVR, although differences exist according to different settings (4). Although comorbidities and cardiac function were mostly strongly associated with mortality, the inclusion of other factors makes the scores less dependent on the training cohort and more extendable to other settings (20). The model is based on pre-procedural factors only and can therefore be used prospectively to inform clinical decision making. Importantly, the score is based on contemporary practice and not on cohorts from earlier trials (16,21), which is relevant with respect to the continuing extension of indications for TAVR.
Based on our previous analyses (6,7), we added TAC to the predictive variables. We used continuous TAC because categorization of a continuous variable is usually associated with a loss of power and inaccurate estimation of the effect of the covariate (22,23). TAC significantly increased the predictive ability of the model as shown by the likelihood test (17) and further highlighted by the IDI index (18). TAC appeared poorly correlated with other markers of atherosclerosis or risk factors in keeping with the Multi-Ethnic Study of Atherosclerosis (24). This suggests that TAC, a surrogate of aortic stiffening and vascular aging, may predict future HF after TAVR, in keeping with our previous reports (6,7). Only thoracic aortic calcification was assessed, as our previous study showed that most of the prognostic information was embedded there (7) and also because not all centers performed whole aorta CT scans. TAC is precisely computable, highly reproducible, and may also encompass some aspect of frailty including fragility fracture (8), muscle loss (25), and cerebrovascular events (9). As frailty is complex to assess with no unequivocal definition (26), this additional facet of TAC may be interesting.
The score was well calibrated for cardiovascular and all-cause mortality in both cohorts, and its prognostic discriminative performance was also good. A score with a good calibration means that the model has a good ability to predict risk of poor outcome with the intervention.
The CAPRI score appears highly relevant for objective decision making in individual patients. Although no formal decision threshold can be proposed, we tried to estimate a meaningful threshold corresponding to the probability of 1-year death with medical treatment. In a cohort comparable to C4CAPRI, the probability of dying was 23.7%, which could increase to >35% in the presence of symptoms (19). As in our study, most deaths were attributable to cardiovascular causes. In our cohort, CAPRI scores predicted a 1-year cardiovascular and all-cause mortality risk after TAVR similar to or greater than that without the intervention in a substantial number of patients (Figure 5). For example, CAPRI scores above 0.70 and 0.78 for cardiovascular and all-cause mortality, respectively, indicate a 50% risk of all-cause death and a 30% risk of cardiovascular death within the first year after TAVR, markedly higher than the estimated risk without intervention. On top of other markers (frailty or cognitive function), such a level of risk may represent an important argument against further proceeding with TAVR. Thus, the CAPRI score represents an objective way for the heart team to improve patient selection and for the patient to make an informed decision.
Other risk scores
Many scores are dedicated to perioperative mortality prediction such as the surgical scores Euroscore, Euroscore II, or the STS score. Their discriminative ability for 1-year mortality is poor (14), in keeping with our study. The same applies to the France 2 score, which has also been proposed to predict early post-TAVR mortality (15). The PARTNER and the TARIS scores have been proposed to predict mid-term all-cause deaths (27). These scores mainly represent various combinations of the same variables collected in registries or study cohorts. Despite the addition of some specific variables of functional and cognitive capacity, the discrimination of the PARTNER score remains moderate (21): much lower than the CAPRI score. The addition of “frailty syndrome” to the other risk factors appears not to improve 1-year predictions of the score (16). Importantly, these previous scores are based on extreme-risk and high-risk patients for standard aortic valve surgery; thus, their performance in lower-risk patients remains unknown. In this respect, the CAPRI score is at present the only score that has been validated in lower-risk patients (test cohort).
The training cohort comprised patients treated between 2010 and 2014 and may not be representative of today’s populations and technologies. It is conceivable that technological refinement and inclusion of patients with lower-risk profiles may improve outcome further in the future. In this respect, it is notable that the mortality rate in the test cohort, treated in 2015, was dramatically lower than in the training cohort. However, the CAPRI scores had very similar performance (AUC) in these more recently treated patients. The outcomes assessed in the C4CAPRI multicenter study were limited to cardiovascular and all-cause mortality; its usefulness in predicting other relevant outcomes, such as residual HF and quality of life, is untested. The score would need to be prospectively evaluated for its power to predict functional improvement and/or persisting HF.
The model might have been improved by the inclusion of frailty-associated variables such as daily-life activities. However, such markers were not available in the C4CAPRI cohorts, and their prognostic impact remains to be determined.
Objective, evidence-based decision making is critical to avoid futile interventions and to make optimal use of finite medical resources. The current study proposes a new specific TAVR-risk score based on pathophysiological variables. The inclusion of TAC, an independent unbiased variable, markedly increased the predictive power of the score. The implementation of a TAC-based score into daily practice is facilitated by the fact that CT scans are systematically performed before TAVR. This study also emphasizes the critical role of aortic biomechanics in determining outcomes after TAVR.
COMPETENCY IN MEDICAL KNOWLEDGE: A substantial number of patients exhibit a poor outcome despite TAVR; they should be identified. TAC is associated with mortality after TAVR.
COMPETENCY IN PATIENT CARE AND PROCEDURAL SKILLS: The CAPRI score, which encompasses TAC combined with classical prognostic factors, has a good ability to predict 1-year cardiovascular and all-cause mortality. Calculation of CAPRI scores should be part of the initial work-up for a more personalized evaluation of the patients who are candidates for TAVR procedures.
TRANSLATIONAL OUTLOOK: The impact of aortic biomechanics and vascular aging on the outcome after TAVR are important but overlooked. Additional research is warranted in this field.
The authors thank Lakdhar Benyahya, PhD, who provided help with data management.
All authors have reported that they have no relationships relevant to the contents of this paper to disclose.
- Abbreviations and Acronyms
- aortic stenosis
- area under the receiver-operating characteristic curve
- confidence interval
- chronic obstructive pulmonary disease
- computed tomography
- glomerular filtration rate estimated by the Cockcroft formula
- heart failure
- integrated discrimination improvement
- hazard ratio
- left ventricular ejection fraction
- receiver-operating characteristic
- thoracic aortic calcification
- transcatheter aortic valve replacement
- Received November 26, 2017.
- Revision received March 6, 2018.
- Accepted March 20, 2018.
- 2019 American College of Cardiology Foundation
- Holmes D.R. Jr..,
- Nishimura R.A.,
- Grover F.L.,
- et al.
- Yotti R.,
- Bermejo J.,
- Gutierrez-Ibanes E.,
- et al.
- Harbaoui B.,
- Courand P.Y.,
- Charles P.,
- et al.
- Harbaoui B.,
- Montoy M.,
- Charles P.,
- et al.
- Szulc P.
- Gilard M.,
- Eltchaninoff H.,
- Donzeau-Gouge P.,
- et al.
- Auffret V.,
- Lefevre T.,
- Van Belle E.,
- et al.
- ↵Creatis. Available at: https://www.creatis.insa-lyon.fr/site7/fr/creatools. Accessed April 18, 2018.
- Kappetein A.P.,
- Head S.J.,
- Genereux P.,
- et al.
- Ludman P.F.,
- Moat N.,
- de Belder M.A.,
- et al.
- Iung B.,
- Laouenan C.,
- Himbert D.,
- et al.
- Arnold S.V.,
- Afilalo J.,
- Spertus J.A.,
- et al.
- Gonzalez-Saldivar H.,
- Rodriguez-Pascual C.,
- de la Morena G.,
- et al.
- Steyerberg E.W.
- Arnold S.V.,
- Reynolds M.R.,
- Lei Y.,
- et al.
- Ong K.L.,
- McClelland R.L.,
- Rye K.A.,
- et al.
- Idoate F.,
- Cadore E.L.,
- Casas-Herrero A.,
- et al.
- Afilalo J.,
- Alexander K.P.,
- Mack M.J.,
- et al.