Author + information
- Received October 31, 2013
- Revision received January 2, 2014
- Accepted February 4, 2014
- Published online April 29, 2014.
- Kenneth C. Bilchick, MD, MS∗∗ (, )
- Sujith Kuruvilla, MD∗,
- Yasmin S. Hamirani, MD∗,
- Raghav Ramachandran, MS†,
- Samantha A. Clarke, BS†,
- Katherine M. Parker, PhD†,
- George J. Stukenborg, PhD, MA‡,
- Pamela Mason, MD∗,
- John D. Ferguson, MBChB, MD∗,
- J. Randall Moorman, MD∗,
- Rohit Malhotra, MD∗,
- J. Michael Mangrum, MD∗,
- Andrew E. Darby, MD∗,
- John DiMarco, MD, PhD∗,
- Jeffrey W. Holmes, MD, PhD∗,†,
- Michael Salerno, MD, PhD∗,†,§,
- Christopher M. Kramer, MD∗,§ and
- Frederick H. Epstein, PhD†,§
- ∗Department of Medicine, Cardiovascular Division, University of Virginia Health System, Charlottesville, Virginia
- †Department of Biomedical Engineering, University of Virginia Health System, Charlottesville, Virginia
- ‡Department of Public Health Sciences, University of Virginia Health System, Charlottesville, Virginia
- §Department of Radiology and Medical Imaging, University of Virginia Health System, Charlottesville, Virginia
- ↵∗Reprint requests and correspondence:
Dr. Kenneth C. Bilchick, Department of Medicine, Cardiology/Electrophysiology, University of Virginia Health System, P.O. Box 800158, Charlottesville, Virginia 22908.
Objectives Using cardiac magnetic resonance (CMR), we sought to evaluate the relative influences of mechanical, electrical, and scar properties at the left ventricular lead position (LVLP) on cardiac resynchronization therapy (CRT) response and clinical events.
Background CMR cine displacement encoding with stimulated echoes (DENSE) provides high-quality strain for overall dyssynchrony (circumferential uniformity ratio estimate [CURE] 0 to 1) and timing of onset of circumferential contraction at the LVLP. CMR DENSE, late gadolinium enhancement, and electrical timing together could improve upon other imaging modalities for evaluating the optimal LVLP.
Methods Patients had complete CMR studies and echocardiography before CRT. CRT response was defined as a 15% reduction in left ventricular end-systolic volume. Electrical activation was assessed as the time from QRS onset to LVLP electrogram (QLV). Patients were then followed for clinical events.
Results In 75 patients, multivariable logistic modeling accurately identified the 40 patients (53%) with CRT response (area under the curve: 0.95 [p < 0.0001]) based on CURE (odds ratio [OR]: 2.59/0.1 decrease), delayed circumferential contraction onset at LVLP (OR: 6.55), absent LVLP scar (OR: 14.9), and QLV (OR: 1.31/10 ms increase). The 33% of patients with CURE <0.70, absence of LVLP scar, and delayed LVLP contraction onset had a 100% response rate, whereas those with CURE ≥0.70 had a 0% CRT response rate and a 12-fold increased risk of death; the remaining patients had a mixed response profile.
Conclusions Mechanical, electrical, and scar properties at the LVLP together with CMR mechanical dyssynchrony are strongly associated with echocardiographic CRT response and clinical events after CRT. Modeling these findings holds promise for improving CRT outcomes.
- cardiac magnetic resonance
- cardiac resynchronization therapy
- heart failure
- myocardial infarction
- ventricular tachycardia
Outcomes after cardiac resynchronization therapy (CRT) are influenced by a complex interaction between the myocardial substrate and the left ventricular lead position (LVLP). The myocardial substrate may be characterized both by the pattern of mechanical activation (1) and the distribution of scar (2). Recent echocardiographic methods such as 3-dimensional echocardiography and speckle tracking (3,4) offer the potential for better performance than previous methods, as do dyssynchrony assessments based on cardiac magnetic resonance (CMR) (5) and the circumferential uniformity ratio estimate (CURE) (6–8). Scar in the posterolateral left ventricle (LV), a common location for the LV lead, has been associated with CRT nonresponse (9), whereas late-activated sites based on electrical parameters (LV lead electrical delay and QRS to left ventricular intrinsic activation interval [QLV]) (10,11) or mechanical criteria (12,13) appear to be better locations for LV leads.
CMR is the gold standard for assessment of myocardial scar. We have recently shown that CMR displacement encoding with stimulated echoes (DENSE) generates high-quality circumferential strain data (5,8,14–16) that can precisely describe the state of mechanical dyssynchrony using the CURE parameter (6), which does not require manual detection of regional time to peak strain (8). We now report the results of a cohort study of patients referred for CRT based on the hypothesis that favorable CMR findings (lower CURE from CMR DENSE, no scar at the LVLP, and delayed onset of circumferential contraction at the LVLP) and late electrical activation at the LVLP are strongly associated with CRT response and clinical events during follow-up. The clinical significance is that CMR applied this way could improve upon current criteria for patient selection (17,18) and facilitate more effective implementation of CRT.
The study was approved by the Institutional Review Board for Human Subjects Research at the University of Virginia. Patients were required to have a clinical indication for CRT based on established guidelines (18) and a glomerular filtration rate of at least 45 ml/min/1.73 m2 in order to receive gadolinium.
Prior to the CRT procedure, patients underwent a research CMR protocol including steady-state free precession imaging, cine DENSE imaging, and late gadolinium enhancement (LGE) on a 1.5-T Avanto scanner (Siemens Healthcare, Erlangen, Germany) with a 4-channel phased-array chest radiofrequency coil. Cine DENSE imaging (previously validated by comparison with myocardial tagging in heart failure) (8) was performed in 4 short-axis and 3 long-axis planes with displacement encoding applied in 2 orthogonal in-plane directions for each plane with the following parameters (14,15): interleaved spiral readout with 6 interleaves per image; repetition time/echo time 17 ms/1.9 ms; slice thickness 8 mm; field of view 350 × 350 mm; flip angle 15°; pixel size 2.8 × 2.8 mm; fat suppression; and displacement-encoding frequency 0.1 cycles/mm.
Determination of echocardiographic volumes before and after CRT
Standard 2D echocardiographic images with Doppler were obtained for all patients at baseline and 3 months and 6 months after CRT with standard short- and long-axis views. The left ventricular end-systolic volume (LVESV), left ventricular end-diastolic volume (LVEDV), and left ventricular ejection fraction (LVEF) before and after CRT were determined using Simpson's rule for 2- and 4-chamber long-axis views using EchoPAC software (GE, Fairfield, Connecticut).
Clinical CRT procedure
Patients then underwent the clinical CRT procedure. During the procedure, venograms of the coronary sinus were recorded in 2 projections. Final cine images of the leads were recorded in the usual left anterior oblique, anterior-posterior, and right anterior oblique projections.
Clinical follow-up and determination of CRT response
The echocardiographic evaluation at 3 months included standard A-V and V-V optimization. CRT response was defined as a 15% reduction in LVESV at 6 months (or the last follow-up echocardiogram prior to death if the patient died prior to 6 months after implantation). After the procedure, subsequent clinic notes, device interrogations, and discharge summaries for inpatient hospitalizations were reviewed for all study patients. Sustained ventricular tachyarrhythmia events were defined as episodes of ventricular tachycardia or ventricular fibrillation requiring implantable cardioverter-defibrillator therapies or untreated ventricular tachyarrhythmia episodes >30 s detected by the implantable cardioverter-defibrillator, and these events were also recorded in the database.
CMR DENSE image processing and strain analysis
Following image acquisition, segmentation of the LV myocardium was performed semiautomatically for cine DENSE images (19), a phase-unwrapping algorithm was applied to LV myocardium pixels, and displacements were calculated (5). Lagrangian strain was computed from displacements in 24 short-axis segments in multiple slices and was then projected in both the radial and circumferential (circumferential strain [ECC]) directions relative to the LV center of mass. LV volumes, mass, and ejection fraction were calculated from cine steady-state free precession images using Argus software (Siemens, Erlangen, Germany).
Evaluation of CMR dyssynchrony, strain onset at the LVLP, and electrical activation at the LVLP
Dyssynchrony was assessed from 4 short-axis cine DENSE slices at basal, midbasal, midapical, and apical levels (with additional weight given to basal and midbasal slices) using CURE, which is based on the Fourier transform (FT) of the spatial distribution of strain as previously described (6–8). Briefly, CURE makes use of the zero-order power and first-order power from the Fourier analysis of this function to index dyssynchrony on a scale between 0 (dyssynchrony) and 1 (synchrony) (6). The characteristics of the ECC curve at the LV lead implantation site were then determined with respect to time to peak ECC and onset of ECC (onset of circumferential contraction), which was defined as the time from QRS detection by the scanner's gating software (when DENSE encoding pulse are applied) to the onset of a negative slope of the ECC curve. Regarding electrical activation, the QLV was calculated, as previously described, as the time from QRS onset to the electrogram at the LVLP, with a value of at least 95 ms associated with greater rates of CRT response in the SMART-AV (SmartDelay determined AV Optimization) trial (11).
Evaluation of myocardial scar from LGE and lead position relative to scar
The lead position relative to scar from CMR LGE was determined using the “o'clock” method in all patients (20). In a subset of these patients, results from this method were also confirmed using a quantitative algorithm we developed and validated for lead localization, as previously described (21). Regarding the latter, we pre-calibrated standard fluoroscopy suites by imaging a phantom at multiple camera positions, then reconstructed and registered 3-dimensional lead positions with pre-procedure CMR. With this method, the LVLP was identified as the point on the epicardial surface at which the distances between the lead position and each of the 3 landmarks (coronary sinus ostium, RV apex, and anterolateral mitral annulus) in the CMR coordinate space were most similar to the equivalent distances in the fluoroscopic space. This algorithm was implemented in custom software written using Matlab version 7.14 (The Mathworks, Natick, Massachusetts).
In patients with previous myocardial infarction, scar was manually segmented from short-axis LGE images using the segmentation software Segment described previously. LGE tissue (scar tissue) had a signal intensity at least 2 SDs above the mean signal intensity in remote areas. Scar transmurality was measured as a fraction of wall thickness calculated over the circumference with a 5° moving average window, using custom software implemented in Matlab. Scar distribution and transmurality were displayed on a Hammer projection map of the epicardial surface (22) along with the LVLP, as shown in Figures 1 and 2.
Statistical analysis was performed using SAS version 9.3 (SAS Institute, Cary, North Carolina). The Wilcoxon 2-sample test (Mann-Whitney U test) was used for univariate comparisons between continuous variables, and the Fisher exact test was used for univariate comparisons between categorical variables (as in Table 1).
Based on the hypothesis that CRT response would be strongly associated with overall dyssynchrony with CURE (continuous), mechanical stretch (delayed ECC onset) at the LVLP (categorical), scar at the LVLP (categorical), and late electrical activation at the LVLP (continuous), bivariable logistic regression was used to estimate the probability of a 15% reduction in LVESV associated with these variables. Based on prior associations between scar burden and CRT outcomes (2), LV percent scar volume (continuous) was also analyzed using bivariable logistic regression, as was the LV mass index (continuous) parameter (based on a proposed mechanistic association between LV mass and CRT outcomes). Receiver-operating characteristic (ROC) analysis was also performed, and the statistical significance of the area under the curve (AUC) was determined based on comparison with chance. Multivariable logistic regression was then performed based on this hypothesis-driven model with CURE, mechanical stretch at the LVLP (delayed ECC onset), scar at the LVLP, and late electrical activation at the LVLP. Multivariable linear regression was then performed to estimate the percent change in LVESV as a variable function of these same 4 selected covariates from the multivariable logistic regression model.
Overfitting was evaluated based on the heuristic shrinkage estimator of van Houwelingen and le Cessiej, which should be ≥0.90 to rule out overfitting (23). For the multivariable logistic model, the shrinkage estimator was calculated as: (model likelihood chi-square statistic − number of covariates)/(model likelihood chi-square statistic). For the multivariable linear model, the shrinkage estimator was calculated as the ratio of the adjusted R2 to the raw R2 (23).
Kaplan-Meier plots, the log-rank statistic, and Cox proportional hazards regression were used to analyze the associations for mechanical and scar findings with the clinical outcomes of death and sustained ventricular tachyarrhythmia events (as defined in the previous text). For the clinical outcome of death, patients having favorable values for the CMR parameters in the multivariable logistic model (CURE <0.70, no scar at the LVLP, and mechanical stretch at the LVLP) were compared with patients without this optimal CMR profile. The threshold value for CURE was based on a prior smaller study in a completely different cohort of patients (6).
The cohort included 75 patients who had either a Class I or Class IIa indication for CRT according to current guidelines (18) and underwent implantation of a CRT defibrillator with subsequent clinical follow-up (median follow-up 2.6 years). The baseline characteristics are given in Table 1 for the entire cohort, as well as responders and nonresponders (LVESV improvement of at least 15%). With respect to LV structural characteristics based on CMR, the baseline LVEF (23.2% [interquartile range (IQR): 15.0% to 28.4%]) was similar in responders and nonresponders, but the baseline LVESV index, baseline LVEDV index, stroke volume index, and baseline LV mass index were all greater in nonresponders versus responders. The frequency of comorbid medical disease was also similar among responders and nonresponders.
With respect to events during follow up, 21.3% of patients died during a median follow-up of 2.6 years (IQR: 1.6 to 3.8 years), whereas 16.0% had sustained ventricular tachycardia or fibrillation, and 26.7% were hospitalized with heart failure. As shown in Table 1, the rates of all of these events were much higher for nonresponders compared with responders (p < 0.009 to 0.0001).
Based on echocardiography before and after CRT, favorable changes in LVESV, LVEDV, and LVEF were confirmed in responders but not in nonresponders. Although significant differences in LVESV are expected based on the definition of CRT response, results for all 3 parameters by group are reported for completeness. In responders, LVESV decreased (LVESV percent change −32.5% [IQR: −49.5% to −22.2%]) and LVEF increased (absolute LVEF change of 16.5% [IQR: 9% to 23.5%]), whereas LVESV and LVEF failed to improve in nonresponders (LVESV percent change 3.3% [IQR: −0.7% to 22.1%], median LVEF absolute change −4% [IQR: −8% to 0%]) (p < 0.001 for comparisons between responders and nonresponders). The LVEDV also decreased in responders (LVEDV percent change −12.9% [IQR: −27.6% to −4.4%]) but remained about the same in nonresponders (LVEDV percent change 0.2% [IQR: −3.7% to 9.0%]) (p < 0.001 for comparisons between responders and nonresponders).
As shown in Figures 1 and 2, mechanical and scar characteristics of the LV as a whole and at the LVLP were characterized in detail with cine DENSE and LGE. Figure 1 shows 3 examples of CRT responders, whereas Figure 2 shows 3 examples of CRT nonresponders. Consistent with our hypothesis, the comparison of nonresponders and responders showed significant differences not only in overall mechanical dyssynchrony with CURE but also in mechanical activation, electrical activation, and scar at the LVLP.
Bivariable logistic regression results for the 4 parameters hypothesized to have strong associations with CRT response (as well as the additional 2 variables of interest) are shown in Table 2. The corresponding multivariable model with these 4 parameters identified in our hypothesis is shown in Table 3. This model had an AUC of 0.95 (p < 0.0001) without evidence of overfitting (shrinkage estimator = 0.931) (Fig. 3). The 4 covariates and corresponding odds ratios (ORs) were: CURE for overall dyssynchrony (OR: 2.59 per 0.1 decrease in CURE [95% confidence interval (CI): 1.58 to 4.23], absence of scar at the LVLP (OR: 14.9 [95% CI: 2.56 to 86.6]), delayed onset of ECC at the LVLP (OR: 6.55 [95% CI: 1.18 to 36.4]), and delayed electrical timing at the LVLP based on the QLV (OR: 1.31 [95% CI: 1.04 to 1.65] per 10 ms increase in QLV), which was determined as the time from the QRS onset to the intraprocedural electrogram at the LVLP. The Nagelkerke maximum rescaled R2 for the model was 0.72.
As shown in Table 4, the original covariates shown in the multivariable logistic model in Table 3 were also strongly associated with the percent change in LVESV (R2 = 0.53) in a multivariable linear model, again without overfitting (shrinkage estimator = 0.948) (23), consistent with an association not only with the presence but also with the degree of LV functional improvement. As shown in Figure 4, the LVESV decreased by 23.4% (IQR: 17.1% to 44.9%) in the 52 patients with CURE <0.70 but increased by 7.2% (IQR: 2.9% to 24.7%) in 23 patients with CURE ≥0.70 (p < 0.0001) (Fig. 4A). Regarding mechanical characteristics at the LV lead site, in the 52 patients with significant dyssynchrony by CURE (<0.70), the 33 patients with delayed ECC onset (mechanical stretch) at the LVLP had a median decrease in the LVESV of 27.4% (IQR: 18.0% to 51.9%) after CRT, compared with 18.2% (IQR: 1.5% to 31.7%) in the 19 patients without delayed ECC onset at the LVLP (p = 0.01) (Fig. 4B). Furthermore, in the 38 patients with CURE <0.70 and no scar at the LVLP, all 25 patients with delayed ECC onset at the LVLP had a CRT response with a decrease of 37.1% (IQR: 22.1% to 52.0%) in LVESV compared with a decrease of only 17.6% (IQR: 2.5% to 22.4%) in the remaining patients without delayed ECC onset at the LVLP (p = 0.002) (Fig. 4D).
The multivariable logistic model for CRT response with CMR parameters only (CURE, delayed onset of ECC at the LVLP, absence of LVLP scar, and LV mass index) also performed very well (p < 0.05 for maximum likelihood estimates for all parameters), with an overall AUC of 0.94 (p < 0.0001) and Nagelkerke R2 = 0.72 (p < 0.05) (Online Table 1). In addition, the multivariable linear model for percent change in LVESV including these same CMR parameters (Online Table 2) also demonstrated similar performance compared with the original model reported in Table 4. Regarding scar burden, the LV percent scar volume was not included in these models because it was no longer associated with CRT response after adjustment for the presence of scar at the LVLP.
In addition to the associations between CMR findings and echocardiographic CRT response, favorable CMR findings were also associated with better clinical outcomes, as shown in the Kaplan-Meier curves for overall survival (log-rank p = 0.006) and ventricular tachyarrhythmias (p = 0.01) in Figures 5 and 6, respectively. Patients with CURE ≥0.70 had a 12-fold increased risk of death (median follow-up of 2.6 years) compared with the group with the favorable CMR findings of CURE <0.70, absence of LVLP scar, and delayed onset of ECC at the LVLP (hazard ratio: 11.9 [95% CI: 1.5 to 93.5]) (Fig. 5). In addition, patients with CURE ≥0.60 had an increased risk of sustained ventricular tachyarrhythmias after CRT (hazard ratio: 8.24 [95% CI: 1.06 to 63.9]) compared with patients with CURE <0.60 (Fig. 6).
The principal finding of this study was that both echocardiographic CRT response and clinical outcomes such as death and sustained ventricular tachyarrhythmia after CRT can be explained with an integrated model based on mechanical and scar-related characterization of the substrate for resynchronization from the pre-procedure CMR, with some additional discrimination provided by intraprocedural characterization of electrical timing at the LVLP. In addition, using models based on logistic, linear, and Cox proportional hazards regression, we have shown strong associations between pre-procedure imaging findings and both echocardiographic CRT response and clinical events. For example, the third of the patients with a favorable CMR profile (CURE <0.70, delayed onset of ECC at the LVLP, and absence of scar at the LVLP) enjoyed a 12-fold higher survival rate than the group of patients with CURE ≥0.70. In addition, none of the patients with CURE ≥0.70 had a CRT response (100% negative predictive value). The remaining patients had intermediate outcomes. In this way, this analysis has identified 3 distinct groups of patients expected to have a 0% CRT response and decreased survival after CRT, 100% CRT response and improved survival after CRT, or intermediate outcomes. With respect to other clinical events, increased ventricular tachyarrhythmia events in patients with higher CURE may have occurred because these patients did not have the beneficial antiarrhythmic effect of LV functional improvement to outweigh the potentially proarrhythmic effects of LV epicardial pacing (24).
The association between delayed onset of ECC at the LVLP and CRT response was likely due to resynchronization of the most dysfunctional LV myocardium, allowing that tissue to contribute much more effectively to overall pump function. Furthermore, the use of time to strain onset rather than time to peak strain at the LVLP may be beneficial considering that the corresponding assessments of electrical timing are also measured during early systole rather than near the end of systole where late strain peaks occur. The independent contributions from regional mechanical and electrical activation are noteworthy. Furthermore, the finding that scar at the LVLP was associated with CRT nonresponse is distinguished from prior reports (9,25) by the strength of association between LVLP scar and CRT response even after adjustment for overall dyssynchrony and other mechanical and electrical characteristics at the LVLP, which was determined using quantitative methods.
Last, the success of our models may be attributed to some extent to the high-quality strain data that was obtained with CMR cine DENSE, as demonstrated in previous studies (5,8,15). In addition, use of the CURE parameter is advantageous because it does not introduce potential errors in the manual detection of regional time to peak strain, as seen with other dyssynchrony parameters, as we have shown previously (8). Of note, other echocardiographic modalities, such as 3-dimensional echocardiography, offer high-quality ECC data, and we have demonstrated that CURE can be effectively determined from 3-dimensional echocardiography (26), such that its use is not limited to CMR. From a more general perspective, these results have high clinical significance for improving patient selection and outcomes after CRT, with potential associated cost savings.
We did not prospectively test the effect of altering the lead position based on CMR findings. Additional factors could have been analyzed, but the purpose of the study was to develop a parsimonious model including key factors related to scar, mechanical activation, and electrical timing. Regarding follow-up, although the primary outcomes measure was reduction in LVESV at 6 months, there was significant variation in the follow-up durations for clinical outcomes such as overall survival, and this was addressed with censoring where appropriate in the survival analysis. Regarding the study cohort, patients were enrolled at a single institution, such that there may be differences between patients at this institution and other institutions receiving CRT. In addition, although this patient cohort provides strong evidence for this model, a large, prospective, multicenter trial would be appropriate prior to widespread clinical use.
Mechanical, electrical, and scar properties at the LVLP together with CMR mechanical dyssynchrony are strongly associated with echocardiographic CRT response and clinical events after CRT. Modeling these findings in patients referred for CRT holds promise for improving outcomes after the procedure.
The authors thank clinical research coordinators Aly Blake and Irene Harvey, echocardiography technologists Kim Chadwell and Dale Fowler, and CMR technologist John Christopher for their contributions.
Dr. Bilchick has received support from the National Institutes of Health (NIH) (K23 grant HL094761); and has served as a consultant to Biosense Webster. Dr. Parker is an employee of Sorin Group USA, Inc. Dr. Mason has received grant support from Medtronic and Boston Scientific; and has received support from St. Jude Medical and Johnson & Johnson. Dr. Ferguson has received consulting support from Biosense Webster and St. Jude Medical; and has received honoraria from Medtronic. Dr. Malhotra has received grant support from Medtronic and Boston Scientific. Dr. Mangrum has received grant and consulting support from St. Jude Medical; and has received research funding from St. Jude Medical, Hansen Medical, EndoSense, CardioFocus, and Boston Scientific. Dr. Darby has served as a consultant for Biosense Webster; has received speaking honoraria from Medtronic; and has received research support from Boston Scientific. Dr. DiMarco has received consulting support from Medtronic, St. Jude Medical, and Boston Scientific. Dr. Holmes has received support from the NIH (R01 grant HL085160). Dr. Kramer has received grant support from Siemens Healthcare; and has served as a consultant to Synarc and St. Jude Medical. Dr. Epstein has received support from the NIH (R01 grant EB001763); and has received grant support from Siemens Healthcare. All other authors have reported that they have no relationships relevant to the contents of this paper to disclose.
- Abbreviations and Acronyms
- cardiac magnetic resonance
- cardiac resynchronization therapy
- circumferential uniformity ratio estimate
- displacement encoding with stimulated echoes
- circumferential strain
- late gadolinium enhancement
- left ventricle/ventricular
- left ventricular end-diastolic volume
- left ventricular ejection fraction
- left ventricular end-systolic volume
- left ventricular lead position
- QRS to left ventricular electrogram interval
- Received October 31, 2013.
- Revision received January 2, 2014.
- Accepted February 4, 2014.
- American College of Cardiology Foundation
- White J.A.,
- Yee R.,
- Yuan X.,
- et al.
- Gorcsan J. III.,
- Tanabe M.,
- Bleeker G.B.,
- et al.
- Tanaka H.,
- Nesser H.J.,
- Buck T.,
- et al.
- Bilchick K.C.,
- Dimaano V.,
- Wu K.C.,
- et al.
- Leclercq C.,
- Faris O.,
- Tunin R.,
- et al.
- Bleeker G.B.,
- Kaandorp T.A.,
- Lamb H.J.,
- et al.
- Gold M.R.,
- Birgersdotter-Green U.,
- Singh J.P.,
- et al.
- Saba S.,
- Marek J.,
- Schwartzman D.,
- et al.
- Khan F.Z.,
- Virdee M.S.,
- Palmer C.R.,
- et al.
- Tracy C.M.,
- Epstein A.E.,
- Darbar D.,
- et al.
- Parker K.M.,
- Bunting E.,
- Malhotra R.,
- et al.
- Medina-Ravell V.A.,
- Lankipalli R.S.,
- Yan G.X.,
- et al.
- Ypenburg C.,
- van Bommel R.J.,
- Delgado V.,
- et al.
- Driver K.,
- Ramachandran R.,
- Kuruvilla S.,
- et al.