Author + information
- Received July 13, 2010
- Revision received October 28, 2010
- Accepted November 8, 2010
- Published online July 26, 2011.
- Fay Y. Lin, MD⁎,
- Leslee J. Shaw, PhD†,
- Allison M. Dunning, MSc⁎,
- Troy M. LaBounty, MD‡,
- Jin-Ho Choi, MD, PhD⁎,
- Jonathan W. Weinsaft, MD⁎,
- Sunaina Koduru, MD⁎,
- Millie J. Gomez, MD⁎,
- Augustin J. Delago, MD§,
- Tracy Q. Callister, MD∥,
- Daniel S. Berman, MD‡ and
- James K. Min, MD‡,⁎ ()
- ↵⁎Reprint requests and correspondence:
Dr. James K. Min, Departments of Medicine, Imaging, and Biomedical Sciences, Cedars-Sinai Heart Institute, Cedars-Sinai Medical Center, S. Taper Building, Rm 1258, Los Angeles, California 90048
Objectives We examined mortality risk in relation to extent and composition of nonobstructive plaques by 64-detector row coronary computed tomographic angiography (CCTA).
Background The prognostic significance of nonobstructive coronary artery plaques by CCTA is poorly understood.
Methods We prospectively evaluated consecutive adults from 2 centers undergoing 64-detector row CCTA without prior documented coronary artery disease (CAD) and without obstructive (≥50%) CAD by CCTA. Luminal diameter stenosis severity was classified for each segment as none (0%) or mild (1% to 49%), and plaque composition was classified as noncalcified, calcified, or mixed.
Results During 3.1 ± 0.5 years, 54 intermediate-term (≥90 days) deaths occurred among 2,583 patients (2.09%), with 4 early (<90 days) deaths. Adjusted for CAD risk factors, the presence of any nonobstructive plaque was associated with higher mortality (hazard ratio [HR]: 1.98, 95% confidence Interval [CI]: 1.06 to 3.69, p = 0.03), with the highest risk among those exhibiting nonobstructive CAD in 3 epicardial vessels (HR: 4.75, 95% CI: 2.10 to 10.75, p = 0.0002) or ≥5 segments (HR: 5.12, 95% CI: 2.16 to 12.10, p = 0.0002). Higher mortality for nonobstructive CAD was observed even among patients with low 10-year Framingham risk (3.4%, p < 0.0001) as well as those with no traditional, medically treatable CAD risk factors, including diabetes mellitus, hypertension, and dyslipidemia (6.7%, p < 0.0001). No independent relationship between plaque composition and incident mortality was observed. Importantly, patients without evident plaque experienced a low rate of incident death during follow-up (0.34%/year).
Conclusions The presence and extent of nonobstructive plaques augment prediction of incident mortality beyond conventional clinical risk assessment.
Coronary computed tomographic angiography (CCTA) has emerged as an accurate noninvasive anatomic imaging modality for the detection and exclusion of obstructive coronary artery disease (CAD) (1–3). Early reports have revealed that obstructive CAD—as identified by CCTA and defined by coronary plaques causing ≥50% reduction in luminal diameter—is valuable for prognosis of individuals at risk for incident death or major adverse cardiovascular events (4–10).
Nevertheless, individuals undergoing CCTA commonly exhibit nonobstructive plaque. Prior invasive ultrasound and autopsy studies have implicated nonobstructive plaques as central to the pathophysiologic processes of sudden cardiac death and myocardial infarction (11–13). Early electron beam computed tomography (EBCT) data suggest that 3-vessel nonobstructive CAD predicts long-term mortality. However, the independent predictive value of nonobstructive plaque by CCTA performed by multidetector computed tomography scanners—with higher spatial resolution to detect lower thresholds of plaque—is unknown (6).
We examined the predictive value of nonobstructive coronary artery plaque extent and composition as detected by 64-detector row CCTA for 3-year mortality risk among symptomatic patients referred for suspected CAD without prior known disease.
We evaluated consecutive adults ≥18 years of age between August 2005 and December 2007 who underwent 64-detector row CCTA at 2 centers (Tennessee Heart and Vascular Institute, Hendersonville, Tennessee; Capital Cardiology Associates, Albany, New York), which we have previously described (10). Institutional review board approval was obtained at each center. Patients from Tennessee were a distinct and nonoverlapping cohort from those undergoing 16-slice CCTA, which we have previously described (4). Patients were referred for evaluation by CCTA for chest pain of varying degrees of typicality; a minority of asymptomatic patients was referred for evaluation of CAD in the setting of peripheral arterial disease, cerebrovascular disease, or multiple CAD risk factors. Individuals with known CAD, including prior coronary revascularization, and individuals with obstructive plaque by CCTA, as defined by 50% stenosis or greater, were excluded from the study analysis.
All patients were in normal sinus rhythm and were capable of the breath-hold needed for CCTA. Patients with heart rates >70 beats/min were given 5 mg intravenous metoprolol at 5-min intervals to a total dose of 25 mg. If the heart rate of the patient did not drop below 70 beats/min, CCTA was performed at the lowest heart rate.
Before the initiation of the scan, 2 investigators (T.C., A.D.) prospectively collected information on the presence of categorical cardiac risk factors in each individual. Systemic arterial hypertension was defined as a documented history of high blood pressure or treatment with antihypertensive medications. Diabetes mellitus was defined by diagnosis of diabetes made previously by a physician and/or use of insulin or oral hypoglycemic agents. Dyslipidemia was defined as known but untreated dyslipidemia or current treatment with lipid-lowering medications. A positive smoking history was defined as current smoking or cessation of smoking within 3 months of testing. Family history of coronary heart disease was determined by patient query. Ten-year risk of hard coronary heart disease events were calculated with a modified Framingham risk score, as we have previously described (14).
Scan protocol and image reconstruction
All scans were performed with a 64-detector row computed tomography scanner (Lightspeed VCT, GE Healthcare, Milwaukee, Wisconsin). Imaging of a test-bolus of contrast was performed at 2 mm superior to the take-off of the left main coronary artery for precise timing of contrast injection. During the CCTA acquisition, 100 cc of iodinated contrast (Isovue 370, Bracco Diagnostics, Princeton, New Jersey, or Visipaque, GE Healthcare, Princeton, New Jersey) was injected, followed by a 50-cc saline flush. The scan parameters were: 64 × 0.625 mm collimation, tube voltage 120 mV, effective 400 to 650 mA. For radiation-sparing purposes, coronary artery calcium scores (CACS) were not performed. Estimated radiation doses ranged from 3 to 18 mSv.
Helical or axial scan data were obtained with retrospective or prospective electrocardiography gating, respectively. Images were reconstructed with 0.625-mm slice thickness immediately after completion of the scan to identify motion-free coronary artery images. Electrocardiographically gated datasets were reconstructed at 70%, 75%, and 80% of the cardiac cycle after the QRS complex to identify ventricular diastole, with additional datasets reconstructed at 40%, 45%, and 50% of the cardiac cycle to identify early diastole if necessary. For CCTA with suboptimal image quality, multisegment reconstruction algorithms were employed (15). Optimal phase reconstruction was assessed by comparison of different phases, if available, and the phase with the least amount of coronary artery motion was chosen for analysis. Multiple phases were used for image interpretation, if minimal coronary artery motion was different for different arteries. The CCTAs were evaluated on 2-dimensional maximum intensity projections in oblique views that focused on coronary arterial segments for optimal viewing. Two orthogonal thin maximal intensity projection views approximating traditional coronary angiography angles were used for the left anterior descending, left circumflex, and right coronary artery circulations, respectively. Three-dimensional rotation was performed, when necessary, to focus on diagonal and marginal branch vessels.
Three-dimensional views with curved multiplanar reformation and short-axis cross-sectional viewing techniques were additionally used to enhance detection of obstructive coronary plaque, if necessary. In all individuals, irrespective of image quality, every arterial segment was scored in an intent-to-diagnose fashion. If a coronary artery segment was not interpretable, despite these multiple techniques, the unevaluable segment was scored similarly to the most proximal segment that was evaluable.
Coronary artery analysis
All scans were analyzed by Level III experienced cardiologists with experience interpreting several thousand CCTA scans (T.C., A.D.). Coronary segments were visually scored for the presence of coronary plaque with a 16-segment coronary artery model (left main artery; proximal, mid, and distal left anterior descending artery; proximal, mid, and distal diagonal/intermediate branch; proximal, mid, and distal left circumflex artery; proximal, mid, and distal obtuse marginal branch; proximal, mid, and distal right coronary artery) and were included for analysis. In cases where multiple diagonal or obtuse marginal branches were present, plaques were graded with the branch with the largest luminal diameter and/or the largest area of myocardium subtended.
In each coronary artery segment, coronary atherosclerosis was defined as tissue structures >1 mm3 that existed either within the coronary artery lumen or adjacent to the coronary artery lumen that could be discriminated from surrounding pericardial tissue, epicardial fat, or the vessel lumen itself. Coronary atherosclerotic lesions were quantified for stenosis by visual estimation. Luminal diameter stenosis severity was scored as none (0% luminal stenosis) or mild (1% to 49% luminal stenosis) (Fig. 1). Percent obstruction of coronary artery lumen was based on a comparison of the luminal diameter of the segment exhibiting obstruction to the luminal diameter of the most normal-appearing site immediately proximal to the plaque. In instances in which plaque was highly calcified, 2-dimensional oblique images were also visualized without maximal intensity projection (i.e., 0.625-mm isotropic voxel resolution) or multiplanar reformats with cross-sectional views to minimize partial volume averaging artifact of calcium.
Plaque severity was graded on a per-patient level by the maximum intraluminal stenosis in any of the 16 scored coronary segments. We further evaluated the plaque extent by summing the number of epicardial vessels and the number of vessel segments exhibiting plaque (maximum 3 and 16, respectively). Left main plaque was included within the left anterior descending distribution.
Plaque composition for each coronary segment was classified as noncalcified, calcified, or mixed, as we have previously described (20). The optimal image display setting was chosen individually—in general, at a window between 600 and 900 Hounsfield units (HU) and at a level between 40 and 250 HU—in keeping with our clinical practice as well as prior published studies (16). Because the HU densities of fibrous, lipoid, or thrombotic plaque are known to overlap, these plaque types were aggregated into a “noncalcified” group, which was defined as any plaque 70% or more of which exhibited HU densities below the luminal contrast density (16). Calcified plaque was defined as any plaque 70% or more of which exhibited a HU density above the luminal contrast density and at least 130 HU. “Mixed” plaque was designated as plaque that possessed 30% to 70% calcified and noncalcified plaque volume.
The primary endpoint was time to death from all causes. Follow-up procedures were approved by the institutional review board of the study center. Death status was ascertained by querying the Social Security Death Index. To account for acute events in which nonobstructive CAD might have represented unstable lesions, all mortality analyses were performed censoring early events within the first 90 days after CCTA testing (4 deaths).
The SPSS software (version 12.0, SPSS, Chicago, Illinois) and SAS software (version 9.2, SAS, Cary, North Carolina) were used for all statistical analyses. Categorical variables are presented as frequencies, and continuous variables are presented as mean ± 1 SD. Variables were compared with chi-square statistic for categorical variables and by Student unpaired t test for continuous variables. The impact of CCTA and clinical findings on time to death was calculated with univariate Cox proportional-hazards models. In each case, the proportional hazards assumption was met. Backward stepwise multivariate regression was performed, accounting for significant interactions to determine the independent predictive value of CCTA findings. Net reclassification improvement of nonobstructive plaque above and beyond Framingham estimated risk was calculated with reclassification tables of logistic regression by tertiles of risk (0% to 1.4%, 1.4% to 1.8%, and >1.8%), because mean Framingham estimated risk for this nonobstructive cohort was very low (17). A 2-tailed p value <0.05 was considered statistically significant.
Among 4,661 consecutive patients undergoing 64-detector row CCTA, 654 were excluded for prior revascularization or known CAD, 1,201 were excluded for CCTA-identified obstructive CAD, 142 were excluded for nonverifiable Social Security numbers, and 81 were excluded for missing coronary segment data, leaving 2,583 patients for whom comprehensive plaque assessment was available with a 16-segment coronary tree model. During a mean follow-up of 3.1 ± 0.5 years (median 3.1 years, interquartile range [IQR]: 2.8 to 3.5 years), 54 intermediate-term (≥90 days) deaths were recorded (2.09%). The study cohort comprised persons of middle-age (52.7 ± 13.9 years, 58% female) with a high prevalence of cardiovascular risk factors (Table 1) and a wide distribution of disease (Table 2).
Plaque involvement, severity, and overall burden
Within the study cohort, over one-half of all individuals (58.5%) did not possess any coronary artery plaque, and 22.6% possessed only 1-vessel nonobstructive CAD. There were a median of 0 (IQR: 0 to 2) segments with plaques/patient; among patients with plaques, there were a median of 0 (IQR: 0 to 1) noncalcified, 0 (IQR: 0 to 1) mixed, and 1 (IQR: 0 to 1) calcified segments/patient (Table 3).
The number of segments with nonobstructive plaques was associated with age, male sex, hypertension, diabetes, dyslipidemia, and past smoking (Table 4). A 10% increase in Framingham estimated 10-year risk was associated with 0.7 more segments of nonobstructive plaque (95% confidence interval [CI]: 0.63 to 0.77, p < 0.001). Typicality of angina symptoms had no consistent association with nonobstructive plaque burden (Table 4).
Prognosis of individuals with no CAD versus with nonobstructive plaques
Individuals with no CCTA-identified coronary artery plaque had a favorable intermediate-term prognosis, with a mortality risk of 1.2% over the study period (annualized mortality rate 0.34%). A higher risk was observed for individuals with any nonobstructive plaque, with a graded relationship by the number of affected vessels (log-rank p < 0.001) (Figs. 2 and 3⇓⇓, Table 5). Mortality was also related to age, diabetes, dyslipidemia, and smoking history (p < 0.01 for all). Even for patients with <10% Framingham estimated 10-year risk, the presence of nonobstructive plaque conferred an increased intermediate-term mortality risk (3.4%, log-rank p < 0.0001) (Fig. 4). Similarly, among patients with no medically treatable risk factors—including diabetes, dyslipidemia, and hypertension—the presence of any nonobstructive plaque conferred an increased probability of death over the study follow-up period (6.7%, log-rank p < 0.0001) (Fig. 5).
The number of involved segments had a nonlinear relationship with intermediate-term mortality, with an increase in risk for those with >5 segments of 9.4% (p = 0.03) (Fig. 6). As compared with individuals with <5 segments with nonobstructive plaque, individuals possessing ≥5 segments with involved plaque experienced increased intermediate-term mortality hazard even after adjustment for CAD risk factors (hazard ratio: 5.12, 95% CI: 2.16 to 12.10, p = 0.0002) or Framingham risk score (hazard ratio: 7.10, 95% CI: 2.88 to 17.50, p < 0.0001).
Compared with estimated Framingham risk, the presence of any plaque showed an improved net reclassification improvement (20.5%, p = 0.04) for prediction of all-cause death over the follow-up period. Sensitivity analyses were performed for all comparisons with inclusion of events that occurred within the first 90 days after CCTA testing and demonstrated similar results (data not shown).
Plaque composition and plaque burden
The relationship of plaque composition to plaque distribution is shown in Figure 7. Patients with ≥5 coronary segments with plaque—as compared with patients with <5 coronary segments with plaques—had a higher percentage of coronary segments with calcified plaque (52% vs. 25% calcified, p < 0.001) or mixed plaque (24% vs. 13%, p = 0.002), with no difference in the percentage with noncalcified plaques (7% vs. 13%, p = 0.12). Of those with 1 to 4 involved coronary segments, 14.4% had only mixed plaques and 14.4% had only noncalcified plaques. In contrast, of those with ≥5 involved coronary segments, 12% had only mixed plaques, and 0% had only noncalcified plaques. There was no significant relationship between plaque composition and intermediate-term mortality after adjustment for CAD risk factors or presence of plaque of any type (Table 6).
In this 2-center study of 2,583 consecutive patients without prior known CAD and without obstructive CAD, nonobstructive coronary artery plaque presence and extent as identified by 64-detector row CCTA are associated with heightened mortality risk in a 3-year follow-up period. The CCTA nonobstructive plaque assessment added significant risk prediction beyond patient demographic data, traditional CAD risk factors, and Framingham risk score.
The results of the present study suggest a potential utility for diagnosis of nonobstructive CAD by CCTA. Such patients experience heightened mortality risk, even though they represent a patient population for whom functional stress testing would be expected to be negative and who might not be referred for evaluation by invasive coronary angiography (ICA) after CCTA. Our results confirm prior observations of a high negative predictive value of a normal CCTA for later adverse clinical events but are additive to the prior published reports by identifying a distinction in mortality risk for individuals with CCTAs demonstrating no plaque versus nonobstructive plaque (4,5,18). Because CCTA-visualized nonobstructive plaques precede the majority of deaths in this cohort, congruent to prior observations from invasive angiography, their population impact is great despite their lower risk profile relative to obstructive plaque.
Importantly, mortality is increased even among patients in whom medical therapy for primary prevention of CAD would not have been warranted. The 3-year mortality for CCTA-visualized plaque among patients with low Framingham risk score and absence of medically treatable risk factors of 3.7% and 6.6%, respectively, might be sufficiently elevated to merit risk reduction by medical therapy over the longer term. Additionally, a graded, nonlinear increase in risk of mortality was observed with increasing plaque extent and burden, raising the question of the appropriate threshold for consideration of medical therapy. Future controlled studies should be performed to test these hypotheses.
The prognostic outcome of nonobstructive coronary plaque by CCTA has not been well-characterized to date. Cross-sectional studies using ICA, intravascular ultrasound, and histopathology have long suggested the important role of nonobstructive “vulnerable” plaques as precedents of later myocardial infarction, but their prognostic impact with hard events has not been characterized in longitudinal studies—due, in part, to their invasive nature (11–13). Among asymptomatic patients, the prognostic impact of subclinical coronary atherosclerosis with noncontrast-enhanced CACS has been well-established (19). Among symptomatic patients, although the diagnostic accuracy and prognostic impact of CCTA-visualized obstructive CAD has been well studied, there has been a relative neglect of nonobstructive CAD, leaving a wide range of ambiguity for the prognosis of mild and moderate intraluminal lesions (2,4,6). Most prognostic studies of ICA and CCTA have grouped together patients with nonobstructive plaques and completely normal coronary arteries, in part to accommodate smaller sample sizes (5,7,20).
Ostrom et al. (6), with older-generation EBCT, identified an independent predictive value of nonobstructive CAD in all 3 major epicardial vessels, with a risk of death comparable to single-vessel obstructive CAD. These findings were observed as independent and incremental to traditional cardiac risk factors and CACS. The prognostic impact of lesser degrees of plaque severity, extent, overall burden, and composition could not be discriminated, possibly due to the technologically inferior spatial resolution of EBCT. In contrast, we employed current-generation 64-detector CCTA with improved spatial resolution and performed more detailed segmental characterization of plaque severity, involvement, and composition. We observed that nonobstructive coronary artery plaque presence and extent impart increased mortality risk, even at lower thresholds of disease (i.e., <50% stenosis).
Our study results do not solidly establish the role of plaque composition as it relates to prognosis for individuals with nonobstructive CAD undergoing CCTA. Although we identified a dominance of calcified and mixed plaque composition for individuals with greater overall CAD plaque burden (≥5 segments), we did not observe a relationship of these plaque composition types to incident mortality. It remains possible that, given a high prevalence of calcified plaque in these individuals, CACS might be sufficient to recognize and quantify risk in this symptomatic population. We have previously noted, in distinction to this premise, a heightened risk of global reversible myocardial ischemia in patients with increasing segments with mixed plaque, and mixed composition plaques have been recently demonstrated to be additive to stenosis severity for detecting regional myocardial hypoperfusion (21). However, these studies were performed in more inclusive cohorts than that within the present study, including those with obstructive CAD or even restricted to those with obstructive CAD. In our study, CCTA was performed as part of clinical evaluation of patients with suspected CAD, and CACS was not performed for radiation-sparing reasons. It remains a clinical challenge, given that symptom presentation in our cohort was not significantly additive for prediction of global plaque extent and burden, to identify a priori patients who will benefit from contrast-enhanced CCTA versus patients for whom CACS will suffice. Future carefully designed studies should be performed to address this issue.
It is noteworthy to emphasize that our study was a cross-sectional sample of individuals presenting with symptoms suggestive of CAD, and the differential proportion of calcified, mixed, and noncalcified plaques represent a single “point-in-time” for which assertions regarding plaque progression are not possible. In our cohort, individuals with greater extent and burden of nonobstructive plaque were more likely to exhibit calcified and mixed plaques and less likely to exhibit noncalcified plaques alone. Indeed, among individuals with ≥5 segments with plaque, 12% exhibited only mixed plaque type, whereas 0% exhibited only noncalcified plaque type. It is possible that nonobstructive, noncalcified components of mixed plaques include a high-risk subgroup of plaques with transformative potential to beget incident adverse CAD events, whereas purely noncalcified plaques are less likely to exhibit this property. Noncalcified components of plaques as visualized by CCTA have greater odds of disease progression on serial angiographic studies, and the presence of nonobstructive plaques containing low-density components, “spotty” calcifications, and positive arterial remodeling have been observed to predict incident acute coronary syndrome (9,22,23). Future studies examining plaque progression by CCTA will be useful to characterize further the natural history of plaques in relation to their composition.
This study is not without limitations. Although the use of all-cause mortality provides a robust endpoint free of ascertainment bias, it nevertheless limits the risk predictive power of CCTA findings, due to the relatively low event rate and inclusion of noncardiac deaths. Second, treatment goals and medications were left to the discretion of the caring physicians, and downstream patient care might have been influenced by CCTA results in this open-label study. Third, there might be residual or unmeasured confounding due to health behaviors, which might account for the apparent protective effect for mortality of dyslipidemia by either history or treatment. Fourth, certain adverse CCTA plaque characteristics that have been described among culprit plaques in acute coronary syndrome–such as positive remodeling, spotty calcifications, low attenuation density, or finer gradations of plaque composition–were not measured and could not be evaluated for risk-predictive ability. Finally, this study examined individuals referred for clinically indicated CCTA. Although we observed inconsistent relationships among symptom typicality, plaque burden, and mortality, the present findings should not be generalized to asymptomatic patients, who might have disease prevalence and outcomes that differ from our symptomatic cohort.
Among individuals without known CAD and obstructive CAD, nonobstructive plaque presence and extent as identified by 64-detector row CCTA enhances risk prediction of incident mortality. Improved risk stratification is evident for nonobstructive CAD, even for patients with low Framingham risk scores as well as for patients without traditional modifiable CAD risk factors for whom medical therapy for primary prevention of CAD would not be warranted. Future studies examining the impact of treatment of individuals with nonobstructive CAD should be performed.
Drs. Delgado and Callister have served on the Speakers' Bureau for GE Healthcare. Dr. Berman has received research support from Lantheus, Astellas Healthcare, GE/Amersham, and Siemens. Dr. Min has served on the Speakers' Bureau and medical advisory board for and received research support from GE Healthcare. All other authors have reported that they have no relationships to disclose.
- Abbreviations and Acronyms
- coronary artery calcium score
- coronary artery disease
- coronary computed tomographic angiography
- confidence interval
- electron beam computed tomography
- Hounsfield unit
- invasive coronary angiography
- interquartile range
- Received July 13, 2010.
- Revision received October 28, 2010.
- Accepted November 8, 2010.
- American College of Cardiology Foundation
- Budoff M.J.,
- Dowe D.,
- Jollis J.G.,
- et al.
- Meijboom W.B.,
- Meijs M.F.,
- Schuijf J.D.,
- et al.
- Hausleiter J.,
- Meyer T.,
- Hadamitzky M.,
- et al.
- Min J.K.,
- Shaw L.J.,
- Devereux R.B.,
- et al.
- Pundziute G.,
- Schuijf J.D.,
- Jukema J.W.,
- et al.
- Ostrom M.P.,
- Gopal A.,
- Ahmadi N.,
- et al.
- Hadamitzky M.,
- Freissmuth B.,
- Meyer T.,
- et al.
- van Werkhoven J.M.,
- Schuijf J.D.,
- Gaemperli O.,
- et al.
- Min J.K.,
- Lin F.Y.,
- Dunning A.M.,
- et al.
- Falk E.,
- Shah P.K.,
- Fuster V.
- Virmani R.,
- Burke A.P.,
- Farb A.,
- Kolodgie F.D.
- Ehara S.,
- Kobayashi Y.,
- Yoshiyama M.,
- et al.
- Leber A.W.,
- Becker A.,
- Knez A.,
- et al.
- Hoffmann U.,
- Nagurney J.T.,
- Moselewski F.,
- et al.
- Greenland P.,
- Bonow R.O.,
- Brundage B.H.,
- et al.
- Motoyama S.,
- Sarai M.,
- Harigaya H.,
- et al.