Author + information
- Received February 11, 2015
- Revision received June 2, 2015
- Accepted June 12, 2015
- Published online November 1, 2015.
- Amanda R. Vest, MBBS∗,
- Yuping Wu, PhD†,
- Rory Hachamovitch, MD, MSc‡,
- James B. Young, MD∗,§ and
- Leslie Cho, MD‖∗ ()
- ∗Kaufman Center for Heart Failure, Heart and Vascular Institute, Cleveland Clinic, Cleveland, Ohio
- †Department of Mathematics, Cleveland State University, Cleveland, Ohio
- ‡Section of Cardiovascular Imaging, Cleveland Clinic, Cleveland, Ohio
- §Endocrinology and Metabolism Institute, Cleveland Clinic, Cleveland, Ohio
- ‖Section of Preventative Cardiology and Rehabilitation, Cleveland Clinic, Cleveland, Ohio
- ↵∗Reprint requests and correspondence:
Dr. Leslie S. Cho, Section of Preventative Cardiology and Rehabilitation, Cleveland Clinic, Mail Code JB-1, 9500 Euclid Avenue, Cleveland, Ohio 44195.
Objectives This study sought to determine whether body mass index (BMI) has a differential impact on survival for females versus males with advanced systolic heart failure (HF).
Background Females have a survival advantage in HF, the mechanisms of which are unclear. There is also a proposed “obesity survival paradox” in which excess adiposity promotes HF survival.
Methods We reviewed 3,811 patients with left ventricular ejection fraction ≤40% who had undergone cardiopulmonary exercise testing between 1995 and 2011. The endpoint was all-cause mortality. Multivariable analysis was performed using a Cox proportional hazards model. Because of the nonlinearity of BMI, a restricted cubic spline was used. An interaction term was added to investigate the impact of BMI on mortality by sex.
Results The unadjusted data demonstrated an overall obesity survival paradox in HF. This survival paradox disappeared for males after adjustment for potential confounders, with overweight and obese males showing higher adjusted mortality hazard ratios compared with normal weight males. Conversely, females in the overweight BMI range (25.0 to 29.9 kg/m2) had the lowest adjusted mortality (hazard ratio: 0.84; 95% confidence interval: 0.77 to 0.93; p = 0.0005 compared with normal weight females) with a nadir in mortality hazard just below BMI 30 kg/m2. The multivariable model supported a differential impact of BMI on mortality in males versus females (p for interaction <0.0001).
Conclusions In this advanced HF cohort, an unadjusted obesity survival paradox disappeared after adjustment for confounders. Overweight and obese males had higher adjusted mortality than normal weight males, whereas a BMI in the overweight range was associated with a significant survival benefit in females.
Obesity is a key determinant of cardiovascular health and an independent risk factor for the development of heart failure (HF) (1,2). The “overweight” and “obese” states are defined by body mass index (BMI), with overweightness diagnosed in the BMI range ≥25 to <30 kg/m2 and obesity ≥30 kg/m2. Among 59,178 adults followed for a mean of 18 years, the adjusted hazard ratios for incident HF at BMIs <25, 25 to 29.9, and ≥30 kg/m2 were 1.00, 1.25, and 1.99 (p < 0.001) for men and 1.00, 1.33, and 2.06 (p < 0.001) for women, respectively (3). However, multiple investigators have demonstrated an “obesity survival paradox” in HF with reduced (and preserved) ejection fraction, whereby overweight and obese patients have either no increased mortality risk compared with normal weight counterparts, or even a lower mortality risk (4–10). Several potential explanations have been postulated to explain these unexpected survival outcomes, including the potential confounding of cigarette smoking or undiagnosed systemic illness. There is also the possibility of “lead time bias” whereby obese individuals present with HF symptoms earlier in their disease course, or a “healthy survivor effect,” whereby the most comorbid obese individuals die before developing HF, leaving the surviving obese HF patients with disproportionately favorable outcomes. The paradox could alternatively be explained by the protection from cardiac cachexia afforded by baseline excess adiposity or by myocardial effects of adipokines secreted from adipose tissue. Both the biological mechanisms of the proposed paradoxical relationship between BMI and mortality in HF, and the role of sex in this relationship, remain incompletely understood. Given the female survival advantage in HF (11–13), and the recognition that female myocardium shows greater fatty acid metabolism and lower glucose utilization (14), we hypothesized that females with HF may derive a greater degree of protection from excess adiposity than males.
We identified 4,380 consecutive adult patients with systolic HF who underwent cardiopulmonary exercise testing at Cleveland Clinic between January 1, 1995, and November 1, 2011. Institutional review board approval was granted both for the prospective recording of exercise testing data and the retrospective collection of additional data specific to this project; the requirement for informed consent was waived. We removed 253 patients from the cohort because of incomplete data, with either a missing stress test date (n = 4), missing mortality follow-up data (n = 46), no information in the electronic medical record to verify clinical data (n = 116), or missing key cardiopulmonary stress test parameters (n = 87). Patients with a baseline left ventricular ejection fraction (LVEF) in the 41% to 50% range were removed (n = 130) to restrict analysis to individuals with LVEF ≤40%. We filtered out patients who had received a heart transplant (n = 15) or left ventricular assist device (LVAD) (n = 8) before their stress test date. Patients with a primary valvular cardiomyopathy etiology (n = 85) or severe congenital heart disease (n = 27) were also excluded from this analysis. We also excluded 51 patients with a BMI <18.5 kg/m2 (below the “normal weight” range). Thus, the final cohort contained 3,811 subjects. If a patient underwent multiple cardiopulmonary stress tests, only the initial test was considered.
Baseline characteristics were prospectively recorded in the stress test database by the exercise physiologist conducting the test. Parameters such as HF etiology, presence of coronary artery disease (CAD), diabetes status, smoking status, and HF medications were ascertained by the physiologist through a combination of verbal history-taking and medical chart review. The patient’s weight was always measured on the day of the test. Smoking and medication status were documented as current at the time of the test. The presence of CAD was defined as a prior myocardial infarction or any degree of obstruction on coronary imaging. Retrospective chart review was performed for >20% of database subjects to confirm the accuracy of the prospectively entered data.
Cardiopulmonary exercise testing
Symptom-limited exercise stress testing was conducted by trained exercise physiologists and supervised by a physician. Exercise testing was performed using a treadmill stress in the majority of patients; the alternate option was a stationary bike. The exercise physiologist assigned the patient to the Bruce, modified Bruce, Cornell 0%, Cornell 5%, Cornell 10%, Naughton, or modified Naughton protocols, appropriate to the patient’s physical abilities. Gas exchange data were collected throughout the test using a MedGraphics cardiopulmonary metabolic cart (St. Paul, Minnesota). Heart rate targets were not used as an endpoint or to judge the adequacy of the test. Blood pressure was manually measured every minute and the heart rate was recorded from an electrocardiogram printed each minute during the test. Electrocardiographic changes and symptoms were also recorded at the end of each stage. Heart rate recovery (HRR) was calculated as peak exercise heart rate minus the heart rate at 1 minute post-exercise. A standard walking cool-down was used during recovery.
The oxygen consumption (VO2) was averaged over 30-s intervals throughout the test and the peak VO2 was determined as the highest 30-s interval in the last 2 min of the test. The ventilatory threshold was defined as the VO2 at which expired carbon dioxide increased nonlinearly relative to VO2 (V-slope method). The ratio of the increase in ventilation to the increase in CO2 output (VCO2) was recorded at peak exercise (15). Estimated functional capacity was calculated in metabolic equivalents of task (MET), where 1 MET = 3.5 ml/kg/min of oxygen consumption. The test result parameters were prospectively entered into an institutional database. Peak VO2 was calculated per ml/kg/min of total body mass. Estimated lean body mass was calculated by the Boer formula: Men: estimated LBM = 0.407 weight (kg) + 0.267 height (cm) − 19.2; women: estimated LBM = 0.252 weight (kg) + 0.473 height (cm) − 48.3 (16). An additional version of peak VO2 was calculated corrected to estimated LBM to adjust for the differential oxygen consumption of muscle versus adipose tissue (17).
We retrospectively collected outcomes data for all subjects up until the date of death or to censor at November 1, 2011. All-cause mortality status and date of death were determined by linking our database with the U.S. Social Security Death Index. We also recorded the occurrence of heart transplantation or LVAD implantation during the follow-up period. We did not consider transplantation or LVAD implantation as an endpoint because physicians use peak VO2 to determine advanced therapy eligibility. However, because these therapies do change the hazard of death, they were handled as time-dependent covariates to appropriately account for the impact of transplantation or LVAD implantation on the patient’s subsequent survival.
Baseline demographic and clinical data were stratified by both sex and BMI category, and comparisons were made between groups. Continuous data were evaluated for normality, and accordingly, between-group comparisons with Student t or Mann-Whitney testing were performed. Categorical data were compared with chi-square tests. Unadjusted survival was stratified by sex and normal weight/overweight/obese status. These weight categories were based on the World Health Organization classification of normal weight (NW) BMI as 18.5 to 24.99 kg/m2, overweight (OW) as 25 to 29.99 kg/m2, and obese (OB) as ≥30 kg/m2.
Unadjusted and adjusted hazard ratios
Hazard ratios (HRs) were tabulated by sex and BMI category. Both the unadjusted and adjusted HRs were calculated to better appreciate the relationship between BMI and mortality in each sex group and the role of potential confounders. Adjustment was performed for age, race, HF etiology, New York Heart Association (NYHA) status, digoxin use, angiotensin-converting enzyme (ACE) inhibitor/angiotensin receptor blocker (ARB) use, beta-blocker use, diabetes, smoking, hypertension (HTN), hypercholesterolemia, atrial fibrillation (AF), resting systolic blood pressure (SBP), heart rate recovery (HRR), peak VO2, peak respiratory exchange ratio (RER), peak tidal volume (Vt), and subsequent receipt of a heart transplant or LVAD as a time-dependent covariable. Not all patients in the cohort attained an acceptable RER; therefore, a subgroup analysis was performed to determine if the mortality HRs were consistent for subjects with peak RER ≥1.05.
Cox model with restricted cubic spline
Multivariable analysis was performed using a Cox proportional hazards model for all-cause mortality. Because of the nonlinearity of the BMI-mortality relationship, a restricted cubic spline was used. This model permits appreciation of differential effects of BMI on mortality through the recorded spectrum of low to high BMI measurements and so enables more accurate characterization of the influence of covariates on the BMI-mortality relationship. Use of a proportional hazards model that assumes linearity will give a hazard ratio for BMI that implies a constant and incremental effect on mortality throughout the BMI continuum, which is likely to miss effects that are not uniform throughout the normal weight, overweight, and obese ranges. The aim of this model was to define the association between sex- and mortality-adjusted for key potential confounders, and then examine for an interaction between sex and BMI. Model covariates were predefined within the study design based on their clinical and pathophysiologic relevance as a confounder, significance in prior literature, and frequency of occurrence in this cohort of patients. Covariates were excluded if they were found to be collinear with another key variable (e.g., both maximum METs and ventilatory threshold showed colinearity with peak VO2 and therefore were excluded from the model). Sensitivity analyses were performed with the substitution of peak VO2 by estimated LBM-adjusted peak VO2 and then VE/VCO2, and restriction of the model to only subjects with peak RER ≥1.05. All statistical analyses were performed using R 3.0.2 (R Foundation for Statistical Computing, Wien, Austria). Values of p < 0.05 were considered statistically significant and all tests were 2-tailed.
To address the study hypothesis of a differential response to the overweight and obese states in males versus females, an interaction term was introduced to the restricted cubic spline model for all-cause mortality, adjusted for sex, BMI, age, race, LVEF, etiology, NYHA functional class, digoxin use, ACE inhibitor/ARB use, beta-blocker use, diabetes status, smoking status, hypertension history, AF history, resting SBP, HRR, peak VO2, peak RER, peak Vt, and subsequent heart transplantation or LVAD. The adjusted HR for all-cause mortality was plotted to a set of reference values based on median or mode values, as follows: age 54 years, white race, LVEF 21%, NYHA functional class 3, taking digoxin, taking ACE-inhibitor/ARB, taking beta-blocker, nonsmoker, no HTN, no hyperlipidemia, no AF, resting SBP 110.9 mm Hg, HRR 12.5, peak VO2 16.8 ml/kg/min, peak RER 1.16, peak Vt 1,650.8 ml, and no transplantation or LVAD during follow-up. Cardiomyopathy etiology (ischemic/nonischemic) and diabetes status (yes/no) was varied to assess the persistence of the interaction under different conditions.
Tables 1, 2, and 3 depict the baseline clinical characteristics and exercise test parameters from the study cohort. Within the cohort of 3,811 subjects, 3,765 (99%) underwent treadmill stress testing, with the most common exercise protocol being the modified Naughton (2,602 subjects, 68% of cohort). The median follow-up was 2,252 days (interquartile range: 955 to 3,821 days), during which time there were 1,537 mortality events (40.3% crude mortality). Females had a slight but significantly lower BMI (27.2 vs. 28.0 kg/m2, p < 0.0015), were younger (52.5 vs. 54.6 years, p < 0.0001), had a lower burden of CAD (30% vs. 57%, p < 0.0001), and less ischemic etiology (26% vs. 54%, p < 0.0001) (Table 1) compared with males. Medication regimens and diabetes prevalence were equivalent between the sexes. Females attained significantly lower peak VO2 and peak Vt than males (p < 0.0001 for both).
As expected, relative to the OW and OB groups NW patients had a lower prevalence of obesity-related conditions including diabetes, HTN, hyperlipidemia, and CAD, but there was no difference in smoking across BMI categories (Table 2). OB patients had a lower mean age than NW and OW counterparts, raising the possibility of a “healthy survivor” effect. There was also an incremental increase in the proportion of patients tolerating beta-blockers across the BMI categories (63% NW vs. 69% OW vs. 73% OB; p < 0.0001). When stratified by both sex and BMI category, it can be seen that the 6 groups are dissimilar by many characteristics, including OW and OB females attaining lower median peak VO2 levels than their male counterparts (Table 3).
Unadjusted and adjusted HRs
Over a median 6.2-year follow-up period, females had a lower crude mortality rate than males (32.9% vs. 42.9%). The crude mortality hazard ratio for male sex was 1.42 (95% confidence interval [CI]: 1.25 to 1.60, p < 0.0001). Compared with NW subjects, the unadjusted all-cause mortality was significantly lower in the OB group (HR: 0.88; 95% CI: 0.85 to 0.92; p < 0.0001), supporting the presence of an overall unadjusted “obesity survival paradox” (Table 4). When examined by sex, the unadjusted survival paradox was evident in both females and males, although the risk was distributed differently; the OW group was associated with lower mortality in females, whereas the OB group was associated with lower mortality in males (Figure 1). After adjustment for all relevant confounders, the only BMI subgroup to retain a survival benefit was the OW female group (adjusted HR: 0.84; 95% CI: 0.77 to 0.93; p = 0.0005 compared with NW females). Conversely, the OW and OB males showed increased adjusted mortality compared with NW males. A subgroup analysis was conducted to confirm this pattern was consistent when restricted to subjects with RER ≥1.05, indicating attainment of anaerobic threshold and a VO2 representing peak oxygen consumption (Table 5).
Cox model with restricted cubic spline
Female sex was a significant predictor of survival on multivariable Cox modeling using the restricted cubic spline (Table 6) (p < 0.0001). There was a statistically significant interaction between BMI and sex (p < 0.0001), supporting the previous observation of a differential impact of BMI on all-cause mortality in males versus females with HF. Figures 2A and 2B illustrate the adjusted mortality HR along the continuum of BMI for males and females. A nadir in mortality hazard was seen just below 30 kg/m2 in females with both ischemic and nonischemic HF etiologies. This relationship also persisted regardless of diabetes status (data not shown). Conversely, males demonstrated the highest mortality hazard around a BMI of 30 kg/m2. When peak VO2 was substituted by LBM-adjusted peak VO2 in the model, the significant interaction between BMI and sex remained unchanged. The interaction also persisted when the VE/VCO2 ratio replaced peak VO2 in the model and when the analysis was restricted to subjects who attained an RER ≥1.05. The correlation between peak VO2 and VE/VECO2 was –0.54 (p < 0.0001) and so both parameters were not added to the model concurrently; the overall model chi-square value was greatest using peak VO2 rather than VE/VECO2 (6,052 vs. 5,999) and hence this is the model reported.
There are basic science observations that support our clinical finding of a differential response to the overweight/obese state in females versus males with systolic HF. Peterson et al. identified that female sex is independently associated with greater myocardial fatty acid uptake and lower myocardial glucose utilization (14). This may be an effect of estrogen, which reduces glucose oxidation, gluconeogenesis, and glycogenolysis in other organs and inhibits glucose uptake. Chronic estrogen replacement in healthy post-menopausal women enhances myocardial fatty acid uptake and oxidation (18). Nonhormonal mechanisms are also likely involved, such as in the higher turnover of fatty acids in females (19). This raises the possibility that female hearts are more dependent on fatty acids for energy production than male hearts, potentially explaining the survival advantage of females with some excess adiposity. The availability of fatty acids for myocardial utilization would only be beneficial if sufficient oxygen is delivered to the myocardium, because upregulation of fatty acid oxidation and downregulation of glucose utilization reduces myocardial efficiency. Interestingly, this hypothesis may be upheld by the observation in a cohort overlapping with our study in which females with ischemic HF had disproportionately poorer survival than nonischemics (20). Sexual dimorphism has also been recognized in the relationship between adiponectin and cardiovascular mortality, with high circulating adiponectin being associated with increased cardiovascular mortality in males, but not females, with type 2 diabetes (21).
A relationship between excess adiposity and favorable survival in females is plausible given that females have a higher percentage of body fat, and percent body fat has proven protective from mortality in HF (22). A few prior HF publications have divided their cohorts by sex, but do not have the sensitivity to observe differential sex effects along the full spectrum of BMI. An HF cohort containing 680 females was stratified by sex and BMI (dichotomized BMI into normal, 18.5 to 24.9 kg/m2, vs. high, ≥25 kg/m2). The higher BMI category was associated with improved adjusted 2-year survival in both sexes (23). However, there was no statistical comparison between the sexes and no presentation of mortality trends along the BMI continuum >25 kg/m2.
One recent publication did formally compare the mortality risk conferred by increasing BMI in males versus females with acutely decompensated HF. This international analysis supported an obesity survival paradox: there was an 11% decrease in 30-day mortality and 9% decrease in 1-year mortality per 5 kg/m2 BMI increase, p < 0.05 (24). No interaction between sex and BMI was detected, with a hazard ratio for all-cause mortality per 5 kg/m2 BMI of 0.87 (0.79 to 0.96) in males and 0.92 (0.85 to 0.99) in females, p for interaction = 0.92. However, unlike our study, the follow-up period was short, there was no risk adjustment for cardiorespiratory fitness (CRF), and the relationship between BMI and mortality was found to be log-linear.
It is notable that the unadjusted HRs in this analysis suggested an obesity survival paradox in both males and females, but that this disappeared with risk adjustment. This highlights the critical importance of careful covariate selection when constructing a model that aims to determine the “pure” effect of baseline BMI on subsequent outcomes. As demonstrated by Güder and colleagues, adjustment by incrementally more complete models can attenuate the strength of the inverse relationship between BMI and mortality (25). Measures of CRF have been also been shown to attenuate the obesity paradox, in cohorts with and without HF (26–30). There is a danger that adding a wide range of potential confounders may unknowingly insert a factor on the biological pathway between BMI and mortality and negate a true relationship. That the OW females continued to show a significantly lower adjusted HR makes this explanation unlikely. The absence of an overall-adjusted HF obesity survival paradox in this study is not unique (8,29). It may be relevant to note that these prior studies reporting no survival advantage for obese HF patients had some similarities to our current study in terms of their younger HF populations and risks adjustments for CRF.
The contribution of this sex-BMI analysis to the study of the obesity paradox is 3-fold. First, this analysis highlights the importance of recognizing a nonlinear relationship between BMI and mortality to permit detection of the differential effects of BMI on survival within different regions of the BMI spectrum. We propose that the linearity of the BMI-mortality relationship should always be examined and more complex nonlinear modeling options pursued if indicated. Second, it highlights the importance of adequate covariate adjustment given that the 6 sex/BMI subgroups were quite dissimilar in their baseline characteristics and an apparent obesity paradox was replaced by a more nuanced relationship between BMI and survival after covariate adjustment. Third, the observation of a HF survival paradox that is limited to overweight females merits further clinical and basic science investigation. Greater understanding of why modest excess adiposity may have a more favorable biological impact in females may reveal new therapeutic opportunities in advanced HF and also permit accurate counseling of HF patients regarding weight management.
Although the baseline and CRF data were prospectively collected, the findings of this study should be viewed in the context of a retrospectively analyzed cohort study design. There are other data parameters, such as biomarkers of nutritional status and invasive hemodynamics, which were unavailable and may have improved risk adjustment. Importantly, body habitus was represented only by BMI, with no available anthropomorphic measurements (31). In HF, weight may partly reflect fluid congestion, although a higher BMI resulting from a greater volume overload would be expected to increase mortality, which was not uniformly seen across sex groups. In studies involving obesity, CRF, and sex, body fat assessment by skinfold thickness should ideally be used to calculate lean body mass adjusted peak VO2 (32), although a recent study using estimated LBM did support the validity of such methods (33). An additional limitation is that this is a study of systolic HF patients who were well enough to perform exercise testing, but sick enough to require advanced disease evaluation; therefore, the results may not apply to the full systolic HF population.
In this large advanced systolic HF cohort, an unadjusted obesity survival paradox was ameliorated by adjustment for confounders. Overweight or obese males showed higher adjusted mortality than normal weight males; the only group to retain an adjusted survival benefit was overweight females. A more favorable response to modest excess adiposity may partially explain the female HF survival advantage.
COMPETENCY IN MEDICAL KNOWLEDGE: A female survival advantage has consistently been observed in outpatient systolic heart failure cohorts, the reasons for which are unclear.
TRANSLATIONAL OUTLOOK: Preliminary research suggests female hearts have greater myocardial fatty acid uptake and lesser myocardial glucose utilization, which could be protective in the advanced heart failure state. The next step is to better characterize the sex differences in adipokine hormones (adiponectin, resistin, leptin, tumor necrosis factor-alpha) that are thought to signal between adipose tissue and the myocardium. Understanding sex-specific effects of the systemic metabolism on the myocardium may facilitate development of novel pharmacological therapies for heart failure.
The authors thank Kathryn Brock, BA, CCRP, and Lee Pierson, MD, for their reviews of the manuscript.
This research project was conducted under the supervision of the senior author, Dr. Cho. Funding was provided by the Cleveland Clinic Women’s Cardiovascular Research Center and the Karos Chair for Women’s Cardiovascular Research fund. The authors have reported that they have no relationships relevant to the contents of this paper to disclose.
- Abbreviations and Acronyms
- angiotensin-converting enzyme
- atrial fibrillation
- angiotensin receptor blocker
- body mass index
- coronary artery disease
- confidence interval
- cardiorespiratory fitness
- heart failure
- hazard ratio
- heart rate recovery
- lean body mass
- left ventricular assist device
- left ventricular ejection fraction
- metabolic equivalent of task
- normal weight (18.5 to 24.99 kg/m2)
- New York Heart Association
- obese (≥30 kg/m2)
- overweight (25 to 29.99 kg/m2)
- respiratory exchange ratio
- systolic blood pressure
- ratio of ventilation to increase in carbon dioxide output
- oxygen uptake
- tidal volume
- Received February 11, 2015.
- Revision received June 2, 2015.
- Accepted June 12, 2015.
- American College of Cardiology Foundation
- Loehr L.R.,
- Rosamond W.D.,
- Poole C.,
- et al.
- Hu G.,
- Jousilahti P.,
- Antikainen R.,
- Katzmarzyk P.T.,
- Tuomilehto J.
- Horwich T.B.,
- Fonarow G.C.,
- Hamilton M.A.,
- MacLellan W.R.,
- Woo M.A.,
- Tillisch J.H.
- Lavie C.J.,
- Alpert M.A.,
- Arena R.,
- Mehra M.R.,
- Milani R.V.,
- Ventura H.O.
- Ghali J.K.,
- Krause-Steinrauf H.J.,
- Adams K.F.,
- et al.
- Gustafsson F.,
- Torp-Pedersen C.,
- Burchardt H.,
- et al.
- O'Meara E.,
- Clayton T.,
- McEntegart M.B.,
- et al.
- Peterson L.R.,
- Soto P.F.,
- Herrero P.,
- et al.
- Robbins M.,
- Francis G.,
- Pashkow F.J.,
- et al.
- Boer P.
- Osman A.F.,
- Mehra M.R.,
- Lavie C.J.,
- Nunez E.,
- Milani R.V.
- Mittendorfer B.
- Shah R.,
- Gayat E.,
- Januzzi J.L.,
- et al.
- Güder G.,
- Frantz S.,
- Bauersachs J.,
- et al.
- Lavie C.J.,
- McAuley P.A.,
- Church T.S.,
- Milani R.V.,
- Blair S.N.