Author + information
- Received March 24, 2011
- Revision received June 14, 2011
- Accepted June 21, 2011
- Published online September 27, 2011.
- Christopher R. deFilippi, MD⁎,⁎ (, )
- Robert H. Christenson, PhD⁎,
- Willem J. Kop, PhD†,
- John S. Gottdiener, MD⁎,
- Min Zhan, PhD⁎ and
- Stephen L. Seliger, MD, MS⁎
- ↵⁎Reprint requests and correspondence:
Dr. Christopher R. deFilippi, G3K63, Division of Cardiology, University of Maryland, 22 South Greene Street, Baltimore, Maryland 21212
Objectives The goal of this paper was to determine whether assessment of left ventricular ejection fraction (LVEF) enhances prediction of new-onset heart failure (HF) and cardiovascular mortality over and above N-terminal pro–B-type natriuretic peptide (NT-proBNP) level in older adults.
Background Elevated NT-proBNP levels are common in older adults and are associated with increased risk of HF.
Methods NT-proBNP and LVEF were measured in 4,137 older adults free of HF. Repeat measures of NT-proBNP were performed 2 to 3 years later and echocardiography was repeated 5 years later (n = 2,375), with a median follow-up of 10.7 years. The addition of an abnormal (<55%) LVEF (n = 317 [7.7%]) to initially elevated or rising NT-proBNP levels was evaluated to determine risk of HF or cardiovascular mortality. Changes in NT-proBNP levels were also assessed for estimating the risk of conversion from a normal to abnormal LVEF.
Results For participants with a low baseline NT-proBNP level (<190 pg/ml; n = 2,918), addition of an abnormal LVEF did not improve the estimation of risk of HF and identified a moderate increase in adjusted risk for cardiovascular mortality (hazard ratio: 1.69 [95% confidence interval: 1.22 to 2.31]). Among those whose NT-proBNP subsequently increased ≥25% to ≥190 pg/ml, an abnormal LVEF was likewise associated with an increased risk of cardiovascular mortality but not HF. Participants with an initially high NT-proBNP level (≥190 pg/ml) were at greater risk overall for both outcomes, and those with an abnormal LVEF were at the highest risk. However, an abnormal LVEF did not improve model classification or risk stratification for either endpoint when added to demographic factors and change in NT-proBNP. An initially elevated NT-proBNP or rising level was associated with an increased risk of developing an abnormal LVEF.
Conclusions Assessment of LVEF in HF-free older adults based on NT-proBNP levels should be considered on an individual basis, as such assessments do not routinely improve prognostication.
Detection of depressed left ventricular function may improve prevention and treatment of progression to symptomatic heart failure (1). In adults >50 years of age, the presence of even a mildly abnormal left ventricular ejection fraction (LVEF) (i.e., ≤55%) is associated with an approximately 3-fold increased risk of developing heart failure and a 2-fold increased risk of mortality compared with individuals with normal LVEF (2–4). Despite declines in the rates of cardiovascular deaths in the general population, more than 80% of cardiovascular deaths occur in older adults (5). However, with a relatively low prevalence (<8%) of an abnormal LVEF even in those age ≥65 years, it is difficult to advocate a routine imaging strategy in this population (3,6). Elevated natriuretic peptide levels are associated with depressed LVEF in the general population including older adults (7,8). Elevated N-terminal pro–B-type natriuretic peptide (NT-proBNP) levels are also associated with an increased risk of new-onset heart failure in general population studies (9,10). Currently, neither assessment of natriuretic peptides nor LVEF is recommended for general population screening (11). However, a combination of both measures would potentially refine risk stratification to identify subjects who could benefit from therapies to reduce the risk of progression to heart failure (12). Following recommendations from recent guidelines for biomarker assessment of risk, we sought to determine the additional prognostic impact of likely downstream testing with echocardiography based on NT-proBNP results in this population (13). Second, to establish if NT-proBNP levels are a biochemical precursor to left ventricular systolic dysfunction in older adults, we investigated whether an elevated or rising NT-proBNP level identifies individuals at risk of progression from a normal to an abnormal LVEF based on sequential echocardiography.
The CHS (Cardiovascular Health Study) is a multicenter, prospective observational cohort study of cardiovascular disease in independently living older adults (age ≥65 years) recruited from 4 communities. The study population consists of the original cohort recruited in 1989 to 1990 and those enrolled in 1992 to 1993 when the study was expanded to include more African Americans. A detailed description of the study methods has been published previously (14).
Of the 5,888 CHS participants, subjects were included if they had no prevalent heart failure, interpretable echocardiograms, and sufficient serum for NT-proBNP measurement. Ultimately, 4,188 (71.1%) participants were included in this analysis (Fig. 1). Participants with sufficient sera volumes and an initial LVEF assessment were modestly more likely to be female and less likely to be African American and diabetic than those without sufficient sera and/or initial ejection fraction measurement (Online Table 1), but other factors did not differ.
The institutional review boards of the University of Washington and the participating centers approved the CHS. The institutional review board of the University of Maryland, Baltimore, approved the current analysis.
The design for the echocardiographic evaluation of CHS participants has been described previously (15). In summary, 2-dimensional echocardiography was performed in 1989 to 1990 and again in 1994 to 1995. For the original cohort, this corresponded to the baseline visit and 5 years later. For the second cohort, this resulted in a single echocardiogram 2 years after the baseline visit. Global left ventricular systolic function was qualitatively assessed from the 2-dimensional echocardiogram as normal (LVEF ≥5%), borderline (LVEF ≥45% to <55%), or subnormal (LVEF <45%) ejection fraction. LVEF was qualitatively interpreted in 99% of the original CHS cohort, with inter-reader agreement of 94% and intrareader agreement of 98% of paired studies (16). For this analysis, subjects with a borderline or subnormal LVEF were grouped together and classified as having an “abnormal” LVEF. In addition, we report measures of Doppler mitral diastolic inflow peak E (early) and peak A (atrial) velocities and left atrial size measured by linear dimensions based on 2-dimensional directed M-mode imaging (17).
NT-proBNP was measured in serum collected at baseline in the main CHS cohort (1989 to 1990) and the second cohort (1992 to 1993). A second measure of NT-proBNP was performed on sera collected 3 years later for the main cohort (1992 to 1993) and 2 years later for the second cohort (1994 to 1995).
All samples were stored at –70° to –80° C and were thawed before testing (maximum of 3 freeze–thaw cycles). NT-proBNP was measured using electrochemiluminescence immunoassay on the Elecsys 2010 system (Roche Diagnostics, Indianapolis, Indiana). The coefficient of variation for the NT-proBNP assay was 2% to 5% during the testing period, and the analytical measurement range for NT-proBNP was 5 to 35,000 pg/ml. Baseline NT-proBNP levels ≥190 pg/ml (the 70th percentile for the study population) were considered elevated on the basis of previously identified cutoff values best corresponding with increased risk of heart failure in this population (10).
Outcomes were incident heart failure and cardiovascular mortality. Incident heart failure events were ascertained through review of medical records, by participant interview at annual study visits, and semi-annual phone calls. An expert adjudication panel determined potential heart failure events and cause of mortality (18). Cardiovascular mortality was defined as mortality related to atherosclerotic heart disease, mortality after cerebrovascular disease, or mortality from other atherosclerotic and cardiovascular diseases as described in detail previously (18).
Clinical history and the electrocardiogram
Clinical characteristics and cardiovascular risk factors were obtained from the initial CHS study visit for each cohort (for the analysis of baseline NT-proBNP and outcomes) or at the study visit of the follow-up NT-proBNP (for the analysis of change in NT-proBNP and outcomes). The methods for assessing cardiovascular risk factors have been described previously (19).
Coronary heart disease was defined as a history of angina, myocardial infarction, coronary angioplasty, or coronary artery bypass surgery. An electrocardiogram (ECG) was performed annually; left ventricular mass was estimated from the ECG, and major ECG abnormalities, including atrial fibrillation and left ventricular hypertrophy, were defined according to previously described methods (20,21).
Characteristics according to baseline NT-proBNP and left ventricular functional status were compared using chi-square tests or 1-way analysis of variance as appropriate. Cumulative incidence of heart failure and cardiovascular mortality were estimated using the Kaplan-Meier method. Multivariate analyses were performed using Cox proportional hazards models for new-onset heart failure and cardiovascular mortality outcomes, adjusting for demographic characteristics (age, sex, and race), cardiovascular disease history, cardiovascular risk factors (systolic blood pressure, diabetes, cholesterol, creatinine, and body mass index), use of antihypertensive medications, and major ECG abnormalities. Elevated NT-proBNP was defined using a previously validated cutoff value of ≥190 pg/ml (10).
Change in NT-proBNP was considered as a categorical predictor among those with an initial low NT-proBNP level of <190 pg/ml. Risk of heart failure and cardiovascular mortality were examined associated with: 1) a stable or decrease in NT-proBNP level (i.e., no increase >25%); and 2) an increase of at least 25% to a level ≥190 pg/ml. The 25% threshold for change was based on the reported intraindividual variability in NT-proBNP levels in patients with stable heart failure (22). We then evaluated whether baseline echocardiographic information about LVEF (≥55% vs. <55%) added to the predictive value of increases in NT-proBNP. Last, we evaluated the incremental value of LVEF as a semi-quantitative variable (<45%, 45% to 54% and ≥55%) and NT-proBNP as a continuous variable (after log-transformation) for both outcomes. The time-dependent C-statistic was used to examine the added predictive value of the LVEF assessment to: 1) a demographic model with and without baseline NT-proBNP levels; and 2) the combination of baseline and of follow-up NT-proBNP levels for incident heart failure and cardiovascular mortality. The improvement in risk classification by the addition of LVEF measurements to NT-proBNP levels in demographic adjusted models was examined using the net reclassification improvement (NRI), which represents the net percentage of subjects correctly reclassified to risk categories (23). We categorized individuals according to Cox model–based risk of 10-year heart failure or cardiovascular mortality of <10%, 10% to 20%, or >20%. An exploratory analysis was also performed using echocardiographic measures of diastolic function, including Doppler mitral E/A ratio (categorized as <0.7, 0.8 to 1.5, and >1.5) and left atrial dimension added to LVEF, NT-proBNP, or both.
Association between changes in NT-proBNP and subsequent new-onset left ventricular dysfunction were evaluated using chi-square tests. Statistical analyses were performed with Stata version 10 (Stata Corp., College Station, Texas) and SPSS version 17.0 (SPSS Inc., Chicago, Illinois), and time-dependent C-statistics were generated using R version 2.7.0. (24).
Of the 4,137 participants without prevalent heart failure and a baseline echocardiogram, 107 (2.6%) had subnormal LVEF (<45%) and 210 (5.1%) had a borderline reduced LVEF (45% to 54%). The area under the curve (AUC) for NT-proBNP to diagnose a subnormal LVEF (<45%) was 0.85, and for any abnormal LVEF (<55%), the AUC was 0.69. High-risk NT-proBNP levels (≥190 pg/ml) were observed in 29.5% (n = 1,219). Table 1 displays demographic, clinical, and echocardiographic diastolic information based on the presence of a high or low NT-proBNP value, further subdivided by the presence of a normal versus abnormal LVEF. The median age of the participants was 71 years (range 65 to 100 years). NT-proBNP status (high vs. low) differentiated patients with a higher prevalence of risk factors, ECG abnormalities, history of coronary heart disease, cardiovascular medication use, increased left atrial size, and diastolic abnormalities. An abnormal LVEF was further associated with male sex, diabetes, coronary heart disease, ECG abnormalities, cardiovascular medication use, increased left atrial size, and diastolic abnormalities.
Outcomes based on NT-proBNP levels and LVEF
The median follow-up was 10.7 years (range 0.1 to 14.1 years) from the time of the baseline measure. There were 1,112 participants who developed heart failure, and 893 who died of cardiovascular causes. The unadjusted hazard ratios for heart failure were 2.95 (95% confidence interval [CI]: 2.61 to 3.32) and 2.42 (95% CI: 2.03 to 2.90) for a baseline NT-proBNP level ≥190 pg/ml and LVEF <55%, respectively. Survival functions for heart failure and cardiovascular mortality based on the combination of baseline NT-proBNP level and LVEF assessment are shown in the Kaplan-Meier plots in Figures 2A and 2B. Differentiation of risk occurred within the first year and continued throughout follow-up. As shown in Table 2 in an unadjusted analysis, the increased risks of heart failure or cardiovascular mortality were of significant magnitude among participants with a low NT-proBNP and an abnormal LVEF (a 1.7- to 2.3-fold increased risk) compared with those with a normal LVEF. For participants with high baseline NT-proBNP, risks of heart failure and cardiovascular mortality were higher, and this finding was further stratified by LVEF assessment.
After adjustment for clinical risk factors, body mass index, ECG abnormalities, and cardiovascular medications in those with low or high NT-proBNP levels, the increased risk associated with the presence of an abnormal LVEF was markedly attenuated but remained significant for both outcomes among those with an initially high NT-proBNP level and for cardiovascular mortality among those with an initially low NT-proBNP level. An abnormal LVEF was no longer associated with risk of heart failure among those with an initially low NT-proBNP (Table 2). In contrast, in a statistical model using NT-proBNP as a continuous variable and LVEF as a semi-quantitative variable, LVEF continued to predict both outcomes after multivariate adjustment (Online Table 2). In a separate sex-based analysis, no differences in the combined effects of LVEF and NT-proBNP were observed between men and women (Online Table 3).
To complement the Cox regression analysis, the C-statistic and NRI were used to evaluate the incremental predictive value of LVEF assessment to NT-proBNP measurement for each outcome (Table 3). For both heart failure and cardiovascular death, the addition of LVEF improved prediction compared with demographic characteristics alone and resulted in a modest reclassification of risk. In contrast, the addition of LVEF assessment to demographic information and the NT-proBNP level resulted in minimal, but statistically significant, improvement in the C-statistic for only the outcome of heart failure and no reclassification of risk for either outcome by the NRI statistic. When restricting the analyses to individuals with an initially elevated NT-proBNP, LVEF assessment did not reclassify risk of heart failure or cardiovascular mortality beyond demographic information and NT-proBNP level (cardiovascular mortality: NRI, −0.006, p = 0.7; heart failure: NRI, 0.008; p = 0.2). Adding echocardiographic measures of diastolic function to LVEF resulted in a significant increase in the C-statistic and reclassification by NRI. However, the addition of NT-proBNP still significantly increased the C-statistic and improved reclassification even after accounting for both LVEF and diastolic measures along with demographic characteristics.
As part of a secondary analysis, we also determined the number of participants who would need to undergo echocardiography to detect either one subnormal (<45%) or abnormal (<55%) LVEF based on an initially high NT-proBNP (Online Table 4).
Follow-up echocardiography with repeat NT-proBNP assessment
Echocardiography was available for 2,375 participants with repeat NT-proBNP levels who had not developed heart failure in the interim between measures (Fig. 1). LVEF <55% (n = 202 [8.5%]) was associated with increased risk of subsequent heart failure (n = 505 events; hazard ratio: 2.38 [95% CI: 1.80 to 3.13]) and cardiovascular mortality (n = 390 events; hazard ratio: 2.93 [95% CI: 2.18 to 3.98]).
We then investigated whether LVEF assessments would add to the risk of both outcomes over and above repeated NT-proBNP assessments in participants with initially low NT-proBNP levels (<190 pg/ml, n = 1,840). Participants were subdivided by comparing those whose NT-proBNP levels had increased >25% to ≥190 pg/ml versus those with stable or decreased NT-proBNP levels.
Among participants with initially low NT-proBNP levels, 361 (19.6%) had increased at follow-up. For these participants, the risk of heart failure was highest among those with an abnormal LVEF, with intermediate risk being present in participants with only one characteristic (i.e., either an increase in NT-proBNP or an abnormal LVEF) (Fig. 3A). For cardiovascular mortality, a persistently low NT-proBNP level indicated a low risk, irrespective of LVEF, whereas LVEF assessment further differentiated the risk of cardiovascular death in individuals with an increase in NT-proBNP level (Fig. 3B). By Cox regression analysis, after adjustment for covariates, LVEF only differentiated risk in the cohort of participants with a rising NT-proBNP level, and only for cardiovascular death (Table 4).
The C-statistic and NRI analysis confirmed the findings from the adjusted Cox regression models. The addition of LVEF assessment to demographic information and serial NT-proBNP measurements neither significantly increased the AUC for the C-statistic nor reclassified the risk of having either outcome using the NRI statistic (Table 5). Similar to models with a single measure of NT-proBNP, echocardiographic diastolic parameters provided additional prognostic and reclassification information to LVEF and serial NT-proBNP concentrations.
As part of a secondary analysis based on change in NT-proBNP level, we determined the number of participants who would need to undergo echocardiography to detect either one subnormal (<45%) or abnormal (<55%) LVEF based on an initial NT-proBNP level of <190 pg/ml that increased >25% to ≥190 pg/ml at follow-up (Online Table 4).
Predicting a decline in LVEF based on serial NT-proBNP levels
Participants in the main cohort with an NT-proBNP level <190 pg/ml and a normal LVEF had repeat echocardiograms 2 years after their second measure of NT-proBNP level (n = 1,486). An abnormal LVEF developed in 95 patients (6.4%). Participants with an increase in NT-proBNP level were significantly more likely to have subsequent decline in their LVEF compared with participants with a stable low NT-proBNP level. Those who started with a high NT-proBNP and a normal LVEF (n = 426) had a similar proportion who developed an abnormal LVEF as those with an initially normal but rising NT-proBNP (Fig. 4).
The results from this study demonstrate that, in ambulatory older adults without heart failure, the addition of LVEF assessment to either a single NT-proBNP assessment or sequential measures adds little to risk assessment for new-onset heart failure or cardiovascular mortality. Furthermore, in contrast to NT-proBNP levels, LVEF alone only modestly reclassifies risk when considering just demographic characteristics. Confirming the limited utility of a natriuretic peptide level to “screen” for subnormal (i.e., <45%) LVEF, 14 participants with a high baseline NT-proBNP level and 34 participants with rising NT-proBNP levels would need to be screened to detect one subnormal LVEF. Despite limited accuracy to detect a subnormal LVEF, a high baseline or an increasing NT-proBNP level identified individuals at greatest risk of developing a new abnormal LVEF on follow-up echocardiography. This latter finding is potentially intriguing because many CHS participants with initially normal LVEF who develop symptoms of heart failure are found to have an abnormal LVEF at the time of presentation (25).
In CHS and other community population studies, an abnormal LVEF is an independent predictor of both new heart failure hospitalizations and cardiovascular mortality (2–4,26). Yet in this analysis, once adjusted for comorbidities, LVEF assessment added little additional predictive benefit beyond the measurement of NT-proBNP. There are several potential reasons for this new finding. First, an abnormal LVEF is a relatively infrequent finding in community-dwelling older adults (<8%) compared with an elevated NT-proBNP level (approximately 30%) (3,6,10). The low prevalence accounts in part for the weak influence of LVEF in reclassifying risk of heart failure or cardiovascular death (Tables 3 and 5). The lack of specificity of natriuretic peptides for increased left ventricular volumes or pressure in asymptomatic subjects can explain the false positive results when using NT-proBNP as a screening tool for an abnormal LVEF in the general population (27–29).
Assessment of LVEF to refine prognostication in community-dwelling older adults on the basis of an elevated natriuretic peptide level should be approached cautiously. Despite a previous study suggesting that natriuretic peptide measurement could be cost-effective in select populations to screen for abnormal LVEF, recent guidelines do not recommend measuring either natriuretic peptides or LVEF as part of a screening strategy (11,12). It may be tempting to consider combining natriuretic peptide levels and LVEF to identify those at greatest risk, and by unadjusted analysis, this seems to be present. With introduction and dispersion of inexpensive handheld ultrasound imaging devices, rapid and less-expensive assessment of LVEF will become prevalent (30). However, once comorbidities are considered, the additional prognostication of LVEF to an NT-proBNP level is markedly attenuated. Furthermore, the addition of LVEF provides insignificant information to improve discrimination and reclassify individuals into lower- or high-risk groups even when considering only participants with initially high NT-proBNP. Our findings should be contrasted to earlier findings in the post-myocardial infarction setting in which natriuretic peptide levels and LVEF have prognostic synergism for both heart failure and death (31). However, reflective of the differences between a post-myocardial infarction population and screening “at-risk” community-based subjects, the prevalence of an abnormal LVEF was approximately 10 times higher in the post-myocardial infarction setting (31). In older adults without known heart failure, clinicians will need to individualize decision making with respect to echocardiography even in the presence of a high NT-proBNP level indicating an increased risk of developing heart failure symptoms, while also considering the importance of knowing diastolic filling patterns, left atrial size, or other cardiac pathologic conditions in specific cases.
This was a large, well-characterized cohort of community-dwelling older adults with serial NT-proBNP levels and echocardiography. However, there are limitations to the study design. The addition of the second cohort of African-American older adults provides for a balanced demographic reflective of older adults in the United States. For this group, baseline NT-proBNP was measured 2 years before an echocardiogram. We have shown that LVEF will change over time but only in a minority of participants even in the presence of an abnormal NT-proBNP level.
In CHS, LVEF was not quantified as a percentage. Interpretation was performed in a semi-quantitative manner, with excellent intrareader and inter-reader reproducibility (16). It is noteworthy that poor outcomes have been associated with even a borderline LVEF (estimated at 45% to 55%) (3). Last, this study does not incorporate all echocardiographic measures of diastolic function, but we do show that diastolic measures can assist in reclassifying risk beyond LVEF and NT-proBNP levels. It remains complex as to how best to integrate diastolic measures into an individual's care.
Older adults comprise the majority of new cases of heart failure, yet most live many years without the diagnosis. Elevated NT-proBNP levels likely reflect an ongoing pathologic process that can initially manifest as progression to an abnormal LVEF before symptoms or as symptoms in the presence of preserved LVEF. Once adjusting for the multiple comorbidities often present in ambulatory older adults, we were unable to demonstrate that an assessment of LVEF could further stratify prognosis after measurement of NT-proBNP. In the presence of an elevated NT-proBNP level in this population, a tailored approach to cardiac imaging appears most appropriate.
The authors thank Simona Barlera, MSc, Medical Statistician, Laboratory of Medical Statistics, Department of Cardiovascular Research Istituto MARIO NEGRI, for the programming code to run the time-dependent C-statistic.
For supplemental tables, please see the online version of this article.
The research reported in this paper was supported by contract numbers N01-HC-85079 through N01-HC-85086, N01-HC-35129, N01 HC-15103, N01 HC-55222, N01-HC-75150, and N01-HC-45133 and grant number U01 HL080295 from the National Heart, Lung, and Blood Institute, with additional contribution from the National Institute of Neurological Disorders and Stroke. Additional support was provided through R01 AG-15928, R01 AG-20098, and AG-027058 from the National Institute on Aging and R01 HL-075366 from the National Heart, Lung, and Blood Institute. Additional funding was provided by Roche Diagnostics. Dr. deFilippi receives grant support as well as consulting and speaking honoraria from Roche Diagnostics and Siemens Healthcare Diagnostics, manufacturers of the NT-proBNP assay; and has received funding or consulted for the following other in-vitro diagnostic companies: Siemens Healthcare Diagnostics, Beckman Coulter, Alere, BG Medicine, and Critical Diagnostics. Dr. Christenson receives grant support and consulting honorarium from Siemens Healthcare Diagnostics as well as grant support and consulting and speaking honoraria from Roche Diagnostics, manufacturers of the NT-proBNP assay. Dr. Seliger has received grant support though Roche Diagnostics in support of this research and other Cardiovascular Health Study ancillary biomarker studies; and consulting fees from Roche Diagnostics. All other authors have reported that they have no relationships relevant to the contents of this paper to disclose.
- Abbreviations and Acronyms
- area under the curve
- confidence interval
- left ventricular ejection fraction
- net reclassification improvement
- N-terminal pro–B-type natriuretic peptide
- Received March 24, 2011.
- Revision received June 14, 2011.
- Accepted June 21, 2011.
- American College of Cardiology Foundation
- Hobbs F.D.,
- Roalfe A.K.,
- Davis R.C.,
- Davies M.K.,
- Hare R.
- Wang T.J.,
- Evans J.C.,
- Benjamin E.J.,
- Levy D.,
- LeRoy E.C.,
- Vasan R.S.
- Rosamond W.,
- Flegal K.,
- Friday G.,
- et al.
- Costello-Boerrigter L.C.,
- Boerrigter G.,
- Redfield M.M.,
- et al.
- Smith J.G.,
- Newton-Cheh C.,
- Almgren P.,
- et al.
- deFilippi C.R.,
- Christenson R.H.,
- Gottdiener J.S.,
- Kop W.J.,
- Seliger S.L.
- Greenland P.,
- Alpert J.S.,
- Beller G.A.,
- et al.
- Heidenreich P.A.,
- Gubens M.A.,
- Fonarow G.C.,
- Konstam M.A.,
- Stevenson L.W.,
- Shekelle P.G.
- Hlatky M.A.,
- Greenland P.,
- Arnett D.K.,
- et al.
- Gardin J.M.,
- Siscovick D.,
- Anton-Culver H.,
- et al.
- Schou M.,
- Gustafsson F.,
- Kjaer A.,
- Hildebrandt P.R.
- ↵The R Project for Statistical Computing. http://www.r-project.org. Accessed August 15, 2011.
- Aurigemma G.P.,
- Gottdiener J.S.,
- Shemanski L.,
- Gardin J.,
- Kitzman D.
- Di Angelantonio E.,
- Chowdhury R.,
- Sarwar N.,
- et al.
- Redfield M.M.,
- Rodeheffer R.J.,
- Jacobsen S.J.,
- Mahoney D.W.,
- Bailey K.R.,
- Burnett J.C. Jr.
- Atherton J.J.
- Richards A.M.,
- Nicholls M.G.,
- Espiner E.A.,
- et al.