Author + information
- Received April 2, 2004
- Revision received May 6, 2004
- Accepted May 6, 2004
- Published online September 1, 2004.
- Meenakshi A. Bhalla, MD⁎,
- Audrey Chiang, MD⁎,
- Victoria A. Epshteyn, MD⁎,
- Radmila Kazanegra, MD⁎,
- Vikas Bhalla, MD⁎,
- Paul Clopton, MS⁎,
- Padma Krishnaswamy, MD⁎,
- L.K. Morrison, BS⁎,
- Albert Chiu, BS⁎,
- Nancy Gardetto, NPN⁎,
- Sunder Mudaliar, MD†,
- Steven V. Edelman, MD†,
- Robert R. Henry, MD† and
- Alan S. Maisel, MD‡,⁎ ()
- ↵⁎Reprint requests and correspondence:
Dr. Alan S. Maisel, VAMC Cardiology 111-A, 3350 La Jolla Village Drive, San Diego, California 92161
Objectives We hypothesized that B-type natriuretic peptide (BNP) levels can predict cardiac mortality in diabetic patients.
Background Detection of cardiovascular disease in diabetics can be difficult until overt events occur.
Methods A total of 482 diabetics (majority male with type 2 diabetes) at the Veterans Affairs Medical Center San Diego were divided into two groups: 1) referred patients for echocardiogram on the basis of clinical suspicion of cardiac dysfunction (referred [R], n = 180); 2) patients randomly selected from the diabetic clinic without any suspicion of cardiac dysfunction (not referred [N-R], n = 302). We examined cardiac events and all-cause mortality in relation to initial BNP levels during the follow-up.
Results A total of 71 (14.7%) patients died during this period: 52 of 180 (29%) in the R group (30 of 52 [58%] cardiac, 10 of 52 [19%] non-cardiac, 2 of 52 [4%] renal, 10 of 52 [19%] unknown cause) and 19 of 302 (6%) in N-R group (6 of 19 [32%] cardiac). The median BNP level in the R and N-R groups who died of cardiac, non-cardiac, and unknown cause was 537 and 87, 80 and 53, and 343 and 38 pg/ml, respectively. The receiver-operating characteristic (ROC) values for mortality in two groups in relation to BNP revealed the area under the curve to be 0.720 and 0.691, respectively (p < 0.01 in both). Among commonly used prognostic indicators in diabetics, only the ROC for triglycerides was significant. The most accurate cut-point in both the N-R group (87%) and R group (61%) was 120 pg/ml of BNP. Cox regression analysis showed BNP to be the most significant predictor of all-cause mortality in the R group. There was a marked decrease in survival in the patient group with BNP >120 pg/ml.
Conclusions B-type natriuretic peptide appears to be a reliable predictor of future cardiac and all-cause mortality in diabetic patients.
There were 12.1 million people in the U.S. in 2002 who were diagnosed with diabetes, and by 2020 that number will grow to 17.4 million (1). Diabetes is a major risk factor for increased cardiovascular events, and diabetic individuals have a 200% to 400% greater risk for vascular disease than non-diabetics. Recent advances in the treatment of coronary disease have improved survival for diabetics and non-diabetics, but diabetics still have double the case fatality rate compared with non-diabetics (2). Diabetic individuals also have an increased frequency of silent ischemia, systolic and diastolic left ventricular (LV) dysfunction, and cardiac autonomic neuropathy (3). The high frequency of modifiable risk factors provides great opportunities for prevention, the cornerstones of therapy being glycemic control, aggressive risk factor modification, and ongoing patient surveillance and monitoring to facilitate early disease detection and prompt intervention.
B-type natriuretic peptide (BNP) is a cardiac neurohormone predominantly released from the ventricles in response to LV volume expansion and pressure overload. B-natriuretic peptide levels are known to be elevated in patients with LV dysfunction and correlate with New York Heart Association functional class, prognosis, and systolic and diastolic LV dysfunctionfindings on echocardiography (4–8). The availability of rapid BNP measurements can help reliably detect the presence or absence of LV dysfunction in patients with diabetes (4,9–11). Because BNP has also been shown to be an effective tool in screening in diabetes, we designed this prospective study in diabetic patients to see whether BNP can also help predict morbidity and mortality.
The University of California San Diego Institutional Review Board approved this study. The participants included 482 diabetic patients recruited between June 1999 and February of 2001 from San Diego Veterans Affairs Medical Center. These patients were divided into two groups:
1. Referred (R): those subjects referred by physician or nurse practitioner for echocardiography on the basis of clinical suspicion of cardiac dysfunction, most commonly shortness of breath, n = 180.
2. Not referred (N-R): those subjects randomly selected and recruited from the diabetic clinic. In this group of patients, there was no suspicion of cardiac dysfunction, no referrals to cardiologists, and no previous records of echocardiography with abnormalities of LV function (systolic or diastolic), n = 302.
Patients were followed for mean time period of 827 ± 384 days and 864 ± 207 days (mean ± SD), respectively, at San Diego Veterans Affairs Medical Center. Most were males with type 2 diabetes (95% in R and 84% in N-R groups, respectively). Death certificates were used to determine the cause of mortality in patients who did not die in the hospital. Patients designated to “unknown cause of death” group, died outside San Diego county, and death certificates were not available. For simplicity, heart failure (HF) with preserved systolic function is referred to as “diastolic dysfunction.”
Measurement of BNP levels
During initial evaluations, a small sample (5 cc) was collected into tubes containing potassium EDTA (1 mg/ml blood). B-natriuretic peptide was measured using the Triage B-Type Natriuretic Peptide test (Biosite Incorporated, San Diego, California). The Triage BNP test is a fluorescence immunoassay for the quantitative determination of BNP in whole blood and plasma specimens and has been recently characterized with regard to precision, analytical sensitivity, and stability (12–14) B-natriuretic peptide values were determined on site utilizing the point-of-care method with either whole blood or plasma samples.
Descriptive statistics were computed for each group. Median BNP levels are reported for subgroups of patients and compared with Mann-Whitney Utests. Box and whisker plots were used to show BNP distributions of subgroups based on outcomes. Receiver-operating characteristic (ROC) curves were calculated for prediction of all-cause death by plasma BNP concentration. In addition, summary statistics for ROC curves were calculated for prediction of all-cause mortality by other markers including triglycerides, high-density lipoprotein, low-density lipoprotein, and HbA1c. Sensitivity, specificity, positive predictive value, negative predictive value, and accuracy were computed for BNP using a selection of possible cut-points. Accuracy is defined as true-positive plus true-negative cases divided by the total sample size. The Kaplan-Meier survival analysis was used to estimate survival for patients with BNP levels above and below 120 pg/ml, and the significance of differences was calculated by the log-rank test. Cox regression was used to test for the combined predictive power of BNP and other variables.
Table 1shows the characteristics of all 482 patients subdivided by whether or not they were initially referred for cardiac work-up. Both groups had a high preponderance of males, as they were recruited from the Veterans Affairs Medical Center in San Diego. Referred patients were younger with a higher frequency of type 2 diabetes. As expected, the referred group had a higher incidence of previous cardiac disease (51% vs. 12%) and, except for angiotensin-converting enzyme inhibitors, angiotensin receptor blockers, and statins, were taking more cardiac medications. Table 2shows the mortality in the study population. There were 52 (29%) deaths during the follow-up period in the referred group and 19 (6%) in the not-referred group. In the referred group, 30 (58%) deaths were due to cardiac causes, 10 (19%) were due to non-cardiac causes, 2 (4%) were renal, and in 10 patients death certificates could not be located (death occurred outside of San Diego County). Similarly 6 (32%) died of cardiac causes in the not-referred group, 8 (42%) from non-cardiac causes, and 5 (26%) were unknown.
Figure 1shows box-whisker representations of BNP levels for survivors and non-survivors in the referred and not-referred groups. The initial median BNP level in patients who were alive at the end of follow-up period was higher in the referred than the not-referred group (76 vs. 14 pg/ml, respectively; p < 0.001). Similarly, the initial median BNP level in patients who had died by the end of follow-up period was higher in the referred than the not-referred group patients (312 vs. 40 pg/ml, respectively, p < 0.0001 for all deaths and 537 vs. 87 pg/ml, respectively, p < 0.009 for cardiac deaths). Patients who died during the follow-up period had higher initial BNP levels than those that survived in both the referred (312 vs. 76 pg/ml; p < 0.001) and the not-referred groups (40 vs. 14 pg/ml; p < 0.005).
Figure 2shows the ROC curve for initial BNP levels predicting all-cause death during the follow-up period. In both the referred and not-referred populations, the area under the curve was significant at 0.720 and 0.691, respectively (p < 0.01 in both cases).
Table 3shows different statistical values and 95% confidence intervals predicting mortality in diabetic patients using four different cut-points of initial BNP levels ≥60, ≥80, ≥100, ≥120 pg/ml in both the referred and not-referred subgroups. The sensitivity and positive predictive value were higher in the referred versus not-referred groups across the range of BNP cut-points. A BNP level of 120 pg/ml gave a 69% sensitivity and a 40% positive predictive value.
The specificity and negative predictive value was higher in the not-referred versus referred group across the range of BNP cut-points. The most accurate cut-point in the not-referred group was also a BNP level of 120 pg/ml (87%). This gave a 91% specificity and a 95% negative predictive value. Table 3also shows the positive and negative likelihood ratio of predicting all-cause mortality in non-referred group at the BNP levels of 20, 40, 60, 80, 100, and 120 pg/ml; the likelihood ratio in this group at BNP level of 120 pg/ml was 5.66.
Table 4shows the values for the area under the ROC curve for different markers (cholesterol subtypes, triglycerides, and HbA1c), as compared with BNP, in predicting: 1) all-cause mortality; and 2) cardiac mortality in diabetics. In the referred group, BNP was the most accurate at predicting all-cause mortality as well as cardiac death, with only triglyceride level showing a significant predictive value. In the not-referred population, BNP also had the highest area under the ROC curve (p < 0.005 for all-cause mortality).
Cox regression analysis provided a multivariable approach to predict all-cause mortality from BNP (<120 pg/ml vs. ≥120 pg/ml), age, gender, and type of diabetes mellitus. In the referred patients, BNP was the only significant predictor of the all-cause mortality (p = 0.003). In the not-referred group, only age (p = 0.012) and BNP (p = 0.042) were significant predictors of all-cause mortality.
Echocardiographic data was only available on a subset of the subjects in the not-referred group (91 of 302 cases). A Cox regression in this subset of patients including LV function, ejection fraction (EF), and BNP yielded no significant predictors of survival. By contrast, in the data for the referred group, BNP was significant (p = 0.02), and no other predictors were significant. Some subjects were lost from this second analysis due to missing quantitative values for EF. If EF is removed from the model, BNP is still significant (p = 0.004), and no other predictors are significant.
Figure 3shows the Kaplan-Meier survival curves of patient subgroups with BNP values <120 pg/ml and ≥120 pg/ml in both the referred and not-referred groups of patients. All-cause mortality is shown. In both the subgroups, survival was significantly reduced over time as a function of BNP level, especially in patients referred for cardiac work-up.
The Framingham study firmly established the epidemiologic link between diabetes and LV dysfunction (15). In the presence of coronary disease, diabetes is an independent factor for deterioration of cardiac function (16). While early detection of LV dysfunction enables administration of treatment that can improve survival and increase well-being (17–19), ventricular dysfunction may be difficult to diagnose because patients may be asymptomatic or have non-specific symptoms, and abnormal findings on physical examination are often absent (20,21). A test that not only elucidates the presence of LV dysfunction but also predicts future risk may be a valuable tool in reducing diabetic heart disease.
B-natriuretic peptide is a 32-aa polypeptide containing a 17-aa ring structure common to all natriuretic peptides (22). It is found mainly in the cardiac ventricles, and its release appears to be directly proportional to ventricular volume expansion and pressure overload (4,23,24) B-natriuretic peptide is an independent predictor of high LV end-diastolic pressure (4) and correlates with both New York Heart Association functional classification and prognosis (25). A rapid assay for BNP has been cleared by the Food and Drug Administration as an aid in the diagnosis of congestive HF, and its use in this capacity has already been delineated in recently published HF guidelines (26).
Despite the association between diabetes mellitus and increased cardiovascular morbidity and mortality rates, the prevalence of myocardial systolic and diastolic functional abnormalities in asymptomatic diabetic patients is not well-defined. A recent echocardiographic study of 66 normotensive, asymptomatic subjects with type 2 diabetes showed reduced LV systolic and diastolic function compared with healthy subjects (27). Another study of 86 patients with type 2 diabetes free of cardiovascular disease found diastolic dysfunction in 41 of 86 patients (47%) (28). Diastolic dysfunction in type 2 diabetic patients is often found despite adequate metabolic control and freedom from clinically detectable disease (28). We found that 32 of 91 diabetic patients with no history of cardiac disease had abnormal LV function on subsequent echocardiography (mean BNP level = 161 ± 40 pg/ml) (9). Of these 32 patients, 21 had diastolic dysfunction (BNP = 190 ± 60 pg/ml). A ROC curve revealed the area under the curve was 0.81 for patients without previous suspicion of cardiac dysfunction (p < 0.001).
The National Institutes of Health sponsored Studies Of Left Ventricular Dysfunction (SOLVD) in patients with asymptomatic LV dysfunction demonstrated humoral activation characterized by increases in the natriuretic peptides without activation of the circulating renin-angiotensin system (29). It has been shown that myocardial ischemia augments the synthesis and release of BNP even in the absence of myocardial necrosis or preexisting LV dysfunction. Recently, BNP levels have been shown to be prognostically important in acute coronary syndromes (28,30). Reversible ischemia may transiently increase LV wall stress, which may be sufficient to cause an elevation in BNP levels. Thus, elevated BNP levels may not only indicate LV dysfunction but also be a marker for myocardial ischemia and, therefore, prognosis.
Recently, the landmark paper by Wang et al. (31) found that, after adjusting for traditional risk factors, the level of B-type natriuretic peptide was independently predictive of the risk of death, HF, atrial fibrillation, and stroke over a mean follow-up period of about five years. Levels of B-type natriuretic peptide above the 80th percentile in this cohort (i.e., higher than 20 pg/ml) were associated with an increase by more than 60% in the long-term risk of death. Furthermore, there was a significant prognostic gradient with respect to the risk of HF, atrial fibrillation, and stroke among the three levels of B-type natriuretic peptide (low, intermediate, and high) examined. This remarkable finding strongly suggests that there are important prognostic data even in the range of B-type natriuretic peptide levels thought, on the basis of previous studies, to rule out HF (i.e., levels below 100 pg/ml). Our results are consistent with these findings. We also saw a clear difference in survival in asymptomatic patients with < or >20 pg/ml in our follow-up (p = 0.004) with likelihood ratio of =1.7 (LR+). But the decision statistics indicated a cut-point of 120 pg/ml was better in decision-making in this subgroup of patients (LR+ = 5.66). This difference could be attributed to being the predominantly male diabetic population in our study as compared with a community-based population in the Framingham study. Our patients undoubtedly are a “sicker” group than the Framingham group, especially the group referred for possible cardiac disease. There is a high incidence of ventricular dysfunction at the start of follow-up in this group of patients. Even though the randomly selected 302 patients had no history of heart dysfunction, many not only had other risk factors such as hypertension, but had underlying diastolic dysfunction on echocardiography. Nonetheless, the magnitude of risk that a high BNP implies is greater than implied by other markers such as high-density lipoprotein, low-density lipoprotein, and HgA1C.
This was a prospective study done at a single veterans hospital, so one must be careful about generalizing the results to the entire population. Both the areas under the ROC curves and the negative predictive values are dependent on the patient population studied. Our population represents generally older, predominantly male veterans with a high prevalence of cardiac disease.
In diabetic patients, whether referred for echocardiography because of possible cardiac disease or completely asymptomatic, BNP levels >120 pg/ml should alert the caregiver to the likelihood of heart disease and a higher risk of morbidity and mortality. We believe that BNP may be a cost-effective, easy-to-obtain, and useful prognostic tool for risk stratification in diabetic patients and can be used to assist in monitoring and preventing complications, which in turn may result in optimization of management and increased patient survival.
Dr. Maisel is a consultant for Biosite Inc., and Dr. Kazanegra and Paul Clopton have received salary support from Biosite. Dr. Marc A. Silver acted as guest editor for this article.
- Abbreviations and acronyms
- B-type natriuretic peptide
- ejection fraction
- heart failure
- left ventricular
- receiver-operating characteristic
- Received April 2, 2004.
- Revision received May 6, 2004.
- Accepted May 6, 2004.
- American College of Cardiology Foundation
- Hogan P.,
- Dall T.,
- Nikolov P.,
- American Diabetes Association
- Stamler J.,
- Vaccaro O.,
- Neaton J.D.,
- Wentworth D.
- Lubien E.,
- DeMaria A.,
- Krishnaswamy P,
- et al.
- Epshteyn V.,
- Morrison K.,
- Krishnaswamy P,
- et al.
- Cheng V.,
- Kazanagra R.,
- Garcia A,
- et al.
- Pfeifer M.A.,
- Braunwald E.,
- Moye L.A,
- et al.
- Deveraux R.B.,
- Liebson P.R.,
- Horan M.J.
- Remes J.,
- Miettinen H.,
- Reunanen A,
- et al.
- Cheung B.M.Y.,
- Kumana C.R.
- Yoshimura M.,
- Yasue H.,
- Okamura K,
- et al.
- Wallen T.,
- Landahl S.,
- Hedner T,
- et al.
- Remme W.J.,
- Swedberg K.,
- Task Force for the Diagnosis and Treatment of Chronic Heart Failure, European Society of Cardiology
- De Lemos J.,
- Morrow D.,
- Bentley J,
- et al.