Author + information
- Received August 24, 2010
- Revision received January 17, 2011
- Accepted February 10, 2011
- Published online June 28, 2011.
- S. Goya Wannamethee, PhD⁎,⁎ (, )
- Paul Welsh, PhD†,
- Gordon D. Lowe, DSc‡,
- Vilmundur Gudnason, MD, PhD§,
- Emanuele Di Angelantonio, MD∥,
- Lucy Lennon, MSc⁎,
- Ann Rumley, PhD‡,
- Peter H. Whincup, PhD¶ and
- Naveed Sattar, MD‡
- ↵⁎Reprint requests and correspondence
: Prof. S. Goya Wannamethee, Department of Primary Care and Population Health, UCL Medical School, Royal Free Campus, Rowland Hill Street, London NW32PF, United Kingdom
Objectives We aimed to compare the predictive capabilities of N-terminal pro-brain natriuretic peptide (NT-proBNP) and C-reactive protein (CRP) for risk of cardiovascular disease (CVD) in older men with and without pre-existing CVD.
Background The clinical utility of NT-proBNP in CVD risk stratification in the general population remains unclear.
Methods A prospective study of 3,649 men age 60 to 79 years were followed for a mean of 9 years during which there were 608 major CVD events (major fatal and nonfatal coronary heart disease, stroke, and CVD death).
Results NT-proBNP was significantly associated with risk of all major CVD outcomes after adjustment for CV risk factors in both men with and without CVD. The adjusted standardized hazard ratios for CVD events in those without pre-existing CVD and those with pre-existing CVD were 1.49 (95% confidence interval [CI]: 1.33 to 1.65) and 1.52 (95% CI: 1.33 to 1.75), respectively. CRP was associated with CVD outcomes only in men without pre-existing CVD (adjusted standardized hazard ratios: 1.22 [95% CI: 1.10 to 1.34] and 1.00 [95% CI: 0.86 to 1.38], respectively). NT-proBNP was more strongly associated with CVD outcome than CRP, particularly among those with pre-existing CVD. Inclusion of NT-proBNP in a Framingham-based model yielded significant improvement in C-statistics in both men with and without CVD and resulted in a net reclassification improvement of 8.8% (p = 0.0009) and 8.2% (p < 0.05), respectively, for major CVD events. Inclusion of CRP in the Framingham-based model did not improve prediction in either group (net reclassification improvement 3.8% and 0.6%, respectively).
Conclusions NT-proBNP, but not CRP, improved prediction of major CVD events in older men with and without pre-existing CVD.
There is intense interest in the use of “novel” blood marker tests for assessing cardiovascular disease (CVD) risk both in the general population and in high-risk groups such as those with CVD (1). Two such markers include N-terminal pro-brain natriuretic peptide (NT-proBNP) and C-reactive protein (CRP). NT-proBNP (2–16) and CRP (17,18) have been shown to be strongly associated with risk of CVD in both high-risk patients with established CVD and in the general population. A recent systematic review based on published findings showed that adding NT-proBNP or BNP to predictive models containing conventional CVD risk factors generally led to modest improvement in measures of risk discrimination (C-statistic) for subsequent CVD (2), but the available data were sparse for generally healthy populations (3,6,9). Furthermore, measures of risk discrimination are only one way of assessing the performance of a statistical prediction model (19), and more clinical and descriptive measures, such as reclassification metrics, can help to assess the clinical potential of adding a new biomarker to a prediction model (19,20). Only a limited number of studies, however, have evaluated the performance of NT-proBNP for the purpose of reclassifying CVD risk in generally healthy people, and the clinical utility of NT-proBNP in CVD risk stratification remains unclear (3,5,10). Similar considerations apply to CRP, with conflicting data showing modest or no clinical utility (18). One study attempted to directly compare CRP and NT-proBNP for clinical utility but found that neither marker increased discrimination as measured by only the C-statistic (9). Finally, no studies have examined separately people with and without baseline CVD within the same general population, which may be important because of reverse causation bias (particularly for NT-proBNP).
We therefore aimed to resolve outstanding uncertainties in 2 ways. First we examined the independent relationship between NT-proBNP and risk of CVD in the British Regional Heart Study separately in men with and without CVD with adjustment for a wide range of CV risk factors. Then we directly compared NT-proBNP and CRP as CVD risk markers by: 1) comparing marker associations with CVD and coronary heart disease (CHD) (including fatal events) outcomes separately in men with and without baseline CVD; 2) comparing marker ability to improve the C-statistic in a basic clinical CVD risk model in both men with and without baseline CVD; and 3) comparing markers in similar reclassification models.
The British Regional Heart Study is a prospective study of CVD involving 7,735 men, screened between 1978 and 1980, age 40 to 59 years drawn from 1 general practice in each of 24 British towns (21). The population studied was socioeconomically representative of British men but consisted almost entirely of white Europeans (>99%). In 1998 to 2000, all surviving men, now age 60 to 79 years, were invited for a 20th year follow-up examination. All men completed a mailed questionnaire providing information on their lifestyle and medical history, had a physical examination, and provided a fasting blood sample collected using the Sarstedt Monovette system (Sarstedt, Numbrecht, Germany). The samples were frozen and stored at −20°C on the day of collection and transferred in batches for storage at −70°C until analysis. Twelve-lead electrocardiograms were recorded using a Siemens Sicard 460 instrument (Siemens, Erlangen, Germany) and analyzed and coded in accordance with Minnesota Coding definitions at the University of Glasgow. The men were asked whether a doctor had ever told them that they had angina or myocardial infarction (MI) (heart attack, coronary thrombosis), heart failure, or stroke; they were also asked to bring their medications to the examination session. A total of 4,252 men (77% of survivors) attended an examination, and blood serum samples were available from 4,088 men. Because of sample attrition, NT-proBNP measurement on serum was conducted in 3,761 men. Of these men, 112 with pre-existing self-reported heart failure at baseline were excluded to reduce the potential for nonatherothrombotic reverse causal associations with CVD risk, leaving 3,649 men for the present analyses.
CV risk factors
Anthropometric measurements including body weight, height, and waist circumference were performed. Details of measurement and classification methods for smoking status, physical activity, body mass index (BMI), waist circumference, social class, alcohol intake, blood pressure, and blood lipids in this cohort have been described (22–24). Prevalent diabetes included men with a diagnosis of diabetes or men with fasting blood glucose ≥7 mmol/l. CRP was assayed by ultra-sensitive nephelometry (Dade Behring, Milton Keynes, United Kingdom) (24). Estimated glomerular filtration rate (eGFR), estimated from serum creatinine using the modification of diet in renal disease equation developed by Levy et al. (25) was used as a measure of renal function. Anemia was defined as hemoglobin levels <13 g/dl. Self-reported previous CVD events were used to classify those with pre-existing CVD at baseline.
NT-proBNP levels were determined using the Elecsys 2010 electrochemiluminescence method (Roche Diagnostics, Burgess Hill, United Kingdom). Samples were snap-thawed at 37°C and assayed on the analyzer, which was calibrated using the manufacturer's reagents. Manufacturer's controls were used to monitor assay drift, using both a high and low control, with limits of acceptability defined by the manufacturer. Low control coefficient of variance was 6.7%, and high control coefficient of variance was 4.9%.
All men have been followed from the initial examination (1978 to 1980) for CV morbidity, and follow-up has been achieved for 99% of the cohort (26). In the present analyses, all-cause mortality and morbidity events are based on follow-up from rescreening in 1998 to 2000 at age 60 to 79 years to July 2008, a mean follow-up period of 9 years (range 8 to 10 years). Information on death was collected through the established “tagging” procedures provided by the National Health Service registers. Fatal stroke episodes were those coded on the death certificate to International Classification of Diseases (ICD) 430 to 438. Nonfatal stroke events were those that produced a neurological deficit that was present for more than 24 h. Fatal CHD was defined as death with CHD (ICD-9 410 to 414) as the underlying code. A nonfatal MI was diagnosed according to World Health Organization criteria (27). CV deaths included all of those with ICD-9 401 to 459. Evidence regarding nonfatal MI and nonfatal stroke was obtained from ongoing reports from general practitioners, biennial reviews of the patients' practice records (including hospital and clinic correspondence) through the end of the study period, and repeated personal questionnaires to surviving patients after the initial examination. Outcomes assessed in the current analyses were major CHD (defined as fatal or nonfatal MI), major stroke events (fatal or nonfatal), CVD death, and all major CVD events (major CHD events, stroke events, or CVD death).
The distribution of NT-proBNP and CRP was skewed, and log transformation was used. The Cox proportional hazards model was used to assess the multivariate-adjusted hazard ratio (relative risk [RR]) in a comparison of quartiles of NT-proBNP and for a 1-SD increase in NT-proBNP and CRP levels. In multivariate analyses, smoking (never, long-term ex-smokers [≥15 years], recent ex-smokers [<15 years], and current smokers), social class (manual vs. nonmanual labor), physical activity (4 groups), alcohol intake (5 groups), diabetes (yes/no), BMI (<25, 25 to 27.5, 27.5 to 29.9, and ≥30 kg/m2), anemia (yes/no), and eGFR (<60, 60 to 69, and ≥70 ml/min/1.73 m2) were fitted as categorical variables; NT-proBNP (log), forced expiratory volume in 1 s (FEV1), high-density lipoprotein (HDL) cholesterol, and CRP (log) were fitted as continuous variables. To assess the clinical utility of NT-proBNP (or CRP) in CVD risk prediction, we performed a series of analyses. Receiver-operating characteristic (ROC) curves and areas under the curve (C-statistics) were used to assess the ability of NT-proBNP (and CRP) to predict CVD beyond established risk factors included in the Framingham score (28). We calculated risk function estimates based on the regression coefficients of the Framingham-based Cox models with and without NT-proBNP (CRP) in the model. We constructed ROC curves and computed the areas under the curve (C-statistics) for predicting CVD events for the risk function estimates. Tests for differences between the C-statistics for established-based risk function models with and without NT-proBNP (CRP) were performed using an SAS macro (%ROC) with SAS software (version 9.1, SAS Institute, Inc., Cary, North Carolina). Then we addressed the issue of how NT-proBNP (or CRP) evaluation may alter the risk stratification of men. All men were categorized according to 3 risk groups based on their 10-year predicted probabilities (<10%, 10% to 19%, and ≥20%) obtained from risk function models for CVD events with and without NT-proBNP (or CRP). We evaluated the ability of NT-proBNP (and CRP) to reclassify risk using methods suggested by Pencina et al. (29) by calculating the net reclassification improvement (NRI) and the integrated discrimination improvement (IDI).
Baseline characteristics by NT-proBNP levels
Table 1 shows the baseline characteristics according to quartiles of NT-proBNP concentration in men with and without CVD. Raised NT-proBNP levels were strongly associated with many adverse CV risk factors including age, physical inactivity, atrial fibrillation, systolic blood pressure, lung function (FEV1), CRP level, and eGFR in both groups of men. However, NT-proBNP was inversely associated with BMI, cholesterol level, and triglyceride level.
Follow-up CVD events in men with and without baseline CVD
During the mean follow-up time of 9 years, there were 194 major CHD events (119 fatal; 61.3% case fatality), 158 major stroke events, 223 CVD deaths, and 402 major CVD events (stroke, CHD, and CVD deaths) in the 2,893 men with no pre-existing CVD. In men with pre-existing CVD, there were 110 major CHD events (85 fatal; 77.3% case fatality), 74 major stroke events, 150 CVD deaths, and 206 major CVD events.
NT-proBNP associations with CVD risk
Men with pre-existing CVD (MI, angina, or stroke; n = 756) had higher levels of NT-proBNP than those without (n = 2,893); geometric means were 181.3 (interquartile range: 85 to 379) pg/ml and 81.5 (interquartile range: 41 to 151) pg/ml, respectively (p < 0.0001). Table 2 shows the relationship between NT-proBNP and major CVD events, major CHD events (nonfatal MI or CHD death), fatal CHD events, and all CV mortality in men without CVD. High NT-proBNP level (highest quartile) was associated with significantly increased risk of all major CVD events, CVD mortality, major CHD events, fatal CHD events, and major stroke events even after adjustment for CV risk factors (age, smoking status, physical activity, alcohol intake, BMI, systolic blood pressure, HDL cholesterol, total cholesterol, FEV1, and diabetes). Further adjustment for CRP, renal dysfunction, atrial fibrillation, and anemia made minor differences to the strength of the association (Table 2). In a sensitivity analysis, no significant association was seen with nonfatal MI (n = 75 events) in a model adjusted for CV risk factors (adjusted RR: 1.00, 1.08 [95% confidence interval (CI): 0.55 to 2.60], 1.40 [95% CI: 0.73 to 2.68], and 1.27 [95% CI: 0.63 to 2.58] for the 4 quartiles, respectively). No association was seen between elevated NT-proBNP level and non-CVD mortality (adjusted RR: 1.00, 1.00 [95% CI: 0.73 to 1.37], 1.04 [95% CI: 0.76 to 1.42], and 1.22 [95% CI: 0.89 to 1.66]). Like NT-proBNP, CRP level was significantly associated with all CV endpoints (except nonfatal MI) after adjustment for risk factors including NT-proBNP (data not shown). However, in contrast to NT-proBNP, CRP related significantly to non-CVD mortality (adjusted RR: 1.00, 1.19 [95% CI: 0.87 to 1.63], 1.34 [95% CI: 0.98 to 1.83], and 1.49 [95% CI: 1.10 to 2.03] for the 4 quartiles of CRP, respectively).
In men with CVD but no diagnosed heart failure (Table 3), risk of major CHD and CVD events, fatal CHD events, and CVD mortality increased significantly with increasing levels of NT-proBNP, even after full adjustment. No association was seen with non-CVD mortality (p = 0.51). CRP showed no significant association with any CVD outcome or mortality after adjustment for CV risk factors in these men.
For comparison with CRP and assessment of the risk prediction of CVD beyond Framingham risk score, Table 4 shows the standardized hazard ratio for a 1-SD change in log CRP and NT-proBNP after adjustment for established risk factors included in the Framingham risk score in men with and without pre-existing CVD and the improvement in C-statistics in models with and without NT-proBNP (or CRP). NT-proBNP was a stronger predictor of major CHD and CVD mortality than CRP, particularly in those with pre-existing CVD. In men without CVD, NT-proBNP added significantly to CVD events and CVD mortality prediction but not to major CHD events after inclusion of risk factors. In men with CVD, NT-proBNP added significantly to prediction of both major CHD and CVD events and CVD mortality. By contrast, CRP did not add significantly to prediction after inclusion of traditional risk factors in men with or without CVD. When we tested CRP and NT-proBNP jointly in predicting all CVD events, we noted that adding CRP to a model that included NT-proBNP provided no further improvement in prediction (p = 0.12). By contrast, adding NT-proBNP to a model that included CRP improved prediction significantly (p = 0.03).
Table 5 shows the cross-tabulation between predicted risk obtained using the risk function based on established risk markers used in Framingham with and without NT-proBNP (or CRP) in cases and noncases in men with and without pre-existing CVD. In men with no pre-existing CVD, the overall NRI was estimated at 8.8% for NT-proBNP (p = 0.01) (NRI for events was 7.3%, and NRI for non-CV events was 1.5%). The IDI was 2.33 and was significant (p < 0.0001). For CRP, the overall NRI was small (3.8%; p = 0.07) and the IDI was nonsignificant (0.32; p = 0.14). In men with pre-existing CVD, the net gain reclassification for NT-proBNP was estimated at 8.2% (p = 0.049) for all CVD events, and the IDI was also significant (4.38; p < 0.0001); for CRP, the NRI for all CVD events was 0.6% (p = 0.71) and the IDI was significant but negative (−0.88; p < 0.001).
We have conducted an in-depth analysis of the predictive ability of NT-proBNP in a large population-based study of men with and without CVD; given the large number of events, results add usefully to the findings of a recent meta-analysis on this topic (2). Our results suggest that NT-proBNP is more strongly associated than CRP with CVD risk (in those with and without prior CVD), provides a more specific signal for CVD risk (seen for both major CHD and stroke events), and provides greater incremental clinical utility than CRP for CVD risk assessment in older men. Among men with no history of CVD, the C-statistic for risk of all CVD events improved significantly when NT-proBNP was added to a (Framingham-based) model; NRI also improved, whereas CRP did not significantly improve either. These results suggest a genuine potential for NT-proBNP to offer potential prognostic value beyond established clinically applicable markers, and this question therefore deserves further detailed study, including cost effectiveness of such an approach, in other cohorts.
As a CVD risk marker, NT-proBNP is pragmatically attractive for many of the same reasons as is CRP in that it can be routinely measured (currently to rule out heart failure diagnosis on admission) and there are robust and reproducible assays available commercially. In addition, from a biological point of view, release of BNP in patients with vascular disease is considered to be primarily a result of myocardial stretch and physiological stress to produce a natriuretic and diuretic effect (via the physiologically active BNP fragment, along with the inactive NT-proBNP fragment). To what extent subclinical cardiac stress also explains more subtle elevations in NT-proBNP level needs to be properly examined, however, because other pathways related to the more complex physiological functions/interactions of the natriuretic peptides may also play a role. Nevertheless, it is of interest that increased NT-proBNP level is associated with increased risk of CVD and CHD events after adjustment for classical risk factors in older men (both with and without baseline CVD) but not with non-CVD mortality. By contrast, CRP is associated with both CVD and non-CVD mortality and predicts noncardiac endpoints, including several cancers and respiratory diseases (17).
That NT-proBNP is more predictive of CVD risk than CRP, particularly among men with prevalent baseline CVD (in whom CRP showed no predictive value), has also been noted in other studies (7,9,10,14,15); NT-proBNP levels will generally be elevated proportionally to the severity of existing disease. However, the practical relevance of this observation among those who have had an event previously is uncertain. Clearly, most individuals who have previously experienced a CVD event will be on a range of medications (including statins, antihypertensives, beta blockers, and aspirin). However, it is possible that those who are at greatest risk may benefit from higher doses to meet more stringent blood pressure and lipid targets or be most suitable to test novel CVD protective modalities in future studies. Because the relative severity of underlying CVD (including silent infarcts) in an apparently healthy patient may not always be obvious to an examining clinician, the information provided by NT-proBNP may also be useful in primary prevention risk prediction. It is also possible that NT-proBNP levels may in part reflect increased vascular risk associated with other comorbidities such as rheumatoid arthritis. These possibilities require further study. However, adjustment for medication for joint disorders made little difference to the findings.
NT-proBNP showed trends to be more strongly associated with fatal CVD and CHD events compared with nonfatal CHD events in men with and without baseline CVD, which is similar to trends previously noted for inflammatory markers (30). This observation suggests that studies with only fatal endpoints are more likely to report stronger associations of BNP or NT-proBNP with incident events than studies with combined fatal and nonfatal events. This observation may also explain inconsistencies among studies and why significant improvement in risk stratification for primary CVD events tend to be observed in studies with older patients in whom case fatality is high (10) or in studies reporting on only fatal CVD endpoints (6).
Study strengths and limitations
Strengths and limitations of the study require consideration. This study is based on a cohort of older (age 60 to 79 years) men. Although this is a group of considerable clinical interest because they constitute a high-risk group in whom traditional risk factors become less predictive (31), our results need further confirmation in similar study populations, middle-aged populations, and women. The study population is socially representative of the United Kingdom, and follow-up rates in the British Regional Heart Study were exceptionally high. Ascertainment of CHD death and MI was based on standard methods, and both CHD mortality and MI incidence rates corresponded closely with national data.
The results of this study suggest that NT-proBNP has a greater potential to improve prediction of major CVD events than does CRP in older men with and without pre-existing CVD. These results may have clinical relevance, and other groups should now extend our observations in other prospective cohorts to establish whether NT-proBNP is indeed able to enhance risk prediction across a range of populations in a clinically meaningful, and cost-effective, manner. If so, future collaborative ventures may also be able to ascertain how to optimally introduce NT-proBNP into clinical CVD risk prediction.
The British Regional Heart Study is a BHF research group and receives support from the BHF Programme grant (RG/08/013/25942). Dr. Welsh is supported by BHF fellowship FS/10/005/28147. The authors have reported that they have no relationships to disclose.
- Abbreviations and Acronyms
- body mass index
- coronary heart disease
- C-reactive protein
- cardiovascular disease
- estimated glomerular filtration rate
- forced expiratory volume in 1 s
- high-density lipoprotein
- integrated discrimination improvement
- myocardial infarction
- net reclassification improvement
- N-terminal pro brain natriuretic peptide
- Received August 24, 2010.
- Revision received January 17, 2011.
- Accepted February 10, 2011.
- American College of Cardiology Foundation
- Di Angelantonio E.,
- Chowdhury R.,
- Sarwar N.,
- et al.
- Linssen G.C.,
- Bakker S.J.,
- Voors A.A.,
- et al.
- Wang T.J.,
- Larson M.G.,
- Levy D.,
- et al.
- Olsen M.H.,
- Hansen T.W.,
- Christensen M.K.,
- et al.
- Lorgis L.,
- Zeller M.,
- Dentan G.,
- et al.,
- RICO Survey Working Group
- Blankenberg S.,
- McQueen M.J.,
- Smieja M.for the H.O.P.E. Study Investigators.
- Shah T.,
- Casas J.P.,
- Cooper J.A.,
- et al.
- Cook N.R.
- Shaper A.G.,
- Pocock S.J.,
- Walker M.,
- Cohen N.M.,
- Wale C.J.,
- Thomson A.G.
- Wannamethee S.G.,
- Lowe G.D.O.,
- Whincup P.H.,
- Rumley A.,
- Walker M.,
- Lennon L.
- Emberson J.,
- Whincup P.H.,
- Walker M.,
- Thomas M.,
- Alberti K.G.M.
- Walker M.,
- Shaper A.G.,
- Lennon L.,
- Whincup P.H.
- Rose G.,
- Blackburn H.,
- Gillum R.F.,
- Prineas R.J.
- Wilson P.W.F.,
- D'Agostino R.B.,
- Levy D.,
- Belanger A.M.,
- Silbershatz H.,
- Kannel W.B.