Author + information
- Received January 14, 2011
- Revision received February 22, 2011
- Accepted April 12, 2011
- Published online July 26, 2011.
- Reitze N. Rodseth, MBChB, MMed⁎,⁎ (, )
- Giovana A. Lurati Buse, MD†,
- Daniel Bolliger, MD†,
- Christoph S. Burkhart, MD†,
- Brian H. Cuthbertson, MBChB, MD‡,
- Simon C. Gibson, MBChB, MD§,
- Elisabeth Mahla, MD∥,
- David W. Leibowitz, MD¶ and
- Bruce M. Biccard, MBChB, MMed Sci, PhD⁎
- ↵⁎Reprint requests and correspondence:
Dr. Reitze N. Rodseth, Department of Anaesthetics, Nelson R Mandela School of Medicine, Private Bag 7, Congella 4013, South Africa
Objectives The aims of this study were to perform an individual patient data meta-analysis of studies using B-type natriuretic peptides (BNPs) to predict the primary composite endpoint of cardiac death and nonfatal myocardial infarction (MI) within 30 days of vascular surgery and to determine: 1) the cut points for a natriuretic peptide (NP) diagnostic, optimal, and screening test; and 2) if pre-operative NPs improve the predictive accuracy of the revised cardiac risk index (RCRI).
Background NPs are independent predictors of cardiovascular events in noncardiac and vascular surgery. Their addition to clinical risk indexes may improve pre-operative risk stratification.
Methods Studies reporting the association of pre-operative NP concentrations and the primary study endpoint, post-operative major adverse cardiovascular events (defined as cardiovascular death and nonfatal MI) in vascular surgery, were identified by electronic database search. Secondary study endpoints included all-cause mortality, cardiac death, and nonfatal MI.
Results Six data sets were obtained, 5 for BNP (n = 632) and 1 for N-terminal pro-BNP (n = 218). An NP level higher than the optimal cut point was an independent predictor for the primary composite endpoint (odds ratio: 7.9; 95% confidence interval: 4.7 to 13.3). BNP cut points were 30 pg/ml for screening (95% sensitivity, 44% specificity), 116 pg/ml for optimal (highest accuracy point; 66% sensitivity, 82% specificity), and 372 pg/ml for diagnostic (32% sensitivity, 95% specificity). Subsequent to revised cardiac risk index stratification, reclassification using the optimal cut point significantly improved risk prediction in all groups (net reclassification improvement 58%, p < 0.000001), particularly in the intermediate-risk group (net reclassification improvement 84%, p < 0.001).
Conclusions Pre-operative NP levels can be used to independently predict cardiovascular events in the first 30 days after vascular surgery and to significantly improve the predictive performance of the revised cardiac risk index.
More than 200 million major surgical procedures are performed annually worldwide. This is equivalent to 1 procedure for every 25 human beings and increases to 1 procedure for every 10 persons in developed countries (1), with the majority of these procedures being noncardiac surgery. In a recent international randomized controlled study (8,351 patients, 190 hospitals in 23 countries), 6.9% of patients age 45 years or older with or at risk of cardiovascular disease who were hospitalized for noncardiac surgery had cardiovascular events (cardiovascular death, nonfatal myocardial infarction [MI], nonfatal cardiac arrest) within 30 days (2). Patients presenting for vascular surgery have a particularly high cardiovascular disease burden. As a result, they experience higher rates of perioperative mortality, adverse cardiovascular events, and rehospitalizations than patients undergoing other noncardiac procedures (3,4).
Pre-operative risk stratification enables both patients and physicians to make informed decisions regarding the appropriateness of surgery when considering the risk for a perioperative cardiovascular event. In addition, the identification of high-risk patients allows targeted resource allocation during the perioperative period by directing additional pre-operative testing and perioperative monitoring.
Current guidelines make use of clinical risk factors, exercise tolerance, and type of surgery to estimate perioperative cardiovascular risk and direct pre-operative investigation (5). These clinical risk factors, which include a history of ischemic heart disease, compensated or prior heart failure, cerebrovascular events, diabetes mellitus, and renal insufficiency, have been derived from a set of risk factors known as the revised cardiac risk index (RCRI) (6). To date, the use of the RCRI and the performance of noninvasive tests and imaging studies as directed by the guidelines have not provided good discrimination when applied to patients undergoing vascular surgery (7–9).
Ventricular cardiomyocytes secrete B-type natriuretic peptide (BNP), a prohormone, and its inactive cleavage product N-terminal fragment (N-terminal pro-B-type natriuretic peptide [NT-proBNP]), into the blood in response to atrial or ventricular wall stretch. Pre-operative elevations of BNP or NT-proBNP have consistently and independently been associated with adverse cardiovascular events in noncardiac and particularly major vascular surgery (10–23). We aimed to study the following questions: 1) What is the optimal BNP cutoff to predict cardiovascular events after vascular surgery? 2) Does the use of pre-operative natriuretic peptides (NPs), BNP or NT-proBNP, improve upon current risk stratification tools?
We aimed to perform an individual patient data meta-analysis of studies using NPs to predict major adverse cardiac events (MACE) and all-cause mortality within 30 days of vascular surgery. MACEs were defined as the composite of cardiac death and nonfatal MI. In addition, we aimed to determine the NP cutoffs for: 1) a diagnostic test; 2) a general optimal test; and 3) a screening test (24), as well as to determine if the pre-operative use of NP assessment improves the predictive performance of the RCRI (6).
Study identification and selection
Studies reporting on the association of pre-operative NP concentrations and post-operative cardiovascular events in adults undergoing noncardiac vascular surgery were identified by electronic searches of the MEDLINE (July 5, 2010) and EMBASE (week 26 of 2010) databases. The electronic searches were completed by manual search of the reports' reference lists. The terms used in the search strategy were “natriuretic peptides,” “surgery or surgical procedures,” and validated combinations of prognostic terms (25) and diagnostic terms (26,27). No language restriction was applied.
Congress abstracts, studies in cardiac surgery populations, and studies in which BNP administration was part of an interventional algorithm were excluded. To avoid the inclusion of duplicate study data from reports publishing partial results, the study with the most complete follow-up or largest sample size was included. Study quality issues in the study selection process were not considered. Working as 2 groups (C.S.B. and G.A.L.B., R.N.R. and G.A.L.B.), we independently selected studies according to predefined eligibility criteria. Selections inconsistencies were resolved by consensus.
The investigators of eligible studies were contacted by e-mail a maximum of 3 times to obtain individual patient data for pre-operative BNP or NT-proBNP concentrations and the type of noncardiac surgery conducted, history of coronary artery disease, congestive cardiac failure, cerebrovascular disease, diabetes mellitus, and renal failure (creatinine >2 mg/dl) to obtain the individual RCRI for each patient. Information on all patients who had undergone vascular surgery was extracted from the individual databases as supplied by the investigators of each study and subsequently merged. After merging, a random sample of 20% of the cases were checked for accuracy against the original data sets provided by the investigators, and no errors were detected (kappa = 1).
Study quality assessment
All studies included for methodological and reporting quality were evaluated according to the Quality Assessment of Diagnostic Accuracy Studies checklist (28), adapting the checklist for the purposes of this review because all the included studies were prognostic in nature (Online Appendix). In the adapted checklist's formulation, “index test,” “target condition,” and “reference standard” were replaced by “natriuretic peptide concentrations,” “all-cause mortality,” and “outcome,” respectively. Criteria 3, 4, 7, and 13 of the original checklist (28) were regarded as not applicable in this context. Criterion 9 (execution of outcome assessment) of the original checklist was considered as not applicable for the studies addressing in-hospital all-cause mortality only. Two authors (G.A.L.B. and C.S.B.) independently rated study quality.
Frequencies are described as numbers and/or percents. Age is described as a mean and agreement between the authors for eligibility of the retrieved studies as a kappa value.
Before merging study data, the association between BNP concentration and MACE heterogeneity across studies was assessed using chi-square analysis, and a meta-analysis was conducted using Review Manager version 4.3 for Windows (The Cochrane Collaboration, Copenhagen, Denmark). Random-effects or fixed-effects models were used according to the presence or absence of significant heterogeneity between studies, respectively.
The association between NP concentration and MACEs at 30 days was determined using backward stepwise logistic regression, and pooled dichotomous outcomes are reported as odds ratios (ORs) with 95% confidence intervals (CIs). Statistical analysis was performed using SPSS version 15.0 for Windows (SPSS, Inc., Chicago, Illinois).
The general optimal test cutoff value, also known as the optimal diagnostic point, is the point that optimizes the rate of true-positive results while minimizing the rate of false-positive results, thereby reflecting the point with the highest accuracy for the prediction of study endpoints. For both NT-proBNP and BNP, this was defined by receiver-operating characteristic (ROC) statistics using a 1:1 weighting of sensitivity and specificity and the point determined by the value with the minimal distance when using the formula distance = (1 − sensitivity)2 + (1 − specificity)2 (24). The screening cutoff point was chosen at a sensitivity of 95% while optimizing specificity. Similarly, the diagnostic cutoff point was chosen at a specificity of 95% while optimizing sensitivity (24).
Patients were categorized according to their RCRI risk groups (0 = low risk; 1 or 2 = intermediate risk; 3, 4, or 5 = high risk) and then reclassified according to their pre-operative NP concentrations (above or below the general optimal test cutoff) (29). The reclassification by net reclassification improvement (NRI) was then tested for discrimination and reclassification calibration statistics.
Two-sided p values were used in all analyses, and values <0.05 were considered significant.
EpiCalc 2000 version 1.02 (Brixton Books, London, U.K.), SAS version 9.1 (SAS Institute Inc., Cary, North Carolina) and Excel 2007 (Microsoft Corporation, Redmond, Washington) were used for statistical analysis.
Study identification and selection
The literature search retrieved 1,648 citations, of which 15 noncardiac surgery studies fulfilled the eligibility criteria (Fig. 1). The kappa value for eligibility was 0.809.
Of these 15 studies, 10 were identified as containing vascular surgery patients (10,12,15–17,19,20,30–32) (Table 1). Individual datasets were obtained from 6 studies, 5 datasets reporting BNP values in 632 vascular patients (10,13,16,19,30) and 1 study measuring NT-proBNP concentrations in 218 vascular patients (20), for a total dataset of 850 patients having undergone both open and catheter-based vascular surgery. A study that has recently been accepted for publication and fulfilled the criteria was also included (33). The characteristics of the studies for which data were received are shown in Table 2. On analysis, the included studies showed significant heterogeneity (chi-square = 13.37, I2 = 70.1%) in the unadjusted association between BNP and 30-day MACEs. The characteristics for the merged patient population are shown in Table 3.
All of the 11 included studies fulfilled the requirements of a representative spectrum of patients by having clearly defined inclusion and exclusion criteria, outcome verification of the whole cohort, equal outcome evaluation regardless of the NP results, sufficient description of NP measurement, and availability of clinical data. We considered the description of the NP measurement (index test) sufficient for replication in 9 of the studies (10,15–17,19,20,30,31,33). Of the 6 studies (10,15,16,20,30,32) that monitored for MACEs after discharge, 3 provided detailed descriptions of their follow-up methods (10,20,30). In only 2 studies were the NP results interpreted without knowledge of outcome (15,16), and only 5 stated that outcomes were determined without knowledge of the NP results (10,16,20,30,31). Two of the 7 studies that lost patients from follow-up provided reasons for this loss or withdrawal (30,34).
Predictive value of NPs
Figure 2 indicates the results of the meta-analysis of the individual studies in predicting MACEs before the merging of the datasets using a random-effects model. The 3 patients from the study by Leibowitz et al. (19) were not included in the analysis, because they provided only true-positive results.
Using the merged dataset, the general optimal test cut points for the BNP (116 pg/ml) and NT-proBNP (277.5 pg/ml) groups were determined as described in our “Methods” section. Patients were then classified as falling above or below this point.
The following independent predictors of MACEs were identified by backward stepwise logistical regression analysis: NP level higher than the optimal cut point (OR: 7.9; 95% CI: 4.7 to 13.3), aortoiliac surgery (OR: 2.1; 95% CI: 1.2 to 3.6), and diabetes mellitus (OR: 1.9; 95% CI: 1.1 to 3.3). The ORs for NP higher than the threshold were 4.3 (95% CI: 1.7 to 11.3) for cardiac death, 7.5 (95% CI: 4.1 to 13.6) for nonfatal MI, and 3.1 (95% CI: 1.4 to 6.7) for all-cause mortality within 30 days of vascular surgery.
Because only 1 NT-proBNP dataset was available, a ROC analysis for pre-operative BNP and the RCRI (n = 632) in predicting perioperative events was performed on the BNP dataset alone (Table 4).
Determination of BNP screening and diagnostic cutoff points
Because there was only a single study in the NT-proBNP group, we calculated screening and diagnostic cut points for the BNP group only (n = 632). Having determined the optimal general cutoff point for BNP to be 116 pg/ml, the BNP level for a screening test with 95% sensitivity was determined to be 30 pg/ml, and a BNP level of 372 pg/ml was determined for a diagnostic test with 95% specificity. Table 5 shows the test characteristics at these 3 cutoff concentrations in predicting MACE at 30 days.
The area under the ROC curve (AUC) for BNP as a continuum in predicting MACE was 80.5% (95% CI: 75.1% to 85.8%). A reduced ROC curve using only these 3 cutoff points resulted in an AUC of 80.1% (95% CI: 74.3% to 80%). The incidence of MACE stratified according to these 3 cut points is shown in Table 6.
Per the American College of Cardiology and American Heart Association guidelines on perioperative cardiovascular evaluation and care for noncardiac surgery (5), all 850 patients were classified into 3 risk groups according to their RCRI scores. A reclassification was performed on the basis of the patients' NP levels. If levels fell below the optimal general cut point, patients were moved down 1 risk category, and if levels fell above the optimal general cut point, patients were moved up 1 risk category (29).
Table 7 shows the results of the reclassification process. In patients classified as low risk by the RCRI, 20% were reclassified as intermediate by the use of NP concentration. In patients classified as intermediate risk by the RCRI, 71% were reclassified as low risk and 28.5% as high risk. In those classified as high risk by the RCRI, 54% were reclassified as intermediate risk. Overall, the use of NP resulted in a statistically significant improvement in discrimination, with an NRI of 58% (z = 5.48, p < 0.001). In patients classified as intermediate risk by the RCRI, the NRI was 84% (z = 5.37, p < 0.001). Applying this cutoff point to the entire population, without predefining risk categories, results in a “continuous” NRI. This can be used to compare the predictive performance of BNP with other pre-operative risk predictors that may be identified in the future. The continuous NRI was 96.4% (z = 6.89, p < 0.000001).
Predictive value of NP
This meta-analysis shows that among patients undergoing vascular surgery, elevated NPs were independently predictive of MACEs at 30 days in patients undergoing vascular surgery and that the addition of BNP to the widely used RCRI risk stratification system significantly improves risk discrimination in a large proportion of these patients.
This finding supports previous evidence of the independent significant association between pre-operative BNP concentrations and the occurrence of cardiovascular events after vascular surgery (18). The validity of this association between NP and MACEs was supported by evidence of a biological gradient, with increasing concentrations of NP being associated with an increase in the risk for MACEs.
As an individual patient data meta-analysis, this study enabled us to determine cutoff values for pre-operative BNP, thereby overcoming the limitations of previous meta-analyses (18,22,23). Previous studies have focused on the identification of a single optimal discrimination cut point with which to direct patient management. Although this single value is appealing, it may be more logical to make use of a categorical classification system. When used as a continuous variable to predict MACEs, BNP has an AUC of 80.5% which falls to 74.1% when the single optimal cutoff point is used.
We propose that a categorical classification system based on BNP cutoff points reflecting the clinical goals of screening and general optimal and diagnostic testing (24) be investigated for use in pre-operative risk stratification. The use of a categorical system would allow the maximization of sensitivity within the lower risk groups while maximizing specificity in the higher risk groups (35), while maintaining a high degree of diagnostic accuracy. Future studies should not focus on the identification of a single cutoff point alone. Further work to define these cutoff points for NT-proBNP is required.
Risk stratification in vascular surgery
Previously, the RCRI has shown only modest performance in predicting perioperative cardiac events in vascular surgery (36). Similarly, this study has shown similar performance of the RCRI in predicting both MACEs (AUC: 61.6%; 95% CI: 54.6% to 68.6%) and all-cause mortality (AUC: 65.8%; 95% CI: 55.7% to 75.9%) in patients undergoing vascular surgery. This is probably due to the RCRI's having been derived from a population of predominantly noncardiac nonvascular surgery patients. In fact, in the original study in which the RCRI was derived, the index did not perform well in patients undergoing abdominal aortic aneurysm surgery (AUC: 54.3 ± 9.2%) (6).
The addition of pre-operative NP concentration to the RCRI risk stratification resulted in the correct reclassification of 58% of patients. A correct reclassification occurs when a patient who had an event moves up into a higher risk category once restratified with NP concentration, or a patient who did not have an event is moved down a risk category. These results suggest that in patients risk stratified with the RCRI, the optimal BNP cutoff point should be used to reclassify patients, thereby obtaining a more accurate risk assessment. This would improve not only risk assessment accuracy but also the identification of high-risk patients who may benefit from further noninvasive testing. Further work should be undertaken to determine whether the RCRI improves pre-operative risk stratification in patients primarily risk stratified using NPs and to examine the role of the individual RCRI factors together with NPs in improving pre-operative risk stratification. The findings of this meta-analysis, together with the other studies in this area (36), raise concerns regarding the use of the RCRI in isolation in vascular surgery populations as proposed by the American College of Cardiology and American Heart Association algorithm.
First, individual patient data could not be obtained for all studies that the search strategy retrieved. In particular, the available datasets that measured NT-proBNP were under-represented; as such, we chose to limit the calculation of screening and diagnostic cutoff points to the BNP dataset only. Second, 3 different BNP assay methods were used in the studies included in this analysis (Table 2), with a lack of standardization between assays. The degree of imprecision is 3.5% to 4.4% for the ADVIA system (Bayer Diagnostics, Leverkusen, Germany), 5.5% for the AxSYM system (Abbott Laboratories, Abbott Park, Illinois), and 8% for the Shionogi system (Shionogi & Company, Ltd., Osaka, Japan) (37–39) The ADVIA and Shionogi systems recognize similar BNP epitopes that differ from those identified by the AxSYM system. As a result, when compared with the AxSYM system, the ADVIA system on average results in lower BNP values (38). However, all 3 BNP assays make use of a cut point of 100 pg/ml (39), and the degree of imprecision around this shared cut point is consistently acceptable (37–39).
Third, the I2 statistic of 70.1% indicates significant heterogeneity in the unadjusted OR between the studies. The incidence of MACEs is significantly different between the NP study groups but correlated with the degree of disease burden, as indicated by those patients with scores of 3 or more on the RCRI (Pearson's correlation p = 0.01). It would seem that as the event rate within the surgical population increases, so does the predictive value of NP concentration.
Pre-operative NP levels are able to independently predict MACEs (OR: 7.9; 95% CI: 4.7 to 13.3), cardiac death (OR: 4.3; 95% CI: 1.7 to 11.3), and nonfatal MI (OR: 7.5; 95% CI: 4.1 to 13.6) in the first 30 days after vascular surgery. The cutoff points for pre-operative BNP when used as a screening, optimal general, and diagnostic test for MACEs in a vascular surgical population are 30, 116, and 372 pg/ml, respectively. Pre-operative NP levels are able to significantly improve on the predictive performance of the RCRI; their inclusion in existing pre-operative evaluation algorithms should be considered, and the role of the RCRI in vascular surgery should be reviewed.
For a description of study quality assessment, please see the online version of this article.
The Predictive Ability of Preoperative B-Type Natriuretic Peptide in Vascular Patients for Major Adverse Cardiac Events: An Individual Patient Data Meta-Analysis
Dr. Rodseth is supported by a CIHR Scholarship (the Canada-HOPE Scholarship). Dr. Lurati Buse is supported by a scholarship by the Swiss National Science Foundation. Dr. Bolliger has received an unrestricted research grant from CSL Behring (Berne, Switzerland). Dr. Mahla has received N-terminal pro-B-type natriuretic peptide kits from Roche Diagnostics GmbH (Mannheim, Germany), study grants from Novo Nordisk Pharma GmbH (Vienna, Austria) and CSL Behring Biotherapies for Life (Vienna, Austria), and speaker honoraria and consulting fees from CSL Behring Biotherapies for Life. Dr. Biccard is supported by a self-initiated research grant from the Medical Research Council. All other authors have reported that they have no relationships to disclose.
- Abbreviations and Acronyms
- area under the receiver-operating characteristic curve
- B-type natriuretic peptide
- confidence interval
- major adverse cardiac event(s)
- myocardial infarction
- natriuretic peptide
- net reclassification improvement
- N-terminal pro-B-type natriuretic peptide
- odds ratio
- revised cardiac risk index
- receiver-operating characteristic
- Received January 14, 2011.
- Revision received February 22, 2011.
- Accepted April 12, 2011.
- American College of Cardiology Foundation
- Fleisher L.A.,
- Beckman J.A.,
- Brown K.A.,
- et al.
- Lee T.H.,
- Marcantonio E.R.,
- Mangione C.M.,
- et al.
- Kertai M.D.,
- Boersma E.,
- Bax J.J.,
- et al.
- Choi J.H.,
- Cho D.K.,
- Song Y.B.,
- et al.
- Feringa H.H.,
- Schouten O.,
- Dunkelgrun M.,
- et al.
- Karthikeyan G.,
- Moncur R.A.,
- Levine O.,
- et al.
- Peat J.K.,
- Barton B.
- Wilczynski N.L.,
- Haynes R.B.,
- Lavis J.N.,
- Ramkissoonsingh R.,
- Arnold-Oatley A.E.
- Haynes R.B.,
- Wilczynski N.L.
- Cuthbertson B.H.,
- Amiri A.R.,
- Croal B.L.,
- et al.
- Riemersma M.,
- Dijkstra P.U.,
- van Veldhuisen D.J.,
- Muskiet F.A.,
- van den Dungen J.A.,
- Geertzen J.H.
- Mahla E.,
- Vicenzi M.,
- Toller W.
- Wu A.H.,
- Packer M.,
- Smith A.,
- et al.
- Rawlins M.L.,
- Owen W.E.,
- Roberts W.L.