Postoperative B-type Natriuretic Peptide for Prediction of Major Cardiac Events in Patients Undergoing Noncardiac Surgery: Systematic Review and Individual Patient Meta-analysis

MORE than 200 million surgeries were performed in adults globally each year and this number is increasing.^1,2  More than 10 million of these patients will suffer a major perioperative cardiovascular event (i.e., cardiovascular mortality or myocardial infarction [MI]) within 30 days of surgery.³  Effective optimization and intervention are only possible when these at-risk patients are accurately identified.

B-type natriuretic peptide (BNP) and its inactive cleavage product N-terminal fragment BNP (NT-proBNP) are hormones secreted from ventricular myocytes in response to ventricular wall stretch or myocardial ischemia.⁴  Preoperative concentrations of BNPs are powerful predictors of mortality and MI in patients undergoing noncardiac surgery.⁵  Postoperative NPs may have similar prognostic abilities,⁶  but not all studies have demonstrated this signal.⁷  Many of these studies have small sample sizes, ranging from 22 to 400 patients, and have been conducted on focused patient populations, such as vascular surgery. All these factors limit the generalizability of these individual studies. Furthermore, the previous individual studies have not established specific postoperative NP thresholds that define a patient at risk of an adverse postoperative cardiac event.

We undertook a systematic review and individual patient data meta-analysis to determine whether NPs, sampled less than 7 days postoperatively, are independently associated with the individual outcomes of mortality, cardiac mortality, nonfatal MI, cardiac arrest, coronary revascularization, or heart failure within 30 days and 180 days or more of adult noncardiac surgery. The study protocol (CRD42012002054) was registered with an international prospective register of systematic reviews (PROSPERO).

Studies of noncardiac surgery patients, where postoperative BNP or NT-proBNP was measured upto 7 days after surgery, were considered eligible for inclusion. Studies were included regardless of language, design, sample size, publication status, or date of publication. Studies were excluded if patients had cardiac surgery, included pediatric patients, or used NPs (e.g., nesiritide) as therapy. Studies that met the inclusion criteria but did not report an outcome of interest (i.e., mortality, cardiac mortality, nonfatal MI, cardiac arrest, coronary revascularization, or cardiac failure) were included if authors were able to provide the unpublished data for one or more of these outcomes.

We searched six databases (EMBASE, OVID Health Star, Ovid Medline, Cochrane Central Register of Controlled Trials, Cochrane Database of Systematic Reviews, and ProQuest Dissertations and Theses A&I), used abstracts from meetings of the American Heart Association and the American Society of Anesthesiologists, consulted with experts, reviewed reference lists from identified articles, and searched for cited references of key publications in Web of Science. The search terms, including validated prognostic search terms, and databases used are listed in Appendix 1. No language filters were used. To avoid inclusion of duplicate study data from reports publishing partial results, only the article with the largest most complete follow-up was included.

Database searches were conducted on January 14, 2012 using the OvidSP search engine (Ovid Technologies, Inc., New York, NY) for the following databases:

1. EMBASE 1980–2012 Week 3

2. OVID Health Star (1966 to November 2011)

3. Ovid MEDLINE(R) In-Process & Other Non-Indexed Citations and OVID MEDLINE(R) 1946 to present

4. Cochrane Central Register of Controlled Trials (January 2012)

5. Cochrane Database of Systematic Reviews (January 2012)

6. ProQuest Dissertations and Theses A&I (January 2012)

Example of search conducted on MEDLINE

Drs. Biccard and Rodseth independently screened the title and abstract of each citation identifying those that potentially could fulfill the eligibility criteria. Full texts of citations identified to undergo full-text review during the screening process were obtained, and eligibility was independently evaluated by Drs. Biccard and Rodseth. Disagreements were solved by consensus, and where this could not be reached a third reviewer (Dr. Lurati Buse) made the final eligibility decision.

Using a standardized extraction form, we recorded the following data: study design, year of publication, sample size, type of surgery, length of follow-up, method of follow-up, type of NP assay used, and measurement frequency. Study quality was assessed using the modified Quality Assessment of Diagnostic Accuracy Studies assessment tool.⁸ 

All authors from eligible studies were contacted and invited to provide anonymous individual patient data using standardized Excel (Microsoft Corp., Redmond, WA) spreadsheets. Age, sex, individual Revised Cardiac Risk Index (RCRI) risk factors⁹  or if this was not available the cumulative RCRI score, type and urgency of surgery, postoperative NPs value, and predefined 30- and more than 180-day outcomes were obtained. Data sets were checked for accuracy and completeness. Only studies supplying individual patient data were included in this review. We assessed the risk of publication bias by constructing a funnel plot for the composite outcome of mortality and nonfatal MI.

Our a priori individual outcomes of interest included: mortality, cardiac mortality, nonfatal MI, cardiac arrest, coronary revascularization, and cardiac failure. Four studies did not differentiate between nonfatal MI and fatal MI and did not conduct routine postoperative troponin surveillance.^10–13  For this reason, we were not able to report nonfatal MI; however, given that all trials collected data on mortality and MI, we were able to report this composite outcome of mortality or nonfatal MI. We deemed this outcome as the most clinically relevant and used it as our primary outcome.¹⁴  The MI definitions used are shown in Appendix 2. Two studies collected data on cardiac mortality (Appendix 3).^6,15  Eleven studies collected data on cardiac failure,^{7,10–13,16–20}  and three of these studies explicitly defined this outcome (Appendix 4).^19,21,22  No studies collected data on coronary revascularization or cardiac arrest.

Interobserver agreement was tested using κ statistics for study eligibility. An a priori decision was made that when studies measured more than one postoperative NP value within the first 48 h, the highest NP measurement would be used in the analysis. For studies measuring NPs after the first 48 h, we used the NP value closest to the time of surgery. Separate data sets were created for BNP and NT-proBNP. Using receiver operating characteristic (ROC) curves, we identified the highest ROC discriminatory threshold, using Youden Index (J = sensitivity + specificity − 1 = sensitivity − false positive rate), together with its associated 95% CIs for the composite outcome of mortality and nonfatal MI at 30 days in both data sets.^23,24  We then classified patients as falling on, above, or below this threshold, and merged the data sets.

Generalized estimating equations²⁵  were used to analyze each outcome. All analyses were adjusted for age, cumulative RCRI score (as a categorical variable), type of surgery (vascular vs. nonvascular), study as a clustering variable, and NP (above or below the highest ROC discriminatory threshold as determined above). Generalized estimating equations are special generalized linear models technique for clustered or correlated data. It allows for the specification of a correlation structure among patients within a study. We used an exchangeable correlation structure, which assumes the same correlation between any two patients within a study. The results are reported as adjusted odds ratio (AOR), corresponding 95% CI and associated P values. As a sensitivity analysis, we repeated this analysis using only those studies where NPs were sampled on the first day after surgery. We assessed collinearity using the variance inflation factor. Variables with a variance inflation factor greater than 10 were considered to be collinear, and if present we excluded one of these variables from the analysis.

The NT-proBNP thresholds of less than 300 pg/ml and greater than 900 pg/ml in patients aged between 50–75 years, and BNP thresholds of 250 and greater than 400 pg/ml, have been used in the diagnosis of acute cardiac failure.^26–28  We explored these thresholds to determine whether they separated patients into clinically useful risk groups for the outcome of mortality or nonfatal MI at 30 days after surgery. To determine the clinical utility of postoperative NP reclassification, we used patient age, RCRI score, and type of surgery (vascular or nonvascular) to classify them into four preoperative risk categories (<5, 5–10, >10–15, and >15) for the outcome of 30-day mortality or nonfatal MI. We then reclassified patients using these postoperative NP thresholds and tested the results using reclassification statistics.²⁹ 

The criterion for statistical significance was set a priori at α = 0.05. We used IBM SPSS Statistics 19.0.0 (Chicago, IL) for descriptive analyses and SAS 9.2 2008 (Cary, NC) for generalized estimating equations and logistic regression.

Our literature search identified 876 citations from which our screening process identified 53 to undergo full-text evaluation and from these we identified 28 eligible studies.^{6,7,10–13,15–22,30–43}  Data from five of these studies^33,39–42  were contained in larger subsequent publications that we included,^6,19,20,30  leaving 23 cohorts as shown in figure 1.^{6,7,10–13,15–22,30–32,34–38,43}  Five studies were not included in the analysis as we were unable to contact authors of three studies;^31,32,34  one study did not collect data on any of our outcomes of interest;³⁶  and data from one group is under review by an institutional research committee and could not be shared.³⁰  Authors of the remaining 18 studies provided individual patient data, and these data are included in this systematic review. Interobserver agreement for study eligibility was high (κ = 0.84). The funnel plot is shown in figure 2.

The characteristics of the 18 study cohorts are shown in table 1. All studies were prospective cohort studies of small sample sizes (smallest, 22 patients and largest, 400 patients). The type of surgery evaluated within studies included: six vascular surgery studies (679 vascular surgery patients), three thoracic studies (471 thoracic surgery patients), two orthopedic studies (122 orthopedic surgery patients), two urological studies (77 urology surgery patients), and five mixed or major general surgery studies (844 total patients). Most studies measured NP on day 1 after surgery. The quality of eligible studies was generally high (Appendix 5).

A total of 2,051 patients were included across the 19 studies. Ten studies evaluated BNP (n = 627) and nine studies evaluated NT-proBNP (n = 1,424). The mean age of the patients was 67 ± 12 (SD) years and 66% were men. The most common surgery was vascular surgery (43% underwent this surgery), and 32% of patients had a history of coronary artery disease. Table 2 shows the clinical characteristics of all 2,051 patients and for the subgroups of patients who did and did not suffer the primary outcome.

The 30-day event rates were as follows: mortality or nonfatal MI, 11.6% (n = 238 of 2,051); mortality, 3.3% (n = 67 of 2,051); cardiac mortality, 1.5% (n = 5 of 337); and cardiac failure, 3% (n = 63 of 2,051). The corresponding greater than or equal to 180-day outcomes (mean follow-up of 212 days) were 33.5% (n = 480 of 1,432) for mortality or nonfatal MI, 11.1% (n = 160 of 1,432) for mortality, 9.2% (n = 31 of 337) for cardiac mortality, and 22.7% (n = 163 of 717) for cardiac failure.

For the composite outcome of mortality or nonfatal MI at 30 days, the highest ROC postoperative NP discrimination point was 245 pg/ml for BNP with a 95% CI ranging from 195 to 468 pg/ml (ROC area under the curve, 0.71; 95% CI, 0.64–0.78) and 718 pg/ml for NT-proBNP with a 95% CI ranging from 656 to 994 pg/ml (ROC area under the curve, 0.80; 95% CI, 0.77–0.84). For the merged data set with both NPs, the ROC area under the curve was 0.76; 95% CI, 0.73–0.80.

Table 3 shows the odds ratios—adjusted for age, RCRI score, and type of surgery (vascular or nonvascular)—associated with postoperative NPs increased above the highest ROC threshold for each of the study outcomes. Patients with an increased postoperative NP measurement were at increased risk of 30-day mortality or nonfatal MI (AOR, 4.5; 95% CI, 2.74–7.4; P < 0.001) and cardiac failure (AOR, 18.5; 95% CI, 4.55–75.29; P < 0.001). NP increases remained predictive for greater than or equal to 180-day mortality or nonfatal MI (AOR, 3.3; 95% CI, 2.58–4.3; P < 0.001) and cardiac failure (AOR, 3.5; 95% CI, 1.0–9.34; P = 0.022).

Testing the robustness of these results by performing logistic regression in which we ignored study clustering did not appreciably change the AORs (Appendix 6). For the outcomes of mortality, and the composite of mortality and nonfatal MI, the results of the sensitivity analysis that excluded measurements obtained after the first postoperative day did not differ appreciably from the primary results (Appendix 7). For the outcomes of cardiac mortality and cardiac failure, the AOR was lower but remained significant. No variables were found to be collinear.

Patients with BNP values of 0–250, greater than 250–400, and greater than 400 pg/ml suffered the composite of 30-day mortality or nonfatal MI at a rate of 6.6, 15.7, and 29.5%, respectively. Patients with NT-proBNP values of 0–300, greater than 300–900, and greater than 900 pg/ml suffered the same composite of 30-day mortality or nonfatal MI at a rate of 1.8, 8.7, and 27%, respectively. In the NT-proBNP group, 32% (460 of 1,421) of patients had a value greater than 900 pg/ml. In a post hoc analysis, postoperative NP measurement improved overall net reclassification index by 20% (P < 0.001), with 6% of patients having an event being reclassified as higher risk and 14% of patients without an event being reclassified as lower risk (table 4). Among the 895 patients with a preoperative risk between 5 and 15% the net reclassification index improvement was 70% (P < 0.001) with 46% of patients having an event being reclassified to a high-risk category, and 25% of patient without an event being reclassified to a low-risk category.

Due to the number of patients with an NT-proBNP greater than 900 pg/ml we decided to undertake a post hoc analysis to determine the results for an additional NT-proBNP threshold of 3,000 pg/ml, which has been used in previous publications.⁴⁴  Patients with NT-proBNP values of 900–3,000 and greater than 3,000 pg/ml had an incidence of 30-day mortality or nonfatal MI of 20.9 and 38.4%, respectively. Using 3,000 pg/ml as a threshold ensured the CIs of the two new risk groups did not overlap. These NT-proBNP and BNP results together with the associated AOR and multilevel likelihood ratios are shown in tables 5 and 6.

Our systematic review and individual patient level meta-analysis included more than 2,000 patients who had a variety of different noncardiac surgeries. Meta-analysis indicates that an increased postoperative NP is an independent predictor of mortality, cardiac mortality, mortality or nonfatal MI, and cardiac failure at 30 and greater than or equal to 180 days after noncardiac surgery.

The strengths of this review lie in the: (1) rigorous systematic review methodology, (2) success in obtaining individual patient data, and (3) the quality of the included studies. Simulation studies show that logistic regression models require at least 10 events per predictor variable to produce stable estimates of association.⁴⁵  In our meta-analyses, we evaluated four independent variables, that is age, RCRI score, type of surgery (vascular vs. nonvascular), and peak postoperative NP measurement. We surpassed 10 events per variable for all outcomes except the 30- and greater than or equal to 180-day cardiac mortality outcomes.

Our systematic review is limited by the inability to adjust for postoperative troponin measurements, which are known to be strongly associated with postoperative mortality.^46,47  The majority of troponin increases occur within the first 48–72 h after surgery;¹⁴  the same time period during which 16 of the studies included in this review sampled NPs. Unfortunately, the data were not available to allow us to determine the temporal relationship between NP and troponin increases. As we were limited by the inability to obtain the individual preoperative RCRI risk factors for each study it is possible that our model suffers from residual confounding. The funnel plot suggests the possibility of publication bias toward studies reporting a stronger association between increased postoperative NPs and the composite outcome of mortality and nonfatal MI. This weakens our inferences.

Preoperative NPs are useful for preoperative risk stratification. This analysis did not evaluate the interaction between preoperative and postoperative NPs. It is possible that postoperative NP increases may allow clinicians to identify the most vulnerable patients among those with high preoperative NP concentrations.³⁷ 

The large number of patients and deaths allowed us to evaluate all-cause mortality as an outcome. Unfortunately, cardiac mortality was only collected in two studies.^6,15  The small number of events analyzed for this outcome make overfitting of the results possible. This, together with the results of the sensitivity analysis, demonstrates that readers should interpret the cardiac mortality and heart failure data with caution. Evaluation of postoperative MI is hampered by the varying definitions used among the studies; however, all these studies included troponin increase as part of their definitions. The definitions of cardiac failure varied widely among studies, which may limit the validity of these results.

Converting a continuous variable into a categorical or dichotomous variable results in a loss of information, but it may make results more practical for clinical use. The thresholds we explored separated patients into clinically useful risk groups, but in many cases CIs overlapped. This may be due to the small sample size in some of the groups. The net reclassification results suggest that in this patient population a postoperative NP measurement is of greatest value in patients with a 5–15% preoperative baseline risk of 30-day mortality or nonfatal MI.

Previous studies have shown that NP concentrations are related to the extent of myocardial injury after nonoperative MI and improve prognostic scoring systems.^48,49  Our systematic review demonstrates the potential utility of postoperative NPs to identify patients at high risk of adverse cardiac events. These patients may benefit from close postoperative monitoring and more rigorous heart rate and fluid management; however, clinical trials are needed to determine whether this risk factor is modifiable. Postoperative NPs may also have a role in the diagnosis and management of patients with subclinical postoperative cardiac failure and may alert clinicians to those patients at risk of acute decompensation.

Before NPs can be incorporated into clinical practice they must demonstrate superiority to other more commonly used risk factors.⁵⁰  For this to take place studies are required to understand the relationship between postoperative troponin and NP increases. Two questions should be addressed: (1) do increased NPs in patients without early postoperative troponin increases predict postoperative cardiac complications? and (2) do increased NPs in patients with postoperative troponin increases add additional prognostic value to troponin measurement alone? Furthermore, studies to determine the optimal timing for sampling postoperative NPs would also be useful.

Postoperative NPs increases are associated with postoperative mortality, cardiac mortality, mortality and nonfatal MI, and cardiac failure at both 30 days and 180 days or more after surgery. Clinicians may thus find postoperative NP measurements useful in stratifying cardiovascular risk after noncardiac surgery.

The authors thank Ian Glenn, M.D., Research Fellow, Department of Cardiology, Barnet and Chase Farm NHS Trust, London, United Kingdom; Ali Ansaripour, Research Fellow, Department of Cardiology Barnet and Chase Farm NHS Trust; Hence J. M. Verhagen, M.D., Ph.D., Professor and Chief of Vascular Surgery, Erasmus Medical Center, Rotterdam, The Netherlands; Sanne E. Hoeks, M.D., Ph.D., Assistant Professor, Erasmus University Medical Center, Department of Anesthesiology, Erasmus Medical Centre, Rotterdam, The Netherlands; Jan Brozek, M.D., Ph.D., Assistant Professor, Departments of Clinical Epidemiology and Biostatistics and Medicine, McMaster University, Hamilton, Ontario, Canada; Agnieszka Grudniewicz, Ph.D. Candidate, Institute of Health Policy, Management and Evaluation, University of Toronto, Toronto, Ontario, Canada, Li Ka Shing Knowledge Institute, St. Michael’s Hospital, Toronto, Ontario, Canada; David McDonagh, M.D., Associate Professor of Anesthesiology and Medicine (Neurology), Department of Anesthesiology, Duke University Medical Center, Durham, North Carolina; and John Morton, M.D., M.P.H., F.A.C.S., Associate Professor of Surgery, Section Chief, Minimally Invasive Surgery, Department of Surgery, Stanford University School of Medicine, Stanford, California.