The Predictive Value of Preoperative Natriuretic Peptide Concentrations in Adults Undergoing Surgery

A Systematic Review and Meta-Analysis

Abstract

BACKGROUND: 

Several studies have evaluated preoperative B-type natriuretic peptides (NPs) for predicting mortality after surgery; however, the number of deaths in each study was small, limiting the power of these studies. We conducted a systematic review and meta-analysis of studies addressing preoperative NP levels to predict mortality after cardiac and noncardiac surgery.

METHODS: 

We searched MEDLINE and EMBASE using the terms “natriuretic peptides,” “surgery or surgical procedures,” and a validated combination of prognostic and diagnostic terms. Two investigators independently assessed studies for eligibility and extracted data. The end points were all-cause mortality at ≥6 months and at ≤90 days. We used a bivariate model to derive measures of prognostic accuracy and their heterogeneity. We calculated the pooled positive predictive value (PPV) and negative predictive value (NPV) by Bayesian Markov chain Monte Carlo methods.

RESULTS: 

Of the 1558 retrieved articles, 23 studies satisfied the predefined eligibility criteria. After cardiac surgery, the diagnostic odds ratio of NP was 4.11 (95% confidence interval, 2.22–7.60) for ≥6-month mortality, the PPV 0.17 (95% Bayesian confidence interval, 0.07–0.36), and the NPV 0.96 (0.90–0.98). After noncardiac surgery, the diagnostic odds ratio of NP was 4.97 (3.06–8.07) for ≥6-month mortality. The corresponding PPV was 0.24 (0.14–0.38) and the NPV 0.94 (0.88–0.97). Results were similar for ≤90-day mortality.

CONCLUSIONS: 

Preoperative NP concentrations were associated with mortality after cardiac and noncardiac surgery. NP had high NPVs for both types of surgery suggesting that preoperative NP concentrations may be helpful in preoperative risk stratification.

In Brief

Published ahead of print March 3, 2011

Prediction of death or major adverse cardiac events in patients undergoing cardiac and noncardiac surgery is mainly based on clinical risk scores such as the revised cardiac risk index¹ or the EuroSCORE.² However, these scores have a moderate power to discriminate between low and high risk for postoperative cardiac events.³ Accordingly, there is a need for more powerful and objective risk indicators. Natriuretic peptides (NPs) are well-established risk markers in numerous clinical settings including stable⁴ and unstable coronary artery disease,⁵ chronic⁶ and acute^6–8 heart failure, and in patients at risk for cardiac events.⁹ In nonsurgical patients, the evidence on NPs satisfies all acceptable criteria required to establish them as a useful risk marker for adverse events¹⁰ (i.e., prospective validation of additional prognostic value, evidence for changes in clinical management, improvement in outcome on the basis of the changes made, and cost-effectiveness).

Despite the publication of numerous studies in surgical patients, the role of NPs as clinical risk markers for adverse events including mortality is less well established. The individual studies enrolled only a limited number of patients or they often included patients with low cardiovascular risk, leading to a low number of adverse events and consequently low power of the statistical models. Recently published meta-analyses,^11–13 moreover, could only partially overcome this shortcoming because they were focused on short-term events (within 30 days after surgery)¹¹ or on a specific patient group (vascular patients).¹² The meta-analysis by Ryding et al.¹³ included a broader range of surgical patients and evaluated outcome within 6 months of surgery; however, this study revealed only odds ratios without information on positive predictive value (PPV) and negative predictive value (NPV) or likelihood ratios of elevated NP levels. The latter analysis would allow for a more accurate estimation of the posttest probability of an event dependent on the pretest probability.¹⁴ Furthermore, no systematic review or meta-analysis assessing the prognostic association of preoperative brain NP (BNP) and outcome in cardiac surgery patients has been published.

Therefore, the purpose of this study was to perform a systematic review and meta-analysis summarizing data on the association between preoperative NP values and all-cause mortality ≥6 months after cardiac or noncardiac surgery in adults. Our hypothesis was that elevated preoperative NP levels were associated with mortality after cardiac and noncardiac surgery.

METHODS

Study Identification

Using electronic searches of MEDLINE (1966 to November 2009) and EMBASE (1985 to November 2009) databases, we identified studies that evaluated the association between preoperative NP levels and mortality in adults undergoing surgery. We manually completed the electronic search of the articles' reference lists. The search strategy used the terms “natriuretic peptides,” “surgery or surgical procedures,” and validated combinations of prognostic terms^15,16 and diagnostic terms^17,18 without language restriction (Appendix 1, see Supplemental Digital Content 1, https://links.lww.com/AA/A230). We excluded congress abstracts, studies with interventional NP administration, studies that did not report mortality data, studies reporting <2 deaths, and reports lacking specific information about number of patients or events, and follow-up duration. To avoid inclusion of replicated study material, we excluded reports publishing partial results if there were reports with more complete follow-up available. Study quality issues were not considered in the selection process.

Study selection was independently performed by 2 investigators (GLB and CB), who resolved inconsistencies by consensus.

Descriptive Data

We extracted information on the number of patients enrolled, follow-up duration and completeness, the nature of surgical procedures (e.g., cardiac and noncardiac surgery, elective versus urgent/emergency), median age, gender proportion, NP cutoff concentration, timing of blood sample collection, and number of deaths in patients with elevated and with nonelevated preoperative NP concentrations according to the cutoff of the individual studies. In the case of missing data, we sought contact with the study authors by mail. If we were unable to obtain the required data necessary for meta-analysis, we reported the results descriptively as they were published, as well as the adjusted measures of association. All data were independently extracted by 2 investigators and disagreement was resolved by discussion.

Predefined Study End Points

We considered all-cause mortality at ≥6 months after surgery as the primary end point. All-cause mortality ≤90 days after surgery was a secondary end point.

Study Quality Assessment

We evaluated all included work for methodological and reporting quality according to the QUADAS checklist.¹⁹ Checklist questions were adapted to this review, because the included studies were of a prognostic nature (Appendix 2, see Supplemental Digital Content 2, https://links.lww.com/AA/A231). Accordingly, the formulation “test,” “target condition,” and “reference standard” were replaced by “NP concentrations,” “all-cause mortality,” and “outcome,” respectively. Moreover, criteria 3, 4, 7, and 13 of the original checklist¹⁹ were regarded as not applicable in our context. Quality was independently rated by 2 readers.

Statistical Analysis

We described frequencies as the number and/or percent and we calculated agreement between the study reviewers for eligibility by κ statistics. We calculated the likelihood ratios (LRs) of NP for mortality within the individual studies.

A bivariate random-effects model was used to obtain summary estimates of sensitivity and specificity.^20–24 The bivariate model preserves the 2-dimensional nature of the data through the joint analysis of the pairs logarithm transformed sensitivity and specificity of each study. By random-effects approach, the model allows for between-study heterogeneity beyond chance as a consequence of differences in study design and quality of the studies included. With the bivariate model, we thus calculated a random-effect estimate of the mean sensitivity and specificity together with their 95% confidence intervals (CIs). The pooled estimates were plotted within the receiver operating characteristic (ROC) curve space together with the estimates of the individual studies and the summary ROC curve. The pooled sensitivity and 1-specificity together with the estimated correlation between the individual sensitivities and 1-specificities of the estimates allowed constructing a 95% confidence area around the summary estimates. The 95% confidence area can be viewed as 2-dimensional CI. The main axis of the 95% confidence area reflects the correlation between sensitivity and specificity (threshold effect).

Heterogeneity was calculated for sensitivity, specificity, and the diagnostic odds ratio (dOR). We planned to assess threshold effect as a determinant of potential between-study variance should the analysis have revealed any significant heterogeneity.

To provide the pooled PPV and NPV and the distributions thereof, we used Markov chain Monte Carlo methods, fitting the equally specified models. The sensitivity and 1-specificity estimates derived from the Markov chain Monte Carlo algorithm were plotted as marginal histograms on the ROC plane and we illustrated the posterior distribution of the PPV and NPV as boxplots for the different scenarios. The uncertainty of the pooled estimates of PPV and NPV is reported as 95% Bayesian CIs. Appendix 3 (see Supplemental Digital Content 3, https://links.lww.com/AA/A232) reports a more detailed statistics description.

RESULTS

Study Selection

Of the 1558 retrieved articles, 23 studies fulfilled the criteria for inclusion, of which 11 addressed ≤90-day mortality only (Fig. 1). In 1493 (95.8%) citations, the abstract contained sufficient information to fully assess eligibility. The κ value for agreement on inclusion between the 2 reviewers was 0.55, and the observed agreement was 0.98 (1531 of 1558). That is, the agreement proportion for the 2 reviewers was 0.74 (17 of 23) for inclusion and 0.99 (1514 of 1535) for exclusion.

Figure 1: Study selection process. NP = natriuretic peptide.

Description of the Included Studies

Eleven studies^25–35 addressed ≥6-month mortality and ≤90-day mortality in a total of 1696 patients and 919 patients, respectively, undergoing cardiac surgery. Twelve studies^36–47 evaluated ≥6-month mortality and ≤90-day mortality in a total of 930 and 4450 patients undergoing noncardiac surgery, respectively. Tables 1 and 2 summarize the characteristics of the studies.

Table 1: Characteristics of the Studies in Patients Undergoing Cardiac Surgery

Table 2: Characteristics of the Studies Evaluating Patients Undergoing Noncardiac Surgery

In 4 cardiac surgery studies^26,27,34,35 and 2 noncardiac surgery studies,^39,45 all data necessary to construct a 2 × 2 contingency table could not be obtained after contacting the authors. Thus, we included the data of 5 studies for cardiac and 10 studies for noncardiac surgery for meta-analysis. Three of these noncardiac surgery studies available for meta-analysis addressed ≥6-month mortality. The results of the studies in which data for 2 × 2 contingency table construction were not obtainable are reported in Tables 3 and 4.

Table 3: Summary of the Results as Published in the Single Studies Addressing Mortality ≥6 Months After Cardiac Surgery

Table 4: Summary of the Results as Published in the Single Studies Addressing Mortality ≤90 Days After Cardiac Surgery

Study Quality

All of the included studies fulfilled the requirements of a representative spectrum of patients, of clearly defined inclusion/exclusion criteria, of outcome verification in the entire cohort, of equal outcome evaluation regardless of the NP results, and of availability of clinical data. We considered the description of the NP measurement (index test) sufficient for replication in 19 studies (82.6%).^{25,27–36,38,40–46} Of the 20 studies that addressed all-cause mortality beyond the in-hospital period, 12 (60%) gave a detailed description of the follow-up methods (execution of the reference standard).^{26,28,30,31,33–36,38,39,45,46} Only 2 studies (8.7%) stated that NP results were interpreted without knowledge of outcome,^41,42 and only 5 (21.7%) stated that outcome was determined without knowledge of the NP results.^{34,36,39,42,45} Of the 21 studies with loss of follow-up, 10 (50%) described the reasons for withdrawal or loss of follow-up.^{26,28,32–34,37–40,46}

Accuracy of Preoperative NP to Predict ≥6-Month All-Cause Mortality After Cardiac Surgery

The pooled sensitivity (0.58, 95% CI: 0.43–0.72) of NP for detecting ≥6-month all-cause mortality after cardiac surgery was lower than the pooled specificity (0.75, 95% CI: 0.67–0.81) (Fig. 2). The histograms along the margins of the ROC plane (posterior distributions of the pooled sensitivity and specificity) are in close agreement with bivariate models based on the likelihood method. The between-study variance of sensitivity and specificity (on the logit scale) was low (0.18 and 0.12, respectively). The PPV and NPV of NP to predict ≥6-month mortality in cardiac surgery patients were 0.17 (95% CI: 0.07–0.36) and 0.96 (95% CI: 0.90–0.98), respectively (Fig. 3). The dOR of NP above versus below the cutoff summarized across individual studies for ≥6-month all-cause mortality after cardiac surgery was 4.11 (95% CI: 2.22–7.60) and it did not show significant heterogeneity (P = 0.17).

Figure 2: Receiver operating curve (ROC) plane with the pairs of sensitivity and 1-specifity of the individual studies (red circles), the summary ROC curve, and the pooled mean sensitivity and 1-specificity (red diamond) within their bivariate 95% confidence area for mortality ≥6 months after cardiac surgery. The gray points' cloud around the pooled estimates are the values of pairs of sensitivity and 1-specificity generated in a computer simulation (Markov chain Monte Carlo simulation) of studies evaluating natriuretic peptides for the prediction of mortality. The frequency distribution of those values is plotted as a histogram, each for sensitivity and 1-specificity, at the margin of the ROC plane.

Figure 3: Positive predictive value (PPV) and negative predictive value (NPV) of preoperative natriuretic peptide levels for mortality at ≥6 months after cardiac surgery. These estimates and their uncertainty measures (Bayesian confidence intervals) were derived by Markov chain Monte Carlo simulation.

We reported in Table 3 the unpooled results (i.e., the OR data as published in the individual studies) of any multivariable analysis published in the cardiac surgery studies. Table 5 shows the positive and negative LRs of NP for the prediction of mortality after cardiac surgery within the individual studies and Figure 4 plots their positive LRs of NP and mortality ≥6 months after cardiac surgery.

Table 5: Positive and Negative Likelihood Ratios of Natriuretic Peptide for the Prediction of Mortality After Cardiac Surgery

Figure 4: Forrest plot of the likelihood ratio of elevated natriuretic peptide concentrations (likelihood ratio of a positive test [LR+]) for the prediction of mortality at ≥6 months after cardiac surgery in the single studies.

Accuracy of Preoperative NP to Predict ≤90-Day All-Cause Mortality After Cardiac Surgery

There were data from only 2 of the studies addressing ≤90-day mortality after cardiac surgery to complete 2 × 2 contingency tables. One of these studies evaluated patients undergoing emergent repair of aortic type A dissection³²; the other addressed patients undergoing pulmonary thromboendarterectomy for pulmonary hypertension.³³ After study selection and before data extraction, we had planned sensitivity analyses excluding those studies from meta-analysis because of their unique populations, because we expected them to lead to heterogeneity. In this situation, we rejected pooling data of ≤90-day all-cause mortality after cardiac surgery and reported their results only descriptively (Tables 4 and 5).

Accuracy of Preoperative NP to Predict ≥6-Month All-Cause Mortality After Noncardiac Surgery

The pooled sensitivity of NP was higher (0.75, 95% CI: 0.62–0.85) than the pooled specificity (0.62, 95% CI: 0.42–0.79) for predicting ≥6-month all-cause mortality after noncardiac surgery. The bivariate 95% confidence area in the ROC plane around the pooled mean values of sensitivity and specificity is shown in Figure 5. The histograms along the margins of the ROC plane show the posterior distributions of the pooled sensitivity and specificity and are in close agreement with the bivariate models. The between-studies variance (heterogeneity) for sensitivity and specificity was low (0.14 and 0.18, respectively).

Figure 5: Receiver operating curve (ROC) plane with the pairs of sensitivity and 1-specifity of the individual studies (red circles), the summary ROC curve, and the pooled mean sensitivity and 1-specificity (red diamond) within their bivariate 95% confidence area for mortality at ≥6 months after noncardiac surgery. The gray points' cloud around the pooled estimates are the values of pairs of sensitivity and 1-specificity generated in a computer simulation (Markov chain Monte Carlo simulation) of studies evaluating natriuretic peptide for the prediction of mortality. The frequency distribution of those values is plotted as a histogram, each for sensitivity and 1-specificity, at the margin of the ROC plane.

The PPV and NPV of NP to predict ≥6-month mortality in noncardiac surgery patients were 0.24 (95% CI: 0.14–0.38) and 0.94 (95% CI: 0.88–0.97), respectively (Fig. 6). The dOR of NP above versus below the cutoff summarized across individual studies for mortality after noncardiac surgery was 4.97 (95% CI: 3.06–8.07), and it did not show significant heterogeneity (P = 0.983). However, this referred to only 3 noncardiac surgery studies.

Figure 6: Positive predictive value (PPV) and negative predictive value (NPV) of preoperative natriuretic peptide levels for mortality ≥6 months after noncardiac surgery. These estimates and their uncertainty measures (Bayesian confidence intervals) were derived by Markov chain Monte Carlo simulation.

We reported in Table 6 the unpooled results (i.e., the OR data as published in the individual studies) of any multivariable analysis published in the noncardiac surgery studies. Table 7 shows the positive and negative LRs of NP for the prediction of mortality after noncardiac surgery within the individual studies, and Figure 7 plots their positive LRs of NP and ≥6-month mortality after noncardiac surgery.

Table 6: Summary of the Results as Published in the Single Studies Addressing Mortality ≥6 Months After Noncardiac Surgery

Table 7: Positive and Negative Likelihood Ratios of Natriuretic Peptide for the Prediction of Mortality After Noncardiac Surgery

Figure 7: Forrest plot of the likelihood ratio of elevated natriuretic peptide concentrations (likelihood ratio of a positive test [LR+]) for the prediction of mortality at ≥6 months after noncardiac surgery in the single studies.

Accuracy of Preoperative NP to Predict ≤90-Day All-Cause Mortality After Noncardiac Surgery

The pooled sensitivity of elevated NP levels for ≤90-day all-cause mortality was 0.9 (95% CI: 0.72–0.97) and the pooled specificity was 0.65 (95% CI: 0.48–0.79) (Fig. 8). However, the 95% CI area and the variation of the posterior distributions of sensitivity and specificity show that the uncertainty in the estimation of these quantities was large for ≤90-day mortality in patients undergoing noncardiac surgery. Moreover, we found a marked between-study variance (logit sensitivity of 1.93 and logit specificity of 0.83) with these studies; however, this was not related to a significant threshold effect (P = 0.879). The PPV and NPV of NP to predict ≤90-day mortality after noncardiac surgery were 0.12 (95% CI: 0.05–0.30) and 0.99 (95% CI: 0.97–1.0), respectively (Fig. 9).

Figure 8: Receiver operating curve (ROC) plane with the pairs of sensitivity and 1-specifity of the individual studies (red circles), the summary ROC curve, and the pooled mean sensitivity and 1-specificity (red diamond) within their bivariate 95% confidence area for mortality ≤90 days after noncardiac surgery. The gray points' cloud around the pooled estimates are the values of pairs of sensitivity and 1-specificity generated in a computer simulation (Markov chain Monte Carlo simulation) of studies evaluating natriuretic peptide for the prediction of mortality. The frequency distribution of those values is plotted as a histogram, each for sensitivity and 1-specificity, at the margin of the ROC plane.

Figure 9: Positive predictive value (PPV) and negative predictive value (NPV) of preoperative NP levels for mortality ≤90 days after noncardiac surgery. These estimates and their uncertainty measures (Bayesian confidence intervals) were derived by Markov chain Monte Carlo simulation.

Expressed by a measure of effect, the summary estimate of the association (dOR [95% CI]) of NP concentrations above versus below the cutoff in an individual study and ≤90-day all-cause mortality after noncardiac surgery was 16.8 (95% CI: 5.1–55.3). The dOR of the noncardiac surgery studies' ≤90-day mortality strongly tended to be significant heterogeneity (P = 0.09). We reported in Table 8 the unpooled results, i.e., the OR data as published in the individual studies, of any multivariable analysis published in the noncardiac surgery studies. Figure 10 plots the positive LRs of NP and mortality at ≤90 days after noncardiac surgery (Table 7).

Table 8: Summary of the Results as Published in the Single Studies Evaluating Mortality at ≤90 Days After Noncardiac Surgery

Figure 10: Forrest plot of the likelihood ratio of elevated natriuretic peptide concentrations (likelihood ratio of a positive test [LR+]) for the prediction of mortality at ≤90 days after noncardiac surgery in the single studies.

DISCUSSION

The data summarized in this meta-analysis showed that elevated preoperative NP levels are associated with all-cause mortality ≥6 months after cardiac and noncardiac surgery and that preoperative NP values below the cutoff chosen in the individual studies were highly predictive of survival. Similarly, preoperative NP concentrations were associated with mortality at ≤90 days after noncardiac surgery.

NPs and Mortality After Cardiac Surgery

Ventricular myocardium responds to stretching stress and to ischemia by the secretion of B-type NP.^48,49 Both BNP and N-terminal prohormone BNP (NT-proBNP), a byproduct of BNP cleavage activation, were demonstrated to be useful biochemical markers for chronic systolic⁵⁰ and diastolic⁵¹ heart failure, as well as for congestive systolic and diastolic heart failure.⁵² In addition, elevated B-type NP concentrations predict adverse cardiovascular events and death in patients with stable^4,53,54 and unstable^5,55 coronary artery disease, chronic^6,56 and acute^6–8 heart failure, and in patients at risk for cardiac events.⁹ The evidence summarized by this systematic review and meta-analysis broadens the clinical spectrum in which elevated NP levels are predictors of mortality by adding patients undergoing cardiac surgery.

Our meta-analysis showed a consistent association between elevated preoperative NP levels as defined by the individual cutoff used in the different studies, and mortality ≥6 months after cardiac surgery. According to the Bradford Hill criteria,⁵⁷ evidence of a biological gradient or concentration-response relationship enforces the association between biomarker and outcome. However, no cardiac surgery studies that have reported on an association between NP and mortality used NP concentrations as continuous variables in logistic regression. Thus, whether there is a potential concentration-response of NP and mortality after cardiac surgery is unknown. Sodeck et al.³² demonstrated a concentration-dependent response for the association between preoperative NT-proBNP and postoperative outcome after repair of type A aortic dissection. These patients, however, represent a focused population of cardiac surgery patients and the observations cannot necessarily be extrapolated to other cardiac surgery patient populations.

Our analysis preserved the 2 components, i.e., sensitivity and specificity, of the NP diagnostic tests rather than computing only a single measure of association (dOR). We also estimated the PPV and NPV, a measure likely more clinically meaningful than the dOR, of preoperative NP concentrations for mortality. Preoperative NP had a low PPV and a high NPV for mortality ≥6 months after cardiac surgery. Thus, these results suggest that an elevated preoperative NP concentration may not accurately identify mortality with a high degree of certainty. In contrast, a nonelevated preoperative NP level is highly indicative of survival ≥6 months after cardiac surgery. This predictive pattern and its dimension parallel the prediction of NT-proNP in acute heart failure, where for short-term mortality a PPV of 19% and an NPV of 96% were reported.⁸ The low PPV and high NPV reflect the low number of true positives and the high number of true negatives, which is again mirrored by a higher specificity (0.75 [95% CI: 0.67–0.81]) than sensitivity (0.58 [95% CI: 0.43–0.72]). The validity of these results is enhanced by the fact that the between-study variance for specificity was lower than for sensitivity, indicating more homogeneous results in specificity than in sensitivity among the included studies. A possible explanation for this is that most studies calibrated their cutoff values for specificity, aiming at high specificity as for a rule-out test.

NPs and Mortality After Noncardiac Surgery

In addition to expanding the clinical settings in which NPs are predictive of mortality for patients undergoing cardiac surgery, our data confirm the results of recent meta-analyses of the diagnostic utility of preoperative NP levels for predicting mortality for patients undergoing noncardiac surgery.^11,13 The estimates presented of the effect size for the association between preoperative NP concentration and mortality ≥6 months was similar to a previous meta-analysis addressing long-term all-cause mortality after noncardiac surgery.¹³ The association between preoperative NP and postoperative outcome after noncardiac surgery is supported by a concentration-effect response.³⁷ The evidence of a concentration-effect response was reported for only short-term outcome after noncardiac surgery.

Because of a different statistical approach compared with previous meta-analyses on preoperative NP and mortality, our data maintained the 2 components of diagnostic tests, i.e., sensitivity and specificity, and thus allowed recognition that, in parallel to the cardiac surgery setting, preoperative NP had a low PPV and a high NPV mortality ≥6 months after noncardiac surgery. Thus, similar to our findings in cardiac surgery patients, the ability of elevated preoperative NPs to predict death after noncardiac surgery was limited, whereas nonelevated preoperative NPs were highly indicative of survival after noncardiac surgery.

Preoperative NP had a low PPV and a high NPV for ≤90-day mortality after noncardiac surgery. Thus, the ability of elevated preoperative NP concentrations to predict death after noncardiac surgery is limited, whereas nonelevated preoperative NP concentrations are highly indicative of an event-free course.

The sensitivity of NP tended to be higher than its specificity for ≤90-day mortality in noncardiac surgery patients. Moreover, specificity distribution among studies was larger than sensitivity in the studies addressing ≤90-day outcome. A possible interpretation is that most studies addressing ≤90-day outcome after noncardiac surgery calibrated their cutoff values in a way that resulted in high sensitivity, with the goal of using NP concentrations as a screening tool for patients in whom modification in clinical management was more immediately conceivable as the patients remained hospitalized.

Overall, as assessed by the heterogeneity of dOR, the noncardiac surgery studies evaluating preoperative NP levels and ≤90-day mortality strongly tended toward between-study variability (heterogeneity) without reaching statistical significance. Frequent sources of between-study variability in diagnostic or prognostic accuracy parameters include different threshold values used to determine individual study sensitivities and specificities. Caution in the appraisal of the pooled results is thus warranted, even when considering calculation from a random-effects model, because between-study variability may arise from differences in study populations as reflected by marked differences in mortality (ranging from 1% to >20%) and/or from the use of different assays, and the use of different NPs (NTproBNP and BNP).

Limitations

Several limitations of our study require comment. First, even after contacting the authors for data on all-cause mortality, 2 cardiac surgery and 2 noncardiac surgery studies had to be excluded from the meta-analysis because not all data necessary to construct a 2 × 2 contingency table could be obtained. In this context, it is worth noting that one of the noncardiac surgery studies that was excluded did not show a significant difference in NP levels between survivors and deceased patients.³⁹ However, this study suffered from a limited sample size (n = 40); as such, we do not expect these data to undermine the significant association between preoperative NP and mortality after noncardiac surgery.

A second limitation of our study is the reduced number of studies available despite using a sensitive search strategy without language restriction. Third, it should be noted that the presented dORs reflect unadjusted associations between preoperative NP and all-cause mortality after surgery. We opted against pooling the adjusted association measures as the different studies adjusted by different variables. However, the results of the multivariable analyses performed within the individual included studies support the independence of the association between preoperative NP and mortality at ≥6 months both in cardiac and noncardiac surgery. Furthermore, for the noncardiac surgery setting, Choi et al.³⁷ presented data on the incremental effect on risk stratification by the addition of NT-proBNP to the Revised Cardiac Risk Index,¹ a widely used risk stratification tool.

Fifth, in the meta-analysis, we pooled studies independent of the type of NP that was measured, i.e. if BNP or NT-proBNP was measured. In addition, individual studies used different assays that target different epitopes with different levels of precision.^58–60 We did not find evidence for between-study variability; however, we cannot exclude the possibility of a different prognostic value of 2 markers or between assays.

Finally, the individual studies used different threshold values. Given that this was not an individual patients' data meta-analysis, it was not possible to calculate a cutoff value across studies and we listed the cutoff values used within the different studies as a source of information on which concentration can be considered nonelevated preoperatively.

CONCLUSIONS

Preoperative NP concentrations were associated with ≤90-day and ≥6-month mortality after both cardiac and noncardiac surgery. In both surgical settings and for both follow-up durations, NP concentrations had high NPVs suggesting that nonelevated preoperative NP concentrations were highly predictive of survival.

DISCLOSURES

Name: Giovanna A. Lurati Buse, MD.

Contribution: This author helped design the study, conduct the study, analyze the data, and write the manuscript.

Attestation: Giovanna A. Lurati Buse, MD, has seen the original study data, reviewed the analysis of the data, approved the final manuscript, and is the author responsible for archiving the study files.

Name: Michael T. Koller, MD, MSc.

Contribution: This author helped design the study, analyze the data, and write the manuscript.

Attestation: Michael T. Koller, MD, MSc, has seen the original study data, reviewed the analysis of the data, and approved the final manuscript.

Name: Christoph Burkhart, MD.

Contribution: This author helped conduct the study and write the manuscript.

Attestation: Christoph Burkhart, MD, has seen the original study data, reviewed the analysis of the data, and approved the final. manuscript.

Name: Manfred D. Seeberger, MD.

Contribution: This author helped design the study and write the manuscript.

Attestation: Manfred D. Seeberger, MD, has seen the original study data, reviewed the analysis of the data, and approved the final manuscript.

Name: Miodrag Filipovic, MD.

Contribution: This author helped design the study, analyze the data, and write the manuscript.

Attestation: Miodrag Filipovic, MD, has seen the original study data, reviewed the analysis of the data, and approved the final manuscript.