Introduction
Sepsis is a leading cause of death in critically ill patients despite improvements in antimicrobial therapy and supportive care [
1]. The septic response is an extremely complex chain of events involving inflammatory and anti-inflammatory processes, hormonal and cellular reactions, and circulatory abnormalities [
2,
3]. Early identification of patients at high risk of dying after intensive care unit (ICU) admission may help determine therapeutic interventions, such as changes in therapeutic protocols or further diagnostic procedures aiming at preventing shock and multiple organ failure with all their sequels that could have an impact on patients' outcome [
4‐
6]. Therefore, there is a need for a fast simple and cost-effective method to enhance risk stratification in septic patients.
Brian natriuretic peptide (BNP) and its inactive cleavage product N-terminal fragment (NT-proBNP) were secreted into the blood in response to atrial or ventricular wall stretch [
7], or myocardial ischemia [
8] by cardiomyocytes. The half-life of BNP is approximately 20 minutes, and that of NT-proBNP is 1-2 hours [
9]. The term BNPs will be used to represent either BNP or NT-proBNP throughout the rest of the paper unless otherwise stated.
BNPs have been found to be useful markers in the diagnosis, management and prognosis of patients with congestive heart failure [
10]. In addition, BNPs are powerful predictors of death and major adverse cardiovascular events in patients with stable coronary disease [
11], acute coronary syndromes [
12] and pulmonary embolism, [
13] and those who undergo noncardiac surgery [
14]. Several prospective studies have been performed to investigate the potential role of BNPs in predicting mortality in septic patients, but they had limited numbers of patients, used different cut-off points and involved different clinical endpoints. In the present study, we made a systematic review and meta-analysis to evaluate the correlation between elevated levels of BNPs and death in septic patients.
Materials and methods
This systematic review and meta-analysis was performed according to the guidelines of Meta-analysis of Observational Studies in Epidemiology [
15].
Study outcome
The aim of this meta-analysis was to see whether elevated BNPs could predict all-cause mortality in adult patients with sepsis.
Search strategy and eligibility assessment
Pubmed, Embase and the Cochrane Central Register of Controlled Trials (up to February 18, 2011) were searched by using Exploded Medical Subject Headings and the appropriate corresponding keywords, "Brain natriuretic peptide", "B-type natriuretic peptide", "BNP", "pro-brain natriuretic peptide", "amino terminal pro-brain natriuretic peptide", "amino terminal pro-BNP", "N-terminal pro B-type natriuretic peptide", "NT-proBNP", "natriuretic peptide" AND "sepsis", "septicemia", "septicaemia", "septic". No language restrictions were applied. Additionally, the reference lists of the original studies and previous review articles were hand-searched to identify other potentially eligible studies.
Two authors independently determined the eligibility of all studies identified in initial research. Studies were included if they prospectively collected data on all-cause mortality in adult septic patients with either plasma BNP or NT-proBNP measurement. Studies where a 2 × 2 table of results could not be constructed were excluded. In case of disagreements, a third author was consulted. Agreement regarding study inclusion was assessed using the Cohen К statistic [
16].
Data extraction and quality assessment
Two authors independently extracted the following descriptive data from all eligible studies: study design, sample size, patient population, inclusion/exclusion criteria, diagnostic criteria for sepsis, severe sepsis and septic shock, follow-up period, completeness of follow-up, outcome assessment, marker evaluated (i.e., BNP or NT-proBNP), assay manufacturer, the optimal cut-off point, timing of BNP measurement, and the proportion of patients with an elevated BNP measurement. If data needed clarification or were not presented in the publication, the original authors were contacted by E-mail. Extracted data were entered into Microsoft Office Excel 2007 and checked by the third author. Any disagreement was solved by discussion.
Given that the eligible studies were of a prognostic nature, methodological and reporting quality was assessed according to the Quality Assessment of Diagnostic Accuracy Studies checklist (see Additional file
1) [
17].
Statistical analysis
Results were analyzed using Review Manager 5.1 (Cochrane Collaboration, Oxford, UK) and Meta-Disc 1.4 [
18] (Clinical Biostatistics, Ramon y Cajal Hospital, Madrid, Spain). As individual studies used different cut-off points for defining elevated BNPs, Spearman's correlation coefficient between sensitivity and specificity was calculated to detect the presence of a threshold effect, where variations in sensitivity and specificity were related to differences in the cut-off point used to define an elevated BNPs level [
19]. If there was no evidence of a threshold effect, then summary estimates, including odds ratio (OR), sensitivity, specificity, positive likelihood ratio, and negative likelihood ratio, were calculated using the random-effects model based on DerSimonian and Lair's meta-analytic statistical method [
20]. In case a study provided multiple cut-off points for BNPs analysis, the point giving the maximum overall accuracy was chosen. In the case of analyses with empty cells, 0.5 was added to all cells to avoid computational errors. Publication bias was assessed by visually inspecting funnel plot. A p value of less than 0.05 was considered statistically significant.
Cochrane's Chi
2 test and I
2 test for heterogeneity were used to assess inter-study heterogeneity [
21]. The Chi
2 test assesses whether observed differences in results are compatible with chance alone and the I
2 describes the percentage of the variability in effect estimates that is due to heterogeneity rather than sampling error. Statistically significant heterogeneity was considered present at Chi
2 P < 0.10 and I
2 > 50%. To explore the heterogeneity observed, sensitivity analyses were performed by sequential exclusion of each study. A prior subgroup analyses to explain significant heterogeneity were: 1) BNPs type (BNP versus NT-proBNP); and 2) blinding of BNPs measurement to the outcome (yes versus no). Single covariate random-effects meta-regression (inverse variance weights) was also used to explore sources of variation. Patient population (sepsis versus severe sepsis or septic shock), the underlying diseases (inclusion versus exclusion of preexisting conditions known to increase BNPs levels), optimal timing of BNPs measurement, and mortality were considered as variables.
Discussion
The present meta-analysis showed that elevated BNPs were associated with a significantly increased risk of mortality in patients with sepsis. The finding is consistent for both BNP and NT-proBNP. As such, measurement of BNPs may be a simple method of risk stratification in septic patients.
BNPs elevation in patients with sepsis can be considerably high, even though cardiac depression is not obvious. A retrospective study [
34] suggested that BNPs level in septic patients with preserved systolic left ventricular function could be as high as that in patients admitted to the hospital with congestive heart failure because of severely impaired systolic left ventricular function. Aside from neurohormonal activation, several mechanisms are likely to account for increased BNPs levels in sepsis, including sepsis-induced biventricular dilatation [
35], the stimulation of lipopolysaccharide [
36] or proinflammatory cytokines [
37,
38], volume resuscitation [
39] and sepsis-associated acute lung injury or acute respiratory distress syndrome [
40]. Accordingly, elevated BNPs level in the presence of sepsis does not essentially mean cardiac dysfunction due to low specificity, while normal BNPs level could be used to rule out the need for further cardiac investigation, unless there are other clinical grounds that strongly suspect a significant cardiac disorder.
Clinical severity scores such as acute physiology and chronic health evaluation (APACHE) II and sequential organ failure assessment (SOFA) scores have been validated for mortality risk stratification, but are unwieldy and tend to be used more for audit and research than clinical decision making in sepsis. A rapidly available biochemical test that provides similar or better prognostic information could therefore be useful, e.g. to help discussions about prognosis with patients' relatives and decisions regarding earlier interventions. Several studies [
25,
41,
42] showed that BNPs level was related to APACHE II and SOFA scores. In the present meta-analysis, two [
25,
27] of the included studies directly compared the prognostic value of BNPs measurement
vs clinical severity scores and both suggested a better prognostic value of elevated BNPs in predicting mortality. Chen et al [
25] reported a greater area under the ROC curve for the plasma BNP level than for APACHE II score (0.737
vs 0.664). Post et al [
27] reported that regarding predicting 30-day mortality, the area under the ROC curve was greater for BNP than that for APACHE II score and SOFA score (0.648
vs 0.494 and 0.493, respectively). Considering the low sensitivity and high specificity of clinical severity scores, they can be combined with BNPs for aggregate analysis on the septic prognosis. What additional value of BNP brings to severity scores on sepsis risk stratification remains to be determined in further studies.
As a risk stratification tool, BNPs measurement is simple, inexpensive, reproducible, non-invasive, and widely available. These attributes make it highly suitable for serial testing to monitoring outcome risk over time, which may be superior to a single value of BNPs assessed on admission, because dynamic changes in BNPs concentration may reflect the development of sepsis. Furthermore, rather than necessarily dichotomizing risk as high or low, BNPs express risk as a continuum and this may be advantageous.
To date, several biomarkers have been identified to have some prognostic value in the field of sepsis. According to a prospective multicenter observational study [
43] procalcitonin > 0.85 ng/ml was associated with an increased risk of death in a Cox regression analysis (hazard ratio 2.31, 95% CI 1.32-4.05, P = 0.003). In addition, both elevated cardiac troponin-T [
44] and -I [
45] were reported to be associated with higher mortality in septic patients. In the present study, pooled sensitivity and specificity of elevated BNPs for predicting mortality in septic patients were 79% and 60%, respectively. So far, none of the proposed prognostic markers had sufficient (more than 90%) sensitivity and specificity to predict which patients were at greater risk of dying due to sepsis [
46]. Accordingly, a BNPs measurement may provide a better prognostic value in combination with other biomarkers, each mirroring different pathophysiological aspects. Further study was warranted to verify this hypothesis and evaluate the cost-effectiveness in sepsis.
In the past decade, several fully automated, rapid assays for determination of BNPs have become commercially available, including both high-throughput automated platforms and point of care tests. Although these existing BNPs assays correlate closely, BNPs assays are not currently analytical equivalent due to the lack of assay standardization [
47]. Recently, a multicenter study conducted in 90 Italian laboratories demonstrated that there were significant differences in analytical characteristics and measured values among the most popular commercial methods for BNP and NT-proBNP [
48]. Therefore, clinicians should be very careful when comparing results obtained by laboratories that use different methods.
An important concern of using BNPs in septic patients is the impact of renal insufficiency on these tests. Acute renal insufficiency occurs in 11 - 16% of critically ill patients who presented with sepsis [
49,
50]. A study by Goei et al [
51] suggested that NT-proBNP had more favorable discriminative value in patients with a glomerular filtration rate more than 90 mL/min/1.73 m
2, while it lost its prognostic value in patients with a glomerular filtration rate less than 30 mL/min/1.73 m
2. No guidance on how BNPs values are adjusted for renal dysfunction is available today. Given the high prevalence and the impact of renal impairment on the BNPs values, it is preferable to adopt different cut-offs stratified by renal function in patients with sepsis.
There were several limitations in the present study. First, marked heterogeneity existed across the included studies in terms of population characteristics, BNPs assays, optimal cut-off point, follow-up period and definitions of endpoints. However, all the individual ORs favored the prognostic value of elevated BNPs in predicting mortality, indicating that the heterogeneity was entirely quantitative. Although there was uncertainty regarding the strength of the association, current evidence suggested that there was a significant correlation between an elevated BNPs level and an increased risk of mortality in septic patients. Second, the optimal timing of BNPs measurement varied across the studies, including the day on admission, and day 2 and day 5 after admission. It can be partly accounted for by the difficulty in determining the time of onset of sepsis and, hence, the time of patient recruitment. Third, it should be noted that the pooled estimates reflected unadjusted associations between BNPs and all-cause mortality. Five [
22,
25,
27,
29,
31] of the included studies provided adjusted ORs accounting for confounders. Pooled analysis of the adjusted ORs also suggested that elevated BNPs were correlated with a significantly increased risk of mortality in septic patients (random-effects model, OR 1.87, 95% CI 1.30 to 2.69,
P = 0.001). Because the ORs in the five studies were adjusted by different confounders with different regression models and the data reported on adjusted ORs were limited, we thus adopted unadjusted mortality data for pooled analyses in the present study. Finally, we could not determine the ideal cutoff points for BNP and NT-proBNP tests because we did not have the raw data to map out ROC curves. To determine whether there is a single threshold or a few important BNP or NT-proBNP thresholds (e.g., age dependent), further evaluation in prespecified groups of larger numbers of patients is needed.
Competing interests
The authors declare that they have no competing interests.
Authors' contributions
Study concept and design: FW, YW, LT, JL and XD. Acquisition of data: FW, YW and LT. Analysis and interpretation of data: WZ, FC, LB and TX. Drafting of the manuscript: FW, YW and LT. Critical revision of the manuscript for important intellectual content: JL and XD. Statistical analysis: FW and YW. Administrative, technical, and material support: WZ, FC, LB and TX. Study supervision: JL and XD. All authors have read and approved the manuscript for publication.