Introduction
Sepsis is a life-threatening complication of infection and the most common cause of death in intensive care units (ICU) [
1]. Delay in diagnosis and treatment often results in rapid progression to circulatory collapse, multiple organ failure and eventual death [
2]. Therefore, accurate and timely diagnosis of sepsis will limit morbidity, reduce costs and improve patients' outcome [
3‐
5].
Diagnosis of sepsis is based on systemic inflammatory response syndrome (SIRS) in the presence of a known infection. SIRS is very common in critically ill patients, being found in various conditions, including trauma, surgery and pancreatitis [
6,
7]. Microbiological culture as a gold standard is used to distinguish sepsis from non-infectious conditions. However, this method lacks sensitivity, and there is often a substantial time delay. Thus, there is an urgent need for a fast, simple and accurate method to enhance sepsis diagnosis.
The triggering receptor expressed on myeloid cells-1 (TREM-1) was a recently discovered member of the immunoglobulin superfamily, expression of which on phagocytes was up-regulated by exposure to bacteria and fungi [
8]. TREM-1 mediated the acute inflammatory response to microbial products. A soluble form of TREM-1 (sTREM-1) is released from the activated phagocytes and can be found in body fluids, such as plasma [
9], pleural fluid [
10], bronchoalveolar lavage fluid [
11], urine [
12] and cerebrospinal fluid [
13]. Thus, sTREM-1 may act as a potential biomarker of bacterial infection [
14,
15]. Recently, several studies have been performed to investigate the role of plasma sTREM-1 in differentiating sepsis from non-infectious SIRS in different settings [
9,
16‐
25]. Due to the limited patient sample size recruited in the individual studies, we aimed to conduct a systematic review and meta-analysis to assess the role of plasma sTREM-1 for sepsis diagnosis in adult patients with SIRS.
Materials and methods
This systematic review and meta-analysis was performed according to the guidelines of Meta-analysis of Observational Studies in Epidemiology [
26].
Search strategy
PubMed, Embase and Cochrane Controlled Clinical Trials Register Database (up to 20 June 2012) were searched by using Exploded Medical Subject Headings and the appropriate corresponding keywords, ''triggering receptor expressed on myeloid cells-1'', ''soluble triggering receptor expressed on myeloid cells-1'', ''TREM-1'', ''sTREM-1''. We updated the literature search of the above electronic databases on 15 October 2012 to find as many eligible studies as possible. No language restriction was used. Further searches were performed by checking the reference lists from primary and review articles, and manually reviewing abstract booklets and conference proceedings. The authors were contacted for study details if needed.
Eligibility criteria and study selection
Studies were included if they assessed the accuracy of plasma sTREM-1 for sepsis diagnosis in adult patients with SIRS and provided sufficient information to construct a 2 X 2 contingency table. Two reviewers independently judged study eligibility when screening the citations. Disagreements were resolved by consensus. Agreement regarding study inclusion was assessed using the Cohen Κ statistic [
27].
Two reviewers independently abstracted data in each study to obtain information on the year of publication, country of origin, clinical setting, sample size, patients' demographics, sTREM-1 test methods, diagnostic cut-off points, sensitivity, specificity and methodological quality. Each reviewer extracted the data to construct a 2 X 2 contingency table.
Definitions
Sepsis was defined according to the criteria proposed by the American College of Chest Physicians/Society of Critical Care Medicine as the presence of an infection complicated by SIRS [
28]. Patients included in the septic group had either microbiologically (culture-proven) or clinically diagnosed sepsis, whereas the other patients were included in the non-infectious SIRS group.
Quality assessment
The methodological quality of each study was graded independently by two reviewers with the Quality Assessment of Diagnostic Accuracy Studies (QUADAS) tool, a validated tool for the quality assessment of diagnostic accuracy studies [
29]. Furthermore, studies were grouped according to Sackett and Haynes' [
30] classification of diagnostic studies. In this classification, phase 1 studies are those that compare the difference in test results between patients with the target disorder and healthy individuals. Phase 2 studies are those that examine how the index test discriminates between patients with and without the target disorder. Phase 3 studies are those that assess the test's real-life performance in patients suspected of having the disorder.
Statistical analysis
The diagnostic meta-analysis was performed using a bivariate meta-analysis model [
31] to calculate the pooled sensitivity, specificity, positive/negative likelihood ratios, and diagnostic odds ratio (DOR). The summary receiver operator characteristic (SROC) curve that plotted sensitivity versus specificity was constructed to plot the individual and summary points of sensitivity and specificity [
32]. Furthermore, around the pooled estimate, we also plotted a 95% confidence region and a 95% prediction region to illustrate the precision with which the pooled value was estimated (confidence ellipse of a mean) and to show the amount of between study variation (prediction ellipse; the likely range of values for a new study). The presence of statistical between-study heterogeneity was assessed by the I
2 test [
33]. Values of 25, 50 and 75% for the I
2 test were regarded as indicative of low, moderate and high statistical heterogeneity, respectively. Meta-regression analysis using a bivariate model was performed in order to find the effect of potentially confounding covariates. Each covariate had a fixed effect when added to the bivariate model and associated with logit(sensitivity) and/or logit(specificity) [
34]. Publication bias through small study effects was assessed with a regression test on the diagnostic odds ratio [
35,
36]. A
P- value <0.05 was considered as indicative of statistical significance. Stata intercooled version 10.1 (StataCorp, College Station, TX, USA) was used for all statistical analyses.
Discussion
Early identification of infection is of vital importance to the clinical course and outcome of septic patients. An ideal marker of sepsis should be present early in the course of the disease, measurable rapidly and easily, of prognostic significance, sensitive enough to detect infection in patients with minimal host response, and specific enough to discriminate infection from other non-infectious SIRS [
43].
Biological markers such as procalcitonin and CRP have been used in the diagnosis of bacterial infections [
44,
45]. However, since they are "indirect" markers of infection, their sensitivity and specificity are not 100% and vary in different patient groups and indications. TREM-1 was a recently identified molecule involved in inflammatory response. Human tissues infected with bacteria were infiltrated with neutrophils and monocytes that expressed high levels of TREM-1 [
8,
46]. TREM-1 amplifies infection-induced inflammatory response signals primarily through the mediation of adapter protein DAP12 on the cell surface [
47]. sTREM-1, as the soluble form of TREM-1, released by activated phagocytes, may be a more "direct" marker of infection. In the present study, pooled analysis showed that plasma sTREM-1 seemed to have a moderate (0.7 ≤AUC <0.9) diagnostic accuracy for sepsis since the area under the SROC curve was 0.87. The pooled sensitivity of plasma sTREM-1 for the diagnosis of sepsis was 79% and the specificity was 80%. With a hypothetical pretest probability of 62% and a PLR of 4.0, detecting plasma sTREM-1 for sepsis diagnosis would raise the post-test probability to 87%. With a NLR of 0.26, detecting plasma sTREM-1 reduced the post-test probability to 30% (Figure
5), showing that application of plasma sTREM-1 test to sepsis diagnosis had a moderate value. Recently, several studies reported that plasma sTREM-1 level could be elevated in non-infectious disease, such as acute pancreatitis, and non-infectious inflammation following traumatic lung contusion [
48‐
50]. We inferred that in the non-infectious SIRS patients, sTREM-1 level was elevated, which might partly account for its moderate but not high accuracy in distinguishing septic patients from non-infectious SIRS patients. To date, none of the proposed biomarkers as a single test had sufficient (more than 90%) sensitivity and specificity to discriminate sepsis from SIRS in critically ill adult patients. A combination of several markers appears to be a useful approach to improving accuracy in diagnosing sepsis, which was proved by a recent paper from Gibot
et al. [
25]. In their study, although sTREM-1, PCT and polymorphonuclear CD64 index were all found to be independent predictors of sepsis, a combination of them was shown to have a far better diagnostic performance for sepsis with the area under the ROC curve to be 97% (95% CI, 95 to 99).
As a single indicator of diagnostic test performance, DOR is independent of disease prevalence. The DOR of the included studies ranged from 2.17 to 623.33, and the pooled DOR was 16. The disparity noticed in the included studies may result from several reasons. First, patient sample size varied in the included studies. It was suggested that small studies tend to overestimate the effect size [
51] and studies based on small sample sizes may have allowed for a type II error. In the present study, meta-regression analysis showed that patient sample size significantly accounted for the heterogeneity. The pooled DOR of the five studies [
9,
16,
20,
21,
24] with small sample size was 90 (95% CI, 20 to 399), higher than that of the other six studies [
17‐
19,
22,
23,
25] with large sample size (pooled DOR 5, 95% CI, 3 to 9). Therefore, caution should be taken when interpreting the results. Another possible reason was different disease spectrum included. The pooled DOR of the three studies [
16,
20,
21] which enrolled trauma or injured patients was 104 (95% CI, 9 to 1,206), higher than that of the remaining studies which enrolled other kinds of patient spectrum (pooled DOR 8, 95% CI, 3 to 19).
Two prior meta-analyses concluded that sTREM-1 represented a useful biological marker of bacterial infection [
14] or bacterial pleural effusions [
15]. The meta-analysis by Jing
et al. [
14] included studies across a wide range of disease spectrum, and assessed sTREM-1 level from different sample origins, including non-directed bronchial lavage fluids, pleural fluid, plasma, bronchoalveolar lavage fluid, cerebrospinal fluid and urine. The other meta-analysis by Summah
et al. [
15] assessed the diagnostic accuracy of sTREM-1 in the pleural fluid for bacterial pleural effusions. Being different from them, the present meta-analysis assessed the diagnostic accuracy of sTREM-1 for sepsis and focused only on the plasma level of sTREM-1. Although detecting the samples from suspected sites of infection might result in higher diagnostic accuracy than detecting the plasma sample, it is faster, easier and simpler to collect plasma samples.
The present meta-analysis had several limitations. First, although extensive literature search was conducted, the number of included studies was small. Second, in the patients who were clinically diagnosed as having sepsis without microbiological evidence, some degree of misclassification bias may have existed. Third, several studies reported sTREM-1 could reflect the severity of sepsis and predict prognosis, we didn't address this issue. Finally, we could not determine the ideal cut-off point for plasma sTREM-1 test because we did not have the raw data to map out the ROC curve. To determine whether there is a single threshold or a few important thresholds, further studies with a larger number of patients are needed.
Competing interests
The authors declared that they have no competing interests.
Authors' contributions
YW, FW, XF, JL and XD were responsible for study concept and design. YW, FW and XF acquired the data. RB and LB analyzed and interpreted the data. YW, FW and XF drafted the manuscript. JL and XD critically revised the manuscript for important intellectual content. YW and FW performed statistical analysis. RB and LB were responsible for administrative, technical and material support. JL and XD supervised the study. All authors have read and approved the manuscript for publication.