Results
Our database search retrieved 185 articles, 172 of which were eliminated for various reasons based on the title and abstract, leaving 13 studies that were scrutinized in a full-text review. Among the 13 studies, one study investigated the prognostic value of presepsin in sepsis, one could not be used to reconstruct the 2 × 2 table, and three were performed using an ineligible design (i.e., they evaluated the diagnostic accuracy of presepsin based on samples that were collected at multiple time points but not every individual sample). In total, eight studies fulfilled our eligibility criteria and were included in the final analysis (Fig. 1). We did not identify any additional relevant articles in the bibliographies of the original articles.
(Enlarge Image)
Figure 1.
Flow chart of study selection
Characteristics of the Included Studies
The included studies were published between 2012 and 2014. Four studies were conducted in Europe, and four were conducted in Asia. A total of 1,815 patients were included, and SIRS criteria were fulfilled in 1,690 patients, including 1,165 patients with septis and 525 patients with non-infectious SIRS. Four studies recruited a group of well-matched (by age and sex) patients without SIRS (60 patients in Behnes, 25 patients in Kweon, 100 patients in Liu, and 70 patients in Vodnik) as controls, and two studies included the control patients as well as the SIRS patients in the non-sepsis group in a 2 × 2 contingency table when analysing the diagnostic accuracy of presepsin for sepsis. The mean age of the patients varied between 54.4 and 79.42 years, and the proportion of male patients ranged from 50.0 to 66.3. The prevalence of sepsis varied from 16.4 % to 79.2 %. The most frequent source of sepsis was pulmonary infection. Six studies were performed in the ED, one in the ICU, and one in the ED and ICU. All included studies recruited a mix of medical and surgical patients. One study excluded patients with comorbidities (e.g., chronic renal failure or a history of resuscitation and trauma) that could influence presepsin levels. Presepsin levels were measured with a chemiluminescent immunoassay on a PATHFAST immunoanalyzer in all studies. The test threshold ranged from 317 to 729 pg/ml. Details of all eight studies are presented in Table 1. The optimal cutoff point was retrospectively determined based on the ROC curve. The mean cutoff for presepsin in the included studies was 560 pg/ml (IQR 317–729).
Study Quality and Publication Bias
Studies were grouped according to Sackett and Haynes' classification for diagnostic studies: two were phase 2 studies (group 1), and six were phase 3 studies (group 2). All studies included a prospective cohort. Two studies were multicentre trials. One study consecutively enrolled patients. Two studies excluded patients with comorbidities that could influence presepsin levels. The QUADAS checklists are presented in Fig. 2. On average the overall QUADAS scores of all studies met 10 of the 14 criteria, which suggests that the studies were of high quality. Deek's funnel plot is presented in Fig. 3. No significant publication bias was observed (p = 0.31).
(Enlarge Image)
Figure 2.
Proportion of Quality Assessment of Diagnostic Accuracy Studies (QUADAS) tool criteria fulfilled by the included studies
(Enlarge Image)
Figure 3.
Deek's funnel plot asymmetry test for publication bias. No publication bias was detected (p = 0.31)
Data Synthesis and Meta-analysis
The pooled SEN and SPE were 0.86 (95 % CI: 0.79, 0.91) and 0.78 (95 % CI: 0.68, 0.85), respectively (Fig. 4). The PLR and NLR were 3.8 (95 % CI: 2.6, 5.7) and 0.18 (95 % CI: 0.11, 0.28), respectively (Fig. 5). The DOR was 22 (95 % CI: 10, 48). The area under the SROC curve was 0.89 (95 % CI: 0.86, 0.92) (Fig. 6). Figure 7 presents Fagan's nomogram for likelihood ratios, and the results indicate that the use of presepsin in the detection of sepsis increased the post-probability to 48 % when the results were positive and reduced the post-probability to 4 % when the results were negative. The mean cutoff for presepsin in the included studies was 560 pg/ml (IQR 317–729).
(Enlarge Image)
Figure 4.
Forrest plot of the sensitivity and specificity of presepsin for the diagnosis of sepsis
(Enlarge Image)
Figure 5.
Positive and negative likelihood ratios. The positive likelihood ratio and negative likelihood ratio were 3.8 (95 % CI: 2.6, 5.7), 0.18 (95 % CI: 0.11, 0.28), respectively LUQ left upper quadrant, RUQ right upper quadrant, LLQ left lower quadrant, RLQ right lower quadrant, LRP likelihood ratio positive, LRN likelihood ratio negative
(Enlarge Image)
Figure 6.
Summary receiver operating characteristic graph of included studies. SEN sensitivity, SPE specificity
(Enlarge Image)
Figure 7.
Fagan's nomogram for calculation of post-test probabilities. Fagan's nomogram for presepsin illustrates post-test probability with fixed pre-test probability of 20 % for sepsis. LR likelihood ratio, pos positive, neg negative
Six studies comprising 838 patients were included in group 2 studies (Table 1). We performed a subgroup analysis restricted to this group because the studies were restricted to patients who were most likely to be encountered by clinicians and who were more informative for routine clinical practice. The pooled SEN and SPE were 0.89 (95 % CI: 0.81, 0.94) and 0.75 (95 % CI: 0.64, 0.84), respectively. The PLR and NLR were 3.6 (95 % CI: 2.2, 5.8) and 0.15 (95 % CI: 0.08, 0.29), respectively. The DOR was 24 (95 % CI: 8, 71). The area under the SROC curve was 0.90 (95 % CI: 0.87, 0.92). The I test results for the pooled SEN and SPE were 68.22 % and 84.90 %, respectively.
There was substantial heterogeneity among the studies. The I test results for the pooled SEN and SPE were 90.49 % and 91.77 %, respectively. The overall I values for the bivariate model were 94 % (95 % CI: 85, 99). The proportion of heterogeneity likely caused by the threshold effect was small (0.07), whereas the variations in SEN and SPE were related to differences in the cutoff points for presepsin that were used in the included studies.
Univariate meta-regression analysis and subgroup analysis were performed to explore the sources of potential heterogeneity in SEN and SPE. Patient blinding, prevalence, setting, consecutive patient recruitment, and sample size were used as covariates. Meta-regression revealed that consecutive recruitment, sample size, and setting significantly accounted for the heterogeneity of sensitivity (Fig. 8). The subgroup analysis restricted to ED patients revealed that the pooled SEN and SPE were 0.85 (95 % CI: 0.77, 0.92) and 0.78 (95 % CI: 0.69, 0.88), respectively.
(Enlarge Image)
Figure 8.
Univariate meta-regression and subgroup analysis