Literature Search
In all, 10,026 articles in Chinese and English were retrieved through the preliminary screening of the databases. After reading the titles and abstracts, 300 articles were selected. Intensive reading was performed following strictly the inclusion and exclusion criteria. After the screening, a set of 54 articles were included in the study (Figure 1).
Study Characteristics
In all, 9,842 participants were finally enrolled in this meta-analysis, with a sepsis prevalence of 54.8 %. It included 20 studies related to neutrophil CD64, 39 studies related to PCT, and 15 studies related to IL-6. We found 37 articles that reported the average age of the study subjects, ranging from 42.0 to 92.6 years. Four papers focused on patients with specific sepsis, such as patients with acute abdominal sepsis [37], ventilator-associated pneumonia [47], and postoperative sepsis [45, 62]. Two articles addressed elderly patients with sepsis (aged > 65 [55] and > 85 years [60]), whereas another study [52] included patients aged ≤ 65 and > 65 years. Two studies [18, 33] reported separately cases of positive and negative blood cultures. One paper [23] included a study conducted in the medical ICU and surgical ICU patients. The detailed baseline characteristics of the included studies are summarized in Table 1.
Quality Assessment
We used the QUADAS-2 scale to evaluate the quality of the included articles. The results showed that all studies were of high quality and had clinical practicability (Figure 2).
Heterogeneity test
Spearman correlation coefficients of neutrophil CD64, PCT and IL-6 were -0.22 (P = 0.35), -0.054 (P = 0.729,) and 0.326 (P = 0.217), respectively. The SROC curve of the three biomarkers did not show a significant shoulder-arm effect, suggesting that there was no threshold effect (Figure 3).
Pooled effect size result
Of all included articles, 20 of them reported the diagnostic value of neutrophil CD64 for sepsis. The results for these studies were: pooled sensitivity, 0.88 (95 % CI, 0.81–0.92); pooled specificity, 0.88 (95 % CI, 0.83–0.91) (Figure 4); pooled PLR, 7.2 (95 % CI, 5.0–10.3); pooled NLR, 0.14 (95 % CI, 0.09–0.22); pooled DOR, 51 (95 % CI, 25–105); and the AUC was 0.94 (95 % CI, 0.91–0.96) (Figure 3a). Thirty-nine studies reported the diagnostic value of PCT with the following results: pooled sensitivity, 0.82 (95 % CI, 0.78–0.85); pooled specificity, 0.78 (95 % CI, 0.74–0.82) (Figure 5); pooled PLR, 3.7(95 % CI, 3.1–4.50); pooled NLR, 0.23 (95 % CI, 0.19–0.29); pooled DOR, 16 (95 % CI, 11–23); and the AUC was 0.87 (95 % CI, 0.83–0.89) (Figure 3b). We found 15 articles reporting the diagnostic value of IL-6 for sepsis. The results for this set of articles were: pooled sensitivity, 0.72 (95 % CI, 0.65–0.78); pooled specificity, 0.70 (95 % CI, 0.62–0.76) (Figure 6); pooled PLR, 2.4 (95 % CI, 1.9–3.0); pooled NLR, 0.40 (95 % CI, 0.32–0.51); pooled DOR, 6 (95 % CI, 4.0–9.0); and the AUC was 0.77 (95 % CI, 0.73–0.80) (Figure 3c).
Publication bias analysis
Publication bias of studies regarding neutrophil CD64 showed that 20 articles were not evenly distributed on both sides of the regression line (t = 2.45, P = 0.025) (Figure 7a), suggesting a publication bias among the included studies. No significant bias was found for studies addressing PCT (t = 1.17, P = 0.249) (Figure 7b) or IL-6(t = 0.53, P = 0.607) (Figure 7c).
Heterogeneity analysis
Meta-regression
Due to the heterogeneity caused by a non-threshold effect in the included studies, meta-regression was performed when the following criteria were met: a sample size of the study over 100; the patients were Chinese; the average age of patients was over 65 years old; the clinical setting was classified into ICU; and test methods. The meta-regression of neutrophil CD64 showed that the sample size had an influence on the heterogeneity of sensitivity and specificity, and regional difference was one of the factors that caused the heterogeneity of specificity (Figure 8a). The meta-regression of PCT showed that the above five factors are likely to be the sources of heterogeneity (Figure 8b). The meta-regression result of IL-6 indicated that the source of heterogeneity might be the sample size (Figure 8c).
Sensitivity analysis
Concerning the sensitivity analysis of neutrophil CD64, we found that when the article by Gámez-Díaz study was removed from the subset of studies, the overall heterogeneity of specificity of the 19 articles left decreased, suggesting that the Gámez-Díaz study was the cause for the heterogeneity of specificity (Figure 9a). When the other 19 studies were removed one by one, the sensitivity, specificity, and AUC showed no significant change. The sensitivity analysis of PCT and IL-6 showed that the sensitivity, specificity, and AUC did not change significantly when they were removed one by one (Figure 9b, 9c).
Subgroup analysis
Through a sensitivity analysis of neutrophil CD64, it was found that the Gámez-Díaz study had an influence on the heterogeneity of neutrophil CD64, so a subgroup analysis was conducted after excluding such study. The subgroup analysis of three biomarkers (Tables 2, 3, 4) indicated that the sample size might be the source of heterogeneity, since the heterogeneity decreased significantly in the group when a small sample size was analyzed, which might be due to the large number of included cases, and a lack of consistency. The subgroup analysis of neutrophil CD64 indicated that regional differences were also a source of heterogeneity, which was consistent with the meta-regression results. Heterogeneity decreased significantly in the Chinese group but remained high in the non-Chinese group. The subgroup analysis showed that the sensitivity, specificity, and AUC of PCT in ICU patients were 0.82 (95 % CI, 0.77–0.86), 0.78 (95 % CI, 0.72–0.82), 0.86 (95 % CI, 0.83–0.89), respectively; and the SEN, SPE, and AUC of PCT in non-ICU patients were 0.77 (95 % CI, 0.72–0.82), 0.74 (95 % CI, 0.64–0.81), and 0.82 (95 % CI, 0.78–0.85), respectively; suggesting that the diagnostic value of PCT in the ICU group was higher than that in the non-ICU group.
Clinical utility evaluation
We assumed a pre-test probability of 50%. The Fagan's nomogram of neutrophil CD64 (Figure 10a) showed a post-test probability of 88 % positive and 12 % negative. The Fagan's nomogram of PCT (Figure 10b) showed a post-test probability of 79 % positive and of 19 % negative, whereas the Fagan's nomogram of IL-6 (Figure 10c) showed a post-test probability of 70 % positive and of 29 % negative.