3.1 Search results and study selection
The literature screening process is shown in Fig. 1. The search of six databases resulted in 44 studies, including six in Pubmed, 12 in Web of Science, 11 in Embase, 13 in Scopus, two in Ovid, and none in Cochrane library. After eliminating the duplicate studies, 20 studies remained. In addition, 12 studies were excluded by title and abstract screening, and one more was excluded by full-text reading. Finally, seven studies were included in this study.
3.2 Study characteristics
The characteristics of the included studies are shown in Table 1. Among seven studies, three were from China, and the other four were from the Netherlands, Spain, the USA, and the Czech. In this meta-analysis, four studies were prospective, and three were retrospective. A total of 500 individuals were included, 201 individuals in the PJI group and 299 individuals in the aseptic failure group. The age of patients in the PJI group ranged from 65 to 75 years, and in the aseptic failure group ranged from 63 to 73 years. The included patients’ knee-to-hip ratio ranged from 1.08 to 13.5 in the PJI group and from 1.10 to 9 in the aseptic failure group. The diagnostic criteria were MSIS 2011 or MSIS 2013. The true positive(TP), false positive(FP), false negative(FN), true negative(TN), cutoff values, sensitivity, specificity, and AUC extracted from the seven studies are shown in Table 2.
Table 1
Characteristics of studies included in the meta-analysis
Study (year)
|
Country
|
Study design
|
Participants
|
Age
|
Involved joint (knee/hip)
|
Gold standard
|
Testing methods
|
PJI
|
Non-PJI
|
PJI
|
Non-PJI
|
PJI
|
Non-PJI
|
Zida Huang 2022
|
China
|
prospective
|
50
|
28
|
65
|
65
|
26/24(1.08)
|
15/13(1.15)
|
MSIS(2013)
|
ELISA
|
C.Dijkman 2019
|
Netherlands
|
retrospective
|
13
|
73
|
-
|
-
|
-
|
-
|
MSIS(2013)
|
Particle-enhanced turbidimetric immunoassay
|
Hanjiang Liu 2022
|
China
|
prospective
|
38
|
20
|
66.4
|
67.3
|
33/5(6.6)
|
18/2(9)
|
MSIS(2011)
|
ELISA
|
R.Li 2021
|
China
|
retrospective
|
25
|
25
|
68
|
63
|
16/9(1.78)
|
17/8(2.13)
|
MSIS(2013)
|
ELISA
|
Vergara,A 2018
|
Spain
|
retrospective
|
22
|
22
|
75
|
73
|
-
|
-
|
MSIS(2013)
|
Modified enzyme immunoassay coupled with chemiluminescence
|
Deirmengian, C 2014
|
USA
|
prospective
|
29
|
66
|
66
|
67
|
27/2(13.5)
|
57/9(6.33)
|
MSIS(2011)
|
Immunoassay
|
Svoboda,M 2022
|
Czech
|
prospective
|
24
|
65
|
-
|
-
|
22/2(11)
|
34/31(1.10)
|
MSIS(2013)
|
-
|
PJI: periprosthetic joint infection; Non-PJI: none periprosthetic joint infection;
|
MSIS: Musculoskeletal Infection Society; ELISA: enzyme-linked immunosorbent assay.
TP: true positive; FP: false positive; FN: false negative; TN: true negative; AUC: the area under the curve.
Table 2
Data extracted for the construction of 2×2 table
Study (year)
|
TP
|
FP
|
FN
|
TN
|
Total
|
Cutoff
|
Sensitivity
|
Specificity
|
AUC
|
Zida Huang 2022
|
46
|
1
|
4
|
27
|
78
|
263.5ng/ml
|
0.929
|
0.98
|
0.98
|
C.Dijkman 2019
|
12
|
12
|
1
|
61
|
86
|
735.5ng/ml
|
0.92
|
0.83
|
0.932
|
Hanjiang Liu 2022
|
31
|
3
|
7
|
17
|
58
|
105.1ng/ml
|
0.816
|
0.85
|
0.919
|
R.Li 2021
|
25
|
0
|
0
|
25
|
50
|
762.821ng/ml
|
1
|
1
|
1
|
Vergara,A 2018
|
19
|
5
|
3
|
17
|
44
|
152ng/ml
|
0.863
|
0.772
|
0.92
|
Deirmengian, C 2014
|
29
|
0
|
0
|
66
|
95
|
2200ng/ml
|
1
|
1
|
1
|
Svoboda,M 2022
|
24
|
0
|
0
|
65
|
89
|
998ng/ml
|
1
|
1
|
1
|
TP: true positive; FP: false positive; FN: false negative; TN: true negative; AUC: the area under the curve.
3.3 Results of the quality assessment
As shown in Fig. 2, the quality assessment of the seven studies was achieved by QUADAS-2. All seven studies had a bias risk of the index test, mainly because none of the diagnostic thresholds of the markers were pre-determined before the test. In terms of case selection, only the studies of Svoboda, M 2022 and Vergara, A 2018 had case selection risk due to inappropriate case selection. The studies performed well in the clinical applicability concerns.
3.4 Diagnostic value
The diagnostic value of NGLA is shown in Table 3 and Fig. 3. The pooled sensitivity, specificity, PLR, NLR, DOR, and AUSROC of NGLA were 0.93(95%CI, 0.88–0.96), 0.93(95%CI, 0.89–0.96), 13.65(95%CI, 4.64–40.18), 0.09(95%CI, 0.04–0.21), 202.89(95%CI, 39.00-1055.53), 0.973, respectively. The diagnostic value of NGLA was extremely high, according to AUC. The heterogeneity(I2) of sensitivity, specificity, PLR, NLR, and DOR was higher than 50%; therefore, there was heterogeneity in this meta-analysis.
Table 3
Summary results of NGLA diagnostic characteristics
|
Sensitivity
|
Specificity
|
PLR
|
NLR
|
DOR
|
AUSROC
|
Pooled value (95%CI)
|
0.93 (0.88–0.96)
|
0.93 (0.89–0.96)
|
13.65 (4.64–40.18)
|
0.09 (0.04–0.21)
|
202.89 (39.00−1055.53)
|
0.973
|
I2(%)
|
66.6
|
84.1
|
78.8
|
59.8
|
69.7
|
-
|
3.5 Threshold effect of NGLA
The threshold effect was implemented by MetaDisc 1.4 software, and Spearman’s correlation analysis showed a correlation coefficient of−0.786 between the log of sensitivity and the log of (1-specificity), with p(0.036) less than 0.05, indicating the existence of a threshold effect.
3.6 Sensitivity analysis
The sensitivity analysis of NGLA was done through Stata 16 software and was shown in Fig. 4. The bivariate boxplot suggested that no studies were outside the circle, indicating that none of these seven studies had significant sensitivity. Therefore, the combined diagnostic results of this meta-analysis were stable.
3.7 Assessment of publication biases
The publication bias was analyzed through Deeks’ Funnel Plot and shown in Fig. 5. The p value(0.03) was less than 0.05, suggesting a publication bias existed in this study.
3.8 Meta-regression and subgroup analysis
This study had significant heterogeneity, as seen in the forest plot(Fig. 3). Therefore, we did subgroup and meta-regression analyses. The seven studies were divided into prospective and retrospective subgroups by study design, into greater than 60 samples and less than 60 samples subgroups by sample size, into greater than 300 ng/ml and less than 300 ng/ml subgroups by cutoff value, into serum and synovial fluid subgroups by sample source, and into Chinese and other countries subgroups by country. The pooled sensitivity, specificity, PLR, NLR, and DOR of each subgroup are shown in Table 4. Study design, cutoff value, sample size, sample source, and country were included in the meta-regression analysis. According to the regression results, subgroup types with P values greater than 0.05 were excluded one by one. As a result, country, study design, sample size, and sample source were excluded in descending order according to P values. Finally, only the cutoff value remained, which had a statistically significant P of 0.042, indicating that the source of heterogeneity in this study was mainly from the threshold effect. The pooled sensitivity, specificity, PLR, NLR, and DOR of the four studies with cutoff values greater than 300 ng/ml subgroup were 0.99(95%CI, 0.94-1.00), 0.95(95%CI, 0.91–0.97), 43.01(95%CI, 1.39-1335.33), 0.04(95%CI, 0.01–0.12), and 1181.88(95%CI, 74.31-18798.23), respectively, while the three studies with cutoff values less than 300 ng/ml subgroup were 0.87(95%CI, 0.80–0.93), 0.87(95%CI, 0.77–0.94), 6.20(95%CI, 2.28–16.84), 0.15(95%CI, 0.08–0.28), and 44.82(95%CI, 10.48-191.68), respectively.
Table 4
Subgroup analysis of NGLA for PJI diagnosis
Subgroup
|
Number of studies
|
Sensitivity (95%CI)
|
Specificity (95%CI)
|
PLR (95%CI)
|
NLR (95%CI)
|
DOR (95%CI)
|
Overall studies
|
7
|
0.93(0.88–0.96)
|
0.93(0.89–0.96)
|
13.65(4.64–40.18)
|
0.09(0.04–0.21)
|
202.89(39.00−1055.53)
|
Study design
|
|
|
|
|
|
|
Prospective
|
4
|
0.92(0.86–0.96)
|
0.98(0.94–0.99)
|
29.60(4.74−184.85)
|
0.07(0.02–0.30)
|
504.72(31.02−8212.01)
|
Retrospective
|
3
|
0.93(0.84–0.98)
|
0.86(0.78–0.92)
|
5.83(2.39–14.23)
|
0.10(0.03–0.34)
|
80.71(9.27−702.79)
|
Sample size
|
|
|
|
|
|
|
༞60
|
4
|
0.96(0.90–0.99)
|
0.94(0.91–0.97)
|
34.88(2.01−605.28)
|
0.07(0.03–0.14)
|
592.86(64.26−5469.91)
|
༜60
|
3
|
0.88(0.79–0.94)
|
0.88(0.78–0.95)
|
6.08(2.05–18.07)
|
0.15(0.05–0.43)
|
50.55(7.55−338.64)
|
Cutoff value
|
|
|
|
|
|
|
༞300ng/ml
|
4
|
0.99(0.94−1.00)
|
0.95(0.91–0.97)
|
43.01(1.39−1335.33)
|
0.04(0.01–0.12)
|
1181.88(74.31−18798.23)
|
༜300ng/ml
|
3
|
0.87(0.80–0.93)
|
0.87(0.77–0.94)
|
6.20(2.28–16.84)
|
0.15(0.08–0.28)
|
44.82(10.48−191.68)
|
Sample source
|
|
|
|
|
|
|
Serum
|
1
|
0.816(0.71–0.93)
|
0.850(0.746–0.969)
|
3.438(1.895–4.606)
|
0.216(0.109–0.330)
|
25.10(14.40−43.75)
|
Synovial fluid
|
6
|
0.95(0.91–0.98)
|
0.94(0.90–0.96)
|
19.46(4.65–81.47)
|
0.07(0.03–0.17)
|
359.77(50.05−2585.89)
|
Country
|
|
|
|
|
|
|
China
|
3
|
0.90(0.83–0.95)
|
0.95(0.87–0.98)
|
14.09(3.04–65.37)
|
0.11(0.03–0.35)
|
172.95(13.73−2178.05)
|
Other countries
|
4
|
0.95(0.89–0.99)
|
0.92(0.88–0.96)
|
14.77(2.89–75.36)
|
0.07(0.02–0.28)
|
300.49(19.36−4665.06)
|
PLR: positive likelihood ratio; NLR: negative likelihood ratio; DOR: diagnostic odds ratio.