Basic features of the included literature There were 596 papers retrieved from seven databases managed by Endnote software. After removing duplicated papers, there were 462 articles. Then 100 papers, such as reviews and conference papers, were taken off after scanning the title. After that, the 317 unrelated papers were excluded by going through the abstract, including 272 non-conforming articles and 45 animal experiment papers. 34 papers with inconsistent outcome indicators and missing data were withdrawn after reading through the full text. Finally, 7 English and 4 Chinese articles were included (see Figure 1). Among these 11 studies involved 1717 was patients and 1652 normal controls, and outcomes included TP; FP. FN and TN. The basic characteristics were shown in Table 1
Table 1 Basic data of lncRNA in AIS diagnosis
Study
|
Tissue
|
Sample
|
LncRNA
|
expression
|
Diagnosticcriteria
|
Time
|
Detection method
|
TP
|
FP
|
FN
|
TN
|
AIS CI
|
Yang 201815
|
peripheral blood
|
550 550
|
LncRNA ANRIL
|
Upregulated
|
CT/MRI
|
NA
|
q-PCR
|
365
|
254
|
185
|
296
|
Feng 201816
|
plasma
|
126 125
|
LncRNA ANRIL
|
Downregulated
|
CT/MRI
|
24h
|
q-PCR
|
91
|
36
|
35
|
89
|
Guo 201817
|
peripheral blood
|
80 80
|
lncRNA-ENST00000568297
|
Upregulated
|
CT/MRI
|
72h
|
q-PCR
|
52
|
29
|
28
|
51
|
lncRNA-ENST00000568243
|
56
|
24
|
24
|
56
|
lncRNA-NR_046084
|
49
|
25
|
31
|
55
|
lncRNA combination
|
66
|
16
|
14
|
64
|
Zhu 201818
|
Leukocyte
|
189 189
|
lncRNA MIAT
|
Upregulated
|
MRI
|
24h
|
q-PCR
|
140
|
37
|
49
|
152
|
Deng 201819
|
PBMC
|
119 92
|
linc-DHFRL1-4
|
Upregulated
|
CT/MRI
|
48h
|
q-PCR
|
86
|
30
|
33
|
62
|
lncRNA SNHG15
|
78
|
22
|
41
|
70
|
linc-FAM98A-3
|
75
|
27
|
44
|
65
|
lncRNA combination
|
102
|
20
|
17
|
72
|
Hong 201920
|
serum
|
26 26
|
LncRNA CAI2
|
Upregulated
|
CT/MRI
|
48h
|
q-PCR
|
17
|
4
|
9
|
22
|
He 201921
|
plasma
|
140 140
|
lncRNA ANRIL
|
Upregulated
|
CT/MRI
|
2h
|
q-PCR
|
114
|
23
|
46
|
117
|
Guo 201922
|
serum
|
110 100
|
lncRNA TUG1
|
Upregulated
|
AIS Guide
|
2h
|
q-PCR
|
94
|
18
|
16
|
82
|
Li 201923
|
peripheral blood
|
32 32
|
lncRNA-C14orf64
|
Downregulated
|
CT/MRI
|
NA
|
q-PCR
|
20
|
8
|
12
|
24
|
lncRNA-AC136007.2
|
29
|
3
|
3
|
29
|
Li 202024
|
plasma
|
210 210
|
lncRNA NEAT1
|
Upregulated
|
CT/MRI
|
24h
|
q-PCR
|
135
|
36
|
75
|
174
|
Liu 202025
|
serum
|
135 108
|
lncRNA SNHG14
|
Upregulated
|
CT/MRI
|
6h
|
q-PCR
|
109
|
46
|
26
|
62
|
Abbreviations: Peripheral blood mononuclear cell(PBMC);Magnetic resonance imaging(MRI)
Literature quality evaluation Based on QUADAS-2 results: The papers’ scores were all above 3 points, and the quality was good. Due to the 11 articles being all case-control trials, there were high risks in selecting cases. Furthermore, four items 15,19,23,25 did not give critical values leading to all being listed as high risks, and nine items 15,16,18-22,25 used more than two diagnostic methods resulting in being listed as high risks. Most of the articles did not describe whether reference standard judgment was blinded, therefore, it was unclear for inclusion. From the risk of bias map, it was found that the evaluation of the clinical applicability of the papers was high(Figure 2).
Meta-analysis results
Threshold effect evaluation Meta DiSc software r=-0.282 obtained Spearman correlation coefficient between sensitivity logarithm and (1-specificity) logarithm (P=0.257>0.05), which was not significant, indicating that there was no threshold effect in this study. Furthermore, no "shoulder arm shape" appeared by drawing the symmetric sROC curve, demonstrating no heterogeneity caused by the threshold effect in this study (Figure 3).
Evaluation of the non-threshold effect The Cochran-Q test of the diagnostic odds ratio (DOR) showed that Cochran-q =129.67, I2=86.9%, P<0.001, describing that heterogeneity caused by non-threshold effect exists in this study. Furthermore, this study's sensitivity, specificity, positive likelihood ratio, negative likelihood ratio, and I2 of DOR were greater than 50%. Therefore, the random effect model was implemented to combine the above five effect sizes.
Effect size results were combined According to stata16, the AUC of lncRNA for the diagnosis of IS was 80%. The Meta DiSc software showed that the combined sensitivity of combined IS was 0.70 (95%: 0.69-0.72), the specificity was 0.71 (CI95%: 0.69-0.72), and the positive likelihood was 0.71 (CI95%: 0.69-0.72). The ratio was 2.84 (CI95%: 2.25-3.59), the negative likelihood ratio was 0.39 (CI95%: 0.33-0.46), and the diagnostic odds ratio was 7.69 (5.2-11.37) (Figure 4).
Subgroup analysis results Subgroup analysis found that: The heterogeneity of tissue source and blood collection time decreased after grouping, while the heterogeneity of lncRNA expression was almost unchanged after high-low grouping. Among tissue sources, the diagnostic efficiency of serum specimens was 88.5%. The diagnostic efficiency within 24h was 82.6% higher than that within 24 ~ 72h, and low expression was slightly higher than high expression (79.8%). The DOR of combined lncRNA was 20.33, much higher than the diagnostic performance of single lncRNA (Table 2).
Table 2 Subgroup analysis of different categorical variables
Subgroup classification
|
Sensitivity
|
Specificity
|
PLR(95%CI)
|
NLR(95%CI)
|
DOR
|
AUC
|
P
|
I2
|
Tissue
|
|
|
|
|
|
|
|
|
peripheral blood
|
0.68
|
0.62
|
2.40(1.68~3.42)
|
0.43(0.32~0.58)
|
6.10
|
78%
|
<0.001
|
86.6%
|
plasma
|
0.69
|
0.80
|
3.39(2.44~4.72)
|
0.40(0.35~0.45)
|
8.85
|
78.9%
|
0.2437
|
29.2%
|
PBMC
|
0.72
|
0.73
|
2.63(2.02~3.44)
|
0.38(0.26~0.55)
|
7.16
|
79.6%
|
0.0034
|
78.1%
|
serum
|
0.74
|
0.75
|
3.30(1.72~6.22)
|
0.33(0.18~0.61)
|
11.67
|
88.5%
|
0.0085
|
79.0%
|
Collection time
|
|
|
|
|
|
|
|
|
≤24h
|
0.72
|
0.79
|
3.41(2.75~4.22)
|
0.35(0.29~0.41)
|
10.11
|
82.6%
|
0.0110
|
63.8%
|
24~72h
|
0.71
|
0.72
|
2.54(2.08~3.10)
|
0.40(0.32~0.51)
|
6.58
|
78.8%
|
0.0003
|
72.3%
|
circRNA expression
|
|
|
|
|
|
|
|
|
Upregulated
|
0.69
|
0.70
|
2.73(2.07~3.71)
|
0.41(0.34~0.39)
|
6.84
|
77.9%
|
<0.001
|
87.9%
|
Downregulated
|
0.71
|
0.70
|
2.82(2.24~3.54)
|
0.38(0.32~0.45)
|
7.75
|
79.8%
|
<0.001
|
86.9%
|
Note: Abbreviations:Positive likelihood ratio(PLR);Negative likelihoodratio(NLR);Diagnostic odds ratio(DOR);ROC curve(AUC);Peripheral blood mononuclear cell(PBMC)
Sensitivity analysis A sensitivity analysis was performed to understand the combined effect size changes after removing an individual study (Figure.5). The results demonstrated that the bivariate model was moderately robust. Two outliers were identified by impact analysis, and two outliers were found through outlier detection.
Publication bias test Through Deek's quantitative funnel plot test, P=0.83, there was no publication bias in the literature (Figure 6).