Eligible studies
Initial screen identified 37 articles. After reviewing the titles and abstracts, 9 articles were excluded because they obviously did not meet our selection criteria. The remaining 28 articles were further checked by screening the full texts. In this period, 12 studies were excluded because they are reviews or meta-analyses (n=5), insufficiency information (n=3), detecting miRNA methylation (n=2) or data overlapping (n=2). As a result, 16 articles11,13,19-32 were eligible for final meta-anlaysis. Figure 1A showed the process of studies identification. Among 16 articles, 6 articles consisted at least 2 independent studies11,19,22,26,27,32, as a result, 30 studies from 16 articles including 2,447 subjects (1,212 patients with lung cancer and 1,235 controls) were included for analysis. These studies published from 2010 to 2018, and the sample size ranged from 30 to 315. Two studies conducted in Asian24,27, seven studies conducted in Caucasian13,20,25,28-31 and the remaining seven studies11,19,21-23,26,32 conducted in mixed population. Two studies19,26 selected healthy population as controls, 11 studies13,20-23,25,28-32 selected cancer-free subjects as controls, and the remaining three studies11,24,27 applied benign pulmonary disease (BPD) as controls. The detailed information of the included articles was listed in Table 1. Quality assessments results for each eligible study according to QUADAS-2 guidelines are presented in Figure 2B.
Pooled Diagnostic Accuracy of miRNA
The overall predictive accuracy of miRNAs was summarized in Table 1. The pooled sensitivity was 0.77 (95% CI: 0.73–0.81), the specificity was 0.87(95% CI: 0.83–0.90) (Figure 2), the pooled PLR was 5.9 (95% CI: 4.7–7.4), the NLR was 0.26 (95% CI: 0.22–0.30) and the DOR of 23 (95% CI: 18–29) (Table 2). The AUC value was 0.89 (95% CI: 0.86–0.91), and the corresponding SROC curve was shown in Figure 3A. The Fagan diagram was also applied to illustrate the post-test probabilities of sputum miRNAs in lung cancer diagnosis (Figure 3B).
Subgroup analyses
Various subgroup analyses according to race, miRNA profiling, source of control, histology type, sample size, publication year and stage were done. The results were summarized in Table 2
Race
For Asian population, the pooled sensitivity, specificity, PLR, NLR, DOR and AUC were 0.81, 0.85, 5.4, 0.22, 25 and 0.90 respectively. For Caucasian, they were 0.79, 0.87, 6.0, 0.24, 25 and 0.89 respectively. For mixed population, they were 0.73, 0.89, 6.4, 0.30, 21 and 0.90 respectively (Table 2). Subgroup analysis according to race suggested the diagnostic accuracy of sputum miRNA in NSCLC was excellent in all races.
miRNA profiling
The pooled analysis suggested combined miRNA may yield higher diagnostic accuracy in distinguish NSCLC. The pooled sensitivity, specificity, PLR, NLR, DOR and AUC were 0.76 vs 0.74, 0.88 vs 0.74, 6.4 vs 2.8, 0.28 vs 0.36, 23 vs 8 and 0.90 vs 0.80 respectively for multiple miRNA vs single miRNA (Table 2 , Figure 4A and 4B, and Figure 5A and 5B).
Source of control
According the source of control, control subjects could be classified as healthy control, cancer-free control and BPD. Studies with healthy control had a pooled sensitivity of 0.73, specificity of 0.91, PLR of 8.5, NLR of 0.29, DOR of 29 and AUC of 0.77. Studies with cancer-free controls had a pooled sensitivity of 0.76, specificity of 0.86, PLR of 5.5, NLR of 0.28, DOR of 19 and AUC of 0.88. Studies with BPD controls showed a pooled sensitivity of 0.82, specificity of 0.86, PLR of 5.7, NLR of 0.22, DOR of 26 and AUC of 0.91 (Table 2). All these subgroups showed satisfied diagnostic accuracy, suggesting sputum miRNA could not only be a tool for lung cancer screening but also be a tool for differential diagnosis.
Stage
Firstly, according to percentage of early stage disease, we divided studies as I/II≥0.6 and I/II<0.6 subgroups. Both group showed satisfied diagnostic efficacy, with pooled sensitivity of 0.76 and 0.79, specificity of 0.88 and 0.85, PLR of 6.1 and 5.3, NLR of 0.27 and 0.25, DOR of 23 and 21, AUC of 0.90 and 0.87 for I/II≥0.6 and I/II<0.6 subgroup respectively(Table 2).
We also extracted studies only focus on stage I and/or II disease for analysis. For stage I and II disease, the pooled sensitivity, specificity, PLR, NLR, DOR and AUC were 0.76, 0.88, 6.5, 0.27, 23 and 0.90, respectively. For only stage I disease, the pooled sensitivity, specificity, PLR, NLR, DOR and AUC were 0.75, 0.88, 6.5, 0.28, 23 and 0.90, respectively (Table 2, Figure 4C and 4D and Figure 5C and 5D). This suggested sputum miRNA could serve as an efficacy tool in early lung cancer screen.
Histology, sample size and publication year
According to the percentage of adenocarcinoma in overall NSCLC, 0.5 was selected as a cut-off to divided studies as AD≥0.5 and AD<0.5. Both studies showed satisfied diagnostic accuracy. The similar results were found in the subgroup analyses according to sample size and publication year (Table 2).
Meta-regression analyses and publication bias
Potential sources of inter-study heterogeneity were explored by meta-regression analyses. The analyses suggested race (P<0.01), miRNA profiling (P<0.01) and source of control (P<0.01) might be the potential sources of inter-study heterogeneity. The effect of samle size, publication year, age, male ratio and smoking on heterogeneity was not significant (Figure 6A). Deeks’funnel plot asymmetry test was applied to detect publication bias. No publication bias was detected in both overall (Figure 6B) and subgroup analyses (Table 2).
Potential useful sputum miRNAs identified from literature
We sorted the miRNAs reported by at least two studies that could serve as potential biomarkers. In upregulated miRNAs, 11 studies reported miR-210,11,20-23,25-28,31,32 eight studies reported miR-2111,13,19,22,24,25,27,32 and six studies reported miR-31.11,21-23, 32,37 miR-205,11,22,26 miR-182,11,19,22 miR-200b11,19,22 and miR-15511,24,26 had been reported by three studies. Only miR-37211,28 had been reported by two studies. In downregulated studies, miR-126-3p had been reported by five studies,11,19,22,26,31 miR-486 had been reported by three studies11,19,22 and miR-145 had been reported by two studies.19,31 We also found some conflicting result. Two studies reported miR-375 was upregulated,11,19 but another two studies reported it was downregulated.22,31 miR-708 was upregulated in two studies11,26 and downregulated in one studies.22 Further studies should verify these conflicting results(Table 3).