3.3 Synthesis of results and subgroup analysis for association of high/low PD-L1 expression with prognostic factors
Meta-analysis of studies for OS indcated that PD-L1expression on TCs had no significant effect on OS (HR, 1.10; 95% CI: [0.86, 1.40]; P =0.461), and statistical heterogeneity (I2 = 80.7%, p < 0.001) emerged(Supplemental Fig.S11) . Also for DSS (Supplemental Fig.S12) , fifteen studies reported ready-made HR or provided sufficient data to calculate the estimated HR that was synthesised that yield a no statistically significant result (HR, 1.22; 95% CI: [0.86, 1.74]; P =0.258) with significant heterogeneity (I2 = 62.6%, p < 0.001). Likewise, the meta-analysis of studies for disease-free survival (DFS), progression-free survival (PFS) and local-regional progression-free survival (LRFS) were respectively performed, of which pooled estimated HR were achieved correspondingly. Only the result for LRFS showed significant finding (HR, 1.77; 95% CI: [1.20, 2.62]; P =0.004) without significant heterogeneity (I2 = 0.0%, p = 0.590) (Supplemental Fig.S13) . Five studies assessed association between the expression of PD-L1 on TILs and OS, of which the pooled result showed no statistical significance (Supplemental File 2 ). The estimated HR for DSS, DFS, PFS and LRFS were unattainable to synthesize due to only one study reporting prognostic values.
Sources of heterogeneity might be attributed in part to geographic region, scoring systems, antibody type, tumour anatomic location. Due to the a lack of study on the relationship between PD-L1 expression on TILS and prognosis, subgroup analysis was performed only on studies with prognosis associated with TCs.
Geographic region was stratified for Asia and non-Asia with comparison for prognostic value and showed no significant differences for OS, DSS. Nevertheless, DFS showed high expression of PD-L1 in TCs was more likely to have worse prognosis (HR, 1.65; 95% CI: [1.09, 2.49]; P =0.018) (Supplemental Fig.S14) . Various scoring systems including semiquantitative evaluation (SE), percentage of positive cell (PPC), CPS , TPS, H score (combination of the staining distribution and intensity scoring systems), were used to recombine these studies for OS, DSS and DFS. There were no statistically significant results of prognosis showed in TPS or CPS. Both DSS (HR, 1.65; 95% CI: [1.04, 2.64]; P =0.035) and DFS (HR, 1.50; 95% CI: [1.15, 1.96]; P =0.003), at a 5% cut-off of PPC, had positive association with high PD-L1 expression(Supplemental Fig.S15) .
Subgroup analysis decided by antibody type was carried out for OS, DSS and DFS when two or more studies providing prognostic values for one type of antibody. The results of OS showed that a worse prognosis for 5H1(HR, 2.50; 95% CI: [1.08, 5.76]; P =0.032), and a favourable prognosis for 22C3(HR, 0.43; 95% CI: [0.27, 0.69]; P =0.001). For DSS, high PD-L1 expression detecting by E1L3N was shown to be associated with an worse prognosis(HR, 1.78; 95% CI: [1.13, 2.80]; P =0.014). There was no evidence suggested the association of DFS with PD-L1 expression was determined by antibody type(Supplemental Fig.S16) . Significant association of PD-L1 expressions in TCs with staining location was not detected(Supplemental Fig.S17) . Seven studies were viable for synthesising that reported OS in terms of location of tongue. It was shown that high expressions of PD-L1 in TCs had a worse prognosis in OSCC of the tongue (HR, 1.24; 95% CI: [1.03, 1.49]; P =0.023). Four studies assessed the DFS of PD-L1 expressions and results showed that high PD-L1 expression in TCs was associated with worse prognosis (HR, 2.03; 95% CI: [1.28, 3.22]; P =0.003)(Fig.3 ). Only one study was available for PFS, LRFS that was insufficient to combine.
Publication bias of studies in terms of the role of PD-L1 expressions in clinicopathological and prognostic was determined by Egger’s tests and was shown by funnel plots, of which results suggested absence of proof (Supplemental File 3 ).