3.1 Characteristics and quality analysis of enrolled studies
Literature search was performed using the priori determined search strategy. A total of 2658 records were retrieved through searching databases of PubMed, Web of Science, the Cochrane Library and Scopus. After exhaustive screening and careful assessment, 46 studies including 4484 patients were ultimately included in the meta-analysis1-9,11,12,17-51. The detailed process and the reason of excluded studies were demonstrated as flow chart, Fig.1 .
The characteristics of entered studies were expatiated in Table 1 . Among them, two studies {de Vicente, 2019 #4;Foy, 2017 #7}were splitted as two separate data sets on account of different antibodies12,39{de Vicente, 2019 #4;!!! INVALID CITATION !!!, #0;Quan, 2020 #20}. One study was stratified as two separate data sets for prognosis due to two different tissue sources, and also, was tread as three separate data sets for clinical parameters based on three various assay cutoffs9. A study was converted to two separate data sets as a result of two different assay cutoffs. There were two separate data sets for clinical parameters and three separate data sets for prognosis in study reported by Pena-Cardelles et al.11 Six studies focused on PD-L1 expression in TC and IC that were calculated separately as two independent datasets1,23,34,35,47,49. 46 enrolled studies were judged whether to provide clinical variables or prognostic variables, of which results were marked as ā€œYā€ or ā€œNā€. The classification and assessment for variables suitable for synthesis were detailed in Table S1 .
Quality analysis of enrolled studies was performed according to the evaluation criteria adapted from REMARK guidelines, of which results were summarized in Supplemental Fig. S1.