After the systematic retrieval from the abovementioned databases, a total of 914 initial citations were obtained by using the search strategy, and 392 potential citations were left for title and abstract screening following the deletions of duplications (n=285), conference papers (n=219), reviews (n=16) and case reports (n=2). Next, 49 articles remained for full-text assessment due to 343 citations being excluded via title and abstract screening; of these, studies that were reviews (n=2), inconsistent to the criteria of the high-level TILs in our study (n=4), devoid of useful data (n=16) and centered on PD-L1+TILs (n=3) or FOXP3+ TILs (n=2) did not meet the inclusion criteria and hence were excluded. Ultimately, 22 qualified studies were included for meta-analysis (Table 1).(5, 6, 8-27) The procedure of qualified article selection is outlined in Figure 1.
Of those included studies, the publication year ranged from 2010 to 2019, 14 (63.6%) were retrospective studies with a total of 6958 cases, 10 (45.5%) were originally from Asian countries, 9 (40.9%) documented the breast cancer patients with ER-negative status, and the predominately chemotherapy strategy was in the setting of neoadjuvant therapy. Table 2 additionally represented the other details involving the median follow-up, publication year, the median age, the analyzed cases in each analysis, the primary endpoint, and the detailed chemotherapy regimen, as well as the TILs subsets.
Association of each 10% increment of TILs and OS
Four studies recorded each 10% increment of TILs and OS in breast cancers without classification to different molecular subtypes, and the pooled results suggested that each 10% increment of TILs could not significantly improve OS (HR, 0.95; 95% CI, 0.91-1.01). However, there was a significant improvement in OS in terms of the pooled results of multivariate data (HR, 0.92; 95% CI, 0.85-0.98) but not that of univariate data (HR, 1.00; 95% CI, 0.94-1.06) (Figure 2). In the subgroup analysis of different subtypes, the pooled results showed that, although each 10% increment of TILs in luminal tumor phenotype did not significantly improve OS (HR, 1.06; 95% CI, 0.99-1.13) (eFigure 1, Supplementary page 1), the improvements in OS were attained by it in HER2-overexpression (HR, 0.92; 95% CI, 0.89-0.95) (eFigure 2, Supplementary page 1) and TN (HR, 0.90; 95% CI, 0.89-0.92) subtypes (eFigure 3, Supplementary page 2). The results were both statistically significant in pooling the univariate data and the multivariate data of the latter two molecular phenotypes (these data were shown in eFigure2 and eFigure 3, respectively).
Association of each 10% increment of TILs and pCR
Two studies reported each 10% increment of TILs and pCR in breast tumors, and one(25) of them divided patients into the training cohort and the validation cohort. Thus, three independently relevant data existed. The pooled results indicated that there was a significantly positive correlation between each 10% increment of TILs and the increased pCR rate (OR, 1.27; 95% CI, 1.19-1.35). The results of pooling univariate data (OR, 1.33; 95% CI, 1.19-1.47) and multivariate data (OR, 1.21; 95% CI, 1.14-1.28) were still statistically significant (Figure 3).
Association of high-level TILs and pCR
Eleven studies provided sufficient data to the association of high-level TILs and pCR. There was a significant difference in pCR rate between high-level and low-level TILs (OR, 2.73; 95% CI, 2.40-3.01), and the pooled results of univariate data (OR, 2.84; 95% CI, 2.46-3.21) and multivariate data (OR, 2.35; 95% CI, 1.65-3.05) were also both statistically significant (Figure 4). In the subgroup analysis, the pooled results all indicated a higher pCR rate in luminal, HER2-overexpression and TN phenotypes with high-level TILs than those with low-level TILs, respectively (these data were outlined in eFigure 4, Supplementary page 3).
Comparison of high-level TILs expression across different breast cancer subsets
Seven studies were collected to perform the comparison of expression of high-level TILs across the different subsets of breast tumors. The pooled data of analysis showed that the presentation of high-level TILs between HER2-overexpression subtype and TN subtype was not significantly different (OR, 1.30; 95%CI, 0.83-2.04), whereas both subtypes experienced a significantly elevated expression of high-level TILs as compared to luminal phenotype (HER2-overexpression vs. luminal, OR, 3.14; 95% CI, 1.95-5.06; and TN vs. luminal, OR, 4.09; 95% CI, 2.71-6.19; respectively) (Figure 5).
Several meta-analyses manifested moderate-to-considerable heterogeneity, and therefore, the random-effect model was employed to pool the data. With the exception of the impact of each 10% increment of TILs in TN tumor subtype on OS (p=0.001) and that of the high-level TILs on pCR (p=0.007), there was no likelihood of publication bias in others because the Egger’s tests of them were not statistically significant (eTable 1, Supplementary page 4). The funnel plots for both analyses with significant publication bias were presented in eFigure 5 (Supplementary page 5).