This meta-analysis and systematic review adhered to the Preferred Reporting Items for Systematic Reviews and Meta‐Analyses guidelines (27) and the protocol for the analysis was registered prospectively in PROSPERO (CRD42024504778). A comprehensive search was conducted on multiple databases including PubMed, Web of Science, the Cochrane Library, and Embase to identify relevant studies published until February 2024. The search terms included tertiary lymphoid structures (TLSs), tertiary lymphoid organ (TLO), tertiary lymphoid tissue (TLT), ectopic lymphoid-like structures (ELSs), cancer/tumor/solid tumor, immunotherapy, ICI, immune checkpoint inhibitor and prognosis, prognostic or survival outcome. The exact search query is provided as follows to allow reproducibility: (TLSs OR TLO OR TLT OR ELSs) AND (cancer OR tumor OR solid tumor) AND (prognosis OR prognostic OR survival outcome). Two researchers (LLL and YSG) independently screened titles and abstracts based on the inclusion and exclusion criteria. Finally, studies that provide reference data needed by this meta-analysis were selected through full-text reading, and discussions with a third author (BXG) were conducted when disagreements occurred between the two researchers (LLL and YSG).

Original studies eligible for inclusion in this meta-analysis had to meet the following criteria:(1) being limited to English articles, (2) studies investigating the TLSs in situ in tumor tissue by applying immunohistochemistry and H&E staining; (3) studies focusing on evaluating the prognostic value of TLSs in cancer patients treated with ICIs, and (4) reporting clinical outcomes such as overall survival (OS), progression-free survival (PFS), objective response rate (ORR). The exclusion criteria were as follows: (1) conference abstracts, letters to the editor, reviews, comments, and animal trials; (2) studies with sample sizes < 20, since a small sample size induces publication bias; and (3) work without raw data that could be traced.

The data extracted from the included studies encompassed various variables, including the publication year, name of the first author, region, the type of ICIs, the type of tumor, TLS detection methods, number of enrolled patients, clinical outcome measures, the radio of sex, Newcastle-Ottawa scale and type of study ( Table 1 ). The hazard ratios (HRs) and their associated 95% confidence intervals (95% CIs) from univariate or multivariate analysis were extracted. The assessment of each study was conducted by two authors (LLL and GYS) independently using the Newcastle-Ottawa scale (NOS), ranging from 0 to 9. Studies with an NOS score ≥ 6 were classified as high-quality studies. The two authors independently conducted the process and had discussions with a third author (BXG) when disagreements occurred.

Statistical software R software (version 4.1.0) was used to perform the analysis, while HR and 95% CI data were log transformed and pooled. Before performing a meta-analysis containing studies with different types of cancer and different types of immune checkpoint inhibitors (ICIs), heterogeneity was assessed through the implementation of a chi-square test and the I² metric. The I² value serves as an indicator of the proportion of variability across the pooled estimates that can be attributed to statistical heterogeneity. Studies with an I² value exceeding 50% were considered to exhibit significant heterogeneity, where a random effects model was used for the analysis. Potential sources of heterogeneity were identified through the utilization of Baujat plots, and subsequent sensitivity analyses were carried out by systematically excluding individual studies. While a fixed effects model was utilized when heterogeneity was low (I²<50%), where sensitivity analysis is not required. Subsequently, the forest maps were subsequently created, followed by a comprehensive description of the pooled HR or OR accompanied by its corresponding 95% confidence interval (CI). Subgroup analyses of OS, PFS and ORR were performed based on patients’ characteristics. Publication bias was evaluated by an inverted funnel plot and was quantified by Egger’s and Begg’s tests. A two-sided α of less than 0.05 was considered statistically significant.

Using the described search strategy, a total of 461 non-duplicated studies were identified. After screening based on predetermined criteria, 13 studies were selected for further evaluation through abstract review in accordance with the inclusion criteria. Due to small sample sizes, the 4 studies (14, 17–19), were only included in the systematic review. Therefore, this meta-analysis selected 9 studies (11–13, 15, 16, 21, 28–30) investigating the correlation between TLSs and clinical outcomes in cancer patients treated with immune checkpoint inhibitors. Figure 1 shows the flow diagram for literature retrieval and selection.

The basic information and main characteristics of the 9 included studies were shown in Table 1 . Among the 9 included studies, six cohorts had a retrospective study design, and three cohorts were prospectively designed or from prospective trials (12, 21, 29). Among the 9 included studies evaluating the prognostic value of TLSs in cancer patients treated with ICIs: 6 studies evaluated overall survival (OS) (11, 12, 21, 28–30), 3 studies evaluated progression‐free survival (PFS) (13, 16, 21), and 4 studies evaluated objective response rate (ORR) (12, 13, 15, 16). All the studies included in the analysis obtained moderately high scores on quality assessments conducted using the Newcastle-Ottawa Scale. Out of the 9 studies that were included, 4 studies focused on specific types of cancer, with melanoma being the most commonly reported tumor (11, 28–30). The ICIs used in the aforementioned studies included PD-1 antibody (pembrolizumab and nivolumab were commonly used), PD-L1 antibody, CTLA-4 antibody, and combination therapies including immunotherapy.

TLSs were divided into high levels and low levels based on different cut-off criteria. Among the 9 studies included in this analysis, different cut-off criteria corresponded to different HR. In the subsequent investigation of the relationship between TLSs and OS, PFS, and ORR, we established inclusion criteria. If a study employs two distinct TLS cut-off criteria, our preference was given to the hazard ratio (HR) associated with Density, Degree of maturity, or Maximal diameter. In cases where a study applies both Density and Degree of maturity, or Density and Maximal diameter for TLS grading, the HR linked to Density was the one we chose.

Among the 9 studies selected for this meta-analysis, 6 studies reported the association between OS and TLSs (11, 13, 21, 28–30). Considering the low heterogeneity (I² = 0%), a fixed effects model was used for analysis. Results indicated the pooled HR was 0.35 (95% CI: 0.24–0.53, p < 0.001, Figure 2 ), suggesting the prognostic role of TLSs in cancer patients treated with ICIs.

The sensitivity analysis confirmed the pooled results’ credibility and stability. The funnel plots exhibited approximate symmetry ( Figure 3A ). Egger’s and Begg’s tests showed no significant publication bias (p = 0.154, p = 0.851), indicating that high levels of TLSs significantly predicted prolonged OS in cancer patients treated with ICIs.

Among the 9 studies selected for the meta-analysis, 3 studies reported the association between PFS and TLSs (13, 16, 21). Considering the low heterogeneity (I² = 0%), a fixed effects model was used for analysis. Results indicated the pooled HR was 0.47 (95% CI: 0.31–0.72, p < 0.001, Figure 4 ). The sensitivity analysis confirmed the pooled results’ credibility and stability. The funnel plots exhibited approximate symmetry ( Figure 3B ). Egger’s and Begg’s tests showed no significant publication bias (p = 0.489, p = 0.602), indicating that high levels of TLSs significantly predicted better PFS in cancer patients treated with ICIs.

The ORR was reported in 4 studies which included 398 cancer patients treated with ICIs (12, 13, 15, 16). The pooled OR was 3.78 (95% CI: 2.26–6.33, p < 0.001, I² = 32%, Figure 5 ) suggesting that TLSs were associated with lower tumor response. Similar results were obtained after conducting a sensitivity analysis. The funnel plots exhibited approximate symmetry ( Figure 3C ). Egger’s and Begg’s tests showed no significant publication bias (p = 0.478, p = 0.174).