Globally, esophageal squamous cell carcinoma (ESCC) is the most common form of esophageal cancer, comprising approximately 85-90% of all cases worldwide, with the remaining cases being attributed to esophageal adenocarcinoma (EAC) (1, 2), of which approximately 50% present as advanced or metastatic, unresectable disease at diagnosis (3–6). Patients with advanced ESCC face a particularly poor prognosis, with a median estimated overall survival of less than 1 year (7, 8), and 5-year survival rates typically ranging between 10% and 30% (1, 9, 10).

Combining programmed cell death protein-1 (PD-1) inhibitors with chemotherapy (CT) has recently emerged as a promising first-line (1L) treatment to enhance outcomes for patients with advanced or metastatic ESCC who are eligible for initial systemic therapy (11). To date, broad regulatory approval has been granted for 1L treatment with tislelizumab + CT, pembrolizumab + CT, and nivolumab + CT for select patients with positive expression for programmed death-ligand 1 (PD-L1), including by the United States (US) Food and Drug Administration and the European Medicines Agency (12–20). The minimum PD-L1 expression threshold varies across approved 1L indications. Tislelizumab + CT is approved for use in patients whose tumors express PD-L1 ≥1 in the US and a PD-L1 TAP score ≥5% in the European Union (EU), respectively (13, 15). Based on the phase 3 KEYNOTE-590 trial (NCT03189719), pembrolizumab + CT is approved in the US for patients with tumors with a CPS ≥1, and in the EU for those with tumors with a CPS ≥10 (17, 20). Similarly, the phase 3 CheckMate-648 trial (NCT03143153) led to the approval of both nivolumab + CT and nivolumab + ipilimumab in the US for patients whose tumors express PD-L1 (≥1), whereas in the EU, these regimens are restricted for use in patients with a tumor proportion score (TPS) ≥1% (18, 19).

In the randomized, phase 3 RATIONALE-306 trial (NCT03783442), tislelizumab + CT demonstrated more favorable overall survival (OS) outcomes and a manageable safety profile relative to placebo + CT as 1L treatment of advanced or metastatic ESCC in the intent-to-treat (ITT) population, with the greatest OS benefit being observed in patients with a tumor PD-L1 TAP score ≥10% (secondary endpoint analysis) (21). Further, a recent retrospective analysis of RATIONALE-306 also found that tislelizumab + CT was associated with a clinically meaningful improvement in OS compared with placebo + CT among patients with advanced or metastatic, unresectable ESCC and tumor PD-L1 ≥1 at primary analysis (22). There is no standardized methodology for PD-L1 testing across currently indicated PD-1 therapies for ESCC, with testing methods varying regionally by local clinical practices, although there is a relatively high level of concordance between TAP score versus CPS (overall percentage agreement [OPA]: 90%) and TPS (OPA: 78%) (23).

Tislelizumab + CT has not been directly compared with pembrolizumab + CT, nivolumab + CT or ipilimumab, or other PD-1 inhibitor–based immuno-oncology (IO) regimens for the 1L treatment of ESCC. The present study is a network meta-analysis (NMA) conducted to assess the comparative efficacy and safety of tislelizumab + CT versus broadly approved 1L IO regimens for patients with unresectable, locally advanced or metastatic ESCC.

The addition of PD-1 inhibitors to CT represents a major therapeutic development for advanced or metastatic 1L ESCC, as treatment was previously only with CT regimens with limited survival benefits (41, 47). The introduction of anti–PD-1 agents like tislelizumab, nivolumab, and pembrolizumab further demonstrates the growing potential of targeting the PD-1/PD-L1 pathway. Finding the clinically optimum anti-PD-1 agent in this setting is critical to ensure appropriate access and clinical use. The computational power of NMA facilitates the ranking of several PD-1 inhibitors in terms of efficacy and safety to better inform clinical practice.

In this analysis, tislelizumab + CT was the highest ranked anti-PD-1 agent for OS, PFS, and grade ≥3 TRAE outcomes, and was significantly more favorable to nivolumab + CT for PFS. Collectively, these results highlight the favorable risk-benefit profile of tislelizumab + CT and provided a potential rationale for adopting the use of tislelizumab in ESCC.

In PD-L1 subgroup analyses, tislelizumab + CT was significantly more favorable to nivolumab + CT for PFS in patients with PD-L1 ≥1 (TAP score ≥1% or CPS ≥1) and PD-L1 ≥5 (TAP score ≥5% or CPS ≥5). Within both subgroups, tislelizumab + CT performed similarly to nivolumab + CT for OS and pembrolizumab + CT for OS and PFS. Although these results should be considered within the context of the differing PD-L1 expression scoring systems used across the included trials, a relatively high concordance between TAP score and CPS has been reported in terms of OPA (90% [95% CI: 86-93]) in the ESCC trial population (48). These results suggest that the relative efficacies of all three anti-PD-1 agents do not differ relative to the base case analyses based on patient PD-L1 level.

The subgroup analyses in patients receiving platinum + fluoropyrimidine as the CT backbone were consistent with the base case for OS, PFS, and ORR, demonstrating that tislelizumab performed similarly to both nivolumab and pembrolizumab in this patient subgroup. This suggests there may be flexibility in choosing the CT backbone based on individual patient factors and institutional or regional preferences. An additional scenario analysis was conducted to include nivolumab + ipilimumab, with similar results observed. Tislelizumab + CT was found to be comparable to nivolumab + ipilimumab in terms of OS and safety profile (grade ≥3 TRAEs) and was statistically more favorable to nivolumab + ipilimumab for PFS and ORR. Consistent results observed across both subgroup and scenario analyses lend additional confidence to these findings.

However, it should be noted that there were violations of the PH assumption in cross-trial comparisons involving nivolumab + ipilimumab, suggesting the treatment effect may not be constant over time and that the resulting HRs specifically against nivolumab + ipilimumab should be interpreted with caution. When considering within-trial data from RATIONALE-306, additional violations of the PH assumption were observed between tislelizumab + CT and placebo + CT specifically for OS. This violation has the potential to introduce time-dependent effects that may bias treatment estimates, particularly given that there are some differences between the three trials in follow-up time. Future models may consider alternative methods, such as fractional polynomial NMAs, restricted mean survival time analyses, or piecewise Cox PH models, to address these PH assumption violations (49, 50). However, it is not expected that these approaches would deviate substantially from the comparative results observed in this NMA. Further, such methods may be associated with overfitting and would require clinical input and validation of model choice.

Previous NMAs have also compared the efficacy of 1L IO therapies (35, 36), although certain relevant approved or novel regimens were not included for comparison. Ma et al. (36) conducted an NMA of six IO agents, including tislelizumab, pembrolizumab, and nivolumab, all in addition to CT. Of note, Ma et al. did not include nivolumab + ipilimumab (36), as was included in the additional scenario analysis of the current study. The present analysis reported that no significant differences were observed for OS between tislelizumab + CT, nivolumab + CT, and pembrolizumab + CT, similar to Ma et al. (36). Similarly, tislelizumab + CT ranked higher than both nivolumab + CT and pembrolizumab + CT for both OS and PFS in the analysis (36), as was seen in the current study. Moreover, in terms of safety profile, the ranking of tislelizumab + CT with respect to grade ≥3 TRAEs was generally consistent with findings from Ma et al. (36). The same NMA also ranked tislelizumab as having the third lowest rate of grade ≥3 TRAEs behind placebo + CT, while the current analysis ranked tislelizumab + CT as having the second lowest rate of grade ≥3 TRAEs behind placebo + CT (36).

More recently, Nian et al. conducted an NMA of nine IO therapies, including tislelizumab + CT, nivolumab + CT, nivolumab + ipilimumab, and pembrolizumab + CT (51). Similar to Ma et al. and the present analysis, there were no significant differences for OS between tislelizumab + CT, nivolumab + CT, nivolumab + ipilimumab, and pembrolizumab + CT (51, 52). Additionally, tislelizumab + CT was ranked higher than the same three comparator IO regimens for both OS and PFS, as observed in the current NMA and by Ma et al. (51, 52). Of note, the NMA by Nian et al. also found significantly more favorable PFS and ORR outcomes for tislelizumab + CT versus nivolumab + ipilimumab, with a similar magnitude of relative benefit (51).

Another recent NMA of eight IO agents was also conducted by Chen et al., including tislelizumab + CT, nivolumab + CT, nivolumab + ipilimumab, and pembrolizumab + CT (53). Both OS and PFS results were consistent with the present NMA. Similar to findings reported in Ma et al., Nian et al., and the current NMA’s scenario analysis, tislelizumab + CT was found statistically more favorable in terms of PFS than nivolumab + ipilimumab (51–53). The study’s safety findings in terms of grade ≥3 TRAEs were also consistent with those observed within the current analysis for tislelizumab + CT and other broadly approved IO therapies (53).

The present study has several strengths. First, analyses were performed according to best practice for conducting and reporting NMAs as described by the National Institute for Health and Care Excellence to ensure transparency and reproducibility (30). The NMAs were also informed by a recent, comprehensive SLR in adherence with best practices provided by PRISMA guidance. Further, all trials included in the SLR underwent a rigorous feasibility assessment to highlight any sources of inter-trial heterogeneity to ensure the validity of results. Key clinical subpopulations were also identified with clinical expert opinion for analysis to reflect the diversity of indications and reimbursement criteria for comparator treatments. To our knowledge, this is the first analysis to consider key PD-L1 expression subgroups aligned with approved indications for the use of tislelizumab + CT in 1L ESCC in the EU (PD-L1 TAP score ≥5%) and the US (PD-L1 ≥1) (13, 15). Further, the present NMA provides additional evidence confirming the assumption of equivalence of CT backbone regimens.

This study has the following limitations. First, network structures were sparse and connections between treatments were informed by a single trial, which increases the potential for biased treatment effect estimates, increases imprecision, and therefore reduces the robustness of results. Second, the limited number of trials informing each treatment comparison prevented the use of meta regression to adjust for potential sources of inter-trial heterogeneity. Additionally, indirect treatment comparisons, such as NMAs, rely on the assumption that the included trials are sufficiently similar, such that the effect estimate will not be biased by underlying differences in patient populations. In this analysis, minimal between-trial heterogeneity was observed, and potential differences were explored using subgroup analyses. As such, population-adjusted analyses were not conducted. Moreover, differences in CT backbones used across trials may have impacted these safety comparisons due to each CT regimen having distinct adverse event patterns (21, 39, 40).

Further, safety subgroup analyses were precluded by inconsistent reporting of safety outcomes in comparator trials (for example, only RATIONALE-306 defined adverse events of special interest, limiting cross-trial comparisons), which increases uncertainty of any findings in this study related to safety outcomes. Of note, KEYNOTE-590 only summarized safety outcomes for the broader esophageal cancer population, with adverse event data not being reported separately for the ESCC and esophageal adenocarcinoma subgroups. This may have limited comparisons of grade ≥3 TRAEs relative to pembrolizumab + CT, as RATIONALE-306 and CheckMate 648 specifically reported these data for the ESCC population. Safety comparisons may also be limited by the significant variation in treatment exposure and length of follow-up across the included trials; although, median treatment durations in the present NMA were similar across the included trials (5.7-7.7 months) (22, 39, 40). Treatments with longer exposure durations are more likely to exhibit a higher frequency of adverse events, potentially impacting comparisons in the current study of the cumulative incidence of grade ≥3 TRAEs.

In summary, the present study provides evidence on the relative efficacy and safety of relevant 1L IO regimens with broad regulatory approval for patients with unresectable, locally advanced or metastatic ESCC using NMAs. Probabilistically, tislelizumab was the top-ranked IO treatment across survival and safety analyses in the present NMA, although most pairwise comparisons were not statistically significant. Tislelizumab + CT was found to have a significant PFS benefit over nivolumab + CT, and similar efficacy to pembrolizumab + CT for OS and PFS. Tislelizumab + CT had a comparable frequency of grade ≥3 TRAEs to all treatments, including placebo + CT, whereas nivolumab + CT was associated with significantly increased incidence odds of grade ≥3 TRAEs relative to placebo + CT. Key subgroup analyses, including for PD-L1 expression (≥1% and ≥5% TAP score or CPS), Asia versus ROW, and underlying CT treatment, were generally consistent with the primary analysis, and showed comparable outcomes for tislelizumab + CT compared to other IO agents in combination with CT. Overall, based on results from the present analyses, tislelizumab + CT represents an effective treatment option for 1L ESCC.