This systematic review and updated meta-analysis pooled randomized evidence comparing PD-1/PD-L1 inhibitors combined with chemotherapy (or ICI monotherapy in specific adjuvant contexts) against chemotherapy alone or observation in triple-negative breast cancer (TNBC). Ten randomized controlled trials met inclusion criteria, enrolling 5,953 patients across both the metastatic (five RCTs, n = 3,526) and early-stage (five RCTs, n = 2,427) settings. Key additions relative to prior syntheses included the final KEYNOTE-522 overall survival analysis, the 7‑year GeparNuevo follow-up, and TORCHLIGHT phase III results for toripalimab.
Investigators searched PubMed and ClinicalTrials.gov through May 2, 2026, and hand-searched ASCO, ESMO, and SABCS proceedings from 2020–2026. The review followed PRISMA 2020 guidance and was prospectively registered in PROSPERO (CRD420261356428). Eligible studies were randomized phase II or III trials enrolling histologically confirmed TNBC and reporting at least one relevant outcome (OS, PFS, EFS, iDFS, pCR, or grade ≥3 adverse events).
Included metastatic trials were KEYNOTE-355 (pembrolizumab), IMpassion130/131/132 (atezolizumab), and TORCHLIGHT (toripalimab). Chemotherapy backbones varied by trial and included nab‑paclitaxel, solvent-based paclitaxel, gemcitabine plus carboplatin, capecitabine, and investigator-choice regimens. Early-stage trials included KEYNOTE-522 (pembrolizumab), IMpassion031, NeoTRIP, GeparNuevo (durvalumab), and A-BRAVE (avelumab). The dataset enabled stratified analyses by clinical setting, PD-L1 status, ICI class, and chemotherapy backbone.
Risk of bias was assessed using the Cochrane RoB 2 tool across randomization, deviations from intended interventions, missing data, outcome measurement, and selective reporting. Time-to-event outcomes were pooled as hazard ratios (HRs) and binary outcomes as odds ratios (ORs) using random-effects models with the DerSimonian–Laird estimator. The Hartung–Knapp–Sidik–Jonkman (HKSJ) approach was applied to adjust confidence intervals. Heterogeneity was quantified with I2; prespecified subgroup and sensitivity analyses examined PD-L1 strata, ICI class (anti‑PD‑1 vs anti‑PD‑L1), chemotherapy backbone, leave‑one‑out influence, model choice, and exclusion of phase II trials.
Across five metastatic RCTs (n = 3,526), adding an ICI to chemotherapy produced a pooled OS benefit (HR 0.85, 95% CI 0.74–0.96). HKSJ adjustment widened the interval (0.70–1.03). Progression-free survival was also improved (PFS HR 0.82, 95% CI 0.73–0.92). Heterogeneity in metastatic analyses was moderate (I2 values approximately 27%–35%).
The survival advantage was greater in PD-L1+ subgroups (OS HR 0.79, 0.65–0.95; I2 = 51%). The analysis noted numerical differences between anti‑PD‑1 and anti‑PD‑L1 agents favoring anti‑PD‑1 in some comparisons, but these differences were not robust on sensitivity testing.
In the early-stage setting (five RCTs, n = 2,427), ICI-containing regimens improved event-free or invasive disease‑free survival (EFS/iDFS) (HR 0.75, 95% CI 0.63–0.90) and overall survival (OS HR 0.64, 95% CI 0.53–0.78). Pathological complete response (pCR) rates were increased with the addition of ICIs (OR 1.65, 95% CI 1.39–1.98). Heterogeneity in early‑stage analyses was near zero (I2 approximately 0%), indicating consistent effects across included trials.
Notable trial-level findings informing these pooled estimates included KEYNOTE‑522’s improvement in both EFS and final OS and GeparNuevo’s favorable long‑term iDFS and OS at 7‑year follow‑up.
Prespecified subgroup analyses evaluated PD‑L1 expression, ICI class (anti‑PD‑1 vs anti‑PD‑L1), and chemotherapy backbone (taxane‑based vs platinum‑containing vs capecitabine). The chemotherapy backbone emerged as a critical modifier: trials using nab‑paclitaxel (e.g., IMpassion130 and TORCHLIGHT) showed positive signals while the trial substituting solvent‑based paclitaxel (IMpassion131) was negative, motivating backbone‑stratified interpretation.
Sensitivity analyses using leave‑one‑out exclusion, fixed‑effect comparisons, and restriction to phase III evidence maintained the overall pattern of benefit, although some subgroup contrasts (for example anti‑PD‑1 vs anti‑PD‑L1) lost robustness.
Heterogeneity was moderate in metastatic analyses and minimal in early‑stage analyses. Applying HKSJ adjustment broadened confidence intervals around pooled HRs but did not meaningfully change the directionality of effects. Funnel plot inspection was performed; formal Egger testing was not conducted because no outcome had ten or more contributing trials.
The pooled evidence supports a durable survival benefit of ICI + chemotherapy in both metastatic and early‑stage TNBC, with the largest effects observed in PD‑L1‑positive populations. These findings align with current guideline recommendations that incorporate pembrolizumab in selected TNBC settings and reinforce the importance of PD‑L1 assessment in treatment selection. The differential results by chemotherapy backbone underscore that backbone choice can materially affect ICI efficacy. The authors highlight the ongoing need to refine predictive biomarkers to expand benefit to PD‑L1‑negative patients.
The analysis included both phase II and phase III trials; although sensitivity analyses addressed this, heterogeneity in trial designs and chemotherapy regimens remains a limitation. Some subgroup comparisons produced numerical differences that were not robust on sensitivity testing. Trial‑level details not required here (for example patient‑level biomarker distributions or certain safety summaries) were reported in the source tables and supplementary materials; where specific trial subanalyses or individual adverse‑event rates were not summarized in the main text, those details were those were reported in source materials referenced by the authors. Future research priorities include refining backbone selection, improving PD‑L1 and other biomarker stratification, and longer follow‑up to further define magnitude and durability of benefit.