KEYNOTE-975 NCT04210115
ForUntreated unresectable locally advanced ESCC/EAC/GEJ, dCRT candidates
HR 0.97
95% CI 0.77-1.23, P=0.3994; not met (boundary 0.0193)
TL;DREFS HR 0.97 (95% CI 0.77-1.23), median 15.2 vs 14.7 mo in PD-L1 CPS ≥10: no EFS gain adding concurrent pembro to dCRT in unresectable esophageal/GEJ cancer.
The trial failed in the enriched PD-L1 CPS ≥10 group, where IO was most likely to help, and the curves sat on top of each other (EFS HR 0.97, OS HR 1.00) at 47.1 mo median f/u. Grade 5 AEs ran 13% vs 10%, so concurrent pembro with 50-60 Gy dCRT adds toxicity without benefit. This leaves dCRT alone as the backbone.
In untreated unresectable esophageal or GEJ cancer going to dCRT, including PD-L1 CPS ≥10, this does not support concurrent pembrolizumab; it does not address consolidation-only IO after dCRT or the resectable trimodality setting.
Concurrent pembro with 50-60 Gy dCRT did not move EFS (HR 0.97) or OS (HR 1.00) in CPS ≥10 pts, and grade 5 AEs rose to 13% vs 10%. No local control or dose-stratified data in source, so the RT backbone stays dCRT alone with no IO-driven change to dose or volume.
PD-L1 CPS ≥10 enrichment did not rescue concurrent pembro (EFS HR 0.97, OS HR 1.00), and AE discontinuation rose to 34% vs 27%. Biomarker selection does not justify adding PD-1 blockade to dCRT; consolidation-only timing remains untested by this trial.
9 details
International phase 3, randomized 1:1, double-blind, placebo-controlled, multicenter; N=703. Results are from the second interim analysis (IA2), after which the study was stopped. Median follow-up 47.1 mo.
Previously untreated locally advanced unresectable ESCC, EAC or GEJ cancer (cTX N+ M0 or cT2-T4a NX M0), or cervical/upper thoracic EC with supraclavicular nodal M1 only. ECOG 0-1, ineligible for curative surgery but suitable for dCRT.
Pembro 200 mg or placebo Q3W for 8 cycles, then 400 mg or placebo Q6W for 5 cycles, concurrent with dCRT. Chemo backbone FP or FOLFOX by site choice.
50 Gy or 60 Gy with FP, or 50 Gy with FOLFOX, by site choice. Fractionation, technique and target volume not reported in source.
Dual primary: investigator-assessed EFS and OS, tested hierarchically in PD-L1 CPS ≥10, CPS ≥1, then all pts.
EFS not improved in CPS ≥10 (arm-level numbers in table), so every downstream comparison went untested and the study was stopped.
| Endpoint (CPS ≥10) | Pembro + dCRT | Pbo + dCRT | HR (95% CI), P |
|---|---|---|---|
| Median EFS (investigator) | 15.2 mo | 14.7 mo | 0.97 (0.77-1.23), P=0.3994; boundary 0.0193 |
| Median OS | 32.2 mo | 35.0 mo | 1.00 (0.77-1.30), nominal P=0.5003 |
| AE | Pembro + dCRT | Pbo + dCRT |
|---|---|---|
| Grade ≥3 | 100% | 100% |
| Grade 5 | 44 (13%) | 37 (10%) |
| Discontinuation for any AE | 117 (34%) | 97 (27%) |
Grade ≥3 AEs in 100% of both arms. Grade 5 AEs 44 (13%) vs 37 (10%), AE discontinuation 117 (34%) vs 97 (27%). Authors report no new safety signals.
Joins KUNLUN (durvalumab + dCRT) as a negative concurrent IO-dCRT trial in esophageal cancer. Adjuvant nivolumab per CheckMate 577 belongs to the resected trimodality setting and is untouched by this result.
RT dose and chemo backbone by site choice with no outcome breakdown, pooled histologies, and no local control or failure-pattern data in source to show whether IO changed in-field control.
A flat EFS and OS in the PD-L1-enriched group argues against a hidden benefit in the untested CPS ≥1 or all-comer groups. Whether timing (consolidation vs concurrent) or histology matters stays open.
CONSORT flow
Double-blind phase 3 stopped at IA2 after the gatekeeper EFS endpoint failed in the enriched CPS ≥10 group; reaffirms dCRT alone as standard. Downstream populations formally untested.
- Would consolidation-only IO after dCRT differ from concurrent dosing?
- Histology-specific EFS in ESCC vs EAC/GEJ
- Did pembro alter local control or failure patterns?
📚 Sources · 📄 1 paper
Abstract
The longer read
The most informative thing about KEYNOTE-975 is where it failed. The sponsor put the PD-L1 CPS ≥10 population first in the testing hierarchy, the group where checkpoint inhibition had the best chance of showing benefit, and the EFS hazard ratio was 0.97 with an interval reaching from 0.77 to 1.23. OS in the same group was HR 1.00. Formally, the CPS ≥1 and all-comer comparisons were never tested. In practice, a null result in the enriched group makes a real benefit in a diluted population hard to believe. With median follow-up of 47.1 months, this is no premature look that could still turn: it is a mature negative read, stopped at the second interim analysis.
It also fits what the field already had. KUNLUN tested durvalumab with dCRT in the same broad setting and also failed to show an EFS benefit, so two placebo-controlled programs with different PD-(L)1 agents now agree. The hope behind both came partly from PACIFIC in stage III NSCLC, but PACIFIC gave durvalumab as consolidation after chemoradiation was finished. Here pembrolizumab started with dCRT and continued alongside it. KEYNOTE-975 therefore answers the concurrent question and leaves the sequencing question open. A reader should not assume consolidation would behave the same way, and should also not assume it would succeed where concurrent dosing did not. No randomized evidence in this source speaks to it.
For a radiation oncologist, the RT-specific gaps matter. Dose (50 or 60 Gy) and backbone (FP or FOLFOX) were left to each site, and the source gives no results by dose, no local control, and no in-field versus distant failure pattern. One mechanistic hope for adding IO to RT is better locoregional control through immune priming, and this abstract cannot confirm or refute that. A trial can be EFS-negative and still move local control, and we cannot tell here. The pooling of squamous and adenocarcinoma histologies is a second blind spot, since these behave differently and respond differently to IO in other settings. A benefit restricted to ESCC with high PD-L1, offset by none elsewhere, would look much like this, though the CPS ≥10 result makes a large hidden effect unlikely.
The safety numbers deserve more weight than the authors' "no new safety signals" suggests. Grade 5 events were 44 (13%) versus 37 (10%), and AE-driven discontinuation was 117 (34%) versus 97 (27%). Given no efficacy return, that is a cost with no benefit attached. A grade ≥3 rate of 100% in both arms is unusual and likely reflects how events were counted over a long follow-up, which the full paper should clarify.
The decision it settles: dCRT without concurrent PD-1 blockade stays the standard for unresectable esophageal and GEJ cancer, regardless of PD-L1 status. The decisions it leaves open: consolidation timing, histology-specific effects, and whether any RT parameter interacts with IO. The result would only be wrong if a benefit sat in a subgroup this hierarchy never tested and the CPS ≥10 cut failed to enrich for it, and the data give little reason to expect that.