SWOG/NRG S1914 NCT04214262
ForT1-3N0M0 NSCLC ≤7cm, medically inoperable or declined surgery, ≥1 risk factor
HR 1.15
95% CI 0.65-2.01, p=0.63; primary endpoint not met
TL;DROS HR 1.15 (0.65-2.01), p=0.63: atezolizumab added to SBRT failed, with more local failures (13% vs 7%) and G≥3 AEs (12% vs 2%).
The RT read is that adding IO did not just fail to help, it tracked with MORE local failure (13% vs 7%), inside a BED ≥100 Gy 3-8 fraction regimen that is already delivering >90% in-field control. Central review of those events is pending, so treat the local signal as unconfirmed. This closes the question of routinely sequencing atezolizumab around definitive SBRT off-protocol.
In medically inoperable T1-3N0M0 NSCLC with high-risk features (size ≥2 cm, SUV ≥6.2, or poorly differentiated histology), this supports SBRT alone at BED ≥100 Gy without added checkpoint blockade; it does not address IO for node-positive or operable early-stage disease.
SBRT alone at BED ≥100 Gy in 3-8 fractions remains the standard, and adding atezolizumab tracked with MORE local failure (13% vs 7%) rather than better in-field durability. Central review of those events is pending. Regional (2% vs 3%) and distant (4% vs 5%) failure were unchanged, so the out-of-field rationale also went unrewarded.
Neoadjuvant/concurrent/adjuvant atezolizumab 1200 mg Q3W × 8 cycles bought no OS or PFS benefit and cost a six-fold rise in G≥3 AEs (12% vs 2%) including a G5 respiratory failure. This argues against porting PACIFIC-style consolidation logic into node-negative early-stage disease pending the PD-L1 analysis.
11 details
Randomized phase III SWOG/NRG cooperative-group trial, accrual 8/13/20 to 9/6/24, 417 randomized / 403 eligible (201 SBRT alone, 202 atezolizumab + SBRT) against an accrual goal of 432. Closed at the first interim analysis for futility on both OS and PFS. Median follow-up in living pts was 12 months (range 0.03-49).
T1-3N0M0 NSCLC ≤7 cm, medically inoperable or declined surgery, with ≥1 recurrence risk factor: tumor diameter ≥2 cm, ≥6.2 (SUV), or moderately/poorly/undifferentiated histology. Median age 73 (41-91), 89% ECOG 0-1, median tumor diameter 2.3 cm. Stratified by location (central vs peripheral), size (<4 vs ≥4 cm) and PS.
SBRT in 3-8 fractions to a BED ≥100 Gy in both arms, i.e. standard definitive dosing rather than a de-escalated backbone. In the experimental arm SBRT began with cycle 3, so radiation was delivered after two neoadjuvant atezolizumab cycles and concurrently with the third.
Atezolizumab 1200 mg IV Q3W for 8 cycles, given neoadjuvantly, concurrently and adjuvantly around SBRT. No protocol treatment was received by 6 pts on S and 8 on AS.
Primary: overall survival, compared by 1-sided stratified log-rank at the 2.5% level. Secondary: PFS, failure patterns, toxicity, QoL.
Neither OS nor PFS favored the IO arm, and local failure ran higher with atezolizumab. See the failure-pattern table and primary endpoint above.
| Failure site | SBRT alone | Atezo + SBRT |
|---|---|---|
| Local | 7% | 13% |
| Regional | 2% | 3% |
| Distant | 4% | 5% |
G≥3 AEs 12% on AS vs 2% on S (21 G3, 1 G4, and 1 G5 respiratory failure with atezolizumab; 3 G3 and 1 G4 with SBRT alone). A six-fold excess of high-grade toxicity with no efficacy return is the safety read.
The prior randomized phase II (PMID 37478883) suggested benefit from adding immunotherapy to SBRT; this larger phase III does not reproduce it. It also sits against PACIFIC-era logic, where consolidation IO helps after chemoRT for stage III, and shows that result does not port down to node-negative disease treated with ablative SBRT.
Follow-up is short (median 12 mo alive) and central review of local recurrence events is not complete, so the local-failure imbalance is provisional. The smoker subgroup harm signal was not multiplicity-controlled, and PD-L1 status, QoL and correlative blood/tissue analyses are all still pending.
A negative primary endpoint with a directionally worse PFS, worse local control and six-fold more high-grade toxicity is stronger than a null result: it argues against the abscopal/radiosensitization rationale in this setting. It does not exclude benefit in a biomarker-selected subset, which the pending PD-L1 analysis will test.
CONSORT flow
Phase III, primary OS endpoint not met, closed at interim for futility. Reaffirms SBRT alone as SoC and refutes the earlier phase II IO signal.
- Does a PD-L1-defined subset benefit from IO added to SBRT?
- Will central review confirm the excess local failures with atezolizumab?
- Is the worse outcome in former/never smokers real or chance?
📚 Sources · 📄 1 paper
Abstract
The longer read
The value of this trial lies less in its negative primary endpoint than in the direction of everything around it. A truly null result would show overlapping curves and matched failure patterns. Instead OS, PFS, and local control all point the same way, against the experimental arm, with an OS HR of 1.15 and a PFS HR of 1.35, and with local failure at 13% on atezolizumab versus 7% on SBRT alone. None of those individual comparisons is statistically convincing on its own, and the confidence intervals are wide enough to include real benefit. But the consistency of direction across independent endpoints, combined with a futility stop that the design triggered on both OS and PFS, is what should move a reader's confidence, and it should move it toward believing there is no benefit here rather than toward believing the trial was merely underpowered.
The local-failure imbalance deserves the most scrutiny, because it is the one finding that contradicts a mechanistic prior rather than merely failing to confirm one. The rationale for combining checkpoint blockade with ablative radiation rests on the idea that RT-induced antigen release plus PD-L1 blockade improves both in-field durability and out-of-field control. This trial found the opposite on the in-field endpoint and nothing at all on the out-of-field ones: regional failures were 2% versus 3% and distant failures 4% versus 5%, essentially superimposable. Central review of the local recurrence events is still running, and that matters, because immunotherapy-associated pneumonitis and post-SBRT radiographic change are both easy to mistake for local progression on imaging, and both were more likely in the atezolizumab arm. If central review reclassifies a meaningful share of those events, the local-control signal softens considerably. Until it reports, this should be read as an unexplained imbalance, not an established harm.
The result also has to be reconciled with the earlier randomized phase II that motivated it, and with the broader stage III experience where consolidation immunotherapy after chemoradiation is established practice. The most parsimonious reading is that the setting, not the drug, is what changed. Early-stage node-negative disease treated to a BED of at least 100 Gy already achieves in-field control above 90%, so the ceiling on any local benefit is low, and the residual failure risk is largely distant and biologically different from bulky nodal disease. A therapy that adds value when the local problem is unsolved need not add value when it is mostly solved. That framing also predicts that the smoker subgroup finding, where former and never smokers did worse on the combination with OS HR 2.50 and PFS HR 2.16, is more likely noise from an unadjusted subgroup analysis in a small event pool than a real biological interaction, though it is uncomfortable enough that the pending PD-L1 correlative work is worth waiting for.
What would have to be true for this conclusion to be wrong is a specific and testable set of conditions: that the OS comparison is simply immature at a median 12 months of follow-up in living pts and that late immunologic benefit emerges, or that a PD-L1-defined subgroup carries a real effect diluted by an unselected population. Both are plausible. Neither justifies giving checkpoint blockade around definitive SBRT off-protocol, particularly against a G≥3 AE rate of 12% versus 2% and a treatment-related death.