21-gene Recurrence Score and locoregional recurrence in node-positive breast cancer (SWOG 8814 correlative)
ForPostmenopausal ER/PR+ node-positive breast, mastectomy without PMRT
TL;DR10-yr LRR 16.5% vs 9.7% for RS ≥18 vs <18 (P=.02) in ER/PR+ node-positive breast; widest split at 1-3 nodes after mastectomy without RT.
In mastectomy-only pts with 1-3 nodes, 10-yr LRR was 1.5% at RS <18 vs 11.1% at RS ≥18 (P=.051, n=165). That gives the PMRT de-escalation hypothesis TailorRT is testing a biological basis. At ≥4 nodes low RS offered no protection (25.9% vs 27.0%), so a low score is no reason to drop PMRT in pN2a.
In a postmenopausal ER/PR+ pN1 pt after mastectomy with RS ≥18, this argues against treating the score as reassurance when discussing PMRT; it does not support omitting PMRT at RS <18 outside a trial, and does not apply to ≥4 positive nodes.
In mastectomy-only pN1 pts, 10-yr LRR was 1.5% at RS <18 vs 11.1% at RS ≥18 (P=.051), which supports the TailorRT hypothesis and argues for PMRT discussion at higher RS. At ≥4 nodes LRR was 25.9% vs 27.0% irrespective of RS, so pN2a PMRT is unchanged.
10 details
Retrospective translational cohort within phase 3 SWOG S8814, which enrolled 1989-1995. RT use was not randomized: it was expected after BCS and left to physician discretion after mastectomy. Reported under REMARK guidelines.
Postmenopausal, ER/PR+, node-positive women from the tamoxifen-only and CAF-T arms who had RS assayed. Final N=316: BCS + RT or mastectomy without RT. Pts who received PMRT were excluded, so the mastectomy cohort is the group whose physicians judged RT unnecessary.
Tamoxifen 20 mg daily for 5 y, alone or after 6 cycles of CAF (CAF-T). The concurrent CAF + tamoxifen arm was excluded.
Primary: LRR (breast, chest wall, axillary, infraclavicular, supraclavicular or IMN), censored at distant recurrence or death. The primary analysis used continuous RS; the secondary analysis compared RS <18 vs ≥18.
The continuous-RS primary analysis was not significant (P=.11). The categorical split and the subset tables carry the signal.
| Cohort | n | RS 0-17 | RS ≥18 | P |
|---|---|---|---|---|
| All pts | 316 | 9.7% | 16.5% | .02 |
| Mastectomy, no RT | 252 | 7.7% | 16.8% | .03 |
| Mastectomy, no RT, 1-3 nodes | 165 | 1.5% | 11.1% | .051 |
| Mastectomy, no RT, ≥4 nodes | not stated | 25.9% | 27.0% | .27 |
| Variable | HR (95% CI) | P |
|---|---|---|
| Intermediate or high RS | 2.36 (1.02-5.45) | .04 |
| ≥4 nodes | 3.37 (1.69-6.72) | .001 |
| Mastectomy | 1.11 (0.45-2.71) | .82 |
| Chemotherapy | 0.58 (0.29-1.15) | .12 |
Direction agrees with NSABP B-28, where RS also independently predicted LRR. Absolute LRR at 1-3 nodes was higher here than in B-28, which treated every pt with chemo and allowed no PMRT. MA.20 and EORTC 22922 show a nodal RT DFS benefit even where LRR is low, so LRR alone may underestimate what RT buys.
Treatment era: CAF and tamoxifen only, with no taxanes or AIs, both of which may lower baseline LRR. LRR was censored at distant recurrence, which can undercount local failures. The ≥4-node mastectomy subset had no n reported.
RS looks prognostic for LRR beyond nodal count, but it cannot yet select pN1 pts for PMRT omission. That question belongs to TailorRT (MA.39), which randomizes only low-RS pN1 pts. Nodal burden still dominates the risk at ≥4 nodes.
Retrospective correlative of a chemo trial with nonrandomized RT. Prespecified continuous-RS analysis null (P=.11); the cut at 18 was confirmed post hoc; nodal subsets exploratory with few events.
- Is PMRT omission safe in pN1 with low RS?
- Does RS predict LRR under modern systemic therapy?
- Does low LRR risk predict lack of RT DFS benefit?
📚 Sources · 📄 1 paper
The longer read
The appeal of this analysis is that it asks a local-therapy question of a genomic assay built and validated for systemic decisions. The answer is suggestive, not conclusive. Its direction matches NSABP B-28, the other node-positive dataset where RS tracked locoregional failure, and the two cohorts complement each other. B-28 gave everyone chemotherapy and prohibited PMRT, so its mastectomy pts were unselected for local risk. In S8814, physicians could give PMRT to higher-risk pts, and those pts were removed. The mastectomy-alone group here is therefore one that clinicians had already judged low enough risk to skip RT. Even in that group, RS ≥18 identified a 1-3 node subset with 10-yr LRR of 11.1%. That is the most clinically relevant finding: a group many radiation oncologists would consider borderline carried a failure rate that most would treat.
The methodological picture should temper confidence in specific ways. The prespecified primary analysis, RS as a continuous term, did not reach significance. The positive results come from a dichotomy at 18 that the authors chose because of event counts and then confirmed post hoc as the best cut. Optimizing a cut point on the same data inflates the apparent separation. The headline 1-3 node contrast rests on 165 pts with a P value at the threshold, and the low-RS 1.5% figure reflects very few events. A rate that low invites over-reading: it is the estimate most likely to rise on replication. Censoring LRR at distant recurrence also means local failures that came after or alongside distant disease are not counted, which matters in a tamoxifen-era cohort with substantial distant failure.
The era cuts both ways. These pts received CAF or tamoxifen alone, with no taxanes, AIs, or modern nodal staging. Modern systemic therapy likely lowers the absolute LRR in every RS stratum. That could make low-RS omission safer than this cohort suggests, but it could also shrink the gap that makes RS discriminating. Whether RS adds information on top of contemporary pathology and therapy is not something a 1989-1995 cohort can settle.
The ≥4-node result is more robust in practical terms. LRR was about a quarter regardless of RS, so a favorable score does not offset nodal burden. That is consistent with PMRT staying standard for pN2a and aligns with the B-28 high-RS ≥4-node rate.
The last caution comes from the regional RT trials the authors cite. MA.20 and EORTC 22922 showed DFS gains from nodal irradiation despite modest LRR differences, so a low LRR estimate does not by itself show that RT adds nothing. Prognosis for LRR and benefit from RT are different questions. This study speaks only to the first. TailorRT, which randomizes low-RS pN1 pts, is the right test of the second, and this analysis is best read as the rationale for that trial rather than as a reason to act now.