onc brain

The longer read

The value of this analysis is not that tamoxifen reduces ipsilateral recurrence in DCIS, which has been the working assumption for two decades, but that it puts a long-horizon number on that reduction in the exact population where the field's current argument sits: good-risk DCIS in which the clinician has already decided to omit radiotherapy. NRG/RTOG 9804 defined that population prospectively and randomized it to RT or observation. By taking the observation arm of 9804 and welding it to E5194's identically-defined good-risk cohort, the authors get 878 patients followed a median of 14.85 years, which is long enough for DCIS recurrence to actually declare itself. That follow-up is the analysis's real asset, and it is why the 15-year estimates of 11.4% versus 19.0% deserve more weight than the effect size alone would earn.

The design, however, will not carry a causal claim, and the paper is explicit that this is exploratory. Tamoxifen was optional in both parent trials, which means the exposure was chosen rather than assigned, and the authors' own baseline table shows the choice was not random. Patients who took tamoxifen were far more likely to have come from 9804 (55.0% vs 21.8%), to have had a negative re-excision (27.2% vs 10.6%), and to have pathologic size at or under 5 mm (57.0% vs 41.3%). Every one of those imbalances points the same direction, toward lower baseline recurrence risk in the treated group. Multivariable Fine-Gray adjustment can attenuate that, and the fact that HR 0.54 survives adjustment is reassuring, but adjustment can only handle the confounders that were measured. Trial of origin in particular is doing double duty here, standing in for era, referral pattern, and surveillance intensity as much as for anything the model can name.

The more interesting internal signal is the dissociation between endpoint types. Tamoxifen was associated with reduced invasive IBR (HR 0.43, P=.0042) but not with reduced DCIS-IBR (P=.089). If that separation is real, it changes how the drug should be framed to a patient: not as a way to avoid a second lumpectomy, but as a way to lower the odds that the next event is one that requires systemic staging and treatment. That is a different value proposition, and a more defensible one. The caution is that with 52 DCIS events, a null P of .089 is as consistent with insufficient power as with a genuinely absent effect, and the paper does not settle which.

For a radiation oncologist the practical read is comparative rather than substitutive. Nothing here compares tamoxifen against radiotherapy, and the design cannot: every patient in the pooled cohort went without RT. What the analysis does supply is the counterfactual arm's magnitude, so that the conversation about omitting RT in good-risk DCIS can be conducted with two numbers instead of one. A patient declining radiotherapy is not thereby committed to accepting a 19% fifteen-year recurrence risk, and a patient declining both should understand which of the two curves they are on. For that to be wrong, the confounding would have to be strong enough to manufacture a near-halving of invasive recurrence out of a 16-point difference in the proportion with tumors at or under 5 mm, which is possible but not the most economical explanation of the data.