# DESTINY-Breast05: endpoint and lung-toxicity denominator audit

Prepared by **evidence-review-agent** · 22 September 2026 · version 1  
Trial: **NCT04622319** · Primary report: **10.1056/NEJMoa2514661**

DESTINY-Breast05 supports improved invasive disease-free survival (IDFS) with trastuzumab deruxtecan (T-DXd) compared with trastuzumab emtansine (T-DM1) in its enrolled high-risk residual-disease population. Its lung-toxicity findings require three distinctions: randomized versus treated populations, all treated patients versus radiotherapy subgroups, and adjudicated drug-related interstitial lung disease (ILD) versus investigator-reported radiation pneumonitis. The radiotherapy timing comparison was not randomized. [Primary report abstract](https://pubmed.ncbi.nlm.nih.gov/41370739/), [trial design and safety presentation](https://datasourcebydaiichisankyo.com/documents/d/datahub/sabcs-2025_t-dxd_loibl_oral-final).

## Population and endpoint check

The drug comparison was an international, open-label, randomized phase 3 trial. Participants had residual invasive HER2-positive early breast cancer after neoadjuvant treatment and either inoperable disease at presentation or positive axillary nodes after treatment. The registry further specifies ECOG 0–1, at least six neoadjuvant cycles over 16 weeks, and at least nine weeks each of taxane chemotherapy and trastuzumab, with or without pertuzumab. T1N0 presentation was excluded. These results should not be generalized automatically to every residual-disease population. [Registry eligibility](https://clinicaltrials.gov/study/NCT04622319).

At the **2 July 2025 cutoff**, median follow-up was approximately 30 months. Efficacy used all randomized participants: 818 T-DXd and 817 T-DM1. [Primary abstract](https://pubmed.ncbi.nlm.nih.gov/41370739/), [SABCS slide 3](https://datasourcebydaiichisankyo.com/documents/d/datahub/sabcs-2025_t-dxd_loibl_oral-final).

| Endpoint | T-DXd | T-DM1 | Reported effect |
|---|---:|---:|---|
| Primary IDFS events | 51/818 (6.2%) | 102/817 (12.5%) | HR 0.47; 95% CI 0.34–0.66 |
| Three-year IDFS estimate | 92.4% | 83.7% | Derived difference: **8.7 percentage points** |
| Key-secondary DFS events | 52/818 (6.4%) | 103/817 (12.6%) | HR 0.47; 95% CI 0.34–0.66 |
| Three-year DFS estimate | 92.3% | 83.5% | Derived difference: **8.8 percentage points** |

IDFS concerns invasive recurrence or death; DFS additionally includes noninvasive breast cancers and second primary nonbreast cancers. The endpoints share a reported hazard ratio but are not interchangeable. The abstract reports P<0.001 for each; FDA Table 27 gives P<0.0001 and separate interim significance thresholds. These are compatible reporting precisions. Three-year estimates must not be reconstructed as one minus the crude event proportions, because follow-up and censoring matter. [Abstract](https://pubmed.ncbi.nlm.nih.gov/41370739/), [FDA label, Table 27, PDF page 46](https://www.accessdata.fda.gov/drugsatfda_docs/label/2026/761139s041s043lbl.pdf).

The regulatory report records 47 deaths across both arms at this analysis (2.9% of 1,635). This extraction does not establish an overall-survival benefit. [FDA label, PDF page 46](https://www.accessdata.fda.gov/drugsatfda_docs/label/2026/761139s041s043lbl.pdf).

## Lung-toxicity denominator check

The safety population received at least one study dose: **806 T-DXd and 801 T-DM1**. [SABCS slide 9, header and footnote a](https://datasourcebydaiichisankyo.com/documents/d/datahub/sabcs-2025_t-dxd_loibl_oral-final).

| Adjudicated drug-related ILD | T-DXd | T-DM1 |
|---|---:|---:|
| All treated patients | **77/806 (9.6%)** | **13/801 (1.6%)** |
| Sequential radiotherapy | 34/319 (10.7%) | 7/270 (2.6%) |
| Concurrent radiotherapy | 42/438 (9.6%) | 5/480 (1.0%) |
| Any radiotherapy | 76/757 (10.0%) | 12/750 (1.6%) |

Two T-DXd ILD cases were fatal: **2/806 = 0.2%** after rounding in the overall safety population, and **2/757 = 0.3%** in the radiotherapy subgroup. Those percentages describe the same two deaths using different denominators. [SABCS slide 9 and footnote b](https://datasourcebydaiichisankyo.com/documents/d/datahub/sabcs-2025_t-dxd_loibl_oral-final).

Radiotherapy timing was selected by investigators. Our methodological conclusion is that these descriptive subgroup rates cannot prove equivalent timing safety or absence of a causal timing effect. Drug randomization does not randomize radiotherapy timing. [SABCS slide 4](https://datasourcebydaiichisankyo.com/documents/d/datahub/sabcs-2025_t-dxd_loibl_oral-final), [FDA label, PDF page 45](https://www.accessdata.fda.gov/drugsatfda_docs/label/2026/761139s041s043lbl.pdf).

Investigator-reported radiation pneumonitis is a separate grouped measure: 238/757 (31.4%) and 229/750 (30.5%) among radiotherapy recipients. The later ASCO tables also express these counts over the full safety populations as 29.5% and 28.6%. These are denominator changes, not additional cases. Do not sum these counts with adjudicated drug-related ILD without information about patient overlap. [SABCS slide 10](https://datasourcebydaiichisankyo.com/documents/d/datahub/sabcs-2025_t-dxd_loibl_oral-final), [ASCO secondary safety presentation](https://datasourcebydaiichisankyo.com/documents/d/datahub/asco-2026_t-dxd_untch_oral-final).

**The ASCO 2026 presentation retains the 2 July 2025 cutoff** on slide 3. Its later presentation date does not establish longer follow-up. The trial also used scheduled chest CT and pulmonary eligibility exclusions; interpretation outside those conditions requires care. [ASCO slides 3–4](https://datasourcebydaiichisankyo.com/documents/d/datahub/asco-2026_t-dxd_untch_oral-final), [registry](https://clinicaltrials.gov/study/NCT04622319).

## Correction, access and reproducibility

An April 2026 correction restores a garbled Discussion passage concerning IHC/ISH subgroup activity and distant recurrence. The inspected publisher search-index text does not announce a revised primary IDFS result. The correction's identity is independently indexed by PubMed; the full publisher correction page was not retrieved. [Publisher correction](https://www.nejm.org/doi/full/10.1056/NEJMx260006), [PubMed correction record](https://pubmed.ncbi.nlm.nih.gov/41931063/).

This is an extraction from the primary abstract, complete investigator congress PDFs, registry and regulatory report. **The publisher's full trial article, protocol and statistical appendix were not retrieved.** The study was funded by Daiichi Sankyo and AstraZeneca; its sponsor-hosted presentations are further reports of the same study, not independent replication. No patient-level analysis, hazard-ratio reconstruction or full risk-of-bias assessment was performed. No individual treatment recommendation is made.

The [structured evidence](evidence.json) separates source findings, calculations and interpretations. [verify.py](verify.py) checks arithmetic and population reconciliation; with `--source-dir` it also checks the five retained source-file hashes. [Verification results](verification-results.json) record the executed checks. Copyrighted reports are linked, not redistributed. Hashes establish byte integrity, not scientific validity.

```sh
python3 verify.py
python3 verify.py --source-dir /path/to/retained-source-files
```