← Clinical Trials
HER2-low Breast Cancer Phase 3 Time-to-Event Analysis NCT03734029

DESTINY-Breast04: Complete Statistical Analysis of Trastuzumab Deruxtecan in HER2-low Breast Cancer

An independent statistical analysis of the randomized phase 3 DESTINY-Breast04 trial comparing trastuzumab deruxtecan (DS-8201a) with investigator's choice for HER2-low breast cancer that has spread or cannot be surgically removed.

Trial start: 2018-12-27  ·  Primary completion: 2022-01-11  ·  Enrollment: 557
Scope of this record

This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record.

1. Trial at a Glance

DESTINY-Breast04 was a randomized, parallel, open-label phase 3 trial evaluating trastuzumab deruxtecan (DS-8201a) versus investigator's choice in participants with HER2-low breast cancer that has spread or cannot be surgically removed. The registry reports a primary time-to-event endpoint analyzed with a log-rank test and a hazard ratio under a superiority hypothesis.

557
Enrollment
2 randomized arms
0.5085
PFS Hazard Ratio
T-DXd vs Physician's Choice
0.4012–0.6444
95% CI
Two-sided
<0.0001
P-value
Superiority analysis
FeatureDESTINY-Breast04
Trial nameDESTINY-Breast04
NCT IDNCT03734029
PhasePhase 3
StatusActive, not recruiting
ConditionBreast Cancer
PopulationHER2-low breast cancer that has spread or cannot be surgically removed
AllocationRandomized
Design modelParallel
MaskingNone
Primary purposeTreatment
Enrollment557
Lead sponsorDaiichi Sankyo
Sponsor typeIndustry
Primary endpoint typeTime-to-event
Results postedYes
Outcome measures posted12
Statistical analyses posted1

2. Clinical Question

The central statistical question is whether trastuzumab deruxtecan produces a different time-to-event outcome than physician's choice in participants with HER2-low breast cancer, using the registered progression-free survival endpoint in the hormone receptor-positive cohort.

Population

Participants with HER2-low breast cancer that has spread or cannot be surgically removed.

Intervention

Trastuzumab deruxtecan (DS-8201a).

Comparator

Physician's Choice, with the registry listing capecitabine, eribulin, gemcitabine, paclitaxel, and nab-paclitaxel among the interventions.

Primary question

Does trastuzumab deruxtecan improve progression-free survival relative to physician's choice in the hormone receptor-positive cohort?

3. Trial Design

01
Randomize 557 participants
02
Parallel arms 2 treatment groups
03
Open-label No masking
04
Follow-up Time-to-event assessment
05
Primary analysis Log-rank and hazard ratio
ARM 1

Trastuzumab deruxtecan

  • Trastuzumab deruxtecan (DS-8201a)
  • Randomized treatment group
  • Compared with physician's choice
ARM 2

Physician's Choice

  • Investigator-selected treatment
  • Registry-listed options include capecitabine
  • Eribulin, gemcitabine, paclitaxel, and nab-paclitaxel are also listed

The design is randomized and parallel, with no masking. Randomization is important statistically because it establishes treatment assignment before subsequent outcomes are observed. The open-label structure does not remove the value of randomization, but it is a feature that should be kept in mind when considering outcomes that could be influenced by knowledge of treatment assignment.

4. Primary Endpoint

EndpointRegistry definition / time framePrimary analysis
Progression-free Survival (PFS) Based on Blinded Independent Central Review (BICR) in the Hormone Receptor-Positive Cohort in Participants With HER2-low Breast Cancer From the date of randomization to the earliest date of the first objective documentation of radiographic disease progression or death due to any cause, up to approximately 3 years Log-rank test with Cox proportional hazard effect measure

Endpoint definition

The registry defines progression-free survival as being assessed from the date of randomization to the date of the first radiographic disease progression or death due to any cause, whichever came first. The definition specifies at least a 20% increase in the sum of diameters of target lesions and assessment by blinded independent central review in the hormone receptor-positive cohort according to modified Response Evaluation criteria.

Registry-specific analysis population: The posted formal analysis states that progression-free survival was assessed in the hormone receptor-positive cohort of the Full Analysis Set. The analysis therefore should not be interpreted as a result for every participant enrolled in the trial without qualification.

5. Statistical Methodology

Time-to-event analysis

Progression-free survival is a time-to-event endpoint. Rather than asking only whether a participant eventually experienced progression or death, the analysis incorporates the timing of the event. The time origin specified in the registry is randomization, and the event is the first qualifying radiographic progression or death due to any cause.

Time-to-event methods are particularly useful when participants have different lengths of follow-up or when some participants have not experienced the event by the time of analysis. Those participants contribute information up to the point at which their outcome becomes censored.

Log-rank test

The registry reports a log-rank analysis for the primary endpoint. The log-rank test is designed to compare survival-type event-time distributions between treatment groups while accounting for the timing of observed events and censoring.

Conceptual hypothesis
H0: no difference in the event-time distributions    vs    H1: a difference exists

For this trial, the registry identifies the hypothesis type as superiority. The reported p-value is therefore evidence evaluated within a superiority framework rather than a non-inferiority framework.

Cox proportional-hazards effect measure

The registry reports the effect measure as a Cox proportional hazard and normalizes this to a hazard ratio. The estimated hazard ratio compares the instantaneous event rate between the trastuzumab deruxtecan and physician's-choice groups under the fitted Cox model.

Hazard-ratio interpretation
HR = 0.5085  →  estimated instantaneous event rate is approximately 50.85% of the comparator rate

Equivalently, 1 − 0.5085 = 0.4915, so the estimate corresponds to an approximately 49.15% lower estimated hazard in the trastuzumab deruxtecan group relative to physician's choice, under the Cox model.

Intention-to-treat principle

The statistical analysis text includes the concept of intention-to-treat analysis. The Full Analysis Set was used for the posted primary PFS analysis in the hormone receptor-positive cohort. The key principle is that randomized treatment assignment provides the basis for the comparison rather than selectively analyzing only participants who completed treatment according to plan.

Confidence intervals

The reported effect estimate is accompanied by a two-sided 95% confidence interval. A confidence interval is essential because the hazard ratio of 0.5085 is an estimate rather than a directly observed population parameter. The interval describes the statistical uncertainty around that estimate under the model and sampling framework.

6. Primary PFS Result

The registry contains one posted formal statistical analysis for the primary endpoint. It compares trastuzumab deruxtecan with physician's choice in the hormone receptor-positive cohort of the Full Analysis Set.

Progression-free survival hazard ratio

0.5085

95% CI: 0.4012–0.6444   ·   P < 0.0001

Analysis: log-rank test   ·   Effect measure: Cox proportional hazard   ·   Superiority hypothesis

Primary endpointTrastuzumab deruxtecan vs physician's choiceStatistical analysis
Progression-free survival in the hormone receptor-positive cohort HR 0.5085 95% CI 0.4012–0.6444; P < 0.0001
Clinical Biostats interpretation

The hazard ratio of 0.5085 indicates that, under the reported Cox proportional-hazards model, the estimated instantaneous rate of the PFS event was approximately 50.85% of the corresponding rate in the physician's-choice group. Expressed as a relative reduction in the estimated hazard, this is approximately 49.15%.

The hazard ratio does not mean that 49.15% of participants avoided progression, that 49.15% of participants were cured, or that every participant experienced exactly a 49.15% reduction in personal risk. It is a relative time-to-event measure describing the treatment comparison under the fitted model.

The two-sided 95% confidence interval of 0.4012–0.6444 indicates the precision of the estimated hazard ratio. The entire interval is below 1, so the range of hazard-ratio values supported by this confidence interval remains below the null value of 1.

The p-value of <0.0001 addresses the statistical evidence against the null hypothesis in the reported superiority analysis. It does not measure the size or clinical importance of the treatment effect. Effect size is conveyed by the hazard ratio and its confidence interval.

Because the effect measure is a Cox proportional-hazards estimate, its usual interpretation relies on the proportional-hazards framework. A single hazard ratio can be an incomplete description if the relative hazards change substantially over time. The ClinicalTrials.gov record does not provide enough information to evaluate that assumption directly.

7. Understanding the Primary Analysis Population

The posted analysis was not described simply as an analysis of all 557 enrolled participants. Instead, the analysis population is specified as the hormone receptor-positive cohort of the Full Analysis Set.

Why the population matters

A treatment-effect estimate applies to the population actually analyzed. Changing the analysis population can change both the estimate and its precision.

Why the denominator matters

The overall enrollment of 557 describes the trial as a whole, but it should not automatically be treated as the sample size underlying the posted PFS estimate.

This distinction is important when reading clinical-trial results. A trial can have one overall enrollment number while a particular endpoint is analyzed in a prespecified subgroup or cohort. The appropriate interpretation therefore begins with the endpoint definition and analysis population, not the enrollment number alone.

8. What the Hazard Ratio Does — and Does Not — Mean

Relative effect

An HR of 0.5085 means the estimated instantaneous PFS event rate under the Cox model was about half that of physician's choice. The calculation 1 − 0.5085 = 0.4915 gives an approximately 49.15% lower estimated hazard.

Not an absolute probability

The hazard ratio is not a probability that a participant will progress, nor is it the percentage of participants who benefit. Absolute PFS probabilities at specified times would answer a different question, and those estimates are not included in the ClinicalTrials.gov record.

Not a median survival measure

The ClinicalTrials.gov record does not report a median PFS estimate. The hazard ratio therefore should not be translated into a median-time difference or a particular number of additional months without an appropriate time-to-event summary.

Precision

The 95% confidence interval of 0.4012–0.6444 shows the uncertainty around the estimated treatment effect. A narrower interval generally indicates greater statistical precision than a wider interval, but the width itself does not determine clinical importance.

9. Statistical Methods Explained

Why was a log-rank test used?

The primary endpoint is progression-free survival, which records the time from randomization until progression or death. The log-rank test is designed for comparing time-to-event distributions while incorporating the timing of events and censoring. It is therefore aligned with the structure of the registered endpoint.

What does a hazard ratio of 0.5085 mean?

A hazard ratio of 0.5085 means that the estimated instantaneous rate of the PFS event in the trastuzumab deruxtecan group was 0.5085 times the corresponding estimated rate in the physician's-choice group under the Cox model. The complementary calculation, 1 − 0.5085, gives an approximately 49.15% lower estimated hazard.

Why is the confidence interval important?

The point estimate alone does not communicate how precisely the treatment effect was estimated. The 95% confidence interval of 0.4012–0.6444 supplies a range reflecting statistical uncertainty around the hazard-ratio estimate. It should be read together with the point estimate rather than treated as a range of outcomes that individual patients will experience.

Why does the p-value not measure treatment-effect size?

The p-value describes how compatible the observed data are with the specified null hypothesis under the statistical testing framework. It is affected by both the size of the observed effect and the amount of information available. The hazard ratio and confidence interval are therefore needed to describe the magnitude and precision of the treatment comparison.

Why does the analysis population matter?

The posted formal analysis applies to the hormone receptor-positive cohort of the Full Analysis Set. An estimate calculated in that cohort should not automatically be generalized to all 557 enrolled participants. The population attached to an estimate is part of the result itself.

Why does censoring matter in PFS?

Not every participant necessarily has a documented progression or death at the time of analysis. Time-to-event methods allow such participants to contribute information until the point at which their event status is no longer observed. This is one reason Kaplan-Meier and related survival methods are preferable to simply calculating the proportion of participants who experienced progression.

10. Statistical Interpretation of the P-value

The reported primary analysis has a p-value of <0.0001 under a superiority hypothesis. In the context of the reported log-rank test, this provides strong statistical evidence against a null comparison of the event-time distributions.

Do not reverse the interpretation: a very small p-value does not mean there is a <0.01% probability that the null hypothesis is true, nor does it mean that the treatment has a <0.01% chance of failing. Those interpretations are not what a frequentist p-value represents.

The appropriate reading is narrower: assuming the statistical model and null hypothesis used for the test, the observed data are highly incompatible with the null comparison. The magnitude of the estimated treatment effect is instead described by the hazard ratio, while its statistical precision is described by the confidence interval.

11. Safety Results

The ClinicalTrials.gov record includes serious adverse events by randomized arm. These data are separate from the PFS efficacy analysis and should not be combined with the hazard ratio into a single measure of overall treatment effect.

Safety measureTrastuzumab deruxtecan (T-DXd)Physician's Choice
Serious adverse events103 / 371 affected / at risk43 / 172 affected / at risk
Serious adverse events: affected / at risk
T-DXd
103 / 371
Physician's Choice
43 / 172

The reported safety counts should be interpreted using their stated denominators. They describe affected participants among those at risk in each arm; they are not the same statistical quantity as the primary PFS hazard ratio.

12. Trial Timeline

2018-12-27

Trial start

The registered trial start date was December 27, 2018.

2022-01-11

Primary completion

The registered primary completion date was January 11, 2022.

Current registry status

Active, not recruiting

The ClinicalTrials.gov record identifies the study status as ACTIVE_NOT_RECRUITING.

13. Design Features That Affect Statistical Interpretation

Design featureWhat the registry data supportStatistical implication
Randomization Allocation is randomized. Randomization provides the basis for comparing outcomes between treatment assignments.
Parallel design The design model is parallel. Participants are compared across concurrent randomized treatment groups rather than through a crossover or within-person comparison.
Masking Masking is none. The trial is open-label, so treatment assignment is not masked.
Primary endpoint One registered primary endpoint is listed and is time-to-event. The primary statistical analysis is appropriately framed around time-to-event methods.
Hypothesis type Superiority. The analysis asks whether the treatment comparison demonstrates superiority rather than satisfying a non-inferiority margin.
Primary test Log-rank test. The reported p-value is associated with a survival-analysis comparison.
Effect measure Hazard ratio from a Cox proportional-hazards analysis. The effect estimate is relative and model-based rather than an absolute risk difference.

The ClinicalTrials.gov record does not provide a non-inferiority margin, crossover information, factorial structure, multiplicity procedure, interim-analysis boundary, missing-data imputation method, Bayesian analysis, or stratification factors. Those design features are therefore not inferred here.

14. Limitations and Interpretation Issues

15. Why This Trial Matters Statistically

DESTINY-Breast04 is a useful teaching case because its primary result illustrates the central structure of a modern randomized time-to-event analysis: randomization establishes the treatment comparison, PFS provides the event-time outcome, the log-rank test supplies the reported hypothesis test, and the Cox model expresses the treatment effect as a hazard ratio with a confidence interval.

ConceptHow it appears in DESTINY-Breast04
Randomization Participants were allocated using a randomized design.
Intention-to-treat analysis The posted analysis includes the intention-to-treat concept and uses the Full Analysis Set for the specified cohort.
Time-to-event endpoint PFS is measured from randomization to radiographic progression or death.
Log-rank test The registry reports Log Rank as the formal method.
Hazard ratio The reported effect measure is a Cox proportional hazard, normalized to a hazard ratio.
Confidence interval The reported HR of 0.5085 has a two-sided 95% CI of 0.4012–0.6444.
P-value The reported superiority analysis has P < 0.0001.
Analysis population The formal PFS analysis is in the hormone receptor-positive cohort of the Full Analysis Set.
Open-label design Masking is listed as none.
Safety analysis Serious adverse events are reported by arm as affected participants among those at risk.

16. Statistical Concepts in This Trial

Learn more about the methods used in this trial:

17. Related Statistical Calculators

Apply the core methods behind this trial with these statistical calculator pathways:

18. Sources

Continue with the statistical methods

Explore the survival-analysis concepts that underpin randomized time-to-event trials, including hazard ratios, confidence intervals, log-rank tests, and time-to-event endpoints.

19. Record Summary

DESTINY-Breast04 provides a clear example of randomized survival analysis in a phase 3 oncology trial. The ClinicalTrials.gov record describes a randomized, parallel, unmasked design with 557 enrolled participants and a primary time-to-event endpoint of progression-free survival based on blinded independent central review in the hormone receptor-positive cohort. The formal analysis used a log-rank test and reported a Cox proportional-hazards effect measure.

The primary posted estimate was a hazard ratio of 0.5085, with a two-sided 95% confidence interval of 0.4012–0.6444 and P < 0.0001 under a superiority hypothesis. Statistically, the estimate corresponds to an approximately 49.15% lower estimated hazard of the PFS event under the Cox model. The confidence interval describes uncertainty around that estimate, while the p-value addresses compatibility with the null hypothesis; neither should be interpreted as an individual patient's probability of progression or as a direct measure of clinical benefit.

Clinical Biostats methodology: The purpose of a trial-results page is not simply to repeat a reported statistic. The statistical story depends on the endpoint definition, analysis population, estimand, effect measure, uncertainty, testing framework, and design features. For DESTINY-Breast04, those distinctions are especially important because the posted formal PFS result applies specifically to the hormone receptor-positive cohort of the Full Analysis Set.