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
AZURE is a randomized phase 3 treatment trial evaluating chemotherapy and/or hormone therapy with or without zoledronate in women with stage II or stage III breast cancer. The registry identifies disease-free survival as the primary endpoint, assessed annually for 10 years.
| Feature | AZURE |
|---|---|
| Acronym | AZURE |
| Phase | Phase 3 |
| Condition | Breast Cancer; Metastatic Cancer |
| Population described by the study title | Women with stage II or stage III breast cancer |
| Allocation | Randomized |
| Masking | None |
| Primary purpose | Treatment |
| Primary endpoint | Disease-free survival as assessed annually for 10 years |
| Lead sponsor | University of Sheffield |
| Sponsor type | Other |
| Enrollment | Not reported in the ClinicalTrials.gov record |
| ClinicalTrials.gov | NCT00072020 |
2. Clinical Question
The registry describes a randomized comparison of chemotherapy and/or hormone therapy with or without zoledronic acid in women with stage II or stage III breast cancer. Statistically, the central question is whether adding zoledronic acid changes the time until disease-free survival is lost during the 10-year assessment period.
Population
Women with stage II or stage III breast cancer, as described in the trial title. The registry also lists breast cancer and metastatic cancer among the study conditions.
Intervention
Zoledronic acid, evaluated in the context of adjuvant and/or neoadjuvant therapy.
Comparator
The corresponding treatment strategy without zoledronic acid.
Primary question
Does treatment with zoledronic acid alter disease-free survival when assessed annually for 10 years?
3. Trial Design
The ClinicalTrials.gov record classifies AZURE as a phase 3, randomized treatment trial. Masking is listed as none, so the registry describes the study as an open-label rather than masked comparison.
With zoledronic acid
- Zoledronic acid
- Adjuvant and/or neoadjuvant therapy as applicable to the study design
Without zoledronic acid
- Corresponding chemotherapy and/or hormone therapy strategy
- No zoledronic acid
The registry does not report treatment doses, schedules, sample sizes by treatment group, or the total enrollment. Those details are therefore not used to characterize the randomized arms here.
4. Endpoints
| Endpoint | Registry definition and time frame | Statistical role |
|---|---|---|
| Primary endpoint | Disease-free survival as assessed annually for 10 years | Primary efficacy endpoint |
The wording of the registered endpoint is important. The endpoint is disease-free survival, and the registry specifies annual assessment over 10 years. The record does not provide a more detailed operational definition of the disease-free-survival event in the information available here.
5. Statistical Methodology
Disease-free survival as a time-to-event endpoint
Disease-free survival is naturally analysed using methods that retain information about when an event occurs. This is different from simply comparing the percentage of patients who are disease-free at the end of follow-up.
For each patient, the analysis conceptually records a follow-up time and an event indicator. A patient who experiences the disease-free-survival event contributes an event time. A patient who has not experienced the event by the last available assessment contributes follow-up information up to that point and is then censored.
The statistical analysis must distinguish patients who experience an event from patients whose follow-up ends before an event is observed.
Kaplan-Meier estimation
A Kaplan-Meier estimator is a standard descriptive method for estimating the disease-free-survival function in the presence of right censoring. It produces a stepwise estimate of the probability of remaining event-free over time.
Here, di represents the number of events at time ti, while ni represents the number of patients at risk immediately before that time.
Log-rank comparison
For a randomized two-group disease-free-survival comparison, a log-rank test is a standard method for testing whether the observed survival experience differs between treatment groups over follow-up. A stratified version can be used when randomization incorporates prognostic strata, although the registry information available here does not specify any stratification factors.
Cox proportional-hazards model
A Cox proportional-hazards model is another standard approach for estimating a relative treatment effect for a time-to-event endpoint. Its treatment coefficient can be expressed as a hazard ratio.
An HR below 1 would indicate a lower estimated instantaneous event rate in the zoledronic-acid group under the fitted model. An HR above 1 would indicate a higher estimated instantaneous event rate. An HR of 1 would indicate equal estimated hazards.
6. Statistical Methods Explained
What makes disease-free survival different from a binary endpoint?
A binary endpoint asks whether an event has occurred by a specified time. Disease-free survival retains the timing of the event. If one patient experiences an event substantially earlier than another, a time-to-event analysis can use that difference rather than reducing both observations to the same binary outcome.
Why is Kaplan-Meier useful?
Longitudinal clinical trials commonly have patients with different lengths of observed follow-up. Kaplan-Meier estimation accommodates censoring and provides an estimate of the probability of remaining event-free throughout the observed time period.
What does censoring mean?
Censoring means that the exact event time is not observed during the patient's available follow-up. For example, a patient may remain disease-free when follow-up ends. The analysis uses the information available through that patient's last observation without treating the patient as though an event occurred at that time.
What would a hazard ratio tell us?
A hazard ratio would summarize the relative instantaneous event rate between the randomized treatment groups under a Cox model. It is a relative measure of event occurrence over time, not a direct percentage of patients cured, a percentage-point difference in disease-free survival, or a statement about the experience of every individual patient.
Why use a log-rank test?
The log-rank test is designed for comparing time-to-event distributions between groups while accounting for the timing of events and right censoring. It therefore addresses a different statistical structure from a simple chi-square comparison of event proportions.
What is the proportional-hazards assumption?
A conventional Cox proportional-hazards interpretation assumes that the relative hazard between groups is reasonably stable over time. If the treatment effect changes substantially during follow-up, a single hazard ratio may provide an incomplete description of the treatment-time relationship. Long-term follow-up makes examination of the survival curves and the timing of effects particularly important.
7. Planned Analysis
The registry identifies disease-free survival as assessed annually for 10 years as the primary endpoint. Because disease-free survival is a time-to-event endpoint, the principal analysis would typically estimate disease-free-survival curves with the Kaplan-Meier method and compare randomized groups using a log-rank test. A Cox proportional-hazards model would typically be used to estimate a hazard ratio and its confidence interval.
Primary estimand
The treatment comparison concerns the relative timing of disease-free-survival events between the randomized groups over follow-up.
Descriptive display
Kaplan-Meier curves would show how the estimated disease-free-survival probability changes over time.
Hypothesis test
A log-rank test would ordinarily compare the disease-free-survival experience between the randomized groups.
Effect estimate
A Cox model would ordinarily provide a hazard ratio with a confidence interval for the relative event rate.
No formal statistical analyses were posted to ClinicalTrials.gov. Results have not been posted on ClinicalTrials.gov for the registered primary endpoint.
8. Understanding the Primary Endpoint Statistically
Disease-free survival combines the longitudinal nature of follow-up with an event definition. This creates several layers of statistical interpretation.
| Quantity | What it describes | What it does not establish by itself |
|---|---|---|
| Kaplan-Meier curve | Estimated disease-free-survival probability over time | A causal explanation for every change in the curve |
| Median disease-free survival | The time at which the estimated survival probability reaches 0.50, if that occurs | The average disease-free-survival time |
| Hazard ratio | Relative instantaneous event rate under a fitted Cox model | An absolute difference in disease-free survival |
| Confidence interval | Statistical uncertainty around an estimated effect | The range of outcomes for individual patients |
| P-value | Evidence against a specified null hypothesis under the statistical model | The size or clinical importance of an effect |
Relative versus absolute effects
A hazard ratio can summarize a relative treatment effect, but absolute disease-free survival at clinically relevant times provides a different perspective. For example, if a future analysis reports both a hazard ratio and disease-free-survival probabilities at specific years, those measures should not be treated as interchangeable. One describes relative event rates; the other describes absolute event-free probabilities at particular times.
Why a single number is not enough
Two trials can produce the same hazard ratio while having different survival curves and different absolute disease-free-survival probabilities. Conversely, similar time-specific survival probabilities can arise from different patterns of events over time. A statistically complete interpretation therefore benefits from examining the curve, event timing, censoring, effect estimate, and uncertainty together.
9. Randomization and Causal Interpretation
Randomization is central to the statistical interpretation of a treatment comparison. When patients are randomized, treatment assignment is determined by the study design rather than by the investigator's selection of therapy for an individual patient. This helps balance measured and unmeasured prognostic factors in expectation.
The AZURE registry record identifies the allocation as randomized and the masking as none. The absence of masking does not remove the value of randomization, but it is relevant when considering outcomes that could potentially be influenced by knowledge of treatment assignment.
The primary comparison should ordinarily preserve the randomized treatment groups for efficacy analysis. This maintains the treatment contrast created by randomization and avoids allowing post-randomization treatment decisions to redefine the principal comparison.
10. Missing Data and Censoring
For a 10-year disease-free-survival endpoint, incomplete follow-up is an important statistical consideration. Patients may have different observation periods, and the analysis must distinguish ordinary right censoring from an actual disease-free-survival event.
The ClinicalTrials.gov record does not specify an imputation strategy, rules for handling missing endpoint assessments, censoring rules, or a detailed statistical analysis plan. Those methods therefore cannot be characterized as features of the registered analysis from the available registry information.
Censoring
A patient without an observed event at the end of available follow-up can contribute partial information rather than being discarded automatically.
Informative follow-up
If loss to follow-up is related to disease status or prognosis, standard censoring assumptions may become less plausible.
Ten-year horizon
Long follow-up creates additional opportunities for incomplete observations and changes in follow-up intensity.
Analysis specification
The precise censoring and missing-data rules should come from the applicable protocol or statistical analysis plan when available.
11. Multiplicity and Endpoint Interpretation
The registry identifies one registered primary endpoint: disease-free survival as assessed annually for 10 years. No additional primary endpoints, multiplicity procedure, alpha-spending method, or hierarchical testing strategy is reported in the ClinicalTrials.gov record.
| Feature | Registry information | Statistical implication |
|---|---|---|
| Primary endpoint | Disease-free survival as assessed annually for 10 years | The principal efficacy question is a time-to-event comparison |
| Additional primary endpoints | Not reported | No additional primary endpoint hierarchy is described |
| Multiplicity procedure | Not reported | No multiplicity adjustment can be characterized from the registry |
| Interim analysis | Not reported | No interim-analysis error-control strategy is described |
12. Interim Analysis and Alpha Spending
Interim efficacy analyses can affect the interpretation of a clinical trial because repeatedly examining accumulating data can increase the probability of rejecting a true null hypothesis unless the design accounts for those repeated looks.
The ClinicalTrials.gov record does not report an interim-analysis schedule or an alpha-spending procedure for AZURE. Consequently, no particular group-sequential boundary or type I error allocation is described here.
When a trial includes planned interim analyses, the statistical design normally specifies how the total type I error is distributed across those analyses.
13. Subgroups and Effect Modification
The registry information does not identify prespecified subgroup analyses or treatment-by-subgroup interaction tests. Subgroup findings therefore cannot be characterized as part of the registered statistical analysis from the information available here.
From a statistical perspective, a subgroup estimate answers a different question from the overall treatment effect. More importantly, observing a treatment effect in one subgroup and a weaker result in another does not by itself demonstrate that treatment effects truly differ between those groups.
14. Safety and Statistical Interpretation
The ClinicalTrials.gov information in the ClinicalTrials.gov record does not provide serious adverse-event counts, adverse-event rates, treatment-emergent adverse-event summaries, or other arm-specific safety results.
Safety is nevertheless an important statistical component of a randomized treatment trial. Safety analyses are generally summarized according to treatment exposure, with the denominator and follow-up period clearly specified. For uncommon serious events, event counts and confidence intervals can be more informative than relying on p-values alone.
| Safety information | Registry information |
|---|---|
| Serious adverse events by arm | Not reported |
| Adverse-event rates by arm | Not reported |
| Exposure-adjusted safety analysis | Not reported |
| Safety analysis population | Not reported |
Because no arm-specific safety results are posted in the registry information used here, no quantitative safety comparison is made.
15. Bayesian Methods
No Bayesian analysis is reported in the ClinicalTrials.gov information for AZURE. A Bayesian analysis would use a prior distribution together with the observed trial data to obtain a posterior distribution for the treatment effect.
The resulting posterior distribution can be summarized with quantities such as posterior probabilities or credible intervals. None of these Bayesian quantities is reported for this trial in the registry information.
16. Crossover and Treatment Switching
The ClinicalTrials.gov record does not report crossover, treatment switching, or a statistical method for handling treatment changes after randomization.
This distinction matters because an intention-to-treat comparison and an analysis based on treatment actually received answer different questions. In a randomized trial, the intention-to-treat comparison preserves the treatment assignment created by randomization. A treatment-received analysis can be affected by the reasons patients switch treatments, discontinue therapy, or receive additional interventions.
17. Non-Inferiority and Equivalence
AZURE is not identified in the registry information as a non-inferiority or equivalence trial. No non-inferiority margin is reported.
That distinction is important because non-inferiority trials are interpreted against a prespecified margin rather than simply asking whether a conventional superiority test is statistically significant. The statistical framework for a superiority comparison should not be substituted for a non-inferiority framework.
18. ClinicalTrials.gov Record Status and Data Completeness
The registry lists the study status as Unknown. The study start date is 2003-08, and the lead sponsor is the University of Sheffield.
| Registry field | Reported value |
|---|---|
| NCT ID | NCT00072020 |
| Acronym | AZURE |
| Status | Unknown |
| Start date | 2003-08 |
| Lead sponsor | University of Sheffield |
| Sponsor type | Other |
| Enrollment | Not reported |
| Allocation | Randomized |
| Masking | None |
| Primary purpose | Treatment |
19. What a Complete Disease-Free-Survival Result Would Contain
A complete statistical report for the registered primary endpoint would ordinarily contain several complementary pieces of information rather than a single p-value.
| Result component | Purpose |
|---|---|
| Number of patients analyzed | Defines the analysis population and denominator |
| Number of disease-free-survival events | Describes the amount of event information available |
| Kaplan-Meier estimates | Shows estimated disease-free survival over time |
| Median disease-free survival, if estimable | Provides a time-based summary of the survival distribution |
| Hazard ratio | Provides a relative treatment-effect estimate |
| Confidence interval | Quantifies statistical uncertainty around the effect estimate |
| P-value | Assesses evidence against the prespecified null hypothesis |
These measures answer different questions. A p-value does not tell the reader how large the treatment effect is, while a hazard ratio without a confidence interval does not communicate the precision of the estimate. The survival curve adds information about how the treatment comparison evolves through time.
20. Statistical Interpretation of a Future Hazard Ratio
A hazard ratio below 1 would indicate a lower estimated instantaneous disease-free-survival event rate in the zoledronic-acid group under the fitted Cox model. It would not mean that the same percentage of every patient experienced a corresponding reduction in personal risk.
A hazard ratio of 1 would correspond to equal estimated hazards between the randomized groups under the model. It would not, by itself, establish that the entire disease-free-survival distributions are identical in every respect.
A hazard ratio above 1 would indicate a higher estimated instantaneous disease-free-survival event rate in the zoledronic-acid group under the fitted model.
The confidence interval describes statistical uncertainty around the estimated treatment effect. A narrow interval indicates greater statistical precision than a wide interval, all else equal. It does not describe the range of treatment effects experienced by individual patients.
A p-value addresses the compatibility of the observed data with a specified null hypothesis under the statistical model. It does not quantify how large or clinically important the treatment effect is.
21. Important Limitations and Interpretation Issues
- Results availability: the registry does not post formal statistical analyses for the registered primary endpoint.
- Enrollment: the ClinicalTrials.gov record does not report the enrollment number in the information available here.
- Endpoint detail: the registry identifies disease-free survival and its annual assessment over 10 years but does not provide a more detailed event definition in the available record.
- Statistical analysis plan: the registry information does not specify the exact model, stratification factors, censoring rules, missing-data methods, or hypothesis-testing procedure.
- Interim analysis: no interim-analysis strategy or alpha-spending method is reported.
- Multiplicity: no multiplicity-adjustment procedure is reported.
- Subgroups: no prespecified subgroup or interaction analyses are reported in the registry information.
- Safety: arm-specific serious adverse-event and adverse-event results are not reported in the available registry information.
- Long follow-up: a 10-year endpoint requires careful treatment of censoring, follow-up completeness, and changes in event risk over time.
- Open-label design: masking is listed as none, which is relevant when considering outcomes that could potentially be influenced by knowledge of treatment assignment.
22. Why This Trial Matters Statistically
AZURE provides a useful teaching case for understanding how randomized treatment comparisons are translated into long-term time-to-event analyses. The registered primary endpoint is disease-free survival assessed annually for 10 years, making the distinction between a time-to-event analysis and a simple fixed-time proportion especially important.
| Concept | How it applies to AZURE |
|---|---|
| Randomization | The registry classifies allocation as randomized. |
| Open-label design | Masking is listed as none. |
| Time-to-event analysis | Disease-free survival is the registered primary endpoint. |
| Kaplan-Meier estimation | A standard approach for estimating disease-free survival with censoring. |
| Log-rank testing | A standard approach for comparing disease-free-survival distributions. |
| Cox regression | A standard approach for estimating a hazard ratio for a time-to-event endpoint. |
| Censoring | Relevant when patients remain event-free at the end of available follow-up. |
| Long-term follow-up | The primary endpoint is assessed annually for 10 years. |
| Absolute versus relative effects | Survival probabilities and hazard ratios provide complementary perspectives. |
| Statistical precision | Confidence intervals are needed to describe uncertainty around estimated treatment effects. |
23. A Practical Reading Framework for AZURE
When a formal disease-free-survival result is available, a statistically careful reading can proceed in a fixed sequence.
The final step is to consider the p-value within the prespecified hypothesis-testing framework. Statistical significance should not be used as a substitute for describing the magnitude, precision, timing, and clinical context of the disease-free-survival difference.
24. Related Statistical Concepts
The primary endpoint in AZURE illustrates several foundational ideas in clinical-trial statistics: randomization, time-to-event outcomes, censoring, Kaplan-Meier estimation, hazard ratios, confidence intervals, and the distinction between statistical evidence and effect magnitude.
Time-to-event thinking
Disease-free survival uses both whether an event occurs and when it occurs.
Survival curves
Kaplan-Meier estimates provide a visual description of disease-free survival over follow-up.
Relative effects
A Cox hazard ratio summarizes a relative event-rate comparison under a proportional-hazards model.
Uncertainty
Confidence intervals show how precisely a treatment effect has been estimated.
25. Sources
- ClinicalTrials.gov: AZURE — NCT00072020.
Continue through Clinical Biostats
Explore additional clinical-trial results, biostatistics tutorials, and statistical methods for understanding randomized clinical-trial evidence.
26. Record Summary
AZURE is a randomized phase 3 treatment trial of zoledronic acid with adjuvant and/or neoadjuvant therapy in women with stage II or stage III breast cancer. The ClinicalTrials.gov record identifies disease-free survival assessed annually for 10 years as the primary endpoint. From a statistical perspective, this endpoint calls for time-to-event methods that account for differing follow-up and censoring, with Kaplan-Meier estimation, log-rank comparison, and Cox modeling representing standard analytical approaches. The registry does not post formal statistical analyses or quantitative primary-endpoint results, so the statistical interpretation centers on how the registered endpoint would ordinarily be analyzed and how future reported estimates should be read.