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
REACT was a randomized, triple-masked, phase 3 parallel-group trial evaluating celecoxib versus placebo as two years of adjuvant therapy in primary breast cancer. The primary endpoint was disease-free survival (DFS), analyzed as a time-to-event outcome.
| Feature | REACT |
|---|---|
| Trial name | REACT |
| Brief title | European Celecoxib Trial in Primary Breast Cancer |
| Phase | Phase 3 |
| Condition | Breast Cancer |
| Design | Randomized, parallel-group |
| Masking | Triple |
| Primary purpose | Treatment |
| Enrollment | 2639 |
| Interventions | Celecoxib and placebo |
| Primary endpoint | Disease Free Survival (DFS) Benefit of Two Years Adjuvant Therapy With the COX-2 Inhibitor Celecoxib Compared With Placebo in Primary Breast Cancer Patients |
| Primary endpoint type | Time-to-event |
| Primary analysis | Log-rank test with hazard ratio |
| ClinicalTrials.gov | NCT02429427 |
| Lead sponsor | Imperial College London |
| Study period | Start: 2005-12; Primary completion: 2019-03 |
2. Clinical Question
The central statistical question was whether two years of adjuvant celecoxib changed disease-free survival compared with placebo in patients with primary breast cancer.
Population
Patients with primary breast cancer enrolled in the REACT trial.
Intervention
Two years of adjuvant therapy with the COX-2 inhibitor celecoxib.
Comparator
Placebo.
Primary question
Does two years of adjuvant celecoxib affect disease-free survival relative to placebo?
3. Trial Design
Celecoxib
- Two years of adjuvant celecoxib therapy.
- Primary efficacy comparison based on disease-free survival.
- Analysis population: intention-to-treat.
Placebo
- Placebo comparator.
- Primary efficacy comparison based on disease-free survival.
- Analysis population: intention-to-treat.
The combination of randomization, parallel treatment groups, and triple masking creates a design in which the treatment comparison can be anchored to randomized assignment while reducing the opportunity for knowledge of treatment assignment to influence trial conduct and outcome assessment.
4. Endpoints
Primary Endpoint: Disease-Free Survival
| Endpoint | Registry definition and time frame | Statistical role |
|---|---|---|
| Disease Free Survival (DFS) Benefit of Two Years Adjuvant Therapy With the COX-2 Inhibitor Celecoxib Compared With Placebo in Primary Breast Cancer Patients | Patients will be followed up to 10 years. DFS will be calculated from date of randomization until the date of first documented DFS event, this will be assessed at 2 and 5 years. | Primary time-to-event endpoint |
The registry defines DFS from randomization to the first event. Events contributing to the analysis are loco-regional and distant breast cancer recurrence, new primary breast cancer (ipsilateral or contralateral), and death without disease relapse (intercurrent death).
Secondary Endpoint: Overall Survival
| Endpoint | Registry definition and time frame | Statistical role |
|---|---|---|
| Overall Survival | Date of randomisation until the date of death from any cause or censored at the date the patient was last seen alive, this will be assessed at 2 and 5 years. | Secondary time-to-event endpoint |
The endpoint definitions make an important statistical distinction. DFS includes several types of disease-related events as well as intercurrent death, whereas overall survival uses death from any cause as the event. Both are measured from randomization, but they answer different clinical questions.
5. Statistical Methodology
Intention-to-treat analysis
The primary and secondary analyses were conducted in the intention-to-treat population. The registry defines this population as all patients entered into the study in the treatment group to which they were allocated, regardless of whether they did not start treatment, took the wrong treatment, or deviated in any way from the protocol.
The ITT principle preserves the treatment comparison created by randomization. It estimates the effect associated with being assigned to the intervention rather than restricting the analysis to patients who fully adhered to the treatment strategy.
Kaplan-Meier estimation
DFS and overall survival are time-to-event endpoints. For this type of endpoint, Kaplan-Meier estimation is typically used to describe the probability of remaining event-free over time while accounting for patients whose event time is not observed during follow-up.
Here, di is the number of events at time ti, while ni is the number of participants at risk immediately before that event time.
The ClinicalTrials.gov record reports the statistical method as a log-rank test and the effect measure as a hazard ratio. The combination is standard for a randomized comparison of time-to-event outcomes: the log-rank framework compares event-time distributions, while the hazard ratio summarizes the relative event rate between groups under a time-to-event model.
Log-rank testing
The registry reports a log-rank method for both the primary DFS analysis and the secondary overall-survival analysis. Conceptually, the log-rank test compares the observed pattern of events over follow-up with the pattern expected under the null hypothesis of no difference between treatment groups.
Stratified analysis
The primary DFS analysis was stratified by oestrogen receptor status and country. The overall-survival analysis was likewise stratified for oestrogen receptor status and country.
Stratification is useful when important baseline factors are expected to influence the event process. Rather than treating every patient as if those factors were distributed identically, a stratified analysis makes the comparison within the specified strata and then combines information across them according to the statistical method.
Hazard ratio
A hazard ratio is a relative time-to-event measure. It is not a risk ratio, an absolute risk difference, or the probability that an individual patient will experience an event.
The reported hazard ratio compares celecoxib with placebo. An HR of 1 would correspond to no estimated difference in the instantaneous event rate under the model. An HR below 1 favors a lower estimated event rate for celecoxib, whereas an HR above 1 favors a higher estimated event rate. Interpretation should remain tied to the specific endpoint and follow-up period.
6. Primary Result: Disease-Free Survival
The registry reports a formal statistical analysis for the primary DFS endpoint. The analysis used the intention-to-treat population, compared celecoxib with placebo, used a log-rank test, and was stratified by oestrogen receptor status and country.
Hazard ratio for disease-free survival
95% CI: 0.80–1.17 · P = 0.751
Two-sided 95% confidence interval; intention-to-treat analysis.
| Primary DFS result | Reported value |
|---|---|
| Comparison | Celecoxib vs Placebo |
| Analysis population | Intention-to-treat |
| Method | Log Rank |
| Stratification | Oestrogen receptor status and country |
| Effect measure | Hazard Ratio |
| Estimate | 0.97 |
| 95% CI | 0.80–1.17 |
| P-value | 0.751 |
| Hypothesis type | Superiority |
The estimated DFS hazard ratio of 0.97 means that the estimated instantaneous rate of a DFS event in the celecoxib group was approximately 97% of that in the placebo group under the reported time-to-event comparison. Expressed differently, an HR of 0.97 corresponds to an estimated relative hazard reduction of approximately 3%.
That statement does not mean that 3% of patients benefited, that the probability of recurrence was reduced by exactly 3%, or that every patient experienced the same change in risk. The hazard ratio is a population-level relative measure of event rates over time.
The 95% confidence interval of 0.80–1.17 describes uncertainty around the estimated hazard ratio. Importantly, the interval includes 1.00, so the reported estimate is compatible with both a lower and a higher DFS hazard relative to placebo within the interval represented by this statistical uncertainty.
The P-value of 0.751 is a measure of compatibility of the observed data with the null hypothesis under the specified statistical framework. It is not a measure of the size of the treatment effect, the probability that celecoxib is ineffective, or the probability that the null hypothesis is true.
Because DFS is a time-to-event endpoint, the interpretation also depends on the handling of censoring and on the assumptions underlying the hazard-ratio framework. The registry reports a log-rank analysis and hazard ratio but does not provide additional details here that would allow a fuller assessment of the proportional-hazards assumption.
7. Secondary Result: Overall Survival
Overall survival was a secondary time-to-event endpoint. It was measured from randomization until death from any cause, with censoring at the date the patient was last seen alive. The analysis used the intention-to-treat population and a log-rank method stratified by oestrogen receptor status and country.
Hazard ratio for overall survival
95% CI: 0.76–1.22 · P = 0.781
Two-sided 95% confidence interval; intention-to-treat analysis.
| Secondary overall-survival result | Reported value |
|---|---|
| Comparison | Celecoxib vs Placebo |
| Analysis population | Intention-to-treat |
| Method | Log Rank |
| Stratification | Oestrogen receptor status and country |
| Effect measure | Hazard Ratio |
| Estimate | 0.97 |
| 95% CI | 0.76–1.22 |
| P-value | 0.781 |
| Hypothesis type | Superiority |
The overall-survival hazard ratio of 0.97 has the same point estimate as the DFS analysis. In model-based relative terms, this corresponds to an estimated instantaneous rate of death approximately 97% of that in the placebo group, or an estimated relative hazard reduction of approximately 3%.
The important difference is the uncertainty: the 95% CI of 0.76–1.22 is wider than the DFS interval and crosses 1.00. The interval therefore encompasses values corresponding to a lower mortality hazard as well as values corresponding to a higher mortality hazard for celecoxib relative to placebo.
The P-value of 0.781 should not be interpreted as an effect-size measure. A large P-value does not establish that the two treatments have exactly identical survival distributions; it indicates that the observed data do not provide strong evidence against the null hypothesis under the specified testing framework.
Overall survival is also a different endpoint from DFS. Death from any cause is the event, whereas DFS counts the first qualifying disease-free-survival event defined by the registry. Consequently, the two analyses address different aspects of clinical outcome even though both use time-to-event methodology.
8. Comparing the Two Time-to-Event Results
| Endpoint | HR | 95% CI | P-value | Analysis |
|---|---|---|---|---|
| Disease-free survival | 0.97 | 0.80–1.17 | 0.751 | Log-rank; stratified by oestrogen receptor status and country |
| Overall survival | 0.97 | 0.76–1.22 | 0.781 | Log-rank; stratified by oestrogen receptor status and country |
The point estimates are identical at 0.97, while the confidence intervals differ because DFS and overall survival contain different event definitions and therefore provide different statistical information. The overall-survival interval is wider, extending from 0.76 to 1.22, whereas the DFS interval extends from 0.80 to 1.17.
It is important not to interpret the similarity of the two point estimates as evidence that the endpoints are interchangeable. DFS and overall survival have different clinical meanings, different event definitions, and different censoring structures. Their estimates should be interpreted separately.
9. Statistical Methods Explained
Why was a log-rank test used?
DFS and overall survival are time-to-event outcomes, so the timing of events matters rather than simply whether an event occurred. The log-rank test is designed to compare the survival experience of two groups across follow-up while accounting for the timing of observed events and censored observations.
In this trial, the reported log-rank analysis provides a formal comparison of celecoxib and placebo across the observed follow-up rather than reducing each patient to a simple yes/no event indicator.
What does a hazard ratio of 0.97 mean?
An HR of 0.97 means that the estimated instantaneous event rate in the celecoxib group was approximately 0.97 times the corresponding rate in the placebo group under the reported time-to-event framework. Numerically, that is approximately a 3% lower estimated hazard.
The HR should not be translated directly into a 3% lower probability of recurrence or death. Probabilities depend on time, the underlying event process, censoring, and other features of the survival distribution.
Why is the confidence interval more informative than the point estimate alone?
The point estimate is only one estimate from the observed data. The confidence interval communicates how much statistical uncertainty surrounds that estimate. For DFS, the interval is 0.80–1.17; for overall survival, it is 0.76–1.22.
Both intervals include 1.00. Thus, neither interval isolates a direction of effect relative to the no-difference value. The wider overall-survival interval also indicates greater uncertainty around that estimate than would be conveyed by the HR of 0.97 alone.
Why was the analysis stratified by oestrogen receptor status and country?
The registry reports that the primary DFS analysis and the overall-survival analysis were stratified by oestrogen receptor status and country. Stratification allows the treatment comparison to account for these prespecified factors rather than assuming that their influence on the event process can simply be ignored.
Conceptually, the analysis compares treatment groups within the relevant strata and then combines the information across those strata. This is particularly useful when the factors are associated with prognosis or with the structure of the randomized trial.
Why was intention-to-treat analysis used?
ITT analysis keeps participants in the treatment group to which they were randomized. The registry explicitly states that this includes patients regardless of whether they started treatment, took the wrong treatment, or deviated from the protocol.
This approach protects the original randomized comparison from being redefined after treatment begins. It therefore answers a question about the consequences of treatment assignment rather than only the consequences among patients who adhered perfectly to treatment.
What does a P-value of 0.751 tell us?
A P-value of 0.751 does not quantify the magnitude of the observed treatment effect. It describes how compatible the observed result is with the null hypothesis under the specified statistical model and testing procedure.
The HR of 0.97 provides the estimated relative effect, the 95% CI of 0.80–1.17 provides an uncertainty interval around that effect, and the P-value of 0.751 addresses the hypothesis-testing question. These three quantities answer related but distinct statistical questions.
Why are DFS and overall survival analyzed separately?
DFS and overall survival use different event definitions. DFS counts the first qualifying event defined by the registry, including recurrence, new primary breast cancer, and death without disease relapse. Overall survival counts death from any cause.
Because the endpoints represent different clinical outcomes, a treatment can have different effects on them. A statistically similar HR point estimate does not make the endpoints equivalent.
10. Interpreting Time-to-Event Endpoints
The statistical structure of REACT illustrates why time-to-event analysis is more informative than simply comparing the percentage of patients who eventually experience an event. Follow-up duration matters, and patients can contribute information even when their event has not occurred by the time they leave observation.
Event timing
A DFS event is defined by the first qualifying event after randomization. The timing of that event contributes directly to the survival analysis.
Censoring
For overall survival, patients are censored at the date they were last seen alive if death has not been observed during follow-up.
Relative effect
The hazard ratio summarizes the relative event rate between randomized groups within the time-to-event framework.
Absolute interpretation
A hazard ratio alone does not provide an absolute probability of remaining disease-free or alive at a particular time.
This distinction is especially important when communicating a result to a clinical audience. The HR is compact and useful for comparing relative event rates, but it should not be mistaken for an absolute risk measure.
11. Safety Results
The registry reports serious adverse events by treatment arm. These figures are presented as affected participants divided by participants at risk.
| Safety measure | Celecoxib | Placebo |
|---|---|---|
| Serious adverse events | 148 / 1755 | 64 / 868 |
The reported safety figures are counts with corresponding numbers at risk. They should be kept distinct from the time-to-event efficacy analyses. Serious adverse events describe safety outcomes, whereas DFS and overall survival describe clinical time-to-event outcomes.
The unequal numbers at risk shown in the registry also illustrate why the raw event counts alone should not be compared as if they represented equally sized treatment groups. The denominators are an essential part of the reported safety information.
12. Randomization and Blinding
REACT used randomized allocation in a parallel-group design and triple masking. Randomization is central to the interpretation of the treatment comparison because it establishes treatment assignment before subsequent outcomes occur.
| Design feature | REACT | Statistical relevance |
|---|---|---|
| Randomization | Yes | Creates the basis for comparing treatment groups with reduced confounding from measured and unmeasured baseline characteristics. |
| Parallel design | Yes | Participants remain in their randomized treatment groups rather than serving as their own controls. |
| Masking | Triple | Reduces the potential for knowledge of treatment assignment to influence trial conduct and outcome assessment. |
| Primary efficacy population | Intention-to-treat | Keeps efficacy analyses aligned with randomized treatment assignment. |
These design features do not remove every source of uncertainty. Instead, they establish a framework in which the treatment comparison can be interpreted using randomized-group methods.
13. Stratification in the REACT Analysis
The registry explicitly identifies two stratification factors: oestrogen receptor status and country. Both the DFS and overall-survival analyses were stratified using these factors.
Oestrogen receptor status
This was incorporated into the reported stratified analysis rather than ignored when comparing the treatment groups.
Country
Country was also incorporated into the stratified comparison, accounting for the geographic strata identified in the registry analysis.
Stratification should not be confused with subgroup analysis. A stratified primary analysis uses the specified factors as part of the statistical comparison. A subgroup analysis instead asks whether the treatment effect differs across levels of a factor. The registry information here supports the former interpretation.
14. Why This Trial Matters Statistically
REACT is a useful teaching example because its primary and secondary outcomes demonstrate the core structure of randomized time-to-event analysis without requiring a treatment comparison to be reduced to a single number.
| Concept | How it appears in REACT |
|---|---|
| Randomization | Participants were randomized to celecoxib or placebo in a parallel-group design. |
| Blinding | The trial used triple masking. |
| Intention-to-treat analysis | Primary DFS and secondary overall-survival analyses included patients according to randomized treatment assignment. |
| Time-to-event endpoints | DFS and overall survival were both analyzed as time-to-event outcomes. |
| Log-rank test | Reported as the statistical method for both DFS and overall survival. |
| Hazard ratio | Used as the reported effect measure for both time-to-event endpoints. |
| Confidence interval | 95% two-sided confidence intervals were reported for the hazard ratios. |
| Stratified analysis | Analyses were stratified by oestrogen receptor status and country. |
The most important statistical lesson is that a hazard ratio, confidence interval, and P-value should be interpreted together. The REACT DFS result illustrates this clearly: the HR of 0.97 gives the point estimate, the 95% CI of 0.80–1.17 describes uncertainty around it, and the P-value of 0.751 addresses the corresponding hypothesis test.
15. What the Hazard Ratio Does — and Does Not — Mean
The DFS HR of 0.97 represents the estimated relative event rate for celecoxib compared with placebo under the reported time-to-event analysis. Numerically, it corresponds to approximately a 3% lower estimated hazard.
It does not mean that 3% fewer patients experienced a DFS event, nor does it mean that every patient experienced a 3% reduction in recurrence risk.
The DFS 95% CI of 0.80–1.17 describes statistical uncertainty around the HR estimate. Because it includes 1.00, the interval contains the no-difference value as well as values on either side of it.
The overall-survival 95% CI of 0.76–1.22 is similarly centered around the reported point estimate of 0.97 but spans a wider range.
The DFS P-value of 0.751 and overall-survival P-value of 0.781 quantify evidence against their respective null hypotheses under the reported testing framework. Neither P-value measures clinical importance or effect size.
Hazard ratios are model-based relative measures of event rates over time. Their interpretation depends on the structure of the survival analysis and, where applicable, assumptions about how hazards behave over follow-up. A single HR should therefore be read as a summary of the time-to-event comparison rather than as an absolute risk measure.
16. Limitations and Interpretation Issues
- Time-to-event interpretation: DFS and overall survival depend on both whether events occur and when they occur. A hazard ratio therefore does not directly provide an absolute risk difference.
- Confidence interval width: The overall-survival estimate has a 95% CI of 0.76–1.22, illustrating substantial uncertainty around the point estimate of 0.97.
- Endpoint definitions: DFS and overall survival are not interchangeable. DFS includes several qualifying disease-related events and intercurrent death, while overall survival uses death from any cause.
- Stratification: The reported analyses were stratified by oestrogen receptor status and country. These factors are therefore part of the interpretation of the reported treatment comparison.
- Analysis population: The efficacy analyses use the intention-to-treat population, so the estimates describe randomized treatment assignment rather than treatment received only among participants who fully adhered to protocol.
- Safety denominators: Serious adverse events are reported as affected participants divided by participants at risk. Raw event counts should not be compared without considering these denominators.
- Proportional-hazards considerations: A hazard ratio is a summary measure of relative event rates over time. The registry information does not provide additional diagnostics for assessing the proportional-hazards assumption.
- Scope of reported results: The registry reports the primary DFS analysis and a secondary overall-survival analysis, but it does not report additional subgroup estimates, baseline characteristics, or other outcome summaries in the information presented here.
17. Clinical Interpretation vs Statistical Interpretation
Statistical interpretation
The primary DFS analysis produced an HR of 0.97 with a two-sided 95% CI of 0.80–1.17 and a P-value of 0.751. The secondary overall-survival analysis also produced an HR of 0.97, with a 95% CI of 0.76–1.22 and a P-value of 0.781.
Clinical interpretation
The statistical results describe the observed randomized comparison of celecoxib and placebo for the defined DFS and overall-survival endpoints. They should be interpreted in the context of the endpoint definitions, follow-up framework, uncertainty intervals, and intention-to-treat analysis.
The distinction matters because statistical evidence and clinical meaning are related but not identical. A hazard ratio summarizes a relative event-rate comparison; it does not by itself establish the magnitude of absolute benefit experienced by individual patients.
18. Trial Timeline
Trial start
The REACT trial began in December 2005.
Longitudinal outcome assessment
The registry specifies follow-up of patients up to 10 years, with DFS assessed at 2 and 5 years and overall survival assessed at 2 and 5 years.
Primary completion
The registry records primary completion in March 2019.
19. Statistical Summary of the Reported Results
Primary endpoint
95% CI 0.80–1.17 · P = 0.751
Stratified by oestrogen receptor status and country; intention-to-treat population.
Secondary endpoint
95% CI 0.76–1.22 · P = 0.781
Stratified by oestrogen receptor status and country; intention-to-treat population.
Both reported hazard-ratio point estimates are 0.97. The DFS and overall-survival confidence intervals both include 1.00, and the corresponding P-values are 0.751 and 0.781, respectively. The appropriate interpretation is therefore not based on the point estimate alone but on the complete statistical result: endpoint definition, analysis population, stratification, effect estimate, confidence interval, and hypothesis test.
20. Related Tutorials
Learn more about the methods used in this trial:
21. Related Calculators
22. Sources
- ClinicalTrials.gov: NCT02429427.
- PubMed: PMID 34264305.
Continue through Clinical Biostats
Explore statistical tutorials, clinical-trial methods, and analysis calculators for deeper study of time-to-event endpoints and randomized clinical-trial methodology.
23. Record Summary
REACT provides a clear example of randomized clinical-trial analysis for time-to-event outcomes. The trial used randomized parallel-group allocation, triple masking, intention-to-treat efficacy analysis, log-rank testing, hazard ratios, and stratification by oestrogen receptor status and country. The primary DFS analysis reported an HR of 0.97 with a 95% CI of 0.80–1.17 and a P-value of 0.751. The secondary overall-survival analysis reported an HR of 0.97 with a 95% CI of 0.76–1.22 and a P-value of 0.781.
The statistical interpretation is strongest when the effect estimate is considered together with its confidence interval and P-value. The hazard ratio supplies the relative time-to-event estimate, the confidence interval communicates uncertainty, and the P-value addresses the corresponding hypothesis test. None of these quantities should be interpreted as an absolute probability of benefit for an individual patient.