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
BR.21 was a randomized, parallel, double-masked phase 3 treatment trial evaluating OSI-774 (Tarceva, erlotinib HCl) in patients with stage III or stage IV non-small cell lung cancer. The registry records an enrollment of 731 participants and identifies OSI Pharmaceuticals as the lead sponsor.
| Feature | BR.21 |
|---|---|
| Acronym | BR.21 |
| ClinicalTrials.gov identifier | NCT00036647 |
| Phase | Phase 3 |
| Status | Completed |
| Condition | Carcinoma, Non-small-cell Lung |
| Enrollment | 731 |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | Double |
| Primary purpose | Treatment |
| Intervention | Tarceva (erlotinib HCl, OSI-774) |
| Lead sponsor | OSI Pharmaceuticals |
| Sponsor type | Industry |
2. Clinical Question
The registry identifies BR.21 as a phase 3 treatment trial of OSI-774 (Tarceva, erlotinib HCl) in patients with stage III or stage IV non-small cell lung cancer. The statistical question implied by the randomized treatment design is whether outcomes differ between the randomized treatment groups.
Population
Patients with stage III or stage IV non-small cell lung cancer, described in the registry under the condition “Carcinoma, Non-small-cell Lung.”
Intervention
OSI-774 (Tarceva, erlotinib HCl), identified in the registry as the study drug intervention.
Comparator
The registry data identify the intervention and randomized design but do not provide a comparator intervention in the record information summarized here.
Primary question
In a randomized phase 3 treatment trial, does assignment to the study intervention produce different clinical outcomes from the comparison condition?
3. Trial Design
Randomization and masking are important statistical design features because they address different sources of bias. Randomization establishes the mechanism for comparing treatment assignments without relying on prognostic balance created by investigator selection. Double masking can reduce the influence of treatment knowledge on participant behavior, investigator decisions, assessment, and reporting.
4. Endpoints
The ClinicalTrials.gov record does not list any registered primary endpoints in the registry information available for this analysis.
| Endpoint field | Registry record |
|---|---|
| Registered primary endpoints | None listed |
| Endpoint definitions | Not reported in the registry information available here |
| Endpoint time frames | Not reported in the registry information available here |
| Posted statistical analyses | None |
This distinction is statistically important. A clinical trial can have outcomes assessed during its conduct even when the current registry record does not contain registered endpoint definitions. Without a documented endpoint definition and analysis time frame, an endpoint-specific interpretation cannot be tied to this registry record.
5. Statistical Methodology
Randomized comparison
The defining statistical feature of BR.21 is randomized allocation. In a randomized treatment trial, the treatment assignment mechanism is established before outcomes are observed. Under appropriate conduct, randomization provides the basis for comparing outcome distributions between groups while reducing confounding by baseline prognostic factors.
The exact statistical form of this contrast depends on the endpoint: continuous outcomes may use means or adjusted mean differences, binary outcomes may use risks or odds, and time-to-event outcomes may use survival probabilities or hazard ratios.
Parallel-group analysis
The parallel design means that participants are assigned to treatment groups and followed within those groups rather than being deliberately exposed to multiple randomized treatment periods. The principal statistical comparison is therefore between groups defined by randomized assignment.
Double masking
Double masking is a design feature intended to reduce bias associated with knowledge of treatment assignment. From a statistical perspective, masking is especially relevant when treatment knowledge could affect behavior, outcome assessment, treatment discontinuation, or other post-randomization processes.
Phase 3 confirmatory framework
The phase 3 designation places the study in the confirmatory portion of clinical development. Statistical interpretation in a phase 3 trial ordinarily depends on prespecified primary endpoints, a defined analysis population, an estimand or treatment-effect measure appropriate to those endpoints, and an error-control framework. Those endpoint-specific elements are not listed in the registry information available here.
Analysis populations
A randomized trial commonly distinguishes the population defined by randomized assignment from populations defined by treatment exposure or protocol adherence. The former supports the principal randomized comparison, while treated or per-protocol populations can provide additional descriptive or sensitivity analyses. BR.21's registry information available here does not specify its named analysis populations.
6. Planned Analysis
No formal statistical analyses are posted to ClinicalTrials.gov for BR.21, and the registry does not list registered primary endpoints. Consequently, the record does not provide an endpoint-specific analysis plan that can be tied to a named outcome.
What would normally be specified?
For a randomized phase 3 treatment trial, the statistical analysis plan would ordinarily identify each primary and secondary endpoint, define the analysis population, specify the treatment-effect measure, describe the hypothesis test and confidence interval, establish any stratification or covariate adjustment, and define the handling of missing observations and protocol deviations.
| Statistical component | What would normally be defined |
|---|---|
| Estimand | The treatment effect of interest, including the population, endpoint, treatment conditions, and summary measure. |
| Primary endpoint | A precisely defined outcome and assessment time frame. |
| Effect measure | A measure appropriate to the endpoint, such as a mean difference, risk ratio, odds ratio, or hazard ratio. |
| Confidence interval | An interval estimate quantifying uncertainty around the treatment-effect estimate. |
| Hypothesis test | A prespecified statistical test and type I error level. |
| Multiplicity | A strategy for controlling false-positive error when multiple confirmatory hypotheses are tested. |
| Missing data | Rules for censoring, missing observations, treatment discontinuation, and sensitivity analyses appropriate to the endpoint. |
7. Statistical Methods Explained
Why is randomization statistically important?
Randomization creates the treatment groups through a defined allocation mechanism rather than through investigator or participant choice. This is the foundation for causal comparison in a randomized trial. With adequate implementation, measured and unmeasured baseline prognostic factors are expected to be balanced in distribution across randomized groups in the long run.
What does a parallel design change?
In a parallel-group study, treatment groups are compared across participants rather than comparing different treatment periods within the same participant. This makes the randomized treatment assignment the primary basis for the between-group comparison.
Why does double masking matter for statistical interpretation?
Masking can limit systematic differences caused by knowledge of assignment. If participants or investigators know which treatment was received, that knowledge can influence reporting, co-interventions, decisions to discontinue treatment, or assessment of outcomes. Double masking therefore helps protect the internal validity of the randomized comparison.
Why is phase 3 relevant to statistical planning?
A phase 3 study is generally intended to provide confirmatory evidence for a treatment question. That requires the statistical question to be specified before the results are examined. In particular, the primary endpoint and its analysis should be defined in advance so that the observed data are not used to select the most favorable outcome after the fact.
Why must the endpoint be defined before choosing the statistical test?
Different endpoint types contain different information. A continuous measurement can be summarized by a mean or median, a binary outcome by a proportion or risk, and a time-to-event endpoint by a survival function or another event-time measure. Choosing the statistical method after seeing the data can alter the interpretation of statistical evidence.
Why are confidence intervals useful even when a hypothesis test is reported?
A hypothesis test addresses compatibility with a prespecified null hypothesis. A confidence interval additionally shows the range of effect estimates compatible with the statistical model and sampling uncertainty. The interval therefore provides information about precision and the magnitude of effects that a p-value alone does not convey.
8. Interpreting Treatment Effects in a Randomized Trial
Because the registry record does not contain endpoint-specific estimates, the most useful statistical interpretation is to understand how a treatment effect would ordinarily be expressed once the endpoint has been defined.
A relative measure describes the relationship between treatment groups, whereas an absolute measure describes the difference in outcome between groups on the original outcome scale. Neither automatically replaces the other. For example, a relative reduction can correspond to very different absolute differences depending on the baseline event rate.
A confidence interval describes statistical uncertainty around an estimated treatment effect under the specified analysis framework. A narrow interval indicates greater statistical precision than a wide interval, but precision is not the same as clinical importance.
A p-value measures the compatibility of the observed data with a specified null hypothesis under the statistical model. It does not measure the magnitude of a treatment effect, the probability that a treatment works, or the probability that the null hypothesis is true.
The population analyzed matters because post-randomization exclusions, treatment discontinuation, and protocol deviations can change the question being answered. In randomized trials, analysis according to randomized assignment is often central because it preserves the treatment comparison created by randomization.
9. Missing Data and Censoring
Endpoint-specific missing-data procedures are not reported in the registry information available here. The appropriate approach would depend on the endpoint and on why observations became unavailable.
For time-to-event endpoints, censoring is a distinct concept from ordinary missing data. A censored participant contributes information up to the time at which follow-up ends under the prespecified censoring rule. For repeated or fixed-time outcomes, missing observations require a different framework, potentially involving assumptions about the missingness mechanism and sensitivity analyses.
Randomized assignment
Maintains the original treatment comparison when participants are analyzed according to their assigned group.
Post-randomization events
Discontinuation, additional treatment, or loss to follow-up can affect the observed data and must be addressed according to the prespecified analysis strategy.
10. Multiplicity and Multiple Endpoints
The registry information available here does not specify the number or hierarchy of primary and secondary endpoints, nor does it describe an alpha-spending or multiplicity-adjustment strategy.
Multiplicity becomes important when several hypotheses are tested confirmatorily. If each hypothesis is tested at the same nominal significance level without adjustment, the probability of obtaining at least one apparently positive result can exceed the intended familywise type I error rate. A prespecified hierarchy, alpha allocation, or another multiplicity-control procedure can address this issue.
The appropriate multiplicity procedure depends on the trial's prespecified objectives and hypothesis structure. It should not be selected solely after seeing which endpoints produced favorable results.
11. Stratification and Covariate Adjustment
The ClinicalTrials.gov record available here does not report stratification factors or covariates used in the statistical analysis.
When stratification is used at randomization, the analysis can account for those factors in the primary comparison. For time-to-event outcomes, this may involve a stratified log-rank test or a stratified Cox model. For continuous outcomes, analysis of covariance can incorporate prespecified baseline covariates. The choice should follow the endpoint and the statistical analysis plan rather than being inferred from the trial's title or treatment.
12. Interim Analysis and Adaptive Features
The registry information available here does not report an interim-analysis schedule, alpha-spending method, stopping boundary, adaptive design feature, or futility rule.
An interim analysis is a planned examination of accumulating trial data before the final analysis. If an interim look is used for confirmatory decision-making, the statistical design normally specifies how repeated looks will be incorporated into type I error control. Without such prespecification, repeatedly examining the data can increase the false-positive probability.
13. Safety Analysis
The registry information available here identifies the study as a treatment trial but does not provide serious adverse-event counts, adverse-event rates, treatment-emergent event definitions, or safety estimates by randomized group.
Safety analysis in a randomized treatment trial generally differs from efficacy analysis in its treatment of exposure. Investigators commonly summarize adverse events according to actual treatment exposure while retaining randomized-group summaries for other safety perspectives. Exact definitions matter: serious adverse events, treatment-emergent adverse events, grade categories, discontinuations due to adverse events, and deaths are distinct measures.
| Safety component | Registry information available here |
|---|---|
| Serious adverse events by arm | Not reported |
| Adverse-event rates by arm | Not reported |
| Severity or grade distribution | Not reported |
| Safety analysis population | Not reported |
14. Trial Timeline
Study start
The registry lists November 1, 2001 as the study start date.
731 participants
The ClinicalTrials.gov record lists an enrollment of 731 participants.
Primary completion
The registry lists January 30, 2004 as the primary completion date.
Completed
The ClinicalTrials.gov record identifies BR.21 as completed.
15. What the Registry Record Allows Us to Conclude Statistically
The record provides several important design facts: BR.21 was randomized, parallel, double masked, phase 3, and conducted for a treatment purpose. These features define the basic architecture of the statistical comparison.
The record also establishes the study population at the level of stage III or stage IV non-small cell lung cancer, identifies OSI-774 (Tarceva, erlotinib HCl) as the intervention, and reports an enrollment of 731. Those facts are sufficient to characterize the trial design, but they are not sufficient to specify an endpoint-specific treatment effect.
Most importantly, the registry record available here contains neither registered primary endpoints nor posted statistical analyses. Consequently, a numerical efficacy interpretation cannot be attached to a named endpoint from the ClinicalTrials.gov record alone.
16. Why This Trial Matters Statistically
BR.21 provides a useful framework for learning how the design of a randomized phase 3 clinical trial determines the statistical questions that can subsequently be answered. Its registry record contains the fundamental structural elements of a confirmatory treatment comparison even though endpoint-level statistical results are not posted.
| Concept | How it appears in BR.21 |
|---|---|
| Randomization | The registry classifies allocation as randomized. |
| Parallel-group design | The registry classifies the design model as parallel. |
| Masking | The registry classifies masking as double. |
| Phase 3 | The study is identified as phase 3. |
| Treatment purpose | The primary purpose is treatment. |
| Sample size | Enrollment is reported as 731. |
| Endpoint specification | No registered primary endpoints are listed in the registry information available here. |
| Statistical results | No posted statistical analyses are listed. |
| Interpretability | Design features can be described, while endpoint-specific effects require documented endpoint and analysis information. |
17. Important Limitations and Interpretation Issues
- No registered primary endpoints: the ClinicalTrials.gov record available here does not list registered primary endpoints or their definitions and time frames.
- No posted statistical analyses: the registry does not contain posted statistical analyses for the trial.
- No numerical efficacy estimates: endpoint-specific estimates, confidence intervals, and p-values are not contained in the registry information available here.
- No comparator details: the intervention is identified, but the comparator intervention is not included in the ClinicalTrials.gov record.
- No analysis population definitions: the registry information available here does not identify ITT, per-protocol, safety, or other named analysis populations.
- No stratification information: randomization is documented, but the record does not identify stratification factors or covariates used in analysis.
- No multiplicity strategy: the record does not describe a hierarchy of endpoints or a procedure for controlling type I error across multiple hypotheses.
- No interim-analysis details: the registry information available here does not specify interim looks, stopping boundaries, or alpha spending.
- No safety estimates: serious adverse events and other safety outcomes by treatment group are not reported in the available record.
18. Statistical Interpretation vs Clinical Interpretation
Statistical interpretation
The documented design supports a randomized between-group treatment comparison. However, without registered endpoints or posted statistical analyses, no endpoint-specific effect estimate can be interpreted from this registry record.
Clinical interpretation
The registry identifies a phase 3 treatment study in stage III or stage IV non-small cell lung cancer evaluating OSI-774 (Tarceva, erlotinib HCl). Clinical interpretation of efficacy or safety requires documented outcome definitions and results.
19. A Practical Framework for Reading BR.21
A statistically rigorous reading of any randomized clinical trial should proceed from design to endpoint to estimand to analysis and finally to interpretation. BR.21 illustrates why those steps need to remain distinct.
For BR.21, the first step is documented clearly: the trial is randomized and parallel, with double masking. The subsequent endpoint-specific steps cannot be completed from the ClinicalTrials.gov information available here because the registered primary endpoints and statistical analyses are not listed.
20. Statistical Concepts to Watch for in Any Future BR.21 Results
If endpoint-specific results are documented elsewhere in an appropriate public record, several statistical questions are especially important when interpreting them.
- What is the endpoint? The definition determines the appropriate effect measure and analysis.
- What population was analyzed? Results from all randomized participants answer a different question from results restricted by treatment exposure or protocol adherence.
- What is the effect estimate? The numerical magnitude should be considered alongside its confidence interval.
- How was censoring or missingness handled? These decisions can affect time-to-event and other endpoint estimates.
- Were multiple hypotheses tested? Confirmatory interpretation depends on the prespecified error-control strategy.
- Were subgroup findings formally compared? A result in one subgroup and a different result in another do not by themselves establish treatment-effect heterogeneity.
- Were model assumptions appropriate? Regression-based estimates depend on assumptions that should be checked or addressed through sensitivity analyses.
21. Record Summary
BR.21 is documented in ClinicalTrials.gov as a completed, randomized, parallel, double-masked phase 3 treatment trial of OSI-774 (Tarceva, erlotinib HCl) in patients with stage III or stage IV non-small cell lung cancer. The registry reports an enrollment of 731, a study start date of November 1, 2001, and a primary completion date of January 30, 2004.
From a statistical-design perspective, the most important documented features are randomization, parallel-group allocation, double masking, and the phase 3 treatment purpose. These features establish the framework for a controlled treatment comparison.
The registry record available here does not list registered primary endpoints or posted statistical analyses. It therefore supports a detailed discussion of the randomized-trial methodology and the statistical principles required for endpoint analysis, but it does not provide endpoint-specific estimates, confidence intervals, p-values, or safety comparisons to interpret.
Continue through Clinical Biostats
Explore additional statistical tutorials, calculators, and clinical trial analyses across the Clinical Biostats site.
22. Sources
- ClinicalTrials.gov: NCT00036647 — BR.21.