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
The Bypass Angioplasty Revascularization Investigation (BARI) was a completed phase 3 randomized parallel-group treatment trial sponsored by the National Heart, Lung, and Blood Institute. The registry identifies percutaneous coronary angioplasty and coronary artery bypass as the interventions and lists angina pectoris, cardiovascular diseases, coronary disease, diabetes mellitus, heart diseases, and myocardial ischemia among the conditions studied.
| Feature | BARI |
|---|---|
| Trial name | Bypass Angioplasty Revascularization Investigation (BARI) |
| Phase | Phase 3 |
| Status | Completed |
| Allocation | Randomized |
| Design model | Parallel |
| Primary purpose | Treatment |
| Interventions | Angioplasty, transluminal, percutaneous coronary (procedure); coronary artery bypass (procedure) |
| Lead sponsor | National Heart, Lung, and Blood Institute (NHLBI) |
| Sponsor type | NIH |
| ClinicalTrials.gov | NCT00000462 |
2. Clinical Question
The registry identifies two procedural interventions within a randomized treatment comparison: percutaneous coronary angioplasty and coronary artery bypass. The clinical setting includes coronary disease and related cardiovascular conditions, including angina pectoris, myocardial ischemia, heart diseases, cardiovascular diseases, and diabetes mellitus.
Population
The registry lists angina pectoris, cardiovascular diseases, coronary disease, diabetes mellitus, heart diseases, and myocardial ischemia as conditions associated with the study.
Intervention
Angioplasty, transluminal, percutaneous coronary (procedure).
Comparator
Coronary artery bypass (procedure).
Primary question
Within the randomized treatment design, how do the outcomes associated with percutaneous coronary angioplasty compare with those associated with coronary artery bypass?
3. Trial Design
The registry describes BARI as a randomized, parallel-group, phase 3 treatment trial. Randomization is important statistically because it is intended to create treatment groups whose differences in outcomes can be compared within a common experimental framework rather than relying on a purely observational comparison.
Percutaneous coronary angioplasty
- Angioplasty, transluminal, percutaneous coronary
- Procedure-based treatment strategy
- Randomized allocation within a parallel design
Coronary artery bypass
- Coronary artery bypass
- Procedure-based treatment strategy
- Randomized allocation within a parallel design
The registry's designation of parallel means the two randomized intervention groups are treated as concurrent comparison groups rather than as sequential phases of the same participants. That distinction matters because the statistical comparison is anchored to the randomized assignment rather than to a within-person before-and-after contrast.
4. Endpoints
No registered primary endpoints are listed in the ClinicalTrials.gov record represented here. Consequently, the record does not specify a primary outcome definition, assessment time frame, event definition, or endpoint-specific estimand that can be reproduced from the registry.
| Endpoint element | Registry record |
|---|---|
| Registered primary endpoints | None listed |
| Primary endpoint definition | Not reported in the registry record |
| Primary endpoint time frame | Not reported in the registry record |
| Posted statistical analyses | None listed |
| Posted numerical results | None listed |
5. Statistical Methodology
Randomization as the foundation of the comparison
Randomization is the central statistical feature explicitly identified by the registry. In a randomized comparison, treatment assignment is determined by the trial's allocation mechanism rather than by the participant's prognosis or the investigator's preference. This supports causal interpretation of differences between the randomized groups, provided the trial is conducted and analyzed according to its design.
Parallel-group comparison
The parallel design means that participants assigned to one procedural strategy form a comparison group distinct from participants assigned to the other strategy. For a parallel trial, the analysis ordinarily compares outcomes between the randomized groups using an endpoint-appropriate estimator and an analysis population defined in the statistical plan.
Endpoint-dependent analysis
Because no registered primary endpoint is listed in the record, a specific endpoint model cannot be attributed to the registry. The statistical method should follow the outcome's measurement scale and follow-up structure. A continuous outcome might be compared using a difference in means or an adjusted regression model; a binary outcome might be summarized with risks, risk differences, or odds ratios; and a time-to-event outcome would generally require methods that account for censoring and event times.
The appropriate numerical form of this contrast depends on the endpoint. The difference might be expressed as a mean difference, risk difference, ratio, odds ratio, hazard ratio, or another prespecified estimand.
Intention-to-treat principle
For a randomized treatment comparison, intention-to-treat analysis is commonly used to preserve the comparison created by randomization. Participants are analyzed according to their randomized assignment rather than being reassigned solely because their treatment exposure changes. The registry record represented here does not specify an analysis population, so a particular ITT definition should not be attributed to BARI's registry record.
Confidence intervals
For an appropriately specified treatment effect, a confidence interval provides information about statistical precision. It should be interpreted together with the point estimate and the clinical scale of the endpoint. A narrow interval indicates greater statistical precision than a wide interval, while neither interval width nor the p-value alone establishes clinical importance.
6. Planned Analysis
The registry identifies BARI as a randomized phase 3 treatment trial comparing percutaneous coronary angioplasty with coronary artery bypass, but it does not list registered primary endpoints or posted statistical analyses.
For a trial of this design, the planned analysis would ordinarily begin by defining the primary endpoint precisely and then selecting an estimator appropriate to that endpoint. If the primary outcome were a binary event, the analysis could compare event proportions between randomized groups and report an effect measure with a confidence interval. If the outcome were measured continuously, the analysis could compare means or another prespecified location parameter. If the outcome were time-to-event, Kaplan-Meier estimation and a suitable between-group model could be used, with censoring handled according to the prespecified analysis plan.
What would be estimated?
The treatment effect should be defined before analysis. The form of the estimate depends on the primary endpoint rather than on the trial's phase or procedure names.
What would be tested?
The hypothesis would compare the prespecified outcome between the randomized intervention groups using the test appropriate for that endpoint.
What would determine precision?
Precision would depend on the amount of information available for the endpoint, including the number of participants, events or measurements, variability, and follow-up.
What has been posted?
No formal statistical analyses are listed in the ClinicalTrials.gov record represented here, and no numerical outcome results are posted in the ClinicalTrials.gov record.
7. Statistical Methods Explained
Why does randomization matter statistically?
Randomization separates treatment assignment from the investigators' selection of therapy. If allocation is properly implemented, prognostic factors—both measured and unmeasured—are expected to be distributed between groups by the randomization process. This is the principal reason a randomized comparison can support causal inference more directly than a simple comparison of patients who chose different procedures.
Why is a parallel design different from a crossover design?
In a parallel trial, each participant contributes primarily to the treatment group to which that participant was randomized. A crossover design instead exposes participants to multiple interventions in sequence. The parallel structure avoids the need to model treatment-period effects and carryover effects that are central to crossover studies.
How would a binary endpoint usually be analyzed?
A binary endpoint has two possible states, such as occurrence or non-occurrence of a prespecified event. Common effect measures include a risk difference, risk ratio, or odds ratio. The preferred measure should be specified before the results are examined because these measures have different interpretations.
How would a time-to-event endpoint usually be analyzed?
When the timing of an event matters and some participants have not experienced the event by the end of observation, ordinary proportions can discard important information. Kaplan-Meier methods can estimate the event-free probability over time, while models such as the Cox proportional-hazards model can estimate a hazard ratio when its assumptions are appropriate.
Why are confidence intervals more informative than a p-value alone?
A p-value addresses compatibility of the observed data with a specified null hypothesis. It does not measure the magnitude of the treatment effect. A confidence interval places the estimated effect on its clinical scale and shows the range of values compatible with the statistical model and sampling framework at the stated confidence level.
Why must the primary endpoint be defined before choosing the model?
Different outcomes contain different statistical information. A continuous measurement, a binary event, and a censored survival time cannot be treated as interchangeable data structures. The endpoint definition determines the estimand, the appropriate analysis population, the handling of missing observations, and the interpretation of the resulting treatment effect.
8. Missing Data and Analysis Populations
The ClinicalTrials.gov record represented here does not specify an analysis population, missing-data method, imputation strategy, censoring rules, or estimand. These details can materially affect the interpretation of a clinical-trial result and ordinarily belong in the protocol and statistical analysis plan.
| Statistical issue | Why it matters | Registry record |
|---|---|---|
| Analysis population | Defines which randomized participants contribute to the primary analysis. | Not reported |
| Missing data | Determines how incomplete outcome information affects estimation. | Not reported |
| Imputation | Can affect estimates when outcome information is unavailable. | Not reported |
| Censoring | Important when the outcome is measured as time to an event. | Not reported |
| Estimand | Defines precisely the treatment effect being targeted. | Not reported |
These are not merely technical details. For example, two analyses of the same randomized population can produce different estimates if they target different treatment-policy questions or handle intercurrent events differently. A statistically rigorous interpretation therefore requires knowing what effect the analysis was designed to estimate.
9. Multiplicity and Hypothesis Testing
No multiplicity strategy is reported in the ClinicalTrials.gov record represented here. The issue becomes important whenever a trial has multiple primary endpoints, multiple time points, multiple treatment comparisons, or multiple confirmatory subgroup hypotheses.
Familywise error
Testing several hypotheses at an unadjusted significance level can increase the probability of obtaining at least one statistically significant result by chance.
Prespecification
A prespecified hierarchy or multiplicity procedure establishes which hypotheses are confirmatory and how the type I error budget is allocated.
Without a documented endpoint hierarchy, an isolated p-value cannot establish whether a finding represents a confirmatory result or one of several exploratory comparisons.
10. Interim Analysis
No interim-analysis plan is listed in the ClinicalTrials.gov record represented here. An interim analysis is a formal statistical examination of accumulating trial data before the final analysis and should be distinguished from an informal decision to look at results.
If interim efficacy testing is planned, the statistical design generally specifies the information times, stopping boundaries, and overall type I error control before unblinded comparative results are examined. Methods such as alpha spending or group-sequential boundaries can preserve the intended error rate while allowing prespecified interim decisions.
A valid sequential design accounts for the fact that the same accumulating evidence may be examined more than once.
11. Stratification and Covariate Adjustment
The registry record represented here does not identify randomization strata or prespecified adjustment covariates. Randomization itself establishes the primary treatment comparison; additional covariate adjustment may improve precision when it is prespecified and the model is appropriate.
Adjustment should not be confused with replacing randomization. In a well-designed randomized trial, the randomized comparison remains the central source of treatment-group comparability, while covariate adjustment can address prognostic imbalance or improve statistical efficiency.
12. Bayesian Methods
No Bayesian statistical method is identified in the ClinicalTrials.gov record represented here. A Bayesian analysis would combine a prior distribution with the observed data to produce a posterior distribution for the treatment effect or another parameter of interest.
A Bayesian analysis would require an explicitly specified prior and likelihood model. Neither is identified in the registry record represented here.
13. Non-Inferiority and Equivalence
No non-inferiority margin or equivalence margin is reported in the ClinicalTrials.gov record represented here. Therefore, the registry record does not characterize the BARI comparison using a non-inferiority or equivalence framework.
Statistically, non-inferiority and superiority are different questions. A non-inferiority analysis asks whether the experimental treatment is not unacceptably worse than the comparator by more than a prespecified margin. That margin must be clinically justified and established before the comparison is interpreted.
14. Safety Analysis
The registry record represented here does not provide posted serious-adverse-event counts by randomized arm, adverse-event rates, treatment-emergent adverse-event definitions, or a statistical safety analysis.
| Safety element | Registry record |
|---|---|
| Serious adverse events by arm | Not reported |
| Adverse-event rates | Not reported |
| Safety analysis population | Not reported |
| Safety statistical method | Not reported |
Safety analysis is conceptually distinct from efficacy analysis. For procedural interventions, a complete statistical assessment would ordinarily distinguish the frequency of adverse events from their severity, timing, relatedness, and clinical consequences.
15. Clinical Interpretation vs Statistical Interpretation
Statistical interpretation
The explicit statistical information in the registry is the randomized parallel-group design. The registry record represented here does not provide a registered primary endpoint or posted statistical result from which a treatment-effect estimate can be interpreted.
Clinical interpretation
The clinical comparison concerns two revascularization procedures in a study associated with coronary and cardiovascular disease, including angina pectoris and myocardial ischemia. The registry does not provide outcome estimates with which to quantify their comparative effects.
16. Important Limitations and Interpretation Issues
- No registered primary endpoint: the registry record represented here does not identify a primary outcome or its assessment time frame.
- No posted statistical analysis: the record does not list a formal statistical analysis for a primary endpoint.
- No posted numerical results: the registry record represented here does not provide numerical treatment-effect estimates, confidence intervals, or p-values.
- No enrollment number: the enrollment field in the registry record does not provide a numerical enrollment value.
- Unknown analysis population: the registry record does not specify whether efficacy analyses used an intention-to-treat, per-protocol, modified intention-to-treat, or another population.
- Unknown missing-data strategy: no imputation or missing-data method is specified in the record represented here.
- Unknown endpoint structure: without a registered primary endpoint, it is not possible to determine whether the principal outcome was continuous, binary, time-to-event, or another data type.
- Unknown multiplicity plan: the record does not specify how multiple endpoints, comparisons, or interim looks would have been handled.
- Unknown stratification plan: randomization strata and prespecified covariates are not identified in the registry record represented here.
17. Why This Trial Matters Statistically
BARI is a useful statistical case study because the registry explicitly identifies a randomized phase 3 parallel treatment design comparing two clinically distinct procedural strategies. That design creates a natural framework for comparative-effect estimation while also illustrating why the endpoint definition and analysis plan are essential to interpretation.
| Concept | How it appears in BARI |
|---|---|
| Randomization | The registry identifies the allocation as randomized. |
| Parallel design | The registry identifies the design model as parallel. |
| Phase 3 | The registry identifies the study as phase 3. |
| Treatment purpose | The primary purpose is listed as treatment. |
| Comparative procedures | Percutaneous coronary angioplasty and coronary artery bypass are the listed interventions. |
| Endpoint specification | No registered primary endpoint is listed in the record represented here. |
| Effect estimation | The appropriate effect measure depends on the endpoint and its statistical structure. |
| Uncertainty | A complete analysis would ordinarily report confidence intervals or another prespecified measure of uncertainty. |
| Analysis population | The registry record represented here does not specify the primary analysis population. |
| Multiplicity | No multiplicity strategy is reported in the registry record represented here. |
| Interim analysis | No interim-analysis plan is reported in the registry record represented here. |
| Bayesian methods | No Bayesian analysis is identified in the registry record represented here. |
18. Understanding the Treatment Effect in a Randomized Procedural Trial
There is no single statistical number called "the BARI treatment effect." The treatment effect must be defined in relation to a particular outcome. For example, a difference in event probabilities answers a different question from a difference in mean measurement or a hazard ratio for time to an event.
A numerical difference cannot be interpreted as favorable or unfavorable until the endpoint and coding direction are known. For an endpoint where higher values represent better outcomes, a positive difference has a different meaning from a positive difference on an endpoint where higher values represent more adverse outcomes.
A point estimate by itself does not communicate how precisely the treatment effect was estimated. Confidence intervals or another prespecified uncertainty measure are therefore essential components of a complete statistical result.
A p-value addresses a statistical hypothesis under a specified testing framework. It does not tell the reader whether an observed difference is clinically important. Clinical importance requires the endpoint's scale, the magnitude of the effect, its uncertainty, and the context in which the outcome matters.
19. A Framework for Reading the BARI Comparison
The randomized design can be understood as a sequence of statistical questions rather than as a single test.
What was randomized?
The registry identifies percutaneous coronary angioplasty and coronary artery bypass as the two procedural interventions in the parallel treatment design.
What outcome was prespecified?
The registry record represented here does not list a registered primary endpoint, so the primary outcome cannot be identified from this record.
What estimand should be used?
The estimand follows from the endpoint. Binary, continuous, and time-to-event outcomes require different statistical summaries.
How uncertain is the estimate?
A complete result should include an appropriate confidence interval or another prespecified measure of uncertainty.
Does the result answer the clinical question?
The final interpretation depends on the prespecified endpoint, analysis population, treatment-effect definition, and clinical meaning of the observed difference.
20. What a Complete Statistical Report Would Contain
For a randomized phase 3 procedural trial, a complete statistical report would ordinarily connect the design to a prespecified estimand and then report the corresponding treatment effect with uncertainty.
| Component | Purpose |
|---|---|
| Primary endpoint | Defines the principal outcome used to evaluate the treatment comparison. |
| Analysis population | Defines which participants contribute to the primary analysis. |
| Estimand | Defines exactly which treatment effect is being estimated. |
| Point estimate | Quantifies the observed treatment contrast on the chosen statistical scale. |
| Confidence interval | Describes statistical precision around the estimate. |
| Hypothesis test | Assesses compatibility with a prespecified null hypothesis. |
| Missing-data strategy | Defines how incomplete observations are handled. |
| Multiplicity strategy | Controls interpretation when multiple confirmatory hypotheses are tested. |
| Sensitivity analyses | Assess whether conclusions are robust to reasonable alternative assumptions. |
The ClinicalTrials.gov record represented here provides the randomized design and intervention comparison but does not supply these endpoint-specific statistical details.
21. Sources
- ClinicalTrials.gov: BARI — NCT00000462.
Continue through Clinical Biostats
Explore additional clinical trial results, statistical tutorials, and analysis resources across the Clinical Biostats site.
22. Record Summary
BARI is registered as a completed phase 3 randomized parallel-group treatment trial sponsored by the National Heart, Lung, and Blood Institute. The registry identifies percutaneous coronary angioplasty and coronary artery bypass as the interventions and associates the study with angina pectoris, cardiovascular diseases, coronary disease, diabetes mellitus, heart diseases, and myocardial ischemia. From a statistical perspective, the randomized parallel design establishes the foundation for a between-group treatment comparison, but the registry record represented here does not identify a registered primary endpoint, analysis population, statistical analysis, or numerical outcome result.