This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record. Numerical trial results and design details on this page are restricted to the ClinicalTrials.gov record.
1. Trial at a Glance
BARI 2D was a randomized, open-label, phase 3 factorial trial with 2368 participants. Its statistical structure separated two principal treatment questions: Revascularization versus Medical Therapy and Insulin Sensitizing versus Insulin Providing glycemic control strategies.
| Feature | BARI 2D |
|---|---|
| Trial name | BARI 2D |
| Brief title | Bypass Angioplasty Revascularization Investigation in Type 2 Diabetes |
| Phase | Phase 3 |
| Status | Completed |
| Start | 2000-09 |
| Primary completion | 2008-11 |
| Lead sponsor | University of Pittsburgh |
| Enrollment | 2368 |
| Allocation | Randomized |
| Design model | Factorial |
| Masking | None |
| Primary purpose | Treatment |
| Primary endpoint | Number of Participants With All-Cause Mortality |
| Primary endpoint time frame | five years |
| ClinicalTrials.gov | NCT00006305 |
2. Clinical Question
The trial contains two principal randomized treatment comparisons within the same factorial framework. The first asks whether a Revascularization strategy differs from Medical Therapy for the primary five-year all-cause mortality endpoint. The second asks whether an Insulin Sensitizing glycemic control strategy differs from an Insulin Providing strategy for the same endpoint.
Population
The registered conditions include Coronary Disease, Cardiovascular Diseases, Heart Diseases, Insulin Resistance, Diabetes Mellitus, and Diabetes Mellitus, Non-Insulin-Dependent.
Revascularization question
Compare Revascularization with Medical Therapy within the factorial design.
Glycemic-strategy question
Compare Insulin Sensitizing with Insulin Providing glycemic control strategies.
Primary question
For each main effect, what is the difference in five-year all-cause mortality between the randomized strategy groups?
3. Trial Design
The design is explicitly identified as factorial, and the primary analyses are described as ITT analyses of the two main effects in a 2×2 factorial design. This structure permits the two principal treatment questions to be evaluated within the same randomized trial population.
Revascularization + Insulin Providing
- Revascularization strategy
- Insulin Providing glycemic control strategy
- Serious adverse events: 230/592
Revascularization + Insulin Sensitizing
- Revascularization strategy
- Insulin Sensitizing glycemic control strategy
- Serious adverse events: 210/584
Medical Therapy + Insulin Providing
- Medical Therapy strategy
- Insulin Providing glycemic control strategy
- Serious adverse events: 230/593
Medical Therapy + Insulin Sensitizing
- Medical Therapy strategy
- Insulin Sensitizing glycemic control strategy
- Serious adverse events: 235/599
Interventions represented in the factorial design
The registry intervention list includes angioplasty, transluminal, percutaneous coronary and other catheter-based interventions; coronary artery bypass; biguanides and thiazolidinediones; insulin and sulfonylurea; and ACE inhibitors, angiotensin receptor blockers, beta blockers, and calcium channel blockers.
4. Endpoints
| Endpoint | Registered definition / wording | Time frame | Type |
|---|---|---|---|
| Primary | Number of Participants With All-Cause Mortality | five years | Binary; analyzed as time-to-event |
| Secondary | Number of Participants With Death, Myocardial Infarction, or Stroke | five years | Time-to-event |
The registry reports one primary endpoint. The posted statistical analyses evaluate that endpoint separately for the two principal factorial main effects. A secondary composite endpoint of Death, Myocardial Infarction, or Stroke is analyzed using the same factorial main-effect framework.
5. Analysis Populations and Factorial Comparisons
The posted primary analyses use an intention-to-treat analysis of the two main effects in the 2×2 factorial design. The analysis therefore preserves the randomized treatment assignments rather than redefining patients according to treatment actually received.
| Analysis | Comparison | Population | Method |
|---|---|---|---|
| Primary main effect | Revascularization vs Medical Therapy | ITT | Log Rank |
| Primary main effect | Insulin Sensitizing vs Insulin Providing | ITT | Log Rank |
| Secondary main effect | Revascularization vs Medical Therapy | ITT | Log Rank |
| Secondary main effect | Insulin Sensitizing vs Insulin Providing | ITT | Log Rank |
6. Statistical Methodology
Intention-to-treat analysis
The registry describes the primary and secondary analyses as intention-to-treat (ITT). Under ITT, participants are analyzed according to the treatment strategy to which they were randomized. This maintains the comparison created by randomization and avoids changing the estimand simply because treatment exposure differs after randomization.
Time-to-event analysis
Although the primary endpoint is registered as a binary outcome—whether participants experienced all-cause mortality within the five-year time frame—the posted statistical analyses identify the endpoint as time-to-event and use a log-rank test. This distinction is important. A time-to-event analysis uses not only whether an event occurred, but also information about when the event occurred and whether follow-up was censored.
Log-rank test
The reported method for all four posted statistical analyses is Log Rank. The log-rank test is a standard nonparametric method for comparing survival or event-time distributions between groups. It evaluates whether the observed pattern of events over follow-up is consistent with the groups having the same underlying event-time distribution.
The registry's formal analysis uses the log-rank method. The posted effect measure is a risk difference, which provides an absolute-scale measure alongside the time-to-event test.
Risk difference
The effect measure reported in each posted analysis is Risk Difference (RD). Conceptually, a risk difference compares the cumulative probability of an event between two groups at the specified time horizon.
An RD below zero indicates a lower estimated event risk in the first-named comparison strategy relative to the reference strategy. An RD of zero corresponds to equal risks at the specified time point.
Confidence intervals
Each posted BARI 2D estimate is accompanied by a two-sided 95% confidence interval. The interval describes uncertainty around the estimated risk difference under the analysis framework. It is not a range containing 95% of individual patient outcomes.
Superiority hypothesis
The registry identifies the hypothesis type for all four analyses as Superiority. This means the statistical question is framed around detecting a difference between the randomized strategies rather than demonstrating that one strategy is merely no worse than another within a prespecified non-inferiority margin.
7. Primary Results: All-Cause Mortality
The primary endpoint was Number of Participants With All-Cause Mortality at five years. Two main-effect analyses were posted: Revascularization versus Medical Therapy and Insulin Sensitizing versus Insulin Providing.
Revascularization vs Medical Therapy
Five-year all-cause mortality risk difference
95% CI: −0.031 to 0.020 · P = 0.97
Analysis: intention-to-treat; Log Rank; superiority hypothesis
The estimated risk difference for all-cause mortality was −0.005 for Revascularization compared with Medical Therapy. On the risk-difference scale, the negative estimate indicates a lower estimated five-year mortality risk for the Revascularization comparison group in the posted analysis.
The estimate does not mean that an individual patient's mortality risk was reduced by exactly 0.005, nor does it establish that every patient benefited. It is a population-level treatment-effect estimate from the randomized comparison.
The 95% confidence interval, −0.031 to 0.020, crosses zero. Thus the interval includes both a possible negative risk difference and a possible positive risk difference under the statistical framework. The interval therefore communicates substantially more information about precision than the point estimate alone.
The P = 0.97 value is evidence against a statistically detectable difference under the specified hypothesis test; it is not a measure of the size or clinical importance of the treatment effect. A p-value should not be interpreted as the probability that the null hypothesis is true.
Because the registry reports a log-rank analysis for a time-to-event endpoint while expressing the effect as a risk difference, the test statistic and effect measure answer related but not identical questions. The ClinicalTrials.gov record does not provide a hazard ratio or Kaplan-Meier estimates for this endpoint, so those measures should not be inferred.
Insulin Sensitizing vs Insulin Providing
Five-year all-cause mortality risk difference
95% CI: −0.029 to 0.022 · P = 0.89
Analysis: intention-to-treat; Log Rank; superiority hypothesis
The estimated risk difference for all-cause mortality was −0.003 for the Insulin Sensitizing glycemic control strategy compared with the Insulin Providing strategy. The negative point estimate indicates a lower estimated five-year mortality risk for the Insulin Sensitizing comparison in the posted analysis.
This estimate is not a statement that an individual participant's probability of death was reduced by exactly 0.003. It describes the estimated difference between randomized strategy groups at the population level.
The 95% confidence interval of −0.029 to 0.022 includes zero. Consequently, the interval is compatible with both a negative and a positive risk difference within the uncertainty represented by this analysis.
The P = 0.89 value does not quantify the magnitude of the observed treatment difference. It describes the evidence against the specified null hypothesis under the test used; it should not be converted into a probability that one strategy is effective or ineffective.
As with the Revascularization analysis, the posted method is a log-rank test while the reported effect measure is a risk difference. The ClinicalTrials.gov record does not report a hazard ratio, Kaplan-Meier estimate, or median time to death, so none should be reconstructed from the risk difference and p-value.
| Primary analysis | Estimate | 95% CI | P-value | Hypothesis |
|---|---|---|---|---|
| Revascularization vs Medical Therapy | RD −0.005 | −0.031 to 0.020 | 0.97 | Superiority |
| Insulin Sensitizing vs Insulin Providing | RD −0.003 | −0.029 to 0.022 | 0.89 | Superiority |
8. Secondary Endpoint Results
The posted secondary endpoint was Number of Participants With Death, Myocardial Infarction, or Stroke, with a time frame of five years. Both posted analyses again evaluate the two principal factorial main effects using ITT and the log-rank method.
Revascularization vs Medical Therapy
Five-year risk difference for Death, MI, or Stroke
95% CI: −0.049 to 0.022 · P = 0.70
The estimated risk difference was −0.013 for Revascularization compared with Medical Therapy. The negative direction indicates a lower estimated event risk for the Revascularization comparison on the reported risk-difference scale.
The 95% confidence interval of −0.049 to 0.022 crosses zero. The data registry-reported therefore do not identify a single direction of the possible population-level risk difference with high precision.
The P = 0.70 value is a hypothesis-test result, not an effect-size measure. It should not be interpreted as saying that the probability of a treatment effect is 30%, 70%, or any other particular percentage.
Insulin Sensitizing vs Insulin Providing
Five-year risk difference for Death, MI, or Stroke
95% CI: −0.060 to 0.012 · P = 0.13
The estimated risk difference was −0.024 for the Insulin Sensitizing strategy compared with the Insulin Providing strategy. The negative point estimate corresponds to a lower estimated risk on the reported absolute scale.
The 95% confidence interval, −0.060 to 0.012, includes zero and therefore permits uncertainty in either direction. The width of the interval also illustrates why a point estimate should not be considered independently of its precision.
The P = 0.13 value does not measure the size of the estimated risk difference. It is the result of the specified superiority hypothesis test and should be interpreted together with the confidence interval and the trial design.
| Secondary analysis | Estimate | 95% CI | P-value | Method |
|---|---|---|---|---|
| Revascularization vs Medical Therapy | RD −0.013 | −0.049 to 0.022 | 0.70 | Log Rank |
| Insulin Sensitizing vs Insulin Providing | RD −0.024 | −0.060 to 0.012 | 0.13 | Log Rank |
9. Safety Results
The ClinicalTrials.gov record reports serious adverse events by each of the four factorial cells. These figures are presented as affected participants divided by participants at risk.
| Factorial cell | Serious adverse events | Affected / at risk |
|---|---|---|
| Revascularization and Insulin Providing | Serious adverse events | 230/592 |
| Revascularization and Insulin Sensitizing | Serious adverse events | 210/584 |
| Medical Therapy and Insulin Providing | Serious adverse events | 230/593 |
| Medical Therapy and Insulin Sensitizing | Serious adverse events | 235/599 |
These are descriptive serious-adverse-event counts by factorial cell. The statistical analyses posted on ClinicalTrials.gov do not report a formal risk difference, confidence interval, or p-value for serious adverse events. Accordingly, the safety data should not be converted into an inferential comparison that is not present in the registry analysis.
10. Factorial Design: Why the Four Arms Matter
A 2×2 factorial trial simultaneously varies two treatment dimensions. In BARI 2D, one dimension was the Revascularization versus Medical Therapy comparison, and the other was the Insulin Sensitizing versus Insulin Providing comparison.
| Insulin Providing | Insulin Sensitizing | |
|---|---|---|
| Revascularization | Revascularization + Insulin Providing 230/592 serious AEs | Revascularization + Insulin Sensitizing 210/584 serious AEs |
| Medical Therapy | Medical Therapy + Insulin Providing 230/593 serious AEs | Medical Therapy + Insulin Sensitizing 235/599 serious AEs |
The statistical advantage of the factorial structure is that the two main questions can be evaluated within the same trial rather than requiring two completely separate randomized studies. The posted analyses explicitly reflect this structure by describing the ITT analysis as an analysis of the two main effects.
Main effect 1
Revascularization is compared with Medical Therapy across the factorial glycemic-strategy dimension.
Main effect 2
Insulin Sensitizing is compared with Insulin Providing across the factorial revascularization dimension.
Four randomized cells
The four combinations make up the complete factorial structure and preserve the joint randomized design.
Interaction question
The ClinicalTrials.gov record identifies the two main effects but do not provide a formal interaction estimate or interaction p-value.
11. Multiplicity and the Two Main Effects
BARI 2D has more than one statistical comparison because the factorial design contains two principal main effects and the registry also posts a secondary endpoint. That creates an important interpretive distinction between the number of analyses reported and the error-control strategy used for those analyses.
The ClinicalTrials.gov record identifies the hypothesis type as superiority and identify the four analyses as two primary and two secondary analyses. However, the ClinicalTrials.gov record does not specify an alpha-allocation procedure, multiplicity-adjustment method, hierarchical testing sequence, or other formal familywise-error strategy.
12. Kaplan-Meier Estimation and Censoring
The ClinicalTrials.gov record classifies the mortality and composite-event endpoints as time-to-event and specify a log-rank test. A typical time-to-event analysis of this type uses Kaplan-Meier methods to describe the event-time distribution while retaining information from participants whose event has not occurred during observed follow-up.
Here, di represents events at an event time and ni represents participants at risk immediately before that time. The registry-reported BARI 2D data do not provide the underlying event and censoring times needed to reconstruct a Kaplan-Meier curve.
This matters because a five-year endpoint can be summarized in more than one statistical way. A simple binary analysis asks whether an event occurred by five years. A time-to-event analysis additionally uses the timing of events and censoring. The registry's classification and log-rank method show that the posted analysis belongs to the latter framework.
13. Why a Risk Difference Is Useful Here
A risk difference is an absolute measure. Unlike a relative measure such as a hazard ratio, it expresses the estimated difference in cumulative event risk between two groups.
A negative risk difference means the estimated risk is lower in the first-named comparison group than in the reference group. A positive value would indicate higher estimated risk, while zero represents equal risk on the risk-difference scale.
If a 95% confidence interval crosses zero, the interval includes the possibility of no absolute risk difference. That does not prove that the two strategies are identical; it indicates that the estimated difference is uncertain enough that zero remains compatible with the interval.
The p-value quantifies how unusual the observed data would be under the null hypothesis represented by the statistical test. It does not tell us how large the treatment effect is, how clinically important it is, or the probability that the treatment hypothesis is true.
14. Statistical Methods Explained
Why was a factorial design used?
A factorial design allows two treatment questions to be studied within the same randomized experiment. BARI 2D's posted analysis explicitly evaluates the two main effects: Revascularization versus Medical Therapy and Insulin Sensitizing versus Insulin Providing. The four treatment combinations form the 2×2 structure.
Why is intention-to-treat important?
ITT maintains the original randomized comparison. If participants later receive a different treatment, discontinue therapy, or otherwise differ in exposure, reclassifying them can undermine the balance produced by randomization. The registry specifically identifies ITT as the population for the posted main-effect analyses.
Why use a log-rank test for a five-year mortality endpoint?
The registry classifies the endpoint as time-to-event for the statistical analysis. The log-rank test is designed to compare event-time distributions over follow-up rather than treating every participant simply as an event/no-event observation without regard to event timing.
What does a risk difference of −0.005 mean?
It is an absolute-scale estimate indicating a lower five-year all-cause mortality risk in the Revascularization comparison than in Medical Therapy by 0.005 on the reported risk-difference scale. It is not a relative risk, hazard ratio, or individual-level probability.
Why does the confidence interval matter more than the point estimate alone?
The point estimate is only one possible estimate supported by the data. The 95% confidence interval communicates the statistical precision around it. For the primary Revascularization analysis, the interval extends from −0.031 to 0.020; for the primary glycemic-strategy analysis, it extends from −0.029 to 0.022. Both intervals cross zero.
Why does a p-value not measure effect size?
A p-value depends on the observed data and the null hypothesis used by the test. It does not directly encode the magnitude of the risk difference. Effect size and uncertainty are better described by the estimate and confidence interval, while the p-value addresses evidence against the null under the specified test.
What does the registry not establish about interaction?
The ClinicalTrials.gov record establishes that BARI 2D used a 2×2 factorial design and analyzed two main effects. They do not provide an interaction estimate or interaction p-value. Therefore, the available data do not support a formal conclusion about whether the effect of one treatment dimension depended on the level of the other treatment dimension.
15. Interpreting the Primary Results Together
| Question | Risk difference | 95% CI | P-value | Statistical reading |
|---|---|---|---|---|
| Revascularization vs Medical Therapy All-cause mortality, five years | −0.005 | −0.031 to 0.020 | 0.97 | CI crosses zero; no statistically detectable difference under the posted test |
| Insulin Sensitizing vs Insulin Providing All-cause mortality, five years | −0.003 | −0.029 to 0.022 | 0.89 | CI crosses zero; no statistically detectable difference under the posted test |
The two primary estimates are both close to zero on the risk-difference scale, and both confidence intervals include zero. The p-values are correspondingly large. The appropriate statistical description is therefore based on the reported estimates, their uncertainty intervals, and the specified log-rank tests—not on a claim that the competing strategies are mathematically identical.
In particular, failure to detect a statistically significant difference is not equivalent to proving equivalence. The ClinicalTrials.gov record identifies the hypotheses as superiority, not equivalence or non-inferiority. A conclusion of equivalence would require a different prespecified statistical framework and margins.
16. Primary vs Secondary Evidence
Primary endpoint
Number of Participants With All-Cause Mortality at five years, with two main-effect analyses.
Secondary endpoint
Number of Participants With Death, Myocardial Infarction, or Stroke at five years, again analyzed for the two main effects.
Primary method
Log-rank test in an intention-to-treat analysis of the 2×2 factorial main effects.
Effect measure
Risk difference with a two-sided 95% confidence interval.
The distinction matters because secondary endpoints should not automatically be treated as interchangeable with the primary endpoint. Death, myocardial infarction, or stroke is a composite outcome, whereas the primary endpoint is all-cause mortality. A treatment effect on a composite endpoint can reflect its individual components in different ways, although the ClinicalTrials.gov record does not provide component-specific results.
17. What the Posted Analyses Do Not Provide
The registry-reported BARI 2D data contain formal estimates and confidence intervals for the primary and secondary main-effect analyses, but they do not report several quantities that are often displayed on a detailed clinical-trial results page.
| Potential result | Availability in the ClinicalTrials.gov record |
|---|---|
| Risk differences | Reported for all four statistical analyses |
| 95% confidence intervals | Reported for all four statistical analyses |
| P-values | Reported for all four statistical analyses |
| Log-rank method | Reported |
| Hazard ratios | Not reported in the ClinicalTrials.gov record |
| Kaplan-Meier estimates | Not reported in the ClinicalTrials.gov record |
| Median survival | Not reported in the ClinicalTrials.gov record |
| Subgroup estimates | Not reported in the ClinicalTrials.gov record |
| Formal interaction analysis | Not reported in the ClinicalTrials.gov record |
| Multiplicity-adjustment procedure | Not reported in the ClinicalTrials.gov record |
| Interim-analysis procedure | Not reported in the ClinicalTrials.gov record |
| Missing-data or imputation procedure | Not reported in the ClinicalTrials.gov record |
18. Limitations
- Registry-level detail: the ClinicalTrials.gov record contains the posted statistical analyses but not the full statistical analysis plan, so some implementation details cannot be established.
- Risk difference with log-rank testing: the registry pairs a time-to-event log-rank method with a risk-difference effect measure. These are complementary but conceptually different statistical quantities.
- No hazard ratio reported: the ClinicalTrials.gov record does not provide a hazard ratio, so relative instantaneous event-rate interpretations should not be substituted for the reported risk differences.
- No Kaplan-Meier estimates reported: although the endpoint is identified as time-to-event, the ClinicalTrials.gov record does not contain the event and censoring information needed to reproduce survival curves.
- No interaction analysis reported: the factorial design supports main-effect analysis, but the ClinicalTrials.gov record does not report a formal test of interaction between the two treatment dimensions.
- Multiplicity details unavailable: the ClinicalTrials.gov record identifies multiple analyses but do not specify an alpha-allocation or multiplicity-adjustment procedure.
- Safety inference is limited: serious adverse events are reported descriptively by factorial cell, without corresponding inferential comparisons in the statistical analyses posted on ClinicalTrials.gov.
- Composite endpoint: Death, Myocardial Infarction, or Stroke combines multiple clinical events, but the ClinicalTrials.gov record does not provide component-specific estimates.
- No non-inferiority framework: the posted hypotheses are superiority hypotheses. The reported results should therefore not be interpreted using non-inferiority margins or equivalence criteria.
19. Why This Trial Matters Statistically
BARI 2D is a useful teaching case because its statistical structure is more complicated than a simple two-arm randomized comparison. The trial simultaneously evaluates two principal treatment dimensions and analyzes a time-to-event endpoint under an intention-to-treat framework.
| Concept | How it appears in BARI 2D |
|---|---|
| Randomization | The trial is randomized with 2368 participants. |
| Factorial design | The design model is factorial, with a 2×2 factorial structure in the posted analyses. |
| Two main effects | Revascularization vs Medical Therapy and Insulin Sensitizing vs Insulin Providing. |
| Intention-to-treat | Primary and secondary analyses use ITT. |
| Time-to-event analysis | The mortality and composite endpoints are classified as time-to-event for the statistical analyses. |
| Log-rank test | Used for all four posted statistical analyses. |
| Risk difference | Reported as the effect measure for all four analyses. |
| Confidence interval | Two-sided 95% intervals accompany every posted estimate. |
| Superiority testing | All four analyses are identified as superiority hypotheses. |
| Composite endpoint | Death, Myocardial Infarction, or Stroke is evaluated as a secondary time-to-event endpoint. |
| Safety by factorial cell | Serious adverse events are reported separately for all four randomized cells. |
20. Statistical Concepts in This Trial
Learn more about the methods used in this trial:
21. Related Statistical Calculators
Use these tools to explore the statistical concepts underlying the BARI 2D analyses:
22. Sources
- ClinicalTrials.gov: BARI 2D — NCT00006305.
- PubMed record: PMID 31590967.
- PubMed record: PMID 30286920.
- PubMed record: PMID 28903941.
- PubMed record: PMID 28126156.
- PubMed record: PMID 27289411.
Continue through the Clinical Biostats statistical pathway
Explore the statistical methods that recur across randomized trials, survival analyses, factorial designs, and clinical endpoint interpretation.
23. Record Summary
BARI 2D is a randomized phase 3 factorial trial with 2368 participants and four treatment cells. Its statistical analysis is centered on two randomized main effects: Revascularization versus Medical Therapy and Insulin Sensitizing versus Insulin Providing. The primary endpoint is all-cause mortality at five years, analyzed using an ITT framework and a log-rank test with risk difference as the reported effect measure.
The posted primary estimates were −0.005 for Revascularization versus Medical Therapy and −0.003 for Insulin Sensitizing versus Insulin Providing. Their respective two-sided 95% confidence intervals were −0.031 to 0.020 and −0.029 to 0.022, with p-values of 0.97 and 0.89. The secondary Death, Myocardial Infarction, or Stroke analyses produced risk differences of −0.013 and −0.024, with corresponding 95% confidence intervals of −0.049 to 0.022 and −0.060 to 0.012.
Statistically, the most instructive feature is the combination of a factorial design, ITT analysis, time-to-event methodology, log-rank testing, and absolute risk differences with confidence intervals. Interpreting the trial correctly requires keeping those components distinct: the p-value addresses the hypothesis test, the risk difference describes the estimated absolute effect, the confidence interval describes its precision, and the factorial design determines how the two principal treatment questions are represented in the analysis.