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
RECORD was a randomized, open-label, parallel phase 3 trial in type 2 diabetes. The trial enrolled 4,447 participants and evaluated cardiovascular outcomes in a comparison of combined rosiglitazone with combined metformin/sulfonylurea therapy.
| Feature | RECORD |
|---|---|
| Phase | Phase 3 |
| Condition | Diabetes Mellitus, Type 2 |
| Design | Randomized, parallel, open-label |
| Allocation | Randomized |
| Primary purpose | Treatment |
| Enrollment | 4,447 |
| Interventions | Rosiglitazone; sulfonylurea; metformin |
| Registered primary endpoints | 10 |
| Primary endpoint type | Binary |
| Statistical analyses posted | 10 |
| ClinicalTrials.gov | NCT00379769 |
2. Clinical Question
The principal cardiovascular question was whether treatment involving rosiglitazone was non-inferior to the combined metformin/sulfonylurea comparison for the time to cardiovascular death or cardiovascular hospitalization. The registry also reports independent re-adjudication analyses addressing all-cause death, major adverse cardiovascular events, cardiovascular or unknown death, myocardial infarction, and stroke.
Population
Participants with type 2 diabetes mellitus enrolled in the randomized phase 3 RECORD trial.
Rosiglitazone strategy
Rosiglitazone was evaluated in combination with background metformin or sulfonylurea therapy.
Comparison strategy
Combined metformin and sulfonylurea therapy served as the comparison group for the posted primary statistical analyses.
Primary question
Is the time to cardiovascular death or cardiovascular hospitalization with combined rosiglitazone non-inferior to combined metformin/sulfonylurea therapy?
3. Trial Design
Combined rosiglitazone
- Rosiglitazone in addition to background metformin
- Rosiglitazone in addition to background sulfonylurea
Combined metformin / sulfonylurea
- Sulfonylurea in addition to background metformin
- Metformin in addition to background sulfonylurea
The posted statistical analyses combine the two rosiglitazone strategies and compare them with the combined metformin/sulfonylurea strategies. This is important because the principal reported hazard ratios do not represent four separate pairwise comparisons.
4. Endpoints
The registry lists 10 primary endpoints. All use the time frame from baseline through the end of the study, up to 7.5 years. Although the registered endpoint wording describes numbers of participants with events, the posted statistical analyses evaluate these outcomes as time-to-event endpoints using hazard ratios.
| # | Registered primary endpoint | Time frame |
|---|---|---|
| 1 | Number of participants with cardiovascular death/cardiovascular hospitalisation events | Baseline through end of study (up to 7.5 years) |
| 2 | Independent re-adjudication outcome: number of participants who died due to any cause | Baseline through end of study (up to 7.5 years) |
| 3 | Independent re-adjudication: first MACE based on original RECORD endpoint definitions | Baseline through end of study (up to 7.5 years) |
| 4 | Independent re-adjudication: first MACE based on contemporary endpoint definitions | Baseline through end of study (up to 7.5 years) |
| 5 | Independent re-adjudication: CV or unknown death based on original RECORD endpoint definitions | Baseline through end of study (up to 7.5 years) |
| 6 | Independent re-adjudication: CV or unknown death based on contemporary endpoint definitions | Baseline through end of study (up to 7.5 years) |
| 7 | Independent re-adjudication: fatal and non-fatal myocardial infarction based on original RECORD endpoint definitions | Baseline through end of study (up to 7.5 years) |
| 8 | Independent re-adjudication: fatal and non-fatal myocardial infarction based on contemporary endpoint definitions | Baseline through end of study (up to 7.5 years) |
| 9 | Independent re-adjudication: fatal and non-fatal stroke based on original RECORD endpoint definitions | Baseline through end of study (up to 7.5 years) |
| 10 | Independent re-adjudication: fatal and non-fatal stroke based on contemporary endpoint definitions | Baseline through end of study (up to 7.5 years) |
Endpoint definitions
The cardiovascular death/cardiovascular hospitalization endpoint counted cardiovascular deaths and cardiovascular hospitalizations, excluding planned admissions not associated with worsening of the participant's disease or condition.
The independent re-adjudication endpoints used either the original RECORD definitions or contemporary definitions from the Standard Data Collection for Cardiovascular Trials Initiative. The two MACE definitions consisted of cardiovascular or unknown death, non-fatal myocardial infarction, and non-fatal stroke.
Under the original RECORD definitions, cardiovascular death was defined as death for which an unequivocal non-cardiovascular cause could not be established. Unknown deaths were counted as cardiovascular deaths for the relevant analysis. Myocardial infarction and stroke were defined using the detailed hospitalization, biomarker, electrocardiographic, neurological, and adjudicated-cause criteria specified in the registry.
The contemporary definitions used myocardial necrosis in a clinical setting consistent with myocardial ischemia for myocardial infarction and acute neurological dysfunction caused by focal or global brain, spinal cord, or retinal vascular injury for stroke. Contemporary cardiovascular death included death following acute myocardial infarction, sudden cardiac death, heart failure, stroke, or other cardiovascular causes; deaths of unknown cause were counted as cardiovascular deaths.
5. Statistical Methodology
The posted primary analyses use the intent-to-treat population, defined in the registry as all randomized participants who were treated with at least one dose of study medication. The comparison is combined rosiglitazone versus combined metformin/sulfonylurea.
The registry reports hazard ratios with 95% confidence intervals for all 10 posted primary analyses. The specific statistical method used to generate the hazard ratios is not reported in the posted analysis record.
| Component | Registry information |
|---|---|
| Analysis population | Intent-to-Treat population: all randomized participants treated with at least one dose of study medication |
| Groups compared | Combined RSG vs Combined MET/SU |
| Effect measure | Hazard ratio |
| Confidence level | 95% |
| Endpoint analysis type | Time-to-event |
| Formal method | Not reported |
| Non-inferiority margin | 1.2 for the upper limit of the 95% confidence interval in the time-to-event analysis for cardiovascular death/cardiovascular hospitalization |
Why the ITT population matters
The ITT definition preserves the randomized treatment assignment as the basis of comparison. In a randomized trial, this helps maintain the comparability created by randomization rather than redefining treatment groups according to what participants subsequently received.
Here, the distinction is especially relevant because the posted analyses combine treatment strategies before calculating the hazard ratio. The interpretation therefore concerns the randomized comparison of the combined strategies represented in the registry analysis rather than an isolated pharmacologic comparison between individual drugs.
6. Primary Results: Cardiovascular Death or Cardiovascular Hospitalization
Combined rosiglitazone vs combined metformin/sulfonylurea
95% CI: 0.85–1.16
Non-inferiority margin: 1.2 for the upper limit of the 95% confidence interval
| Endpoint | Comparison | HR | 95% CI | Hypothesis |
|---|---|---|---|---|
| Cardiovascular death / cardiovascular hospitalization | Combined RSG vs Combined MET/SU | 0.99 | 0.85–1.16 | Non-inferiority or equivalence |
An HR of 0.99 is very close to 1.00. On the hazard-ratio scale, it corresponds to an estimated instantaneous event rate that is approximately 1% lower for combined rosiglitazone than for combined metformin/sulfonylurea. This is a relative time-to-event measure, not a statement that 1% fewer participants experienced an event.
The 95% CI of 0.85–1.16 describes uncertainty around the estimated hazard ratio. It includes 1.00, so the interval is compatible with a range of relative effects extending from a lower estimated hazard to a higher estimated hazard for the rosiglitazone strategy.
The non-inferiority question is different from a conventional superiority question. The registry specifies a margin of 1.2 for the upper limit of the 95% confidence interval. Because the upper confidence limit is 1.16, it remains below the stated margin of 1.2. That is the relevant comparison for the posted non-inferiority criterion.
No p-value is reported in the posted analysis. A p-value would address a particular hypothesis-testing question; it would not replace the confidence-interval comparison against the prespecified non-inferiority margin.
7. Independent Re-adjudication Results
The registry reports nine additional primary statistical analyses based on independent re-adjudication. These analyses provide a useful statistical distinction between the original RECORD endpoint definitions and contemporary cardiovascular endpoint definitions.
All-cause death
Death due to any cause
95% CI: 0.68–1.08
Hypothesis type: Superiority
An HR of 0.86 corresponds to an estimated instantaneous rate of death approximately 14% lower with combined rosiglitazone than with combined metformin/sulfonylurea under the fitted time-to-event comparison. It does not mean that 14% fewer participants died, nor does it describe an individual participant's probability of survival.
The 95% CI of 0.68–1.08 includes 1.00. Thus, the reported interval does not exclude equal hazards under the superiority comparison. The interval also shows substantial uncertainty around the point estimate.
No p-value is reported. The p-value would not measure the magnitude of the estimated effect; the HR and confidence interval provide the effect estimate and its precision.
First MACE — original RECORD definitions
Major adverse cardiovascular event
95% CI: 0.78–1.17
MACE: CV or unknown death, non-fatal MI, and non-fatal stroke
An HR of 0.95 indicates an estimated instantaneous MACE rate approximately 5% lower with combined rosiglitazone. The estimate is relative to the combined metformin/sulfonylurea strategy and does not translate directly into an absolute difference in the number of participants experiencing MACE.
The 95% CI of 0.78–1.17 crosses 1.00. Therefore, the reported interval includes both a potentially lower and a potentially higher hazard for the rosiglitazone strategy under the superiority framework.
No p-value is reported. The confidence interval is particularly important here because it communicates the uncertainty surrounding a relatively modest point estimate.
First MACE — contemporary definitions
Major adverse cardiovascular event
95% CI: 0.79–1.18
MACE: contemporary CV or unknown death, non-fatal MI, and non-fatal stroke definitions
An HR of 0.97 is close to the null value of 1.00 and corresponds to an estimated instantaneous MACE rate approximately 3% lower with combined rosiglitazone.
The 95% CI of 0.79–1.18 includes 1.00. The interval therefore does not establish a superiority effect and remains compatible with a range of lower and higher hazards.
No p-value is reported. The change from the original to contemporary endpoint definitions also illustrates why endpoint definitions matter: the statistical comparison can change when the events included in an endpoint are defined differently.
CV or unknown death — original definitions
Cardiovascular or unknown death
95% CI: 0.68–1.21
Original RECORD endpoint definitions
An HR of 0.90 represents an estimated instantaneous hazard approximately 10% lower with combined rosiglitazone than with combined metformin/sulfonylurea. The estimate concerns the rate of the defined event over time rather than an absolute 10-percentage-point reduction.
The 95% CI of 0.68–1.21 includes 1.00 and extends above the registry's 1.2 non-inferiority margin, although that margin is specifically stated for the cardiovascular death/cardiovascular hospitalization analysis rather than for this superiority endpoint.
No p-value is reported. A superiority analysis should not be interpreted from the point estimate alone; the confidence interval must also be considered.
CV or unknown death — contemporary definitions
Cardiovascular or unknown death
95% CI: 0.68–1.21
Contemporary endpoint definitions
The point estimate again corresponds to an estimated instantaneous hazard approximately 10% lower with combined rosiglitazone. The identical reported HR and confidence interval do not mean that the underlying event definitions are interchangeable; the contemporary and original definitions are distinct registry endpoints.
The interval of 0.68–1.21 includes 1.00 and is therefore compatible with no difference as well as materially lower or higher hazards. This degree of uncertainty is important when interpreting an HR of 0.90.
No p-value is reported. The hazard ratio provides the relative estimate, while the confidence interval communicates its statistical precision.
Myocardial infarction — original definitions
Fatal and non-fatal myocardial infarction
95% CI: 0.80–1.59
Original RECORD endpoint definitions
An HR of 1.13 corresponds to an estimated instantaneous myocardial-infarction hazard approximately 13% higher with combined rosiglitazone. It does not mean that the probability of myocardial infarction was 13% higher for every participant or that there was a 13-percentage-point absolute increase.
The 95% CI of 0.80–1.59 is wide and includes 1.00. Consequently, the point estimate should not be interpreted independently of the substantial uncertainty represented by the interval.
No p-value is reported. The width of the confidence interval is itself informative about the precision of the estimated treatment comparison.
Myocardial infarction — contemporary definitions
Fatal and non-fatal myocardial infarction
95% CI: 0.82–1.62
Contemporary endpoint definitions
An HR of 1.15 represents an estimated instantaneous myocardial-infarction hazard approximately 15% higher with combined rosiglitazone. This is a relative hazard estimate and is not an absolute risk increase.
The 95% CI of 0.82–1.62 includes 1.00 and is relatively wide. The observed estimate therefore has substantial uncertainty, and the interval includes both lower and higher hazards.
No p-value is reported. The contemporary definition also illustrates that endpoint adjudication criteria are part of the statistical estimand: changing the event definition can alter which events enter the analysis.
Stroke — original definitions
Fatal and non-fatal stroke
95% CI: 0.54–1.14
Original RECORD endpoint definitions
An HR of 0.79 corresponds to an estimated instantaneous stroke hazard approximately 21% lower with combined rosiglitazone. The estimate should not be translated into a 21% absolute reduction in stroke risk.
The 95% CI of 0.54–1.14 includes 1.00 and spans a broad range of possible relative effects. The point estimate therefore provides only one part of the statistical picture.
No p-value is reported. For a relatively less frequent endpoint, wider confidence intervals can arise because fewer events provide less information about the underlying treatment contrast.
Stroke — contemporary definitions
Fatal and non-fatal stroke
95% CI: 0.57–1.18
Contemporary endpoint definitions
An HR of 0.82 corresponds to an estimated instantaneous stroke hazard approximately 18% lower with combined rosiglitazone. Again, this is not equivalent to an 18-percentage-point difference in stroke probability.
The 95% CI of 0.57–1.18 includes 1.00. The interval is compatible with lower, similar, or higher instantaneous stroke hazards under the superiority comparison.
No p-value is reported. The appropriate interpretation therefore emphasizes both the direction of the point estimate and the uncertainty shown by the confidence interval.
8. Results Summary Across the Ten Primary Analyses
| Primary endpoint | HR | 95% CI | Hypothesis |
|---|---|---|---|
| CV death / CV hospitalization | 0.99 | 0.85–1.16 | Non-inferiority or equivalence |
| All-cause death | 0.86 | 0.68–1.08 | Superiority |
| MACE — original definitions | 0.95 | 0.78–1.17 | Superiority |
| MACE — contemporary definitions | 0.97 | 0.79–1.18 | Superiority |
| CV or unknown death — original definitions | 0.90 | 0.68–1.21 | Superiority |
| CV or unknown death — contemporary definitions | 0.90 | 0.68–1.21 | Superiority |
| MI — original definitions | 1.13 | 0.80–1.59 | Superiority |
| MI — contemporary definitions | 1.15 | 0.82–1.62 | Superiority |
| Stroke — original definitions | 0.79 | 0.54–1.14 | Superiority |
| Stroke — contemporary definitions | 0.82 | 0.57–1.18 | Superiority |
9. Statistical Methods Explained
Why was a hazard ratio used?
The registered cardiovascular outcomes occur over a follow-up period extending from baseline through the end of study, up to 7.5 years. A time-to-event analysis uses not only whether an event occurred but also when it occurred and accommodates participants whose event time is censored. The registry reports hazard ratios as the effect measure for all 10 posted primary analyses.
What does an HR of 0.99 mean?
An HR of 0.99 means the estimated instantaneous event rate in the combined rosiglitazone group is approximately 99% of that in the combined metformin/sulfonylurea group under the fitted time-to-event comparison. It does not mean that 99% of participants remained event-free or that the two groups had identical individual outcomes.
Why is the non-inferiority margin different from 1.00?
For a superiority analysis, the null reference for a hazard ratio is typically 1.00. A non-inferiority design asks a different question: whether the new strategy can be ruled out as worse than the comparator by more than a prespecified clinically acceptable amount. Here the registry specifies 1.2 as the upper-limit criterion for the 95% confidence interval in the cardiovascular death/cardiovascular hospitalization analysis.
What does the 0.85–1.16 confidence interval tell us?
The interval describes uncertainty around the estimated HR of 0.99. It does not describe the range of individual treatment effects. For the non-inferiority analysis, the important feature is that the upper limit, 1.16, remains below the prespecified margin of 1.2.
Why were original and contemporary endpoint definitions both analyzed?
Independent re-adjudication provides an opportunity to examine cardiovascular outcomes using both the original RECORD definitions and contemporary definitions. This matters because the statistical endpoint is determined partly by which clinical events satisfy the operational definition. The resulting analyses can therefore be compared without treating different endpoint definitions as though they were identical.
Why can an HR above 1 have a confidence interval that includes 1?
The point estimate is the single value most compatible with the observed data under the analysis, whereas the confidence interval reflects statistical uncertainty. For myocardial infarction, the reported HRs are 1.13 and 1.15, but their 95% confidence intervals are 0.80–1.59 and 0.82–1.62. Because both intervals include 1.00, the data are compatible with both lower and higher hazards under the superiority comparison.
Why should the four-arm design not be confused with four separate hazard-ratio comparisons?
The trial had four arms, but the posted statistical analyses compare combined rosiglitazone strategies with combined metformin/sulfonylurea strategies. The reported hazard ratios therefore summarize the prespecified combined comparison represented in the registry analyses rather than four separate treatment contrasts.
10. Re-adjudication as a Statistical Design Feature
The re-adjudication component is important statistically because it creates parallel definitions of several clinically meaningful cardiovascular endpoints. Rather than relying on one endpoint construction, the registry reports analyses based on the original RECORD definitions and contemporary definitions.
| Outcome family | Original definition | Contemporary definition |
|---|---|---|
| MACE | CV or unknown death + non-fatal MI + non-fatal stroke | CV or unknown death + non-fatal MI + non-fatal stroke |
| CV / unknown death | Independent re-adjudication using original RECORD definitions | Independent re-adjudication using contemporary definitions |
| MI | Independent re-adjudication using original RECORD definitions | Independent re-adjudication using contemporary definitions |
| Stroke | Independent re-adjudication using original RECORD definitions | Independent re-adjudication using contemporary definitions |
For an analyst, the important lesson is that endpoint definitions are not merely descriptive labels. They determine which observations are classified as events and therefore influence the event-time data entering the statistical model.
11. Safety Results
The registry reports serious adverse events by treatment strategy and provides affected participants relative to participants at risk. These counts should be read as arm-specific safety summaries rather than as hazard-ratio efficacy analyses.
| Treatment group | Serious adverse events affected | At risk |
|---|---|---|
| RSG in addition to background MET | 424 | 1,117 |
| SU in addition to background MET | 428 | 1,105 |
| RSG in addition to background SU | 427 | 1,103 |
| MET in addition to background SU | 431 | 1,122 |
| Combined RSG: observational follow-up | 99 | 1,280 |
| Combined MET/SU: observational follow-up | 76 | 1,250 |
The serious-adverse-event summaries use counts of affected participants and participants at risk. They answer a different question from the primary hazard-ratio analyses. A safety proportion describes the frequency of affected participants in an arm, whereas a time-to-event hazard ratio describes a relative instantaneous event rate over follow-up.
The observational-follow-up rows should also be distinguished from the randomized treatment-strategy rows. They are labeled separately in the registry and should not automatically be combined with the randomized arm-level summaries.
12. Missing Data, Censoring, and Time-to-Event Interpretation
The primary statistical analyses are identified as time-to-event analyses, so censoring is an inherent component of the analysis framework. A participant who has not experienced the relevant endpoint by the last time at which the event status is known contributes follow-up information up to that point rather than being treated as having experienced the event.
The registry does not report a specific missing-data or imputation method for these hazard-ratio analyses. That distinction matters because conventional missing-data imputation is not interchangeable with time-to-event censoring. Censoring concerns the time at which event information ceases to be observed, whereas imputation generally concerns missing measurements or outcomes that are not directly observed.
The hazard ratio summarizes the relative event rate over the analyzed follow-up. It is not a simple ratio of cumulative event proportions.
13. Confidence Intervals and Statistical Precision
The width of the confidence interval varies across the primary outcomes. The cardiovascular death/cardiovascular hospitalization analysis has a 95% CI of 0.85–1.16, whereas the myocardial infarction analyses have intervals of 0.80–1.59 and 0.82–1.62. The wider intervals for myocardial infarction indicate greater uncertainty around those estimates.
| Endpoint family | HR | 95% CI | Interpretive feature |
|---|---|---|---|
| CV death / CV hospitalization | 0.99 | 0.85–1.16 | Upper limit below the 1.2 non-inferiority margin |
| All-cause death | 0.86 | 0.68–1.08 | Interval includes 1.00 |
| MACE, original | 0.95 | 0.78–1.17 | Interval includes 1.00 |
| MACE, contemporary | 0.97 | 0.79–1.18 | Interval includes 1.00 |
| MI, original | 1.13 | 0.80–1.59 | Wide interval; includes 1.00 |
| MI, contemporary | 1.15 | 0.82–1.62 | Wide interval; includes 1.00 |
| Stroke, original | 0.79 | 0.54–1.14 | Interval includes 1.00 |
| Stroke, contemporary | 0.82 | 0.57–1.18 | Interval includes 1.00 |
A confidence interval should therefore be read together with the point estimate and the hypothesis being tested. In this trial, the same numerical confidence interval can have a different interpretive role depending on whether the analysis is framed as non-inferiority or superiority.
14. Non-Inferiority Logic in the Primary Cardiovascular Analysis
Reference value
For a hazard ratio, 1.00 represents equal instantaneous event rates between the two comparison strategies.
Non-inferiority margin
The registry specifies an upper confidence-limit margin of 1.2 for the primary cardiovascular time-to-event analysis.
Observed estimate
The reported HR is 0.99, with a 95% CI of 0.85–1.16.
Margin comparison
The upper confidence limit of 1.16 is below 1.2, which satisfies the registry-stated confidence-interval criterion for non-inferiority.
This is a useful example of why non-inferiority should not be interpreted simply by asking whether a conventional superiority p-value is below 0.05. The decision criterion is tied to the prespecified margin and the confidence interval. A result can have a confidence interval that includes 1.00 and still satisfy a non-inferiority criterion when the entire relevant interval remains below the non-inferiority margin.
15. Multiplicity and Multiple Primary Analyses
The registry lists 10 primary endpoints and 10 posted primary statistical analyses. They cover the main cardiovascular outcome plus multiple independently re-adjudicated outcomes and alternative endpoint definitions.
| Multiplicity feature | What the registry reports |
|---|---|
| Registered primary endpoints | 10 |
| Primary analyses posted | 10 |
| Analyses with estimate + confidence interval | 10 |
| Effect measure | Hazard ratio |
| Hypothesis types | Non-inferiority or equivalence; superiority |
Multiple primary analyses raise an important statistical distinction: each reported confidence interval describes uncertainty for its own estimated comparison, but the ClinicalTrials.gov record does not describe a multiplicity-adjustment procedure across all 10 primary analyses. Accordingly, the individual confidence intervals should be interpreted according to the hypothesis assigned to each analysis rather than treating the collection as a single unqualified superiority testing framework.
16. Why the Hazard Ratio Is Not an Absolute Risk Difference
An HR of 0.86 for all-cause death means that the estimated instantaneous death rate was approximately 14% lower with combined rosiglitazone than with combined metformin/sulfonylurea under the analysis model.
It does not mean that 14% of participants avoided death, that the absolute mortality difference was 14 percentage points, or that every participant experienced the same proportional change.
Absolute event probabilities depend on the baseline hazard and the follow-up distribution. The hazard ratio therefore provides a relative time-to-event measure rather than a complete description of absolute clinical risk.
17. Comparing the Cardiovascular Outcome Families
The collection of analyses shows why cardiovascular trials often examine several related but distinct endpoints. Cardiovascular hospitalization and cardiovascular death capture a broader composite experience than myocardial infarction alone, while MACE restricts the event definition to a specified collection of major cardiovascular outcomes.
| Outcome family | HR | 95% CI | Statistical role |
|---|---|---|---|
| CV death / CV hospitalization | 0.99 | 0.85–1.16 | Primary non-inferiority/equivalence analysis |
| All-cause death | 0.86 | 0.68–1.08 | Primary superiority analysis |
| MACE | 0.95 / 0.97 | 0.78–1.17 / 0.79–1.18 | Primary superiority analyses under two definitions |
| CV or unknown death | 0.90 / 0.90 | 0.68–1.21 / 0.68–1.21 | Primary superiority analyses under two definitions |
| MI | 1.13 / 1.15 | 0.80–1.59 / 0.82–1.62 | Primary superiority analyses under two definitions |
| Stroke | 0.79 / 0.82 | 0.54–1.14 / 0.57–1.18 | Primary superiority analyses under two definitions |
The point estimates are not directly interchangeable because the event definitions differ. A composite endpoint can also be driven by its more frequent components, so the statistical meaning of a composite should always be considered alongside the individual components when those are separately reported.
18. Study Timeline
Trial start
The registry identifies April 2001 as the study start.
Randomized phase 3 study
The trial evaluated rosiglitazone, sulfonylurea, and metformin strategies in a randomized parallel design.
Primary completion
The registry identifies December 2008 as the primary completion date.
Primary endpoint time frame
The registered primary cardiovascular and re-adjudication endpoints use baseline through the end of study, up to 7.5 years.
19. Important Limitations and Interpretation Issues
- Open-label design: the registry identifies masking as none. This is a relevant design feature when interpreting treatment exposure, clinical management, reporting, and ascertainment processes.
- Combined treatment comparison: the posted hazard ratios compare combined rosiglitazone strategies with combined metformin/sulfonylurea strategies rather than four separate treatment arms.
- Endpoint definitions: several outcomes were independently re-adjudicated under both original RECORD and contemporary definitions. The estimates should therefore be interpreted in the context of their specific endpoint definition.
- Method reporting: the posted analyses identify the hazard ratio and ITT population but do not report the specific statistical method used to generate the hazard ratios.
- Non-inferiority versus superiority: the cardiovascular death/cardiovascular hospitalization analysis has a non-inferiority margin of 1.2, while the other posted analyses are labeled superiority analyses. These hypotheses should not be conflated.
- Confidence-interval interpretation: most superiority-analysis confidence intervals include 1.00. The point estimates alone should not be treated as proof of superiority.
- Multiplicity: the registry reports 10 primary endpoints and 10 primary analyses, but the ClinicalTrials.gov record does not describe a multiplicity-adjustment procedure across those analyses.
- Safety framework: serious adverse-event counts are presented separately from the time-to-event efficacy analyses and include distinct randomized-arm and observational-follow-up groupings.
- Generalizability: the registry information presented here establishes the trial population as participants with type 2 diabetes, but does not provide a detailed baseline-characteristic table in the statistical analysis record.
20. Why This Trial Matters Statistically
RECORD is a useful teaching case because it illustrates several important principles of clinical-trial statistics within a single randomized study: non-inferiority logic, time-to-event analysis, hazard ratios, independent endpoint re-adjudication, competing definitions of cardiovascular events, ITT analysis, and the distinction between randomized and observational safety summaries.
| Concept | How it appears in RECORD |
|---|---|
| Randomization | Randomized phase 3 parallel-group design with 4,447 participants |
| Intention-to-treat analysis | All randomized participants treated with at least one dose of study medication |
| Time-to-event endpoints | All 10 posted primary statistical analyses are analyzed as time-to-event outcomes |
| Hazard ratio | Reported effect measure for all 10 primary analyses |
| Confidence interval | 95% confidence intervals reported for every primary analysis |
| Non-inferiority | Primary cardiovascular endpoint uses a 1.2 upper confidence-limit margin |
| Superiority | Nine additional primary analyses are identified as superiority hypotheses |
| Endpoint adjudication | Multiple outcomes independently re-adjudicated using original and contemporary definitions |
| Safety analysis | Serious adverse events reported by randomized treatment arm and observational follow-up grouping |
The strongest statistical lesson is that an effect estimate cannot be interpreted independently of the question that generated it. An HR of 0.99 is evaluated differently when the question is non-inferiority with a margin of 1.2 than when the question is superiority with a null value of 1.00.
21. Related Tutorials
Learn more about the methods used in this trial:
22. Related Calculators
23. Sources
- ClinicalTrials.gov record: NCT00379769.
- PubMed record: PMID 25524432.
- PubMed record: PMID 21562320.
Continue through Clinical Biostats
Explore statistical tutorials, calculators, and additional clinical trial analyses organized around the methods used in randomized clinical research.
24. Record Summary
RECORD provides a detailed example of how a randomized clinical trial can use time-to-event methods to address a non-inferiority question while also examining multiple cardiovascular outcomes under superiority hypotheses. The principal cardiovascular death/cardiovascular hospitalization analysis reported an HR of 0.99 with a 95% CI of 0.85–1.16; the upper confidence limit was below the registry-specified non-inferiority margin of 1.2. The additional re-adjudication analyses produced HRs ranging from 0.79 for stroke under the original definitions to 1.15 for myocardial infarction under contemporary definitions, with confidence intervals that include 1.00 for the superiority analyses.
The statistical interpretation therefore depends on several layers: the randomized comparison, the ITT population, the time-to-event estimand, the distinction between non-inferiority and superiority, the confidence interval around each hazard ratio, and the precise definition of each cardiovascular endpoint. The re-adjudication analyses further demonstrate how clinical endpoint definitions can influence the statistical outcome being estimated.