This page separates reported trial results from statistical interpretation. The numerical results presented here are restricted to the statistical analyses reported in the ClinicalTrials.gov record.
1. Trial at a Glance
REWIND was a randomized, double-masked, parallel phase 3 trial comparing dulaglutide with placebo in the therapeutic area of endocrinology, with diabetes mellitus, type 2 and cardiovascular disease listed as conditions.
| Feature | REWIND |
|---|---|
| Trial name | Researching Cardiovascular Events With a Weekly Incretin in Diabetes (REWIND) |
| Phase | Phase 3 |
| Conditions | Cardiovascular Disease; Diabetes Mellitus, Type 2 |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | Double |
| Primary purpose | Treatment |
| Enrollment | 9901 |
| Interventions | Dulaglutide; Placebo |
| Primary endpoint type | Time-to-event |
| Primary hypothesis type | Superiority |
| Lead sponsor | Eli Lilly and Company |
| Sponsor type | Industry |
| ClinicalTrials.gov | NCT01394952 |
2. Clinical Question
The statistical question was whether assignment to dulaglutide, compared with placebo, was associated with a difference in the time to first occurrence of the registered composite cardiovascular outcome: cardiovascular death, non-fatal myocardial infarction, or non-fatal stroke.
Population
Participants enrolled in a phase 3 randomized trial addressing diabetes mellitus, type 2 and cardiovascular disease.
Intervention
Dulaglutide.
Comparator
Placebo.
Primary question
Does dulaglutide change the time to first occurrence of the registered composite cardiovascular outcome relative to placebo?
3. Trial Design
Dulaglutide
- Dulaglutide
- Randomized treatment assignment
- Time-to-event outcomes evaluated during follow-up
Placebo
- Placebo
- Randomized treatment assignment
- Time-to-event outcomes evaluated during follow-up
4. Endpoints
The registered primary endpoint was a time-to-event composite. The registry also posted seven secondary statistical analyses covering individual cardiovascular components, all-cause mortality, a composite microvascular endpoint, heart-failure events, and hospitalization for unstable angina.
| Role | Endpoint | Time frame | Analysis |
|---|---|---|---|
| Primary | Number of participants who experienced an event for time, from randomization to first occurrence of cardiovascular death, non-fatal myocardial infarction, or non-fatal stroke (a composite cardiovascular outcome). | From randomization to first occurrence or death from any cause or study completion (Median Follow-Up of 5.4 Years). | Cox proportional-hazards model; hazard ratio. |
| Secondary | Time to first occurrence after randomization of cardiovascular death, non-fatal myocardial infarction, or non-fatal stroke, individually. | From randomization to first occurrence or study completion (Median Follow-Up of 5.4 Years). | Cox proportional-hazards model; hazard ratio. |
| Secondary | Time to all-cause mortality. | From randomization to study completion (Median Follow-Up of 5.4 Years). | Cox proportional-hazards model; hazard ratio. |
| Secondary | Time to first occurrence after randomization of the composite microvascular endpoint. | From randomization to first occurrence or study completion (Median Follow-Up of 5.4 Years). | Cox proportional-hazards model; hazard ratio. |
| Secondary | Time to first occurrence after randomization of heart failure requiring hospitalization or an urgent heart failure clinic visit. | From randomization to first occurrence or study completion (Median Follow-Up of 5.4 Years). | Cox proportional-hazards model; hazard ratio. |
| Secondary | Time to first occurrence after randomization of first hospitalization for unstable angina. | From randomization to first occurrence or study completion (Median Follow-Up of 5.4 Years). | Cox proportional-hazards model; hazard ratio. |
5. Statistical Methodology
Cox proportional-hazards model
The registry identifies a Cox proportional-hazards regression model as the method used for the primary endpoint and all seven posted secondary analyses. The analysis compared placebo with dulaglutide in the all randomized participants population.
For this trial, a hazard ratio below 1 indicates a lower estimated instantaneous event rate under dulaglutide than under placebo within the fitted time-to-event model. The hazard ratio is not an absolute risk difference and does not directly state the probability that an individual participant will experience an event.
Analysis population
The statistical analyses were reported for all randomized participants. This is important because the randomized comparison is defined by treatment assignment rather than by whether participants subsequently remain on assigned treatment.
Superiority framework
The primary hypothesis type was superiority. The primary analysis note specifies that superiority was declared if the upper limit of the two-sided 95.33% confidence interval for the hazard ratio was below 1.0, after adjustment for the interim analysis.
Confidence intervals
The primary endpoint used a 95.33% two-sided confidence interval, whereas the posted secondary analyses used 95% two-sided confidence intervals. The different confidence levels are part of the reported statistical framework and should not be silently replaced with a common confidence level.
Multiplicity
The registry notes that, once superiority was achieved for the primary endpoint, multiplicity adjustments using a graphical approach were performed for secondary efficacy endpoints. The secondary analyses were intended to control the overall Type I error rate at a 2-sided alpha level of 0.0467.
Interim analysis
The primary analysis note explicitly states that the superiority criterion was evaluated after adjustment for an interim analysis. The ClinicalTrials.gov record does not provide additional numerical details about the interim boundary, information fraction, or alpha-spending function, so those features are not inferred here.
6. Statistical Methods Explained
Why was a Cox proportional-hazards model used?
The endpoints are times to clinical events rather than simple binary outcomes at a single fixed time. A Cox model is designed for this setting because it uses the ordering and timing of events while allowing participants without an observed event by the end of follow-up to contribute censored follow-up information.
What does the primary HR of 0.88 mean?
An HR of 0.88 means the fitted model estimated a hazard under dulaglutide that was 88% of the corresponding hazard under placebo. Expressed as a relative complement, this corresponds to an estimated 12% lower hazard for the composite event. It does not mean that 12% fewer participants necessarily experienced an event, nor does it mean that every participant had a 12% reduction in personal risk.
Why is the confidence interval important?
The primary 95.33% confidence interval was 0.79 to 0.99. It describes statistical uncertainty around the estimated hazard ratio under the analysis framework. Because the entire interval is below 1.0, the reported estimate is compatible with a hazard reduction relative to placebo under the prespecified superiority criterion. The interval does not describe the range of individual treatment effects across participants.
Why does the p-value not measure effect size?
The primary p-value was 0.026. A p-value addresses the evidence against the relevant null hypothesis under the specified testing framework; it does not quantify how large the treatment effect is. The hazard ratio and its confidence interval describe the estimated relative effect and its precision, while the p-value addresses statistical evidence.
Why does the 95.33% confidence level matter?
The primary endpoint was evaluated with a two-sided 95.33% confidence interval rather than the 95% interval used for the posted secondary analyses. That distinction reflects the trial's reported multiplicity and interim-analysis framework. Replacing the primary interval with a conventional 95% interval would change the stated statistical procedure.
Why does multiplicity matter for the secondary endpoints?
Multiple efficacy endpoints create multiple opportunities for a statistically unusual result to occur by chance. The registry states that a graphical multiplicity adjustment was used for secondary efficacy endpoints to control the overall Type I error rate at a two-sided alpha level of 0.0467. Consequently, secondary p-values should be interpreted within that prespecified multiplicity framework rather than as isolated tests.
7. Primary Result
Composite Cardiovascular Outcome
The primary endpoint measured the time from randomization to first occurrence of cardiovascular death, non-fatal myocardial infarction, or non-fatal stroke. The analysis population was all randomized participants, and the comparison was placebo versus dulaglutide.
Primary hazard ratio
95.33% two-sided CI: 0.79–0.99 · P = 0.026
Analysis: Cox proportional-hazards model · Hypothesis: Superiority
| Primary endpoint component | Reported analysis |
|---|---|
| Endpoint | Time from randomization to first occurrence of cardiovascular death, non-fatal myocardial infarction, or non-fatal stroke |
| Time frame | From randomization to first occurrence or death from any cause or study completion (Median Follow-Up of 5.4 Years) |
| Population | All randomized participants |
| Groups compared | Placebo vs Dulaglutide |
| Effect measure | Hazard ratio |
| Estimate | 0.88 |
| Confidence interval | 95.33% two-sided CI: 0.79–0.99 |
| P-value | 0.026 |
The estimated hazard ratio of 0.88 indicates that the fitted Cox model estimated approximately a 12% lower hazard for the composite cardiovascular outcome with dulaglutide relative to placebo. This is a relative model-based measure over the analyzed follow-up, not a statement that 12% fewer participants experienced an event.
The 0.79–0.99 95.33% confidence interval describes uncertainty around the estimated hazard ratio. Its upper limit is below 1.0, which is the criterion specified in the registry analysis note for declaring superiority after adjustment for the interim analysis.
The p-value of 0.026 describes statistical evidence under the specified hypothesis-testing framework; it does not measure the magnitude or clinical importance of the treatment effect. The magnitude is described by the hazard ratio, while the confidence interval provides information about its precision.
As with any Cox-model hazard ratio, interpretation depends on the model and its proportional-hazards framework. The ClinicalTrials.gov record does not report a diagnostic assessment of that assumption, so no independent conclusion about proportionality is made here.
8. Secondary Endpoint Results
The registry contains seven posted secondary statistical analyses. Each used a Cox proportional-hazards model in all randomized participants, comparing placebo with dulaglutide. The secondary analyses used two-sided 95% confidence intervals and were subject to the reported graphical multiplicity adjustment.
| Secondary endpoint | HR | 95% CI | P-value | Analysis note |
|---|---|---|---|---|
| Cardiovascular death | 0.91 | 0.78–1.06 | 0.211 | Death from CV causes |
| Non-fatal myocardial infarction | 0.96 | 0.79–1.16 | 0.652 | Nonfatal MI |
| Non-fatal stroke | 0.76 | 0.61–0.95 | 0.017 | Nonfatal stroke |
| All-cause mortality | 0.90 | 0.80–1.01 | 0.067 | Time to all cause mortality |
| Composite microvascular endpoint | 0.86 | 0.79–0.93 | <0.001 | Microvascular endpoint |
| Heart failure requiring hospitalization or an urgent heart failure clinic visit | 0.93 | 0.77–1.12 | 0.456 | Heart failure event |
| First hospitalization for unstable angina | 1.14 | 0.84–1.54 | 0.413 | Hospitalization for unstable angina |
Cardiovascular Death
Hazard ratio
95% two-sided CI: 0.78–1.06 · P = 0.211
The estimated hazard ratio was below 1, but the confidence interval included 1.0. The reported p-value was 0.211. Within the stated superiority framework, this result does not provide the same statistical evidence as the primary composite analysis.
Non-fatal Myocardial Infarction
Hazard ratio
95% two-sided CI: 0.79–1.16 · P = 0.652
The estimated hazard ratio was close to 1, with a confidence interval extending on both sides of 1.0. The p-value was 0.652. This component therefore provides a different statistical pattern from the overall composite endpoint.
Non-fatal Stroke
Hazard ratio
95% two-sided CI: 0.61–0.95 · P = 0.017
The estimated hazard ratio of 0.76 corresponds to an estimated 24% lower hazard for non-fatal stroke with dulaglutide relative to placebo. The confidence interval remained below 1.0, while the reported p-value was 0.017. Because this was a secondary endpoint, its interpretation belongs within the trial's stated graphical multiplicity procedure.
All-cause Mortality
Hazard ratio
95% two-sided CI: 0.80–1.01 · P = 0.067
The estimated hazard ratio indicates a lower estimated hazard under dulaglutide, but the 95% confidence interval extended slightly above 1.0. The reported p-value was 0.067. The hazard ratio should therefore be distinguished from the binary question of whether a conventional significance threshold has been crossed.
Composite Microvascular Endpoint
Hazard ratio
95% two-sided CI: 0.79–0.93 · P < 0.001
The estimated hazard ratio corresponds to an estimated 14% lower hazard for the composite microvascular endpoint with dulaglutide relative to placebo. The confidence interval remained below 1.0 and the reported p-value was <0.001. As a secondary endpoint, this result remains subject to the stated multiplicity framework.
Heart Failure Requiring Hospitalization or an Urgent Heart Failure Clinic Visit
Hazard ratio
95% two-sided CI: 0.77–1.12 · P = 0.456
The estimated hazard ratio was 0.93, with a confidence interval extending from 0.77 to 1.12. The p-value was 0.456. The interval therefore includes both a potentially lower and a potentially higher hazard relative to placebo under the fitted model.
First Hospitalization for Unstable Angina
Hazard ratio
95% two-sided CI: 0.84–1.54 · P = 0.413
The estimated hazard ratio of 1.14 is above 1, corresponding to a higher estimated hazard under dulaglutide relative to placebo in the fitted model. The confidence interval includes 1.0 and is relatively broad, extending from 0.84 to 1.54. The p-value was 0.413.
9. Reading the Secondary Results as a Statistical Set
The seven secondary analyses illustrate why a clinical trial should not be reduced to a list of isolated p-values. The effect estimates range from 0.76 for non-fatal stroke to 1.14 for first hospitalization for unstable angina, and the corresponding confidence intervals differ substantially in width.
| Statistical feature | What the REWIND results illustrate |
|---|---|
| Relative effect | Hazard ratios below 1 indicate lower estimated hazards; a hazard ratio above 1 indicates a higher estimated hazard. |
| Precision | Confidence-interval width varies across endpoints, reflecting different amounts of statistical information. |
| Null value | For a hazard ratio, 1.0 is the relevant no-difference value. |
| Multiplicity | Secondary efficacy endpoints were subject to the reported graphical adjustment controlling the overall Type I error rate at a two-sided alpha level of 0.0467. |
| Endpoint composition | A composite endpoint and its individual components can produce different estimates because they measure different event processes. |
10. Safety Results
The ClinicalTrials.gov record reports serious adverse events by treatment arm as affected participants divided by participants at risk.
| Safety measure | Placebo | Dulaglutide |
|---|---|---|
| Serious adverse events | 2044/4949 | 1991/4943 |
These figures report the number affected and the corresponding number at risk in each arm. They are not presented here as a formal comparative safety effect measure because the statistical analyses posted on ClinicalTrials.gov do not report a hazard ratio, confidence interval, or p-value for serious adverse events.
Placebo
2044/4949 participants experienced serious adverse events among the reported participants at risk.
Dulaglutide
1991/4943 participants experienced serious adverse events among the reported participants at risk.
11. Primary Endpoint: What the Hazard Ratio Does — and Does Not — Mean
The primary HR of 0.88 means the fitted Cox model estimated the hazard for the composite cardiovascular outcome under dulaglutide to be 88% of the hazard under placebo. This is equivalent to an estimated 12% lower hazard in relative terms.
It does not mean that 12% of participants avoided an event, that 12% fewer participants necessarily experienced the event, or that every individual participant had exactly a 12% reduction in risk.
The primary 95.33% confidence interval was 0.79–0.99. This quantifies uncertainty around the estimated hazard ratio within the reported statistical framework. It is not a prediction interval for individual participants and does not describe the distribution of treatment effects across people.
The primary p-value was 0.026. The p-value measures evidence against the relevant null hypothesis under the specified testing procedure. It is not a measure of effect size, clinical importance, or the probability that the null hypothesis is true.
The primary endpoint was followed from randomization until first occurrence, death from any cause, or study completion, with a median follow-up of 5.4 years. Time-to-event methods allow participants with incomplete event observation to contribute their available follow-up rather than requiring every participant to have an observed event.
12. Multiplicity and Interim Analysis
The primary analysis contains two important pieces of statistical design information. First, superiority was declared if the upper limit of the two-sided 95.33% confidence interval was below 1.0, after adjustment for the interim analysis. Second, once superiority was achieved for the primary endpoint, multiplicity adjustments using a graphical approach were performed for secondary efficacy endpoints.
Interim analysis
The primary analysis note explicitly accounts for an interim analysis. This means the primary confidence interval and superiority criterion should be interpreted within a design that anticipated examination of accumulating information.
Secondary multiplicity
The registry states that secondary efficacy endpoints were adjusted using a graphical approach to control the overall Type I error rate at a two-sided alpha level of 0.0467.
These features matter because an unadjusted interpretation of every individual p-value would ignore the trial's prespecified testing structure. The reported primary endpoint and secondary endpoints do not all occupy the same statistical role.
13. Analysis Population and Randomization
The primary and secondary analyses were reported using all randomized participants. This is consistent with the intention-to-treat principle: the treatment comparison is anchored to randomized assignment rather than being redefined according to treatment actually received after randomization.
Randomization establishes the treatment-assignment mechanism. The subsequent statistical model estimates how event times differed between those randomized groups. It does not require that every participant remain continuously exposed to the assigned intervention for the randomized comparison to retain its intended meaning.
The ClinicalTrials.gov record identifies the allocation as randomized and the design model as parallel. They do not provide additional randomization-stratification variables, so no stratification factors are introduced here.
14. Trial Timeline
Trial start
The REWIND trial began on July 22, 2011.
Randomized cardiovascular-outcome evaluation
The study used a randomized, double-masked, parallel design comparing dulaglutide with placebo.
Primary completion
The registry lists August 21, 2018 as the primary completion date.
Results posted
The ClinicalTrials.gov record is marked completed and reports six posted outcome measures and eight posted statistical analyses.
15. What the Primary Result Says Statistically
The primary analysis combines several pieces of information that should be read together rather than separately. The hazard ratio of 0.88 gives the relative effect estimate. The 0.79–0.99 confidence interval describes uncertainty around that estimate. The 0.026 p-value describes the evidence against the relevant null hypothesis under the specified testing framework.
| Quantity | REWIND primary result | Statistical role |
|---|---|---|
| Hazard ratio | 0.88 | Estimated relative treatment effect |
| 95.33% CI | 0.79–0.99 | Uncertainty around the hazard-ratio estimate |
| P-value | 0.026 | Evidence against the null under the specified testing procedure |
| Hypothesis | Superiority | Tests whether the treatment effect is consistent with a hazard ratio below 1 |
| Analysis | Cox proportional-hazards model | Model used to estimate the hazard ratio |
| Population | All randomized participants | Preserves the randomized treatment-assignment comparison |
16. Limitations
- Hazard-ratio interpretation: a single Cox hazard ratio is a model-based relative measure. Its interpretation depends on the proportional-hazards framework and should not be treated as an absolute risk difference.
- Composite endpoint: the primary outcome combines cardiovascular death, non-fatal myocardial infarction, and non-fatal stroke. The composite result does not establish that each component has the same treatment effect.
- Secondary endpoints: individual component and other secondary analyses occupy a different inferential position from the primary endpoint and are subject to the reported graphical multiplicity adjustment.
- Confidence-level difference: the primary analysis uses a 95.33% confidence interval, whereas the posted secondary analyses use 95% confidence intervals. These should not be treated as interchangeable.
- Interim analysis: the primary superiority criterion was adjusted for an interim analysis. The ClinicalTrials.gov record does not provide the detailed interim monitoring scheme.
- Safety comparison: the registry-reported serious-adverse-event data provide affected and at-risk counts but no formal comparative effect estimate or hypothesis test.
- Incomplete statistical detail: the ClinicalTrials.gov record identifies the Cox model and effect measures but do not provide additional model specifications, such as covariate details or diagnostics for the proportional-hazards assumption.
- Analysis population: the posted efficacy analyses are based on all randomized participants. That is appropriate for the randomized comparison but does not answer every question that could be asked about treatment exposure or adherence.
17. Why This Trial Matters Statistically
REWIND is a useful teaching case because the registry-reported statistical record brings together the core elements of modern time-to-event analysis: randomization, masking, a composite cardiovascular endpoint, Cox regression, hazard ratios, confidence intervals, superiority testing, an interim analysis, and multiplicity adjustment across secondary efficacy endpoints.
| Concept | How it appears in REWIND |
|---|---|
| Randomization | The trial used randomized allocation in a parallel-group design. |
| Blinding | The registry records double masking. |
| Time-to-event endpoint | The primary outcome measures time from randomization to first occurrence of a cardiovascular event in a composite endpoint. |
| Cox model | The primary and posted secondary analyses used Cox proportional-hazards regression. |
| Hazard ratio | HR was the reported effect measure for the time-to-event analyses. |
| Confidence interval | The primary analysis used a two-sided 95.33% CI; secondary analyses used two-sided 95% CIs. |
| Superiority | The primary hypothesis type was superiority. |
| Interim analysis | The primary superiority criterion was adjusted for an interim analysis. |
| Multiplicity | A graphical approach was used for secondary efficacy endpoints, with the stated overall two-sided alpha level of 0.0467. |
| Composite endpoint | Cardiovascular death, non-fatal myocardial infarction, and non-fatal stroke formed the primary composite outcome. |
| Component analysis | The three individual cardiovascular components were also analyzed as secondary time-to-event outcomes. |
18. Statistical Concepts in This Trial
Learn more about the methods used in this trial:
19. Related Statistical Calculators
20. Sources
- ClinicalTrials.gov: NCT01394952 — REWIND.
- Linked publication: PubMed PMID 41988866.
- Linked publication: PubMed PMID 39963952.
- Linked publication: PubMed PMID 36897834.
- Linked publication: PubMed PMID 36723445.
- Linked publication: PubMed PMID 35996147.
Continue through the Clinical Biostats statistical library
Use the related tutorials and calculators to explore the survival-analysis concepts illustrated by the REWIND statistical record.
21. Record Summary
REWIND provides a clear example of a randomized phase 3 time-to-event analysis. The primary endpoint was a composite of cardiovascular death, non-fatal myocardial infarction, and non-fatal stroke, analyzed in all randomized participants using a Cox proportional-hazards model. The reported hazard ratio was 0.88, with a two-sided 95.33% confidence interval of 0.79–0.99 and a p-value of 0.026. The registry identifies the hypothesis as superiority and specifies that the upper confidence-limit criterion was evaluated after adjustment for an interim analysis. Secondary efficacy endpoints were analyzed with the same Cox-model framework and were subject to a graphical multiplicity adjustment controlling the overall Type I error rate at a two-sided alpha level of 0.0467.
The statistical lesson is broader than the primary p-value. A time-to-event trial should be interpreted through the combination of randomized treatment assignment, endpoint definition, hazard ratio, confidence interval, p-value, analysis population, and multiplicity and interim-analysis rules. The secondary endpoint results further demonstrate why individual components of a composite outcome can have different effect estimates and different levels of statistical precision.