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
IRIS was a phase 3 randomized trial evaluating pioglitazone versus placebo in participants with stroke, myocardial infarction, or diabetes represented in the registered condition set. The primary endpoint was a composite time-to-event outcome of recurrent fatal or non-fatal stroke or fatal or non-fatal myocardial infarction, assessed for up to 5 years.
| Feature | IRIS |
|---|---|
| Phase | Phase 3 |
| Brief title | Insulin Resistance Intervention After Stroke Trial |
| Conditions | Stroke; Myocardial Infarction; Diabetes |
| Design | Randomized, triple-masked, parallel-group |
| Primary purpose | Treatment |
| Enrollment | 3,876 |
| Interventions | Pioglitazone; placebo |
| Primary endpoint | Recurrent fatal or non-fatal stroke, or fatal or non-fatal myocardial infarction |
| Primary endpoint time frame | Up to 5 years |
| Primary statistical method | Cox proportional-hazards model |
| Additional statistical method | Mixed-effects model |
| Primary hypothesis type | Superiority |
| Trial status | Completed |
| Study period | 2005-02 to 2015-11 primary completion |
| Lead sponsor | Yale University |
2. Clinical Question
The central question was whether treatment with pioglitazone, compared with placebo, changed the time to a recurrent major vascular event defined as fatal or non-fatal stroke or fatal or non-fatal myocardial infarction.
Population
The registered conditions were stroke, myocardial infarction, and diabetes. The registry reports an overall enrollment of 3,876 participants.
Intervention
Pioglitazone.
Comparator
Placebo.
Primary question
Does pioglitazone alter the time to recurrent fatal or non-fatal stroke or fatal or non-fatal myocardial infarction?
3. Trial Design
Pioglitazone
- Pioglitazone
- Randomized comparison against placebo
- Primary efficacy analysis based on the time-to-event endpoint
- Longitudinal cognitive outcome analyzed separately with a mixed-effects model
Placebo
- Placebo
- Randomized comparison against pioglitazone
- Primary efficacy analysis based on the time-to-event endpoint
- Longitudinal cognitive outcome analyzed separately with a mixed-effects model
The parallel-group structure means each participant contributes to one randomized treatment comparison rather than moving between treatment arms. Triple masking reduces the opportunity for knowledge of treatment assignment to influence treatment delivery, outcome assessment, or participant behavior.
4. Endpoints
| Endpoint | Time frame | Endpoint type | Statistical analysis |
|---|---|---|---|
| Recurrent Fatal or Non-fatal Stroke, or Fatal or Non-fatal Myocardial Infarction | Up to 5 years | Time-to-event | Cox proportional-hazards model |
| Fatal or Non-fatal Stroke Alone | 5 years | Time-to-event | Cox proportional-hazards model |
| Acute Coronary Syndrome | 5 years | Time-to-event | Cox proportional-hazards model |
| All Cause Mortality | 5 years | Time-to-event | Cox proportional-hazards model |
| Decline in Cognitive Status | Annual measures from baseline to exit (up to 5 years) | Continuous / longitudinal | Longitudinal repeated-measures mixed-effects model |
| Composite Outcome of Fatal or Non-fatal Stroke, Fatal or Non-fatal MI or Episode of Serious Congestive Heart Failure | 5 years | Time-to-event | Cox proportional-hazards model |
The primary endpoint is a composite of recurrent fatal or non-fatal stroke and fatal or non-fatal myocardial infarction. Composite endpoints require attention to both the overall treatment effect and the clinical components that make up the composite; an overall hazard ratio does not establish that every component is affected to the same degree.
5. Statistical Methodology
Cox proportional-hazards model
The registry reports a Cox regression for the primary time-to-event endpoint, comparing pioglitazone with placebo. The effect measure is a hazard ratio. The primary analysis also incorporated covariate adjustment and an interim-analysis framework, with the confidence interval adjusted for interim looks.
A hazard ratio below 1 indicates a lower estimated instantaneous event hazard in the pioglitazone group relative to the placebo group, under the Cox model.
Covariate adjustment
The registry identifies covariate adjustment as a concept in the primary and secondary Cox analyses. Adjustment can improve precision and account for prespecified prognostic variables, but the adjusted hazard ratio remains a model-based relative comparison rather than a direct measure of absolute risk reduction.
Mixed-effects model for cognitive status
Decline in cognitive status was evaluated using annual modified mini-mental examination measurements from baseline to exit, up to 5 years. The analysis population consisted of participants with a baseline and at least one follow-up modified mini-mental examination score.
A mixed-effects framework is useful when each participant contributes repeated measurements over time because observations from the same participant are correlated rather than independent.
Multiplicity adjustment
The registry states that the confidence intervals for the five secondary outcomes were adjusted for multiplicity. This matters because evaluating several secondary endpoints creates multiple opportunities to observe apparently notable results by chance.
Interim analysis and alpha spending
The primary analysis notes that its confidence interval was adjusted for interim looks. Interim monitoring permits efficacy information to be evaluated before the full follow-up period has accumulated, but the repeated opportunities to inspect accumulating data must be incorporated into the statistical error-control framework.
Superiority hypothesis
The registered analyses use a superiority framework. In a superiority analysis, the objective is to determine whether the treatment groups differ in the prespecified direction or magnitude according to the statistical testing framework, rather than to demonstrate that the treatments are sufficiently similar.
6. Primary Result
Recurrent Fatal or Non-fatal Stroke, or Fatal or Non-fatal Myocardial Infarction
Hazard ratio for the primary composite endpoint
95% CI: 0.62–0.93 · P = .0067
Pioglitazone compared with placebo; confidence interval adjusted for interim looks.
| Primary endpoint | Pioglitazone vs placebo | 95% CI | P-value | Analysis |
|---|---|---|---|---|
| Recurrent fatal or non-fatal stroke, or fatal or non-fatal myocardial infarction | HR 0.76 | 0.62–0.93 | .0067 | Cox proportional-hazards model |
The hazard ratio of 0.76 means that, under the fitted Cox model, the estimated instantaneous rate of experiencing the primary composite event in the pioglitazone group was approximately 24% lower than in the placebo group over the analyzed follow-up.
This does not mean that 24% of participants avoided an event, that an individual participant's probability of an event was reduced by exactly 24%, or that the absolute difference in event risk was 24 percentage points. A hazard ratio is a relative time-to-event measure.
The 95% confidence interval of 0.62–0.93 describes statistical uncertainty around the estimated hazard ratio under the model and analysis framework. Because the interval is entirely below 1, the range of model-compatible relative hazard estimates represented by the interval is below the null value.
The p-value of .0067 addresses the compatibility of the observed result with the null hypothesis under the specified testing framework. It is not a measure of effect size and does not indicate the clinical magnitude of benefit.
The confidence interval was adjusted for interim looks, which is important because the primary analysis incorporated interim monitoring. Interpretation also depends on the Cox model assumptions and on appropriate handling of censoring and follow-up.
7. Secondary Time-to-Event Results
The registry reports four additional time-to-event analyses using Cox proportional-hazards models. The confidence intervals for these five secondary outcomes were adjusted for multiplicity.
| Secondary endpoint | HR | 95% CI | P-value | Interpretive feature |
|---|---|---|---|---|
| Fatal or Non-fatal Stroke Alone | 0.82 | 0.61–1.1 | 0.19 | CI adjusted for multiplicity |
| Acute Coronary Syndrome | 0.75 | 0.52–1.07 | 0.11 | CI adjusted for multiplicity |
| All Cause Mortality | 0.93 | 0.73–1.17 | 0.52 | CI adjusted for multiplicity |
| Composite outcome of fatal or non-fatal stroke, fatal or non-fatal MI or episode of serious congestive heart failure | 0.82 | 0.65–1.05 | 0.11 | CI adjusted for multiplicity |
Fatal or Non-fatal Stroke Alone
The estimated hazard ratio was 0.82, with a 95% confidence interval of 0.61–1.1 and p = 0.19. The estimate is below 1, but the confidence interval extends above 1, so the interval includes the conventional null value for a hazard ratio.
Acute Coronary Syndrome
The estimated hazard ratio was 0.75, with a 95% confidence interval of 0.52–1.07 and p = 0.11. The point estimate is below 1, while the interval includes 1. The multiplicity adjustment is relevant when interpreting this secondary endpoint alongside the other secondary outcomes.
All Cause Mortality
The estimated hazard ratio was 0.93, with a 95% confidence interval of 0.73–1.17 and p = 0.52. The point estimate is close to 1 relative to the other reported secondary estimates, and the confidence interval includes both values below and above 1.
Composite Vascular and Heart-Failure Outcome
The estimated hazard ratio was 0.82, with a 95% confidence interval of 0.65–1.05 and p = 0.11. As with the other secondary endpoints, the confidence interval was adjusted for multiplicity.
These estimates should be read as a set rather than as four independent opportunities to declare success. The registry explicitly identifies multiplicity adjustment for the five secondary outcomes. A hazard ratio below 1 is descriptive of the direction of the estimated treatment effect, while the confidence interval communicates its precision and the p-value addresses the corresponding statistical test.
In particular, a p-value is not a measure of how large or clinically important an effect is. Similarly, a confidence interval crossing 1 does not prove that there is no treatment effect; it indicates that the reported data and statistical framework do not exclude the null hazard ratio at that confidence level.
8. Cognitive Status: Longitudinal Mixed-Effects Analysis
Decline in cognitive status was analyzed using annual modified mini-mental examination measurements from baseline to exit, up to 5 years. The analysis population was participants with a baseline and at least one follow-up modified mini-mental examination score.
Mean difference in cognitive decline
95% CI: −0.326 to 0.280 · P = 0.88
Mean Difference (Net), pioglitazone compared with placebo.
| Endpoint | Effect measure | Estimate | 95% CI | P-value | Model |
|---|---|---|---|---|---|
| Decline in Cognitive Status | Mean Difference (Net) | −0.023 | −0.326 to 0.280 | 0.88 | Longitudinal repeated-measures mixed-effects model |
The estimated net mean difference was −0.023 units on the scale. The negative sign indicates the estimated direction of the between-group difference under the registered model, but the estimate itself is very close to zero relative to the uncertainty represented by its confidence interval.
The 95% confidence interval extends from −0.326 to 0.280, spanning zero. Thus, the interval includes both modest differences in either direction. The p-value of 0.88 is a test statistic for the specified comparison; it is not evidence that the two groups are exactly identical.
The use of a mixed-effects model is important because repeated cognitive measurements from the same participant are correlated. A longitudinal model can account for this within-participant dependence rather than treating every annual observation as if it came from a different independent participant.
9. Safety
The registry reports serious adverse events by randomized treatment arm. The affected and at-risk counts were:
| Treatment group | Participants with serious adverse events | Participants at risk |
|---|---|---|
| Pioglitazone | 950 | 1,939 |
| Placebo | 987 | 1,937 |
These counts provide a descriptive view of serious adverse events among participants at risk in each arm. They should not be interpreted as a substitute for the time-to-event efficacy analyses: serious adverse events and efficacy events have different definitions and different statistical questions.
10. Statistical Methods Explained
Why was a Cox proportional-hazards model used for the primary endpoint?
The primary endpoint is explicitly a time-to-event outcome with follow-up of up to 5 years. A Cox model uses both the occurrence of events and the amount of follow-up contributed by participants who have not yet experienced an event. Participants can therefore contribute information until an event or censoring time rather than requiring every participant to have the same follow-up duration.
What does an HR of 0.76 mean?
An HR of 0.76 means that the estimated instantaneous event hazard under the fitted model was 0.76 times the corresponding hazard in the placebo group. Equivalently, 1 − 0.76 = 0.24, so the model estimates approximately a 24% lower hazard in the pioglitazone group. This is not the same as saying that the probability of an event was reduced by exactly 24%.
Why is the confidence interval important?
The point estimate is only one summary of the treatment effect. The 95% confidence interval of 0.62–0.93 shows the uncertainty around the primary hazard-ratio estimate under the specified model and inferential framework. A narrower interval generally indicates greater precision than a wider interval, although precision and clinical importance are separate questions.
Why does the p-value not measure effect size?
The p-value addresses how compatible the observed data are with a specified null hypothesis under the statistical testing framework. It is affected by the amount of information and variability as well as the effect estimate. A small p-value therefore does not tell the reader whether an effect is large, clinically meaningful, or important to an individual patient.
Why was multiplicity adjustment important for the secondary endpoints?
The registry states that the confidence intervals for the five secondary outcomes were adjusted for multiplicity. When several outcomes are tested, the probability of observing at least one apparently unusual result can increase if each is interpreted in isolation. Multiplicity procedures modify the inferential framework so that the collection of tests is interpreted with the planned error control in mind.
Why use a mixed-effects model for cognitive measurements?
Cognitive status was measured annually from baseline to exit, so the same participant could contribute multiple observations. Those observations are correlated within a person. A mixed-effects model accommodates that longitudinal structure and estimates the treatment-related mean difference while accounting for repeated measurements.
Why does the interim analysis affect the confidence interval?
The primary analysis notes that the confidence interval was adjusted for interim looks. If accumulating trial data are inspected repeatedly, ordinary final-analysis methods can underestimate the overall type I error. Adjustment for the interim monitoring schedule aligns the reported inference with the sequential design.
11. Interpreting Hazard Ratios Correctly
The primary HR of 0.76 is a relative measure of the modeled event hazard. It indicates a lower estimated hazard in the pioglitazone group than in the placebo group under the Cox model.
The HR does not tell us directly how many fewer participants experienced the composite endpoint. Absolute risks require treatment-group event probabilities at a defined time point or another appropriate absolute-effect measure.
A population-level hazard ratio does not mean that each participant has a 24% lower individual probability of experiencing the endpoint. Individual risk depends on baseline characteristics, follow-up, competing events, and other factors.
The Cox model provides a compact relative-effect summary, but its interpretation depends on the model being appropriate for the observed time-to-event process. In particular, the usual hazard-ratio interpretation relies on the proportional-hazards framework. A single HR can be less informative if the relative hazards vary substantially over time.
12. Confidence Intervals and Statistical Precision
The primary estimate of 0.76 has a 95% confidence interval from 0.62 to 0.93. The secondary time-to-event estimates have wider intervals, including 0.61–1.1 for stroke alone, 0.52–1.07 for acute coronary syndrome, 0.73–1.17 for all-cause mortality, and 0.65–1.05 for the composite vascular and heart-failure outcome.
| Estimate | 95% confidence interval | What the interval communicates |
|---|---|---|
| Primary HR 0.76 | 0.62–0.93 | Uncertainty around the primary relative hazard estimate; interval remains below 1 |
| Stroke HR 0.82 | 0.61–1.1 | Interval includes 1 |
| Acute coronary syndrome HR 0.75 | 0.52–1.07 | Interval includes 1 |
| All-cause mortality HR 0.93 | 0.73–1.17 | Interval includes 1 |
| Composite HR 0.82 | 0.65–1.05 | Interval includes 1 |
| Cognitive mean difference −0.023 | −0.326 to 0.280 | Interval includes zero |
For hazard ratios, 1 is the conventional null value because an HR of 1 represents equal estimated hazards between groups. For a mean difference, zero is the conventional null value because a mean difference of zero represents no estimated difference in means.
13. Multiplicity Across Secondary Outcomes
The registry explicitly identifies multiplicity adjustment for the five secondary outcomes. This is an important feature of the statistical interpretation because the secondary analyses include several clinically distinct outcomes.
Five secondary outcomes
The registry identifies five secondary outcomes with multiplicity-adjusted confidence intervals: stroke alone, acute coronary syndrome, all-cause mortality, cognitive status, and the composite vascular/heart-failure outcome.
Why adjustment changes interpretation
When several hypotheses are considered together, the inferential threshold or confidence-interval construction can be adjusted to account for the family of analyses rather than treating every result as an isolated test.
Multiplicity does not change the numerical point estimate itself. It changes how uncertainty and statistical evidence are interpreted across the collection of secondary outcomes.
14. Composite Endpoint Interpretation
The primary outcome combines recurrent fatal or non-fatal stroke with fatal or non-fatal myocardial infarction. Composite endpoints can increase the number of observed events and therefore improve statistical information, but their interpretation depends on the clinical importance and frequency of their individual components.
The reported HR of 0.76 summarizes time to the first qualifying component of the registered composite endpoint. It does not establish that the hazard ratio is identical for stroke and myocardial infarction separately.
The registry separately reports stroke alone and acute coronary syndrome as secondary outcomes. This provides additional endpoint-specific information beyond the primary composite, although those analyses have their own statistical uncertainty and multiplicity considerations.
15. Analysis Population and Missing Follow-up
The registry gives an explicit analysis-population definition for the cognitive endpoint: participants with a baseline and at least one follow-up modified mini-mental examination score. The longitudinal analysis therefore concerns participants who satisfy that measurement criterion.
For the time-to-event endpoints, the registry reports Cox regression analyses and follow-up of up to 5 years for the primary endpoint. Time-to-event methods are designed to use available follow-up until an event or censoring, but the validity of the resulting estimates depends on the censoring process and the appropriateness of the underlying model.
16. Interim Analysis and Sequential Inference
The primary analysis states that its confidence interval was adjusted for interim looks. This indicates that the final inferential result accounts for the fact that the accumulating data were examined during the course of the trial.
Why interim looks matter
If a trial repeatedly examines accumulating efficacy data using an unadjusted fixed-sample threshold, the chance of a false-positive conclusion can be greater than the nominal level.
Why adjusted intervals matter
An adjusted confidence interval incorporates the sequential monitoring framework so that the reported interval is consistent with the planned interim-analysis procedure.
The primary analysis therefore should not be interpreted as though the study were analyzed exactly once at the end with a conventional unadjusted confidence interval. The interim monitoring is part of the inferential design.
17. Blinding and Bias Control
The registry identifies the trial as triple-masked. Masking is a design feature rather than a statistical model, but it can affect the reliability of treatment comparisons by reducing opportunities for knowledge of treatment assignment to influence participant behavior, clinical management, assessment, or other aspects of trial conduct.
Randomization and masking address different sources of potential bias. Randomization creates the treatment comparison, while masking can reduce the influence of treatment knowledge during trial participation and assessment.
18. Statistical Interpretation of the Complete Results
| Outcome | Effect | 95% CI | P-value | Statistical reading |
|---|---|---|---|---|
| Primary composite | HR 0.76 | 0.62–0.93 | .0067 | Estimated hazard lower with pioglitazone; interval below 1 |
| Stroke alone | HR 0.82 | 0.61–1.1 | 0.19 | Point estimate below 1; interval includes 1 |
| Acute coronary syndrome | HR 0.75 | 0.52–1.07 | 0.11 | Point estimate below 1; interval includes 1 |
| All-cause mortality | HR 0.93 | 0.73–1.17 | 0.52 | Point estimate near the null; interval includes 1 |
| Cognitive status | Mean difference −0.023 | −0.326 to 0.280 | 0.88 | Interval includes zero |
| Composite vascular / heart-failure outcome | HR 0.82 | 0.65–1.05 | 0.11 | Point estimate below 1; interval includes 1 |
The primary and secondary analyses illustrate why treatment-effect interpretation should use more than a single p-value. The primary estimate provides the magnitude and direction of the relative time-to-event effect, the confidence interval provides precision, and the p-value provides the corresponding statistical evidence under the specified inferential framework. The secondary endpoints add information about individual clinical outcomes but require multiplicity-aware interpretation.
19. Important Limitations and Interpretation Issues
- Composite endpoint: the primary endpoint combines stroke and myocardial infarction, so the overall hazard ratio should not be interpreted as though it were an endpoint-specific estimate for every component.
- Cox model assumptions: the hazard ratio is model-based and its interpretation depends on the suitability of the proportional-hazards framework.
- Censoring: time-to-event analyses rely on appropriate handling of censoring and follow-up. The registry summary does not provide a complete description of every censoring mechanism.
- Multiplicity: the registry explicitly identifies multiplicity adjustment for the five secondary outcomes, so secondary results should not be treated as unrelated standalone tests.
- Interim analysis: the primary confidence interval was adjusted for interim looks, meaning sequential monitoring is part of the primary inferential framework.
- Cognitive analysis population: the cognitive endpoint was analyzed among participants with a baseline and at least one follow-up modified mini-mental examination score.
- Different endpoint structures: the vascular outcomes are time-to-event endpoints, while cognitive status is a longitudinal continuous outcome. Their effect measures and models therefore answer different statistical questions.
- Serious adverse events: the reported serious-adverse-event counts are descriptive affected/at-risk counts and should not be conflated with the primary efficacy event definition.
- Registry detail: the ClinicalTrials.gov record does not provide every possible component-level, baseline, or graphical analysis that could be used to explore the trial further.
20. Why This Trial Matters Statistically
IRIS is a useful statistical teaching case because it combines randomized treatment comparison, a clinically meaningful composite time-to-event endpoint, Cox regression, interim monitoring, multiplicity-adjusted secondary outcomes, and a longitudinal mixed-effects analysis within one trial.
| Concept | How it appears in IRIS |
|---|---|
| Randomization | Participants were randomized to pioglitazone or placebo in a parallel design. |
| Blinding | The registry identifies the trial as triple-masked. |
| Time-to-event analysis | The primary endpoint and several secondary outcomes were analyzed as time-to-event endpoints. |
| Cox proportional-hazards model | Used for the primary and several secondary time-to-event analyses. |
| Hazard ratio | Used to express the relative treatment effect for time-to-event outcomes. |
| Confidence interval | Quantifies uncertainty around hazard-ratio and mean-difference estimates. |
| Interim analysis / alpha spending | The primary confidence interval was adjusted for interim looks. |
| Multiplicity adjustment | Secondary-outcome confidence intervals were adjusted for five secondary outcomes. |
| Mixed-effects model | Used for annual cognitive measurements from baseline to exit. |
| Analysis populations | The cognitive analysis had an explicit requirement for baseline and at least one follow-up score. |
21. Related Tutorials
Learn more about the methods used in this trial:
22. Related Calculators
23. Sources
- ClinicalTrials.gov: NCT00091949.
- PubMed: PMID 39921499.
- PubMed: PMID 30734043.
- PubMed: PMID 29934374.
- PubMed: PMID 29084736.
- PubMed: PMID 28975241.
Continue through the Clinical Biostats knowledge graph
Explore the statistical methods that connect randomized clinical-trial design, survival analysis, longitudinal modeling, multiplicity, and interpretation of treatment effects.
24. Record Summary
IRIS provides a compact example of how several statistical methods operate together in a randomized phase 3 trial. The primary endpoint was a time-to-event composite analyzed with a Cox proportional-hazards model, producing an HR of 0.76 with a 95% CI of 0.62–0.93 and p = .0067. The primary confidence interval incorporated the trial's interim-look framework. Secondary time-to-event outcomes were also analyzed with Cox models and had multiplicity-adjusted confidence intervals, while decline in cognitive status was evaluated using a longitudinal repeated-measures mixed-effects model.
The most useful statistical reading combines the effect estimate, confidence interval, and p-value with the design features that determine how those quantities should be interpreted. In IRIS, that includes randomization, triple masking, time-to-event follow-up, covariate adjustment, interim analysis, multiplicity adjustment, and a separate longitudinal analysis for cognitive status.