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
DEFUSE 3 was a randomized, parallel-group clinical trial evaluating endovascular thrombectomy therapy versus medical management in patients represented by the registry conditions Stroke, Acute and Cerebral Infarction. The registry reports 182 enrolled participants and one posted formal statistical analysis for the primary endpoint.
| Feature | DEFUSE 3 |
|---|---|
| Trial name | DEFUSE 3 |
| Brief title | Endovascular Therapy Following Imaging Evaluation for Ischemic Stroke 3 |
| Therapeutic area | Neurology |
| Conditions | Stroke, Acute; Cerebral Infarction |
| Phase | NA |
| Status | TERMINATED |
| Allocation | RANDOMIZED |
| Design model | PARALLEL |
| Masking | NONE |
| Primary purpose | TREATMENT |
| Enrollment | 182 |
| Arms | 2 |
| Primary endpoint analyses posted | 1 |
| Formal method | Cochran-Mantel-Haenszel test |
| Effect measure | Odds ratio |
| Hypothesis type | Superiority |
2. Clinical Question
The statistical question is whether patients randomized to endovascular thrombectomy therapy had a different distribution of Modified Rankin Scale scores at Day 90 than patients randomized to medical management. The registry's posted formal analysis treats the comparison as a superiority analysis and reports an odds ratio estimated using a Cochran-Mantel-Haenszel method.
Population
The registry identifies the conditions as Stroke, Acute and Cerebral Infarction. The primary endpoint is assessed at Day 90 using the Modified Rankin Scale.
Intervention
Endovascular thrombectomy therapy. The registered interventions include endovascular thrombectomy and the listed thrombectomy devices.
Comparator
Medical management.
Primary question
Does the Day 90 Modified Rankin Scale outcome differ between endovascular thrombectomy therapy and medical management under the prespecified superiority framework?
3. Trial Design
Endovascular Thrombectomy Therapy
- Endovascular thrombectomy procedure
- Trevo Retriever
- Solitaire™ FR Revascularization Device
- Penumbra thrombectomy system
- Covidien MindFrame Capture Revascularization Device
Medical Management
- Medical management comparator arm
4. Trial Timeline and Registry Status
Trial start
The registry lists 2016-04 as the study start.
Primary completion
The registry lists 2017-08-23 as the primary completion date.
Terminated
The ClinicalTrials.gov record identifies the trial status as TERMINATED.
5. Primary Endpoint
| Endpoint | Time frame | Registry analysis information |
|---|---|---|
| The Distribution of Scores on the Modified Rankin Scale (mRS) at Day 90 | Day 90 | Formal analysis posted; analysis population: intention to treat; Cochran-Mantel-Haenszel method; odds ratio; superiority hypothesis |
The registry describes the Modified Rankin Scale as a commonly used scale for measuring the degree of disability or dependence in daily activities after stroke or other neurological disability. The registry definition states that the scale runs from 0-6, with 0 representing perfect health without symptoms and 6 representing death.
6. Results
The registry contains one formal statistical analysis for the primary endpoint. It compares endovascular thrombectomy therapy with medical management in the intention-to-treat population using the Cochran-Mantel-Haenszel method and reports an odds ratio.
Modified Rankin Scale at Day 90
Odds ratio for the primary endpoint
95% CI: 1.63–4.70 · P < 0.001
Comparison: Endovascular Thrombectomy Therapy vs Medical Management
| Primary endpoint | Analysis population | Method | Effect | 95% CI | P-value |
|---|---|---|---|---|---|
| The Distribution of Scores on the Modified Rankin Scale (mRS) at Day 90 | Intention to treat | Cochran-Mantel-Haenszel | Odds Ratio 2.77 | 1.63–4.70 | <0.001 |
An odds ratio of 2.77 means that the estimated odds associated with the analyzed primary endpoint were 2.77 times as large for endovascular thrombectomy therapy as for medical management, under the binary analysis recorded in the registry.
This does not mean that 2.77 times as many patients improved, nor does it mean that an individual patient's probability of a favorable outcome was multiplied by 2.77. An odds ratio and a risk ratio are different measures, particularly when the underlying event is not rare.
The 95% confidence interval of 1.63–4.70 describes statistical uncertainty around the reported odds-ratio estimate under the analysis framework. It does not describe the range of outcomes for individual patients.
The P-value <0.001 addresses the statistical evidence against the null hypothesis used for the superiority comparison. It does not measure the size, clinical importance, or certainty of benefit. The odds ratio and its confidence interval are needed to describe the estimated magnitude and precision.
Because the registry labels the posted endpoint analysis as binary while the registered endpoint is expressed as a distribution of mRS scores, the exact binary contrast used to generate the odds ratio should not be inferred beyond the information reported by ClinicalTrials.gov.
7. Understanding the Cochran-Mantel-Haenszel Analysis
The Cochran-Mantel-Haenszel (CMH) test is a family of methods for analyzing categorical outcomes while accounting for stratification or other categorical structure. In this trial's posted analysis, the registry identifies CMH as the statistical method and an odds ratio as the effect measure.
The reported 2.77 is therefore an odds-based relative effect, not a direct probability ratio. The registry does not provide enough information in the ClinicalTrials.gov record to reconstruct the underlying 2×2 table or calculate an absolute risk difference.
The CMH framework is especially useful when a categorical treatment comparison is being evaluated while preserving a stratified analysis structure. However, the registry-reported DEFUSE 3 analysis record does not identify additional stratification variables, so none are added here.
8. Intention-to-Treat Analysis
The posted primary analysis uses the intention-to-treat population. This means the statistical comparison is anchored to randomized treatment assignment rather than retrospectively redefining treatment groups according to what participants actually received.
Why ITT matters
Randomization establishes the treatment groups before outcome information is known. Preserving that assignment maintains the basic randomized comparison.
What ITT does not solve
ITT does not automatically eliminate every source of uncertainty. Missing outcomes, protocol deviations, treatment exposure, and other design issues can still affect interpretation if they are present.
9. Odds Ratio: A Careful Interpretation
The reported effect measure is an odds ratio, not a risk ratio, risk difference, hazard ratio, or mean difference.
Reported effect
95% CI: 1.63–4.70
Endovascular Thrombectomy Therapy vs Medical Management
An odds ratio above 1 indicates higher odds for the outcome as it was coded in the registry's binary analysis. Here, the point estimate of 2.77 corresponds to an odds ratio substantially above 1, while the entire reported 95% confidence interval, from 1.63 to 4.70, is also above 1.
The direction of a categorical effect measure must always be tied to the specific outcome coding. Because the ClinicalTrials.gov record does not state the binary cutoff or the exact favorable/unfavorable coding used to construct the endpoint, the safe interpretation is to report the numerical odds ratio and its direction without assigning an unreported clinical category to it.
10. Confidence Interval and P-value
What does the 95% confidence interval mean?
The reported 95% CI of 1.63–4.70 expresses uncertainty around the estimated odds ratio of 2.77. The interval is wide enough to show that the exact magnitude of the treatment-group difference is not known precisely, even though the point estimate is clearly above 1.
The confidence interval answers a precision question: how uncertain is the estimated effect? The P-value addresses a hypothesis-testing question: how incompatible are the observed data with the null hypothesis under the specified test? These are related but different quantities.
Why does P < 0.001 not mean a large effect?
A P-value is not an effect-size measure. It depends on the observed data and the statistical model or test, as well as the amount of information available. The effect size here is described by the odds ratio 2.77, while its uncertainty is described by the 1.63–4.70 confidence interval.
11. Safety Results
The ClinicalTrials.gov record reports serious adverse events by randomized arm as affected participants over participants at risk.
| Safety measure | Endovascular Thrombectomy Therapy | Medical Management |
|---|---|---|
| Serious adverse events | 40/92 | 48/90 |
These figures describe the number affected and the corresponding number at risk in each arm. They should be kept separate from the primary efficacy analysis because the primary endpoint and the serious-adverse-event measure answer different questions.
12. Statistical Methods Explained
Why was an intention-to-treat analysis used?
The registry identifies the primary analysis population as intention to treat. An ITT framework retains randomized treatment assignment, which is important because the purpose of randomization is to establish comparable treatment groups at the point of allocation.
What does an odds ratio of 2.77 mean?
It means that the estimated odds for the binary outcome used in the posted analysis were 2.77 times the corresponding odds in the medical-management group. It does not mean that the probability was 2.77 times higher, because odds and probability are different mathematical quantities.
Why use a Cochran-Mantel-Haenszel test?
The CMH method is designed for categorical comparisons and can incorporate stratified categorical structure. In DEFUSE 3, ClinicalTrials.gov explicitly identifies the CMH test as the posted statistical method. The ClinicalTrials.gov record does not specify additional strata for this analysis, so none are assumed.
What does the confidence interval of 1.63–4.70 tell us?
It quantifies uncertainty around the reported odds-ratio estimate. A narrower interval would represent greater numerical precision, while a wider interval would represent less precision. The interval should not be interpreted as a range containing individual patient outcomes.
Why is P < 0.001 not the same as the effect size?
The P-value describes evidence against a null hypothesis under the specified statistical test. The effect size is the odds ratio, 2.77. The confidence interval supplies additional information about uncertainty around that effect estimate.
Why should the mRS endpoint not automatically be analyzed as an ordinal model here?
The endpoint name refers to a distribution of scores on a 0-6 scale, but the posted statistical-analysis record explicitly labels the endpoint as binary and reports a Cochran-Mantel-Haenszel analysis with an odds ratio. A different ordinal analysis should not be substituted for the analysis actually posted in the registry.
13. What the Primary Result Does — and Does Not — Establish
What is directly reported
The intention-to-treat comparison has an odds ratio of 2.77, a two-sided 95% confidence interval of 1.63–4.70, and a P-value of <0.001 under a Cochran-Mantel-Haenszel superiority analysis.
What is not reported here
The ClinicalTrials.gov record does not provide the binary cutoff used for the mRS analysis, an absolute response probability, a risk difference, a number needed to treat, or the underlying 2×2 outcome table.
This distinction matters because an odds ratio can be statistically clear while still requiring additional information to understand the absolute clinical magnitude. Absolute event rates, the exact outcome coding, and the underlying contingency table would allow a more complete translation of the relative odds into probabilities.
14. Other Trial Design Features
| Design topic | What the ClinicalTrials.gov record supports |
|---|---|
| Randomization | Yes. Allocation is recorded as RANDOMIZED. |
| Parallel design | Yes. Design model is PARALLEL. |
| Masking | NONE. |
| Superiority | Yes. The posted primary analysis identifies the hypothesis type as Superiority. |
| Intention-to-treat | Yes. The posted primary analysis uses the intention-to-treat population. |
| Cochran-Mantel-Haenszel | Yes. This is the reported statistical method. |
| Odds ratio | Yes. This is the reported effect measure. |
| Non-inferiority margin | Not reported in the ClinicalTrials.gov record. |
| Crossover | Not reported in the ClinicalTrials.gov record. |
| Factorial design | Not reported in the ClinicalTrials.gov record; the registry identifies the design model as parallel. |
| Multiplicity procedure | Not reported in the ClinicalTrials.gov record. |
| Interim analysis | Not reported in the ClinicalTrials.gov record. |
| Missing-data or imputation method | Not reported in the ClinicalTrials.gov record. |
| Bayesian methods | Not reported in the ClinicalTrials.gov record. |
| Additional stratification factors | Not reported in the ClinicalTrials.gov record. |
15. Limitations
- Endpoint coding: the registered primary endpoint is described as a distribution of mRS scores, whereas the posted analysis identifies the endpoint as binary. The ClinicalTrials.gov record does not specify the binary cutoff or coding.
- Limited formal analysis: one statistical analysis is posted for the primary endpoint. The ClinicalTrials.gov record does not provide additional formal statistical analyses for other outcome measures.
- No absolute efficacy rates: the registry-reported statistical analysis provides an odds ratio and confidence interval but does not provide the underlying event proportions needed to calculate an absolute risk difference.
- No reconstructed contingency table: the counts underlying the reported odds ratio are not reported, so the result should not be reverse-engineered.
- No additional stratification information: although the method is Cochran-Mantel-Haenszel, the ClinicalTrials.gov record does not identify the strata used, if any.
- Safety comparison: serious adverse events are reported as 40/92 versus 48/90, but no formal statistical comparison is reported.
- Unreported design details: the ClinicalTrials.gov record does not report a non-inferiority margin, crossover, multiplicity strategy, interim-analysis plan, missing-data method, imputation method, or Bayesian analysis.
- Generalizability: the ClinicalTrials.gov record identifies the trial conditions and design but does not provide the detailed baseline characteristics needed to assess how closely the enrolled population corresponds to a particular external population.
16. Why This Trial Matters Statistically
DEFUSE 3 is a useful teaching case because it illustrates how a categorical clinical outcome can be connected to randomization, intention-to-treat analysis, the Cochran-Mantel-Haenszel framework, odds ratios, confidence intervals, and hypothesis testing.
| Statistical concept | How it appears in DEFUSE 3 |
|---|---|
| Randomization | The registry identifies allocation as RANDOMIZED. |
| Parallel-group design | The design model is PARALLEL with two arms. |
| Intention-to-treat analysis | The posted primary analysis uses the intention-to-treat population. |
| Categorical data | The reported method category is categorical data. |
| Cochran-Mantel-Haenszel test | The registry-reported statistical method for the primary endpoint. |
| Odds ratio | The reported effect measure is an odds ratio of 2.77. |
| Confidence interval | The 95% two-sided CI is 1.63–4.70. |
| P-value | The posted primary comparison has P < 0.001. |
| Superiority testing | The posted hypothesis type is Superiority. |
| Safety analysis | Serious adverse events are reported by arm as affected/at risk. |
17. Statistical Interpretation vs Clinical Interpretation
Statistical interpretation
The posted intention-to-treat CMH analysis reports an odds ratio of 2.77 with a two-sided 95% CI of 1.63–4.70 and P < 0.001 for the Day 90 mRS primary endpoint.
Clinical interpretation
The result concerns the Day 90 Modified Rankin Scale endpoint, but the registry analysis does not identify the binary cutoff used to generate the odds ratio. Therefore, the numerical statistical result can be described without assigning an unreported absolute clinical response rate.
This separation is important. Statistical evidence answers whether the observed treatment-group comparison is compatible with the specified null hypothesis. Clinical interpretation requires knowing exactly how the endpoint was coded and understanding the absolute outcome distribution, not just its relative odds.
18. Serious Adverse Events: Statistical Context
| Arm | Affected | At risk |
|---|---|---|
| Endovascular Thrombectomy Therapy | 40 | 92 |
| Medical Management | 48 | 90 |
The safety data provide a straightforward affected/at-risk representation for serious adverse events. They are useful for describing the observed safety burden in each arm, but they should not be substituted for the efficacy endpoint or treated as evidence for or against the primary hypothesis.
The ClinicalTrials.gov record does not report whether the serious-adverse-event comparison was prespecified as a formal hypothesis test, whether confidence intervals were calculated, or whether adjustment for multiple safety endpoints was used. Accordingly, this page reports the counts without adding an unreported inferential analysis.
19. Reading the DEFUSE 3 Result as a Complete Statistical Statement
The registry-posted primary analysis compares endovascular thrombectomy therapy with medical management for the Day 90 Modified Rankin Scale endpoint in the intention-to-treat population using a Cochran-Mantel-Haenszel method, reporting an odds ratio of 2.77 with a two-sided 95% confidence interval of 1.63–4.70 and P < 0.001.
That sentence contains the essential statistical components: who was analyzed, what was compared, which method was used, what effect measure was reported, how uncertain the estimate is, and what the hypothesis test produced.
20. Related Tutorials
Learn more about the methods used in this trial:
21. Related Calculators
22. Sources
- ClinicalTrials.gov: NCT02586415 — DEFUSE 3.
- PubMed: PMID 29364767.
- PubMed: PMID 29986935.
- PubMed: PMID 36472536.
- PubMed: PMID 34276547.
- PubMed: PMID 33596675.
Continue through the Clinical Biostats statistical pathway
Use the related tutorials and calculators to explore the categorical-data methods, effect measures, confidence intervals, hypothesis testing, and randomized-trial principles illustrated by DEFUSE 3.
23. Record Summary
DEFUSE 3 provides a focused example of randomized clinical-trial analysis in which the primary endpoint is the Distribution of Scores on the Modified Rankin Scale (mRS) at Day 90. The registry-posted formal analysis uses the intention-to-treat population, a Cochran-Mantel-Haenszel test, an odds ratio as the effect measure, and a superiority hypothesis. The reported estimate is 2.77, with a two-sided 95% confidence interval of 1.63–4.70 and P < 0.001.
The most important statistical caution is the distinction between the registered endpoint wording and the posted analysis classification: the endpoint describes a distribution of mRS scores, while the statistical-analysis record identifies the analyzed endpoint as binary. Because the ClinicalTrials.gov record does not provide the binary cutoff or underlying outcome table, the odds ratio should be interpreted as the registry-reported binary effect measure without reconstructing an unreported probability comparison.
The trial also reports serious adverse events of 40/92 in the endovascular thrombectomy therapy arm and 48/90 in the medical management arm. These safety counts provide an additional descriptive perspective but are not accompanied in the ClinicalTrials.gov record by a formal statistical comparison.