The ClinicalTrials.gov record identifies DIRECT-MT as a completed randomized trial with a parallel design and single masking. The registry lists a primary endpoint based on the modified Rankin Scale (mRS) at 3 months after the procedure. No statistical analyses are posted in the registry record.
1. Trial at a Glance
DIRECT-MT is a completed randomized clinical trial addressing treatment of acute ischemic stroke associated with large vessel occlusion. The registry records 656 participants, two arms, a parallel design, and a primary endpoint based on the modified Rankin Scale 3 months after the procedure.
| Feature | DIRECT-MT |
|---|---|
| ClinicalTrials.gov identifier | NCT03469206 |
| Acronym | DIRECT-MT |
| Status | Completed |
| Condition | Ischemic Stroke; Large Vessel Occlusion; Thrombectomy; Thrombosis; Alteplase |
| Design | Randomized, parallel-group, single-masked |
| Primary purpose | Treatment |
| Enrollment | 656 |
| Number of arms | 2 |
| Lead sponsor | Changhai Hospital |
| Sponsor type | OTHER |
| Start date | 2018-02-21 |
| Primary completion date | 2019-10-20 |
2. Clinical Question
The clinical question centers on the functional outcome of patients with acute ischemic stroke and large vessel occlusion when treated in a randomized comparison involving intra-arterial thrombectomy and intravenous thrombolysis.
Population
Patients with ischemic stroke and large vessel occlusion treated in Chinese tertiary hospitals, as reflected in the trial title and registry conditions.
Intervention
Intra-arterial thrombectomy, listed in the registry as a procedure.
Comparator
Intravenous thrombolysis, listed in the registry as a drug intervention.
Primary question
How does the randomized treatment comparison relate to the distribution of functional disability measured by the modified Rankin Scale 3 months after the procedure?
3. Trial Design
Intra-arterial thrombectomy
- Intervention type: procedure
- Randomized treatment arm
- Primary outcome assessed using the modified Rankin Scale
Intravenous thrombolysis
- Intervention type: drug
- Randomized comparison arm
- Primary outcome assessed using the modified Rankin Scale
The statistical importance of this design is that the principal treatment comparison is randomized rather than observational. If treatment assignment is properly implemented, differences in outcomes between the randomized groups can be interpreted as evidence about the treatment strategy under study, subject to the usual limitations of trial conduct, missing outcome measurements, and generalizability.
4. Endpoints
| Endpoint | Registry definition | Time frame |
|---|---|---|
| mRS (the modified Rankin Scale) | The mRS is an ordinal hierarchical scale ranging from 0 to 5, with higher scores indicating more severe disability. | 3 months after procedure |
Why the mRS is statistically different from a binary endpoint
The modified Rankin Scale is an ordinal outcome. Its categories have a natural ordering from less severe to more severe disability, but the numerical labels should not automatically be treated as if the distance between adjacent categories were equal.
That distinction is central to the statistical analysis. A mean mRS score, for example, would impose assumptions about the numerical spacing between categories that are not part of the registry definition. An analysis that preserves the ordered structure can instead ask whether treatment shifts the distribution toward less severe disability.
The ordering is meaningful: higher scores represent greater disability. The scale contains six ordered categories, but the registry describes it as an ordinal hierarchical scale rather than a continuous measurement.
5. Statistical Methodology
No statistical analyses are posted in the ClinicalTrials.gov record. The primary endpoint is nevertheless sufficiently defined to identify the statistical problem: comparison of an ordinal functional outcome measured once at a specified follow-up time.
Typical analysis of an ordinal endpoint
For an ordinal endpoint such as the mRS, a common approach is an ordinal logistic regression model, also called a proportional-odds model. This approach uses all ordered outcome categories rather than collapsing the scale into a single binary threshold.
Here, k indexes the ordered mRS cutpoints and X represents treatment assignment. The model estimates a common treatment effect across the cumulative cutpoints when the proportional-odds assumption is appropriate.
The corresponding treatment effect is often expressed as a common odds ratio. Its direction depends on how the cumulative probabilities and treatment indicator are coded, so the coding convention must always be stated before interpreting the number.
Why an ordinal model can be attractive
A binary analysis requires choosing a threshold, such as separating favorable from unfavorable outcomes. That can be clinically useful when a particular threshold is prespecified, but it discards information from the remaining categories. An ordinal model retains the ordering across the entire mRS scale.
For DIRECT-MT, this matters because the registered primary endpoint is explicitly the mRS itself, rather than a binary endpoint derived from the mRS.
Alternative analysis considerations
Other approaches can be used for ordinal outcomes, including proportional-odds models with appropriate covariate adjustment, nonparametric comparisons of distributions, or models that relax the proportional-odds assumption. The registry record does not identify a specific statistical model among these alternatives.
6. Planned Analysis
The registry specifies the primary endpoint as the modified Rankin Scale at 3 months after the procedure. Because the endpoint is ordinal, the primary comparison would typically preserve the ordering of the six mRS categories rather than treating the score as a continuous variable.
Primary outcome structure
The mRS ranges from 0 to 5, with higher values representing more severe disability. The analysis should therefore account for the ordered nature of the categories.
Typical model
An ordinal logistic or proportional-odds model is a standard way to compare two treatment groups across an ordered outcome distribution.
Effect estimate
A proportional-odds analysis produces an odds ratio describing the relative odds of being on one side of the cumulative outcome thresholds, subject to the model's coding and assumptions.
Results status
Results have not been posted on ClinicalTrials.gov in the record represented here.
What a complete analysis would need to establish
- The distribution of participants across the six mRS categories in each randomized group.
- The treatment-effect estimate and its confidence interval.
- The statistical test or model used to compare the randomized groups.
- Whether any covariate adjustment or stratification was prespecified.
- How missing 3-month mRS assessments were handled.
- Whether the proportional-odds assumption was evaluated if an ordinal logistic model was used.
These elements determine how strongly the resulting estimate can be interpreted. In particular, an ordinal treatment effect should be tied to the full outcome distribution and to the assumptions of the model used to obtain it.
7. Statistical Methods Explained
Why is mRS treated as an ordinal endpoint?
The mRS categories have an inherent order from 0 through 5, with increasing disability. However, the scale does not state that the numerical difference between 0 and 1 represents the same amount of change as the difference between 4 and 5. An ordinal analysis respects the ordering without requiring equal spacing between categories.
What does a proportional-odds model do?
It compares the cumulative distribution of the ordinal outcome across treatment groups. Rather than analyzing only one threshold, the model considers multiple cumulative splits of the ordered scale simultaneously and estimates a common treatment effect when the proportional-odds assumption is reasonable.
Why not simply compare mean mRS?
A mean treats the mRS labels as though they were measurements on an interval scale. That can obscure the structure of an ordinal endpoint. Two treatment groups could have the same mean while having meaningfully different distributions across the individual mRS categories.
What does an odds ratio mean for an ordinal endpoint?
The interpretation depends on model coding. In a proportional-odds model, the odds ratio describes the relative odds of being at or below cumulative outcome thresholds, or equivalently above them depending on the chosen formulation. The direction of a favorable treatment effect therefore cannot be interpreted from the number alone without knowing the coding convention.
Why is the proportional-odds assumption important?
The standard proportional-odds model assumes that the treatment effect is common across the cumulative thresholds of the outcome. If treatment has a very different effect at different parts of the mRS distribution, that assumption may be inappropriate and a more flexible ordinal model may be preferable.
Why does randomization matter for the analysis?
Randomization establishes the treatment comparison before the outcome is observed. Under appropriate conduct, this reduces confounding by baseline characteristics because treatment assignment is determined independently of the eventual outcome. The resulting comparison is therefore fundamentally different from an observational comparison in which treatment choice may be related to prognosis.
8. Analysis Population and Missing Data
The registry record identifies the trial as randomized and provides an enrollment of 656, but it does not post a detailed statistical analysis population or a missing-data strategy in the information represented here.
| Issue | Statistical importance |
|---|---|
| Randomized population | The randomized population is the natural foundation for an intention-to-treat comparison because treatment assignment, rather than subsequent treatment behavior, defines the primary groups. |
| 3-month outcome availability | The primary endpoint is measured 3 months after the procedure, so incomplete follow-up can create missing outcome data. |
| Missing-data mechanism | The implications depend on why mRS measurements are missing and whether missingness is related to treatment or prognosis. |
| Imputation | The registry record does not identify an imputation method for missing mRS outcomes. |
Missing ordinal outcomes deserve particular attention because simply excluding participants with missing mRS can change the composition of the analyzed groups. More sophisticated methods may use assumptions about the missing-data mechanism, but those assumptions should be prespecified and assessed rather than selected after seeing the outcome data.
9. Interpreting an Ordinal Treatment Effect
An ordinal treatment effect summarizes a shift in the distribution of mRS outcomes across the randomized groups. It is fundamentally different from reporting the proportion of patients who cross one particular mRS threshold.
An ordinal odds ratio does not mean that every patient experiences the same change in disability. It also does not identify an individual patient's probability of recovery without additional information about the baseline population, model, and covariates.
Two treatment groups can have similar rates at one selected threshold while differing at other parts of the ordinal distribution. Using the complete mRS ordering can therefore provide a broader description of functional outcome than a single dichotomized endpoint.
A confidence interval around an ordinal treatment effect describes statistical uncertainty around the estimated population-level effect under the specified model. It does not describe the range of outcomes that individual patients will experience.
10. Why This Trial Matters Statistically
DIRECT-MT provides a useful teaching case for the statistical analysis of randomized trials with ordinal functional outcomes. The central methodological issue is not simply whether one treatment group has a higher or lower average score, but how the entire ordered distribution of disability should be compared.
| Concept | How it appears in DIRECT-MT |
|---|---|
| Randomization | The registry identifies the allocation as randomized. |
| Parallel-group design | The study uses a parallel design with two arms. |
| Ordinal endpoint | The primary endpoint is the modified Rankin Scale, an ordinal hierarchical scale. |
| Six outcome categories | The registered mRS ranges from 0 to 5. |
| Fixed follow-up | The primary endpoint is assessed 3 months after the procedure. |
| Functional outcome | Higher mRS values indicate more severe disability. |
| Ordinal regression | Proportional-odds modeling is a standard framework for analyzing ordered categorical outcomes. |
| Model assumptions | An ordinal regression analysis requires attention to the assumptions underlying the selected model. |
| Missing data | Incomplete 3-month mRS measurements can affect the validity and precision of the comparison. |
| Effect interpretation | An ordinal treatment effect summarizes a distributional shift rather than a simple difference in means. |
11. Limitations
- No posted statistical analysis: the ClinicalTrials.gov record does not post a formal statistical analysis for the registered primary endpoint.
- Endpoint model not specified: the registry defines the mRS outcome but does not identify the statistical model used for its analysis in the record represented here.
- Proportional-odds uncertainty: if an ordinal logistic model is used, the proportional-odds assumption must be considered rather than taken for granted.
- Missing 3-month outcomes: the effect of incomplete follow-up depends on the amount and mechanism of missingness and on the analytic strategy used to address it.
- Outcome interpretation: a population-level ordinal treatment effect should not be translated directly into an individual patient's expected disability outcome.
- Generalizability: the trial title identifies Chinese tertiary hospitals, so applicability to other healthcare settings depends on whether the enrolled population and treatment context are comparable.
- Registry detail: the available record does not provide the full statistical analysis plan needed to establish every implementation detail of the primary analysis.
12. Statistical Interpretation vs Clinical Interpretation
Statistical interpretation
The registered primary endpoint is an ordered categorical measure of disability at a defined follow-up time. The appropriate analysis should preserve that ordinal structure and quantify uncertainty around the randomized treatment comparison.
Clinical interpretation
The mRS is intended to describe functional disability after stroke. Its clinical meaning comes from the distribution of patients across the ordered disability categories rather than from a numerical score treated as a conventional continuous measurement.
This distinction is important. Statistical significance, if established in a formal analysis, would answer whether the observed treatment comparison is inconsistent with a specified null hypothesis under the chosen model. Clinical interpretation asks a different question: how large and meaningful is the shift in functional outcomes, and how is that shift distributed across patients?
13. Sample Size and Precision
The registry reports 656 participants and two randomized arms. The total enrollment determines the amount of information available for estimating the treatment effect, but enrollment alone does not determine the precision of the primary endpoint estimate.
For an ordinal endpoint, precision also depends on how observations are distributed across the mRS categories, the treatment effect, missing outcome data, and the statistical model used.
Consequently, a sample size should not be interpreted as evidence of treatment effectiveness by itself. The inferential quantity of interest is the estimated difference between randomized groups together with its uncertainty.
14. Non-Inferiority, Superiority, and Hypothesis Testing
The registry information represented here does not identify a non-inferiority margin, a superiority margin, or a formal hypothesis-testing framework for the primary mRS endpoint.
Why a margin matters
In a non-inferiority trial, the conclusion depends on a prespecified clinically justified margin. The observed treatment effect is evaluated against that margin rather than simply against zero.
Why a p-value is insufficient
A p-value addresses evidence against a null hypothesis under a specified statistical model. It does not quantify the clinical importance of the treatment effect or replace an effect estimate with its confidence interval.
Because the registry record does not specify which hypothesis-testing framework was prespecified for DIRECT-MT, no particular superiority or non-inferiority conclusion should be assigned to the registered endpoint from the endpoint definition alone.
15. Multiplicity and Interim Analysis
The registry information represented here does not identify a multiplicity adjustment, hierarchical testing strategy, interim efficacy analysis, alpha-spending procedure, or stopping boundary for the primary mRS endpoint.
These design elements matter when a trial makes multiple formal comparisons or examines accumulating data before the planned final analysis. Without a prespecified error-control framework, multiple statistical looks or endpoints can change the interpretation of nominal p-values.
16. Crossover and Factorial Design
The registry identifies DIRECT-MT as a parallel-group trial. The record represented here does not identify a crossover design or a factorial design.
17. Bayesian Methods
The registry information represented here does not identify a Bayesian statistical analysis for the primary mRS endpoint.
A Bayesian ordinal analysis would require a prior distribution, a likelihood for the ordered outcome, and a prespecified method for summarizing the posterior treatment effect. Those elements are distinct from the endpoint definition itself.
For example, a Bayesian analysis could report a posterior probability that one treatment has a more favorable ordinal outcome distribution than the other. That probability would be a model-based posterior quantity rather than a frequentist p-value. Because the registry does not specify such a model, no Bayesian interpretation is assigned to the primary endpoint.
18. What a Complete mRS Analysis Should Report
A statistically informative presentation of the primary endpoint should go beyond a single p-value. Because the outcome is ordinal, the reader benefits from seeing both the distribution of mRS categories and the treatment-effect summary.
| Component | Why it matters |
|---|---|
| mRS distribution by arm | Shows where participants fall across the complete ordered outcome. |
| Treatment-effect estimate | Quantifies the magnitude and direction of the treatment comparison. |
| 95% confidence interval | Shows the statistical precision of the estimated treatment effect. |
| Statistical test | Provides the formal inferential framework for the primary comparison. |
| Analysis population | Clarifies which randomized participants contributed to the primary analysis. |
| Missing-data handling | Shows how incomplete 3-month outcomes were incorporated or excluded. |
| Model diagnostics | Relevant assumptions should be evaluated when a model such as proportional odds is used. |
This reporting structure is particularly important for an ordinal endpoint because a single summary statistic can hide clinically relevant differences in the shape of the outcome distribution.
19. Why the mRS Distribution Is More Informative Than a Mean Alone
Suppose two randomized groups had identical mean mRS values but different distributions. One group might have more patients concentrated at both ends of the scale, while another might have more patients concentrated around the middle categories. The same mean would conceal those differences.
An ordinal analysis instead uses the ranking of the categories. This allows the treatment comparison to reflect movement toward better or worse functional states without assuming that the numerical distances between mRS categories are equal.
The key statistical question is therefore not simply "Which group has the lower average mRS?" It is "Does the treatment shift the ordered distribution of functional disability toward more favorable categories, and how precisely is that shift estimated?"
20. Primary Endpoint: Statistical Reading Guide
Step 1 — Identify the scale
The mRS is ordinal and ranges from 0 to 5. Higher scores indicate greater disability.
Step 2 — Identify the time point
The registry specifies assessment at 3 months after the procedure.
Step 3 — Preserve ordering
An appropriate primary analysis should use the ordered structure rather than automatically treating the score as continuous.
Step 4 — Quantify uncertainty
The treatment comparison should be accompanied by a confidence interval or other appropriate measure of inferential uncertainty.
21. Trial Timeline
| Milestone | Date |
|---|---|
| Trial start | 2018-02-21 |
| Primary completion | 2019-10-20 |
The registry identifies the study as completed. The primary endpoint is defined at 3 months after the procedure, while the registry dates identify the overall study start and primary completion milestones.
22. Statistical Concepts in This Trial
DIRECT-MT illustrates several core principles in modern clinical-trial statistics:
| Concept | Application to DIRECT-MT |
|---|---|
| Randomization | Creates the basis for a causal comparison between the two treatment arms. |
| Parallel-group trial | Uses two randomized treatment arms without a crossover or factorial structure identified in the registry. |
| Ordinal outcome | The primary endpoint is the mRS, an ordered six-category scale. |
| Fixed-time functional assessment | The primary endpoint is assessed 3 months after the procedure. |
| Ordinal logistic regression | A standard framework for estimating a treatment effect across ordered outcome categories. |
| Proportional-odds assumption | A key model assumption when a common odds ratio is used across cumulative thresholds. |
| Confidence interval | Provides information about the precision of the treatment-effect estimate. |
| Missing data | Incomplete 3-month mRS assessment can affect the primary analysis. |
| Analysis population | Clarifies which randomized participants are included in the inferential comparison. |
23. Related Statistical Calculators
No calculator cards are included because the available topic list does not contain a specific calculator topic for this trial.
24. Limitations of Registry-Based Statistical Interpretation
A registry record can define the trial population, intervention, comparator, design, enrollment, and endpoint without providing every implementation detail of the final statistical analysis. For DIRECT-MT, the available record identifies the primary mRS endpoint and its timing but does not post the statistical analysis itself.
That distinction is important for statistical interpretation. An endpoint definition tells us what was measured; it does not by itself establish how the treatment effect was estimated, how missing observations were handled, which covariates were included, or how multiplicity and interim monitoring were controlled.
For an ordinal endpoint, these details can materially affect the interpretation of the treatment effect. The choice between proportional odds, a more flexible ordinal model, a dichotomized endpoint, or another approach should therefore be treated as an analysis-plan question rather than inferred solely from the mRS definition.
25. Sources
- ClinicalTrials.gov record: DIRECT-MT, NCT03469206.
Continue through Clinical Biostats
Explore statistical methods for clinical trials, survival analysis, regression, and analysis of categorical and ordinal outcomes.
26. Record Summary
DIRECT-MT is a completed randomized, parallel-group clinical trial with 656 participants and two treatment arms involving intra-arterial thrombectomy and intravenous thrombolysis. Its registered primary endpoint is the modified Rankin Scale, an ordinal hierarchical scale ranging from 0 to 5, measured 3 months after the procedure.
Statistically, the defining feature of the trial is therefore the analysis of an ordinal functional outcome following randomization. A proportional-odds model is a standard framework for such an endpoint because it can use the ordering across the complete mRS scale rather than reducing the outcome to a single threshold or treating the category labels as a continuous measurement. However, the ClinicalTrials.gov record does not post the formal statistical analysis for DIRECT-MT, so the specific model, analysis population, missing-data strategy, multiplicity framework, and other implementation details are not established by the registry record represented here.