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
SWIFT DIRECT was a randomized, parallel-design clinical trial evaluating bridging thrombolysis versus direct mechanical thrombectomy in acute ischemic stroke. The registry lists 410 participants, two arms, single masking, and a primary endpoint based on the modified Rankin Scale (mRS) 90 days after randomization.
| Feature | SWIFT DIRECT |
|---|---|
| Clinical question | Bridging thrombolysis versus direct mechanical thrombectomy in acute ischemic stroke |
| Condition | Ischemic Stroke |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | Single |
| Primary purpose | Treatment |
| Enrollment | 410 |
| Number of arms | 2 |
| Phase | NA |
| Status | Completed |
| Lead sponsor | Insel Gruppe AG, University Hospital Bern |
| ClinicalTrials.gov | NCT03192332 |
2. Clinical Question
The central clinical question is whether a strategy of bridging thrombolysis differs from direct mechanical thrombectomy in patients with acute ischemic stroke, using the modified Rankin Scale score at 90 days after randomization as the registered primary endpoint.
Population
Participants enrolled in the trial for the condition listed as ischemic stroke.
Intervention framework
Bridging thrombolysis versus direct mechanical thrombectomy, as specified by the trial title and registered interventions.
Registered interventions
Stent-retriever thrombectomy with a revascularization device of the Solitaire™ type and intravenous thrombolysis with recombinant tissue-type plasminogen activator (IV t-PA).
Primary question
How does the randomized treatment strategy affect the modified Rankin Scale score 90 days after randomization?
3. Trial Design
Bridging thrombolysis
- Intravenous thrombolysis with recombinant tissue-type plasminogen activator (IV t-PA)
- Mechanical thrombectomy using a stent-retriever revascularization device of the Solitaire™ type
- Compared with the direct thrombectomy strategy in the randomized trial
Direct mechanical thrombectomy
- Stent-retriever thrombectomy with a revascularization device of the Solitaire™ type
- Represents the direct mechanical thrombectomy strategy named in the trial title
- Compared with bridging thrombolysis in the randomized trial
The statistical advantage of randomization is that treatment assignment is determined independently of the outcome that will later be measured. In principle, this allows the difference in the distribution of the 90-day mRS outcome between randomized groups to be interpreted as a treatment-group comparison rather than simply an observational association.
4. Trial Timeline
Study begins
The registry lists November 29, 2017 as the study start date.
Primary completion
The registry lists May 14, 2021 as the primary completion date.
Completed
The ClinicalTrials.gov record identifies the study status as completed.
5. Primary Endpoint
| Endpoint | Registry definition | Time frame |
|---|---|---|
| Score in modified Rankin Scale (mRS) | Score in modified Rankin Scale (mRS) | 90 days after randomization |
The modified Rankin Scale is an ordinal clinical outcome used to characterize the degree of functional disability. Unlike a binary endpoint, an ordinal endpoint preserves information across several ordered outcome categories. That distinction is central to the statistical analysis of SWIFT DIRECT: the primary outcome is a score, not simply a yes/no event.
6. What the mRS Endpoint Represents
The modified Rankin Scale is ordered: its categories represent progressively greater levels of functional dependence. The statistical information is therefore contained not only in whether a participant crosses a particular threshold, but in the complete ordered distribution of scores.
Ordinal information
An mRS analysis can use the ordering of the categories rather than collapsing the outcome into a single binary definition.
Fixed assessment
The registered assessment occurs 90 days after randomization, creating a common follow-up landmark for the primary endpoint.
Patient-level outcome
The endpoint is recorded as an individual participant's mRS score rather than as a group-level mean alone.
Treatment comparison
The principal statistical comparison concerns the distribution of mRS scores between the randomized treatment strategies.
7. Statistical Methodology
Ordinal outcome analysis
Because the registered primary endpoint is an mRS score, an analysis that respects the ordered nature of the outcome is particularly relevant. A common approach for a randomized ordinal endpoint is a proportional-odds ordinal logistic regression model. This approach estimates a treatment effect across cumulative thresholds of the ordered outcome rather than reducing the endpoint to one arbitrarily selected cutoff.
Here, Y is the ordinal mRS outcome, j indexes the cumulative thresholds, and X represents treatment assignment. Under the proportional-odds assumption, the treatment coefficient corresponds to a common odds ratio across the modeled cumulative thresholds.
Why an ordinal model can be informative
Suppose two treatment groups have similar proportions at one particular mRS threshold but differ across several adjacent categories. A binary analysis at that one threshold can discard information. An ordinal analysis instead uses the ordering of the entire outcome distribution.
Alternative binary analysis
An mRS endpoint can also be analyzed by defining a clinically meaningful dichotomy, such as the proportion achieving an outcome at or below a prespecified score. That approach is straightforward to interpret but requires the threshold to be specified in advance and does not use all of the ordinal information in the scale.
Mean-score analysis
A simple comparison of mean mRS scores is another possible descriptive summary, but treating an ordinal scale as though the distances between adjacent categories were numerically equivalent requires an additional modeling assumption. For a primary analysis, the ordered structure of the endpoint should be considered explicitly.
Confidence intervals
An effect estimate should be accompanied by a confidence interval whenever a formal treatment effect is reported. The interval communicates the precision of the estimate; it is not a statement that individual patients' outcomes must fall within that numerical range.
8. Planned Analysis
The ClinicalTrials.gov record identifies Score in modified Rankin Scale (mRS) at 90 days after randomization as the primary endpoint. No formal statistical analyses are posted to ClinicalTrials.gov for this endpoint.
Primary analysis framework
Ordinal clinical outcome comparing the randomized treatment strategies
A proportional-odds ordinal logistic regression is a standard method for analyzing an ordered mRS outcome, provided its assumptions are appropriate for the study data.
The statistical question can be expressed as a comparison of the cumulative distributions of mRS scores between the two randomized strategies. An ordinal model evaluates whether assignment to one strategy shifts the distribution toward better or worse categories while retaining the ordering of the scale.
| Component | Planned statistical interpretation |
|---|---|
| Outcome | Modified Rankin Scale score |
| Assessment time | 90 days after randomization |
| Primary comparison | Randomized treatment strategies |
| Common ordinal model | Proportional-odds ordinal logistic regression |
| Effect measure | Odds ratio describing the relative odds across cumulative outcome thresholds, under the proportional-odds assumption |
| Uncertainty | Confidence interval around the treatment effect |
9. How an Ordinal Treatment Effect Is Interpreted
For a proportional-odds model, an odds ratio summarizes the relative odds of being at or below the relevant cumulative outcome threshold across the ordered mRS categories, conditional on the model's assumptions.
The direction of the odds ratio must be defined by the coding of treatment and outcome. Therefore, an odds ratio should never be interpreted in isolation from the model's coding convention and the direction in which the mRS categories are ordered.
An ordinal odds ratio is not a risk ratio, not a difference in mean mRS, and not the probability that an individual patient will have a particular outcome. It is a model-based summary of the ordered outcome distribution.
A confidence interval shows how precisely the randomized comparison estimates the treatment effect under the chosen statistical model. A wide interval indicates greater uncertainty; a narrow interval indicates greater precision. Neither describes the range of individual patient responses.
10. Statistical Methods Explained
Why is mRS analyzed as an ordinal endpoint?
The mRS contains ordered categories. Treating those categories as ordered allows an analysis to use more of the information than a single binary cutoff. The key statistical question becomes whether the treatment assignment is associated with a shift in the distribution of scores.
What does a proportional-odds model assume?
The model assumes that the treatment effect can be represented by a common odds ratio across the cumulative thresholds of the ordinal outcome. In practical terms, the treatment effect is constrained to have the same estimated magnitude when comparing progressively different cumulative definitions of a favorable outcome.
Why not simply compare mean mRS?
A mean is easy to calculate, but the mRS is ordinal rather than a conventional continuous measurement. The numerical spacing between adjacent categories should not automatically be interpreted as equal. An ordinal model avoids requiring that particular assumption.
Why can a binary mRS analysis be useful?
A binary endpoint can correspond to a clinically meaningful distinction and can be especially easy to communicate. Its limitation is that the selected threshold discards information about outcomes on the other side of the cutoff. The threshold should therefore be prespecified rather than selected after seeing the data.
Why is randomization important for the primary analysis?
Randomization establishes the treatment groups before the outcome is observed. When the analysis follows randomized assignment, differences in the primary endpoint can be attributed to the treatment comparison with substantially less concern about confounding than in a nonrandomized comparison.
What does the 90-day time frame mean statistically?
It defines a common assessment landmark. Each participant's primary endpoint is evaluated relative to the same nominal post-randomization time point, rather than analyzing the outcome as an event occurring at an arbitrary time during follow-up.
11. Missing mRS Assessments
Fixed-time clinical outcomes create an important missing-data issue. If an mRS score is unavailable at 90 days for some randomized participants, the analysis must address whether and how those missing outcomes are handled.
Complete-case analysis
A complete-case approach analyzes participants with observed 90-day mRS scores but can lose information and can be biased when missingness is related to prognosis or treatment.
Multiple imputation
Multiple imputation can incorporate observed baseline and follow-up information to represent uncertainty about missing outcomes when its assumptions are appropriate.
Missing-not-at-random sensitivity
If the probability of a missing mRS depends on the unobserved mRS itself, sensitivity analyses can examine how conclusions change under alternative assumptions.
Best practice
The missing-data strategy should be prespecified and reported alongside the primary model because it can affect the estimated treatment effect.
The registry information identifies the primary endpoint and its time frame but does not provide a posted statistical analysis specifying a missing-data or imputation procedure.
12. Analysis Population and the Role of Randomization
For a randomized trial, the primary efficacy comparison is generally anchored to the randomized population. This preserves the treatment contrast created by randomization and avoids redefining treatment groups according to events that occur after randomization.
| Principle | Statistical role |
|---|---|
| Randomized assignment | Defines the treatment comparison before outcomes are observed. |
| Primary endpoint | mRS score measured 90 days after randomization. |
| Ordinal analysis | Uses the ordered structure of the mRS categories. |
| Per-protocol or as-treated analyses | Can be informative as sensitivity analyses but may lose some of the protection against post-randomization selection. |
The distinction is especially important in a treatment-strategy trial. If participants do not receive treatment exactly as originally intended, analyzing them according to what they ultimately received can introduce selection that was absent from the randomized comparison.
13. Multiplicity
The registry lists one primary endpoint: the Score in modified Rankin Scale (mRS) at 90 days after randomization. No additional registered primary endpoints are identified in the record used for this page.
Multiplicity becomes particularly important when investigators examine many secondary endpoints, multiple time points, multiple thresholds, or numerous subgroup analyses. Each additional hypothesis test creates another opportunity for a nominally small p-value to arise by chance.
| Analysis feature | Statistical consideration |
|---|---|
| Primary endpoint | One registered primary endpoint at 90 days. |
| Ordinal thresholds | A proportional-odds model summarizes cumulative thresholds within one model rather than treating every threshold as an independent primary test. |
| Secondary outcomes | Should be distinguished from the primary confirmatory endpoint when interpreting statistical evidence. |
| Subgroups | Multiple subgroup comparisons can generate apparently different effects by chance and generally require interaction testing for claims of heterogeneity. |
14. Interim Analysis and Sequential Testing
The ClinicalTrials.gov information posted on ClinicalTrials.gov for SWIFT DIRECT does not identify a posted interim statistical analysis or a specific alpha-spending procedure.
If a randomized trial examines accumulating outcome data before the planned final analysis, the statistical design must account for repeated looks at the data. Without appropriate adjustment, repeated significance testing can increase the overall type I error rate.
Fixed analysis
A single prespecified final analysis has one primary opportunity to test the treatment hypothesis under the planned significance level.
Sequential analysis
Planned interim looks require a prespecified framework controlling the overall probability of a false-positive conclusion.
15. Non-Inferiority and Superiority Logic
The trial title describes a comparison between bridging thrombolysis and direct mechanical thrombectomy, but the available registry information does not specify a non-inferiority margin or a posted non-inferiority statistical analysis.
This distinction matters because a non-inferiority trial asks a different statistical question from a superiority trial. Non-inferiority requires a prespecified margin defining how much loss of efficacy would be considered acceptable. The confidence interval for the treatment contrast is then compared with that margin rather than simply asking whether a conventional p-value is below a significance threshold.
16. Safety and Adverse Events
The registry information identifies the interventions but does not provide posted serious adverse-event counts by treatment arm in the statistical information used here.
Safety is nevertheless an important component of a treatment-strategy comparison. A complete safety analysis would normally distinguish at least the number of participants experiencing serious adverse events, the number experiencing any adverse event, event severity, event type, and treatment exposure. These measures should be reported by randomized treatment group using clearly defined safety populations.
| Safety component | Registry information available here |
|---|---|
| Serious adverse events by arm | Not reported in the available registry statistical information |
| Adverse-event incidence by arm | Not reported in the available registry statistical information |
| Adverse-event severity | Not reported in the available registry statistical information |
| Specific adverse-event categories | Not reported in the available registry statistical information |
17. Statistical Interpretation of a 90-Day mRS Comparison
The primary endpoint differs fundamentally from the time-to-event endpoints commonly encountered in oncology or cardiovascular survival trials. There is no need to estimate a hazard ratio simply because the endpoint is measured after randomization.
The central question is whether the distribution of functional outcomes at 90 days differs between the randomized strategies. An ordinal model can summarize that difference while retaining the ordered nature of the mRS.
An odds ratio from a proportional-odds model represents a relative comparison of cumulative odds across the ordered mRS scale. Its meaning depends on the coding of the treatment group and the direction of the mRS categories.
A statistically detectable shift in the mRS distribution does not imply that every participant experiences the same magnitude of benefit. Ordinal treatment effects summarize a population-level distributional difference.
A p-value addresses the compatibility of the observed data with a specified null hypothesis under the statistical model. It does not quantify the size of the treatment effect, the probability that a particular patient benefits, or the clinical importance of the observed difference.
18. Statistical Assumptions That Matter
| Assumption or issue | Why it matters |
|---|---|
| Ordinal structure | The mRS categories are ordered, so the statistical model should respect that structure. |
| Proportional odds | An ordinal logistic model assumes a common treatment effect across cumulative thresholds. |
| Missing outcomes | Missing 90-day mRS assessments can bias the comparison if missingness is related to unobserved outcomes. |
| Randomization | Preserving randomized treatment assignment protects the principal causal comparison. |
| Post-randomization treatment | Departures from assigned strategy can affect per-protocol or as-treated analyses. |
| Multiple testing | Numerous secondary, subgroup, or threshold-specific analyses can increase false-positive findings. |
19. Limitations
- No posted statistical analyses: the ClinicalTrials.gov record does not contain formal posted statistical analyses or numerical outcome estimates for the registered primary endpoint.
- Limited endpoint detail: the registry identifies the mRS score and 90-day time frame but does not provide a posted statistical analysis plan describing the precise model specification.
- Missing-data strategy: the available record does not specify how missing 90-day mRS assessments were handled statistically.
- Masking detail: the study is classified as single masked, but the available record does not identify the specific masked party in the information used here.
- Safety detail: serious adverse-event counts by randomized arm are not reported in the available registry statistical information.
- Multiplicity: the available record does not specify a posted multiplicity-adjustment framework for additional analyses.
- Model assumptions: any proportional-odds analysis requires assessment of the proportional-odds assumption; the assumption should not be treated as automatically satisfied.
20. Why This Trial Matters Statistically
SWIFT DIRECT is a useful teaching case because its primary endpoint illustrates a different branch of clinical-trial statistics from the familiar Kaplan-Meier and Cox-model framework. The central outcome is a fixed-time ordinal functional scale, making the preservation of ordered outcome information a central methodological issue.
| Concept | How it appears in SWIFT DIRECT |
|---|---|
| Randomization | Participants were allocated randomly to two treatment strategies. |
| Parallel design | The registry identifies a parallel design with two arms. |
| Ordinal endpoint | The primary outcome is a score on the modified Rankin Scale. |
| Fixed-time outcome | The primary assessment occurs 90 days after randomization. |
| Proportional odds | A standard framework for modeling an ordered mRS outcome when its assumptions are appropriate. |
| Missing data | Incomplete 90-day assessments require an explicit statistical strategy. |
| Multiplicity | Secondary and subgroup analyses require separate interpretation from the registered primary endpoint. |
| Treatment-strategy comparison | The trial compares bridging thrombolysis with direct mechanical thrombectomy. |
21. Sources
- ClinicalTrials.gov: NCT03192332 — SWIFT DIRECT.
Continue through Clinical Biostats
Clinical trial methods connect naturally to broader statistical concepts, including randomized comparisons, ordinal outcomes, regression modeling, missing data, and interpretation of treatment effects.
22. Record Summary
SWIFT DIRECT is a completed randomized, parallel-design trial with 410 participants, two arms, single masking, and a primary endpoint defined as the Score in modified Rankin Scale (mRS) 90 days after randomization. The registered treatment comparison concerns bridging thrombolysis versus direct mechanical thrombectomy in acute ischemic stroke, with interventions including intravenous thrombolysis with recombinant tissue-type plasminogen activator and stent-retriever thrombectomy using a Solitaire™-type revascularization device.
From a statistical perspective, the defining feature of the trial is the ordinal primary outcome. The appropriate analysis should preserve the ordered information in the mRS whenever the modeling assumptions support doing so. Proportional-odds ordinal logistic regression provides one established framework, while binary threshold analyses provide a simpler but less information-rich alternative. Interpretation should account for the fixed 90-day assessment, missing outcomes, treatment assignment, model assumptions, and the distinction between the primary endpoint and any additional analyses.