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
DIAS-2 was a randomized, parallel, triple-masked phase 3 treatment study of desmoteplase in acute ischemic stroke. The study enrolled 193 participants and used three treatment arms: two desmoteplase arms and one placebo arm.
| Feature | DIAS-2 |
|---|---|
| Phase | Phase 3 |
| Condition | Stroke, Acute |
| Allocation | Randomized |
| Design | Parallel |
| Masking | Triple |
| Primary purpose | Treatment |
| Enrollment | 193 |
| Number of arms | 3 |
| Study status | Completed |
| Start date | 2005-04 |
| Primary completion date | 2007-03 |
| Lead sponsor | Forest Laboratories |
| Sponsor type | Industry |
| ClinicalTrials.gov | NCT00111852 |
2. Clinical Question
The central clinical question was whether treatment with desmoteplase could improve functional outcomes after acute ischemic stroke compared with placebo. The registry describes two desmoteplase treatment arms and one placebo arm, with the primary outcomes assessed using neurological impairment and functional-disability scales at day 90.
Population
Patients enrolled in a phase 3 treatment study for acute ischemic stroke.
Intervention
Desmoteplase was evaluated in two randomized study arms.
Comparator
Placebo was the comparator treatment in the third randomized arm.
Primary question
Does desmoteplase treatment improve neurological and functional outcomes at day 90 compared with placebo?
3. Trial Design
Desmoteplase
- Randomized treatment arm
- Triple-masked study design
- Primary outcomes assessed at day 90
Desmoteplase
- Randomized treatment arm
- Triple-masked study design
- Primary outcomes assessed at day 90
Placebo
- Randomized comparator arm
- Triple-masked study design
- Primary outcomes assessed at day 90
4. Randomization and Masking
The study used randomized allocation, a parallel design, and triple masking. Randomization is important because the treatment groups are formed by allocation rather than by investigator choice, providing the foundation for a causal comparison when the randomized groups are analysed according to their assigned treatment.
Participants were assigned to one of three study arms.
Participants remained associated with their randomized treatment arm rather than switching between treatment groups as part of the registered design.
The registry identifies the study as triple-masked, reducing the opportunity for knowledge of treatment assignment to influence trial conduct or assessment.
The study was designed to evaluate treatment in acute ischemic stroke.
5. Primary Endpoints
The registry specifies three primary endpoints, all focused on outcomes at day 90. Unlike a time-to-event endpoint such as survival, these outcomes define clinically meaningful states or changes on neurological and functional scales.
| Endpoint | Time frame | Registry definition |
|---|---|---|
| National Institutes of Health Stroke Scale (NIHSS) | Change from Baseline to day 90 | Improvement of greater than or equal to 8 points from baseline, or NIHSS score less than or equal to 1. The NIHSS score ranges from 0 (least severe) to 42 (more severe). |
| Modified Rankin Scale (MRS) | Day 90 | Improvement on the Modified Rankin Scale, defined as a score of 0-2. The MRS ranges in severity from 0 (no symptoms) to 6 (Dead). |
| Barthel Index (BI) score of 75-100. | Day 90 | The Barthel Index (BI) is a scale used to measure performance in basic Activities of Daily Living and ranges from 0 (most disability) to 100 (no disability). |
Why the endpoint definitions matter
All three endpoints convert neurological or functional status into a prespecified outcome. The NIHSS endpoint is based on either a large improvement from baseline or attainment of a low score. The MRS endpoint uses a threshold of 0-2, while the Barthel Index endpoint uses a threshold of 75-100.
This distinction matters statistically. The underlying scales are ordered or quantitative measurements, but the registered primary endpoints define clinically interpretable categories. A statistical analysis of the registered endpoint therefore needs to respect the way the endpoint is actually defined rather than silently substituting a different analysis of the raw scale.
6. Statistical Methodology
The ClinicalTrials.gov record identifies the three primary outcome definitions and their assessment time frames but does not post formal statistical analyses or numerical efficacy results. For endpoints of this type, the appropriate analysis depends on whether the investigator treats the registry-defined outcome as a binary responder endpoint or analyses the underlying scale directly.
Binary responder analysis
Each registry definition can naturally be represented as a binary outcome: a participant either meets the prespecified NIHSS improvement criterion or does not; either has an MRS score of 0-2 or does not; and either has a Barthel Index score of 75-100 or does not.
For a two-group comparison, the resulting proportions can be compared using a difference in proportions, risk ratio, odds ratio, or an appropriate regression model. With three randomized arms, the analysis must also account for the fact that more than one treatment comparison may be of interest.
Three-arm comparisons
Because DIAS-2 has three randomized arms, a statistical analysis can begin with an overall comparison across all three groups and then examine specific treatment-versus-control contrasts. If multiple pairwise comparisons are formally tested, multiplicity becomes relevant because conducting several hypothesis tests can increase the probability of a false-positive finding.
Analysis of the underlying scales
The registry definitions do not require that the original scale itself be reduced to a binary outcome for every possible analysis. For example, NIHSS and Barthel Index have numerical ranges, while MRS is an ordered disability scale. If raw scale data are available, analyses can preserve more information by modelling the continuous or ordinal outcome rather than only the threshold-defined responder status.
However, those approaches answer somewhat different questions. A responder analysis asks whether the patient crossed a prespecified clinically meaningful threshold. A scale-based analysis asks about the distribution or magnitude of scores. The registered endpoint definition should remain the anchor for interpretation of the primary analysis.
Effect estimates and confidence intervals
For a binary endpoint, an effect estimate such as a risk difference or risk ratio describes the contrast in outcome probability between groups. A confidence interval quantifies uncertainty around that estimated contrast. The interval should be considered alongside the estimated effect rather than treating a p-value as a complete description of the evidence.
A positive risk difference would indicate a higher observed responder probability in the desmoteplase group under that particular comparison. The size of the difference describes magnitude; the confidence interval describes statistical precision.
7. Planned Analysis
NIHSS endpoint
The registered NIHSS endpoint is assessed by change from baseline to day 90 and defines improvement as either an improvement of greater than or equal to 8 points from baseline or an NIHSS score less than or equal to 1. A typical primary analysis would compare the proportion meeting this responder definition across the randomized treatment groups, with the placebo group serving as the comparator. A three-arm analysis would need to distinguish an overall comparison from any prespecified pairwise contrasts.
Modified Rankin Scale endpoint
The registered MRS endpoint is defined as a day-90 score of 0-2. A typical analysis of this binary endpoint would compare the proportion of participants achieving MRS 0-2 between randomized groups. Because MRS is intrinsically ordinal, an alternative scale-level analysis can preserve the ordering of disability categories, but that is a different estimand from the registered 0-2 threshold.
Barthel Index endpoint
The registered Barthel Index endpoint is defined as a score of 75-100 at day 90. A typical analysis of this endpoint would compare the proportion of participants reaching that threshold between treatment groups. An analysis of the underlying 0-to-100 score could provide additional information about the distribution of activities-of-daily-living function, but it would not be identical to the registered threshold endpoint.
Three primary endpoints
Having three primary endpoints creates an important statistical-design issue. If all three are tested formally, the trial's multiplicity strategy determines how evidence should be interpreted across the endpoint family. The registry record summarized here does not report a multiplicity procedure, alpha allocation, or endpoint hierarchy, so no particular multiplicity strategy can be attributed to the study.
8. Statistical Methods Explained
Why does the NIHSS endpoint use two responder criteria?
The NIHSS endpoint is satisfied by either a large improvement from baseline or attainment of an NIHSS score of 1 or less. These two routes recognize that meaningful neurological recovery can occur through substantial improvement from a more severe starting point or through reaching a very low final score.
Why is the MRS endpoint dichotomized at 0-2?
The registry defines improvement on the MRS as a score from 0-2 at day 90. This converts an ordered disability scale into a binary clinical outcome. The advantage is straightforward interpretation: the analysis asks whether participants reached the prespecified functional-status threshold. The trade-off is that scores outside the threshold are grouped together.
Why is the Barthel Index threshold different from the scale range?
The Barthel Index ranges from 0 to 100, but the registered endpoint focuses specifically on scores from 75-100. The analysis therefore targets achievement of a relatively favorable functional category rather than the average change in the entire 0-to-100 scale.
Why does the three-arm design matter statistically?
With two desmoteplase arms and one placebo arm, there are multiple possible comparisons. An overall three-group test asks whether the outcome differs somewhere among the groups. A treatment-versus-placebo comparison asks a more specific question. If several comparisons are formally tested, multiplicity must be addressed to control the false-positive error rate.
Why might an ordinal analysis be preferable for MRS?
MRS contains ordered disability categories. Converting it to 0-2 versus above 2 is clinically simple, but it discards information about distinctions among the remaining categories. An ordinal regression approach can retain more of the ordering, although its estimand and assumptions differ from those of the registered binary endpoint.
Why is a confidence interval useful even without a p-value?
A confidence interval describes the precision of an estimated treatment contrast. It allows readers to see not only whether the estimate differs from a null value under a specified testing framework, but also the range of effect sizes compatible with the data and model. The width of the interval is especially important when sample sizes are limited.
9. Interpreting the Three Functional Endpoints Together
Neurological recovery
The NIHSS endpoint focuses on improvement in neurological impairment, using a change-from-baseline threshold or a low day-90 score.
Global disability
The MRS endpoint focuses on functional disability at day 90 and defines success as a score of 0-2.
Activities of daily living
The Barthel Index endpoint focuses on performance in basic Activities of Daily Living and defines success as a score of 75-100.
Complementary perspectives
The endpoints measure related but distinct dimensions of recovery. Agreement across them would provide a broader picture than any single scale alone.
The three endpoints should not automatically be treated as interchangeable measures of the same outcome. NIHSS primarily characterizes neurological impairment, MRS captures overall disability, and the Barthel Index addresses basic activities of daily living. Their thresholds therefore define different responder populations.
10. Understanding the NIHSS Endpoint
The NIHSS ranges from 0, representing least severe impairment, to 42, representing more severe impairment.
The NIHSS endpoint combines a change criterion and a final-score criterion. That structure is statistically important because two participants can qualify for the endpoint through different clinical paths.
For example, a participant with substantial baseline impairment could meet the endpoint through a large improvement even if the day-90 score remains above 1. Conversely, a participant who reaches an NIHSS score of 1 or less qualifies even if the absolute change from baseline is less than 8 points. The registry definition therefore should be treated as a single composite responder rule rather than as two independent primary endpoints.
The direction of the raw NIHSS scale also matters. Lower scores indicate less severe impairment. An analysis of the underlying continuous change score must therefore distinguish improvement from worsening correctly; a negative change from baseline would generally represent worsening if change is defined as follow-up minus baseline.
11. Understanding the Modified Rankin Scale Endpoint
The MRS ranges in severity from 0, representing no symptoms, to 6, representing Dead.
The MRS endpoint is a classic threshold-based functional outcome. Its statistical interpretation is therefore straightforward: the key quantity is the probability of achieving a score of 0-2 by day 90.
Because the MRS is ordinal, however, the binary threshold does not use all of the information contained in the scale. A participant with a score of 3 and a participant with a score of 6 are both classified as nonresponders under the registered definition, even though their clinical states are not equivalent.
This is not necessarily a flaw. Prespecified binary thresholds can be clinically useful and easy to communicate. The important statistical point is that the estimand is the probability of crossing the 0-2 threshold, not the average MRS score.
12. Understanding the Barthel Index Endpoint
The Barthel Index measures performance in basic Activities of Daily Living and ranges from 0, representing most disability, to 100, representing no disability.
The Barthel Index endpoint similarly transforms a numerical functional scale into a binary outcome. Participants with scores from 75 through 100 are classified as having achieved the registered endpoint, while lower scores do not meet the definition.
The threshold means that a small numerical change can have different statistical consequences depending on where a participant begins and ends relative to 75. Someone moving from 70 to 74 remains a nonresponder under the registered definition, whereas a participant moving from 74 to 75 crosses the threshold.
Consequently, threshold-based analyses are often most naturally interpreted as comparisons of responder probabilities rather than as comparisons of mean score changes.
13. Missing Data and Endpoint Assessment
The primary endpoints are assessed at day 90, making ascertainment of the day-90 outcome important to the analysis. If an outcome is missing, the treatment-group comparison can be affected if missingness differs between randomized groups or is related to prognosis.
Several statistical approaches can be considered in clinical-trial analysis, including complete-case analysis, explicit missing-as-nonresponse rules, multiple imputation, or model-based approaches. Each makes different assumptions about the unobserved outcomes.
14. Multiplicity Across Three Primary Endpoints
DIAS-2 has three registered primary endpoints: NIHSS, MRS, and Barthel Index. This creates a multiplicity issue whenever several formal hypotheses are tested.
| Endpoint | Primary status | Statistical question |
|---|---|---|
| NIHSS | Primary | Does the treatment increase the probability of meeting the registered neurological-improvement definition at day 90? |
| MRS | Primary | Does the treatment increase the probability of achieving MRS 0-2 at day 90? |
| Barthel Index | Primary | Does the treatment increase the probability of achieving a Barthel Index score of 75-100 at day 90? |
If all three hypotheses are treated as confirmatory, the analysis should specify how the familywise type I error rate is controlled. Possible approaches in clinical-trial methodology include Bonferroni-type procedures, hierarchical testing, or other prespecified multiplicity strategies.
The key distinction is between multiple measurements of recovery and multiple independent opportunities to declare statistical significance. The first is a clinical-design issue; the second is a statistical-error-control issue. The registry record does not report the procedure used to connect the three primary endpoints to a formal multiplicity framework.
15. Comparing Two Desmoteplase Arms With Placebo
The three-arm structure provides two distinct ways to frame the statistical question. One approach is to ask whether the three groups differ overall. Another is to focus on each desmoteplase arm relative to placebo.
Overall comparison
Tests whether the distribution of the endpoint differs somewhere among the three randomized groups.
Desmoteplase arm 1 vs placebo
Evaluates the first desmoteplase treatment arm against the randomized placebo comparator.
Desmoteplase arm 2 vs placebo
Evaluates the second desmoteplase treatment arm against the randomized placebo comparator.
Multiplicity
If both treatment-versus-placebo comparisons are confirmatory, the statistical plan needs to account for the multiple comparisons.
The registry does not report the sample size assigned to each arm or distinguish the two desmoteplase arms by dose. It is therefore not appropriate to infer the relative information content of the three treatment groups from the total enrollment alone.
16. Why Randomization Matters
Randomization is the central design feature supporting causal interpretation. In a randomized parallel trial, treatment assignment is determined independently of the participant's subsequent outcome. This helps balance both measured and unmeasured prognostic factors in expectation.
The important analysis consequence is that the primary comparison should remain anchored to randomized treatment assignment. If treatment discontinuation, protocol deviations, or other post-randomization events occur, analysing participants according to their randomized group can preserve the original treatment contrast more effectively than selectively excluding participants after randomization.
The registry record identifies the study as randomized but does not provide detailed analysis-population definitions. Therefore, the specific handling of protocol deviations or treatment exposure should not be assumed.
17. Why Triple Masking Matters
Triple masking reduces opportunities for knowledge of treatment assignment to influence the conduct or assessment of the trial. This is particularly relevant when outcomes involve clinical assessment or classification.
Each of the primary endpoints requires an assessment of neurological or functional status. Although masking cannot eliminate every possible source of measurement variation, it can reduce the risk that expectations about treatment assignment influence assessment.
From a statistical perspective, masking is therefore not itself an estimator or hypothesis test. It is a design feature that helps protect the validity of the outcome measurements entering the statistical analysis.
18. Statistical Interpretation of Responder Endpoints
For a binary endpoint, the most direct clinical quantity is the difference in responder probability between treatment groups. A risk ratio instead describes the responder probability relative to the comparator. An odds ratio compares the odds of response rather than the probabilities themselves.
A group-level effect estimate does not mean that every individual participant experiences the same magnitude of benefit. It summarizes the difference between outcome probabilities under the compared treatment assignments.
A confidence interval expresses statistical precision around the estimated effect under the selected analysis framework. A wide interval indicates greater uncertainty than a narrow interval. The interval does not describe the range of outcomes that individual patients will experience.
A p-value addresses the compatibility of the observed data with a specified null hypothesis under the statistical model. It does not measure the magnitude of the treatment effect, the clinical importance of the effect, or the probability that the treatment works.
19. What a Clinically Meaningful Analysis Would Report
For each primary endpoint, a statistically informative report would ideally present the number and proportion of responders in each randomized arm, the treatment contrast of interest, a confidence interval for that contrast, and the corresponding hypothesis test where appropriate.
| Reporting element | Why it matters |
|---|---|
| Number of responders | Shows how many participants met the endpoint definition. |
| Total analysed | Places the responder count in the context of the available analysis population. |
| Responder percentage | Describes the absolute probability of achieving the endpoint. |
| Treatment contrast | Quantifies the difference between randomized groups. |
| 95% confidence interval | Describes uncertainty around the treatment estimate. |
| P-value | Provides a hypothesis-test result under the specified null and analysis framework. |
| Multiplicity adjustment | Clarifies how several primary endpoints or pairwise comparisons affect type I error. |
Presenting all of these quantities together is preferable to relying on a single significance test. The responder percentage describes absolute outcome frequency, the effect estimate describes the treatment contrast, the confidence interval describes precision, and the p-value addresses the prespecified hypothesis test.
20. Limitations
- No posted numerical results: the ClinicalTrials.gov record does not post numerical efficacy estimates, confidence intervals, or p-values for the three primary endpoints.
- Three-arm complexity: two desmoteplase arms create multiple possible treatment comparisons with the placebo arm.
- Multiplicity: three primary endpoints create a need for a clearly defined error-control strategy if they are all tested formally.
- Thresholding: the registered NIHSS, MRS, and Barthel Index endpoints reduce continuous or ordinal information to clinically defined categories.
- Scale-specific interpretation: NIHSS, MRS, and Barthel Index measure different aspects of neurological or functional recovery and should not be treated as interchangeable.
- Missing outcomes: day-90 endpoints require appropriate handling of participants without an observed day-90 assessment.
- Unreported analysis details: the registry does not report the detailed statistical analysis plan, including a specific multiplicity method, missing-data method, or analysis-population definitions.
- Arm-specific information: the registry identifies two desmoteplase arms but does not report their treatment-specific sample sizes or distinguish the two arms by dose in the record summarized here.
21. Why This Trial Matters Statistically
DIAS-2 is a useful teaching case because its primary outcomes illustrate an important class of clinical-trial endpoints: binary functional responders derived from neurological and disability scales.
| Concept | How it appears in DIAS-2 |
|---|---|
| Randomization | Participants were randomized among three parallel treatment arms. |
| Three-arm design | Two desmoteplase arms were compared within a design that also included placebo. |
| Triple masking | The registry identifies the trial as triple-masked. |
| Binary endpoint construction | Each primary endpoint defines a clinically meaningful threshold or responder state. |
| Neurological outcome | NIHSS measures neurological severity and defines improvement using a change or final-score criterion. |
| Functional disability | MRS defines success as a day-90 score of 0-2. |
| Activities of daily living | Barthel Index defines success as a day-90 score of 75-100. |
| Multiplicity | Three primary endpoints require careful interpretation of the overall type I error rate. |
| Effect measures | Responder probabilities can be summarized using absolute and relative treatment contrasts. |
| Missing data | Day-90 outcomes require a prespecified approach to participants without an observed endpoint. |
22. Clinical Question vs Statistical Estimand
The clinical question asks whether desmoteplase improves recovery after acute ischemic stroke. The statistical estimand makes that question more precise by identifying exactly what outcome is being compared and when it is measured.
NIHSS estimand
Difference in the probability of meeting the registered neurological-improvement definition from baseline to day 90.
MRS estimand
Difference in the probability of achieving MRS 0-2 at day 90.
Barthel Index estimand
Difference in the probability of achieving a Barthel Index score of 75-100 at day 90.
Comparator
The treatment contrast can be framed against the randomized placebo arm, with the exact pairwise comparison depending on the prespecified analysis.
This distinction is useful because changing the outcome definition changes the statistical question. For example, comparing mean Barthel Index scores is not equivalent to comparing the probability of achieving a score of 75-100. Both may be clinically informative, but they estimate different quantities.
23. Record Timeline
Study start
The DIAS-2 study began in April 2005.
Primary completion
The registry lists March 2007 as the primary completion date.
Study status
The ClinicalTrials.gov record identifies the study as completed.
24. Planned Statistical Analysis by Endpoint
| Endpoint | Registered outcome | Typical analysis of this endpoint type | Key statistical issue |
|---|---|---|---|
| NIHSS | Improvement ≥8 points from baseline or NIHSS ≤1 at day 90 | Binary responder comparison across randomized groups | Correct handling of the composite responder definition and baseline-to-day-90 assessment |
| MRS | MRS 0-2 at day 90 | Binary responder comparison; ordinal analysis is possible when the full scale is analysed | Distinguishing the registered binary estimand from an ordinal-scale estimand |
| Barthel Index | Score 75-100 at day 90 | Binary responder comparison; continuous analysis is possible when the full score is analysed | Preserving the distinction between threshold response and score-level change |
The table describes statistical approaches commonly used for endpoints of these types. It does not establish that a particular method was used in DIAS-2. The registry record does not report a formal statistical analysis for these endpoints.
25. What the Registry Establishes — and What It Does Not
This distinction is important in clinical-trial interpretation. A well-defined endpoint and a well-designed randomized comparison establish how evidence should be evaluated, but they do not substitute for the numerical treatment estimates themselves.
26. Summary of the Statistical Framework
DIAS-2 combines three features that are particularly important for clinical-trial statisticians: a three-arm randomized comparison, multiple primary functional endpoints, and threshold-based definitions derived from neurological and disability scales.
| Question | Statistical perspective |
|---|---|
| What was randomized? | Participants were randomized among two desmoteplase arms and one placebo arm. |
| When were outcomes assessed? | The primary endpoints were assessed at day 90, with the NIHSS endpoint defined as change from baseline to day 90. |
| What constitutes success? | NIHSS improvement ≥8 points or NIHSS ≤1; MRS 0-2; Barthel Index 75-100. |
| What type of outcomes are these? | They can be represented as binary responder outcomes, although the underlying scales contain additional information. |
| What is the main design issue? | The presence of three randomized arms creates multiple potential treatment comparisons. |
| What is the main multiplicity issue? | There are three registered primary endpoints, so formal confirmatory testing requires an appropriate error-control strategy. |
| What is needed for interpretation? | Responder counts and proportions, treatment contrasts, confidence intervals, hypothesis tests, and the prespecified analysis framework. |
27. Sources
- ClinicalTrials.gov: NCT00111852 — DIAS-2.
- PubMed: PubMed record.
- PubMed: PubMed record.
- PubMed: PubMed record.
Continue through the Clinical Biostats knowledge graph
Explore the broader statistical concepts that connect randomized clinical-trial design with endpoint definition, effect estimation, uncertainty, and multiplicity.
28. Record Summary
DIAS-2 was a randomized phase 3, parallel, triple-masked treatment study of desmoteplase in acute ischemic stroke. The study enrolled 193 participants across three arms: two desmoteplase arms and one placebo arm. Its three registered primary endpoints were neurological improvement on the NIHSS from baseline to day 90, achievement of MRS 0-2 at day 90, and achievement of a Barthel Index score of 75-100 at day 90.
Statistically, the trial illustrates why endpoint definitions matter. Each primary outcome transforms a neurological or functional scale into a clinically interpretable responder state, while the three-arm design introduces multiple possible treatment comparisons. The presence of three primary endpoints also makes multiplicity an important component of any confirmatory interpretation.
The ClinicalTrials.gov record does not post formal statistical analyses or numerical results for these primary endpoints. The appropriate statistical framework is therefore to focus on the registered estimands, the randomized three-arm structure, the distinction between binary threshold analyses and analyses of the underlying scales, and the need to interpret effect estimates together with their confidence intervals and multiplicity framework.