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
DECIDE was a completed phase 3, randomized, parallel-group trial evaluating BIIB019 (daclizumab high yield process) against interferon beta-1a in participants with relapsing-remitting multiple sclerosis. The registry reports 1841 enrolled participants, two arms, quadruple masking, and a primary endpoint of adjusted annualized relapse rate through 144 weeks.
| Feature | DECIDE |
|---|---|
| Trial name | (DECIDE) |
| ClinicalTrials.gov identifier | NCT01064401 |
| Phase | Phase 3 |
| Status | Completed |
| Therapeutic area | Neurology |
| Condition | Relapsing-Remitting Multiple Sclerosis |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | Quadruple |
| Primary purpose | Treatment |
| Enrollment | 1841 |
| Primary endpoint | Adjusted Annualized Relapse Rate (ARR) |
| Registered primary endpoint type | Count / rate |
| Statistical methods represented | Cox proportional-hazards model; Logistic regression; Negative binomial regression |
| Hypothesis type | Superiority |
| Lead sponsor | Biogen |
| Sponsor type | Industry |
2. Clinical Question
The central question was whether BIIB019 (daclizumab high yield process) differed from interferon beta-1a in the rate of confirmed relapses through 144 weeks in participants with relapsing-remitting multiple sclerosis. The registry also reports secondary analyses addressing MRI lesions, sustained disability progression, relapse-free status, and worsening in the MSIS-29 Physical Impact Score.
Population
Participants with relapsing-remitting multiple sclerosis enrolled in a phase 3 randomized trial.
Intervention
BIIB019, identified in the registry as daclizumab high yield process, with daclizumab high yield process placebo included in the trial intervention structure.
Comparator
Interferon beta-1a, with interferon beta-1a placebo included in the trial intervention structure.
Primary question
What is the relative difference in adjusted annualized relapse rate between interferon beta-1a and daclizumab high yield process through 144 weeks?
3. Trial Design
Interferon beta-1a
- Interferon beta-1a
- Interferon beta-1a placebo
Daclizumab high yield process
- BIIB019 (daclizumab high yield process)
- Daclizumab high yield process placebo
The registry identifies the study as randomized, parallel, quadruple-masked, and intended for treatment. The registry does not report an allocation ratio in the information summarized here.
4. Endpoints
Primary Endpoint
| Endpoint | Registry definition | Time frame | Type |
|---|---|---|---|
| Adjusted Annualized Relapse Rate (ARR) | Relapses are defined as new or recurrent neurological symptoms not associated with fever or infection, lasting at least 24 hours, and accompanied by new objective neurological findings upon examination by the examining neurologist. Only relapses confirmed by Independent Neurology Evaluation Committee (INEC) are included in this analysis. Adjusted ARR was estimated from a negative binomial regression model adjusted for the baseline relapse rate, history of prior IFN beta use, baseline Expanded Disability Status Scale score (EDSS; ≤ 2.5 vs > 2.5) and baseline age (≤ 35 vs > 35 years). Data after participants switched to alternative MS medications are excluded. | Up to 144 weeks | Count / rate |
Secondary Endpoints
| Endpoint | Time frame | Type | Analysis represented |
|---|---|---|---|
| Adjusted Mean Number of New or Newly Enlarging T2 Hyperintense Lesions up to Week 96 | Up to 96 weeks | Continuous | Negative binomial regression |
| Proportion of Participants With Sustained Disability Progression at 144 Weeks | Baseline through 144 weeks | Time-to-event / Binary | Cox proportional-hazards model and a reported percent-reduction measure |
| Proportion of Participants Relapse-free at Week 144 | 144 weeks | Time-to-event / Binary | Cox proportional-hazards model and a reported percent-reduction-in-risk measure |
| Percentage of Participants With a ≥ 7.5 Point Worsening From Baseline in the Multiple Sclerosis Impact Scale (MSIS-29) Physical Impact Score at 96 Weeks | Baseline and 96 weeks | Binary | Logistic regression and a reported percent-reduction-in-odds measure |
5. Statistical Methodology
Negative Binomial Regression for Adjusted ARR
The primary endpoint is a count/rate outcome because relapses can occur repeatedly for the same participant. The registry reports negative binomial regression for adjusted ARR. The model adjusts for baseline relapse rate, history of prior interferon beta use, baseline EDSS category, and baseline age category.
The resulting rate ratio compares the modeled relapse rates between the two randomized treatment groups over the specified follow-up. A rate ratio below 1 indicates a lower estimated relapse rate in the numerator group relative to the reference group.
Cox Proportional-Hazards Model
The registry reports Cox proportional-hazards models for sustained disability progression and relapse-free status. These endpoints are time-to-event outcomes, so the analysis uses information from both participants who experience the event and participants who remain event-free through censoring.
A hazard ratio is a relative time-to-event measure. It is not a probability, an absolute risk difference, or the percentage of participants who benefit.
Logistic Regression
The MSIS-29 endpoint is binary: participants are classified according to whether they experienced a ≥ 7.5 point worsening from baseline in the physical impact score at 96 weeks. Logistic regression expresses the treatment comparison as an odds ratio.
Effect Measures Reported
| Effect measure | Registry use | Relevant endpoints |
|---|---|---|
| Rate ratio | Primary count/rate analysis | Adjusted ARR |
| Percent reduction | Reported complementary treatment-effect expression | Adjusted ARR; sustained disability progression |
| Hazard ratio | Time-to-event analysis | Sustained disability progression; relapse-free status |
| Percent reduction in risk of relapse | Reported complementary treatment-effect expression | Relapse-free status |
| Odds ratio | Binary regression analysis | MSIS-29 physical impact worsening |
| Percent reduction in odds of worsening | Reported complementary treatment-effect expression | MSIS-29 physical impact worsening |
6. Primary Result: Adjusted Annualized Relapse Rate
The registry reports two statistical expressions of the primary endpoint. The formal model is negative binomial regression, with a rate ratio of 0.550 and a two-sided 95% confidence interval of 0.469 to 0.645. The registry also reports a 45.0% reduction with a two-sided 95% confidence interval of 35.5% to 53.1% and a P-value of < 0.0001.
Adjusted annualized relapse rate
95% CI: 0.469–0.645 · P < 0.0001
Reported reduction: 45.0% · 95% CI: 35.5–53.1%
The reported rate ratio of 0.550 means that the modeled relapse rate for the first-listed group in the registry comparison was 0.550 times that of the reference group. Equivalently, the registry reports this as a 45.0% reduction in the adjusted annualized relapse rate. The two expressions describe the same relative treatment comparison rather than two independent primary findings.
The 95% confidence interval of 0.469–0.645 describes the statistical uncertainty around the estimated rate ratio. It does not mean that individual participants experienced relapse-rate reductions restricted to that interval. It also does not provide the distribution of relapse counts among individual participants.
The P < 0.0001 value addresses evidence against the null hypothesis under the specified statistical test; it does not measure the magnitude or clinical importance of the effect. The magnitude is conveyed by the rate ratio and its confidence interval.
Interpretation also depends on the model specification: the analysis adjusts for baseline relapse rate, prior IFN beta use, EDSS category, and age category, and excludes data after participants switched to alternative MS medications. Those features are part of the definition of the reported primary analysis.
7. Secondary Result: New or Newly Enlarging T2 Hyperintense Lesions
The registry reports the adjusted mean number of new or newly enlarging T2 hyperintense lesions through 96 weeks. The analysis used negative binomial regression, with the logarithmic transformation of the scan number of the MRI assessment included in the model as the offset parameter.
Adjusted MRI lesion burden
95% CI: 46.9–60.8% · P < 0.0001
Time frame: up to 96 weeks
The registry's reported 54.4% reduction describes the relative treatment comparison for the adjusted mean number of new or newly enlarging T2 hyperintense lesions. The 95% confidence interval of 46.9% to 60.8% expresses uncertainty around that estimated relative reduction.
The analysis population consisted of participants with a baseline and at least one post-baseline MRI measurement. The use of the logarithmic scan-number offset accounts for the number of MRI assessments represented in the model, so the result should be understood as a model-based comparison rather than a simple comparison of raw lesion counts.
The P-value of < 0.0001 provides evidence against the null hypothesis used for the comparison; it does not itself indicate the size of the MRI effect.
8. Secondary Result: Sustained Disability Progression
The registry reports sustained disability progression from baseline through 144 weeks as both a time-to-event and binary endpoint. For the time-to-event analysis, the Cox proportional-hazards model produced a hazard ratio of 0.84 with a two-sided 95% confidence interval of 0.66 to 1.07 and a P-value of 0.1575. The registry also reports a 16.1% reduction with a 95% confidence interval of -7.0% to 34.2%.
Sustained disability progression
95% CI: 0.66–1.07 · P = 0.1575
Reported reduction: 16.1% · 95% CI: -7.0–34.2%
The hazard ratio of 0.84 represents the estimated relative instantaneous rate of sustained disability progression for the first-listed treatment group compared with the reference group in the Cox analysis. It is not equivalent to a 16% absolute reduction in the probability of disability progression.
The 95% confidence interval of 0.66–1.07 includes 1.00, indicating that the estimated treatment effect is compatible with a range that includes no relative difference under the model. The corresponding P-value of 0.1575 does not measure the size of the estimated effect; it quantifies evidence against the specified null hypothesis.
The reported percent reduction of 16.1% has a 95% confidence interval of -7.0% to 34.2%. The negative lower bound is consistent with the fact that the uncertainty interval includes effects in either direction.
Because this is a time-to-event analysis, censoring and the proportional-hazards model are relevant to interpretation. A single hazard ratio summarizes a relative event-rate relationship rather than providing a direct probability of progression by 144 weeks.
9. Secondary Result: Participants Relapse-free at Week 144
The registry reports relapse-free status at Week 144 using both a Cox proportional-hazards model and a binary percent-reduction measure.
Relapse-free status
95% CI: 0.50–0.69 · P < 0.0001
Reported reduction in risk of relapse: 40.9% · 95% CI: 30.8–49.5%
The hazard ratio of 0.59 indicates a lower estimated instantaneous relapse hazard for the first-listed group relative to the reference group in the Cox model. The registry separately expresses the result as a 40.9% reduction in risk of relapse.
The 95% confidence interval for the hazard ratio, 0.50–0.69, describes uncertainty around the estimated relative hazard. It does not mean that 50% to 69% of participants were relapse-free, nor does it give an individual participant's probability of remaining relapse-free.
The P-value of < 0.0001 is evidence against the specified null hypothesis but is not an effect-size measure. The effect size is communicated by the hazard ratio and the corresponding confidence interval.
Because the analysis uses a Cox model, the usual interpretation depends on the model's proportional-hazards framework. A hazard ratio should not be read as though it were a risk ratio at a single fixed time point.
10. Secondary Result: MSIS-29 Physical Impact Score
The registry reports the percentage of participants with a ≥ 7.5 point worsening from baseline in the MSIS-29 Physical Impact Score at 96 weeks. The analysis population included participants with a baseline and Week 96 assessment.
Odds of MSIS-29 physical impact worsening
95% CI: 0.60–0.95 · P = 0.0176
Reported reduction in odds of worsening: 24.2% · 95% CI: 4.7–39.6%
An odds ratio of 0.76 means that the estimated odds of meeting the specified worsening criterion were 0.76 times the odds in the reference group under the logistic regression analysis. The registry expresses this equivalently as a 24.2% reduction in odds of worsening.
An odds ratio is not a risk ratio. The difference matters especially when the outcome is not rare: an odds reduction of 24.2% should not automatically be interpreted as a 24.2% reduction in the proportion of participants who worsened.
The 95% confidence interval of 0.60–0.95 expresses uncertainty around the odds-ratio estimate. The P-value of 0.0176 addresses evidence against the null odds ratio of 1 and does not measure the magnitude of the treatment effect.
11. Results Summary
| Endpoint | Analysis | Effect estimate | 95% CI | P-value |
|---|---|---|---|---|
| Adjusted Annualized Relapse Rate (ARR), up to 144 weeks | Negative binomial regression | Rate ratio 0.550; 45.0% reduction | 0.469–0.645; 35.5–53.1% | < 0.0001 |
| Adjusted mean number of new or newly enlarging T2 hyperintense lesions, up to 96 weeks | Negative binomial regression | 54.4% reduction | 46.9–60.8% | < 0.0001 |
| Sustained disability progression, baseline through 144 weeks | Cox proportional-hazards model | HR 0.84; 16.1% reduction | 0.66–1.07; -7.0–34.2% | 0.1575 for HR analysis |
| Relapse-free at Week 144 | Cox proportional-hazards model | HR 0.59; 40.9% reduction in risk of relapse | 0.50–0.69; 30.8–49.5% | < 0.0001 |
| MSIS-29 physical impact worsening at 96 weeks | Logistic regression | OR 0.76; 24.2% reduction in odds | 0.60–0.95; 4.7–39.6% | 0.0176 |
The results span several statistical scales. ARR is modeled as a rate, disability progression and relapse-free status are evaluated with time-to-event methods, and MSIS-29 worsening is modeled as a binary outcome. Keeping these effect measures distinct is important because rate ratios, hazard ratios, and odds ratios answer different statistical questions.
12. Statistical Methods Explained
Why was negative binomial regression used for annualized relapse rate?
Relapses are recurrent count events rather than a single yes/no outcome. A negative binomial model is suited to count data when the variability can exceed what a simpler Poisson model assumes. In DECIDE, the registry specifically reports negative binomial regression for adjusted ARR and identifies several baseline covariates used for adjustment.
What does a rate ratio of 0.550 mean?
A rate ratio of 0.550 means that the modeled relapse rate for the first-listed group is estimated to be 0.550 times the rate for the reference group. Expressed as a relative reduction, this corresponds to 45.0%. The rate ratio is not a statement that 55.0% of participants avoided relapse or that every participant experienced the same relative change.
Why does the primary ARR model adjust for baseline factors?
The registry states that the adjusted ARR model includes baseline relapse rate, history of prior IFN beta use, EDSS category, and age category. Adjustment incorporates these prespecified baseline variables into the model rather than relying only on an unadjusted comparison of observed relapse counts. The resulting rate ratio therefore represents the specified adjusted model rather than a simple crude rate ratio.
What does a hazard ratio of 0.59 mean for relapse-free status?
A hazard ratio of 0.59 represents the relative instantaneous rate of relapse under the Cox model. It does not mean that the probability of relapse was 59%, nor does it directly state the proportion of participants who remained relapse-free at Week 144. A fixed-time probability requires an appropriate estimate of the event-free distribution at that time point.
Why is an odds ratio different from a risk ratio?
The MSIS-29 endpoint is binary, and logistic regression reports an odds ratio. Odds are defined as probability divided by one minus probability. Consequently, an odds ratio of 0.76 is not automatically a 24% reduction in the probability of worsening. The registry separately reports the effect as a 24.2% reduction in odds of worsening.
Why are confidence intervals important alongside P-values?
A P-value addresses evidence against a specified null hypothesis. A confidence interval gives an interval representation of statistical uncertainty around an estimated effect. For example, the sustained disability progression hazard ratio is 0.84 with a 95% CI of 0.66–1.07, while the primary ARR rate ratio is 0.550 with a 95% CI of 0.469–0.645. These intervals provide information about precision that a P-value alone does not.
13. Analysis Populations and Endpoint-Specific Scope
| Endpoint | Analysis population reported | Important analysis feature |
|---|---|---|
| Adjusted ARR | Participants with a relapse | Only relapses confirmed by the Independent Neurology Evaluation Committee are included; data after switching to alternative MS medications are excluded. |
| New or newly enlarging T2 hyperintense lesions | Participants with baseline and at least one post-baseline MRI measurement | Logarithmic transformation of the scan number of the MRI assessment was included as the offset parameter. |
| MSIS-29 physical impact worsening | Participants with baseline and Week 96 assessment | Logistic regression; outcome defined as a ≥ 7.5 point worsening from baseline. |
These population definitions matter because an endpoint estimate is tied to the participants who contribute information to that analysis. The denominators therefore should not automatically be assumed to be identical across all reported outcomes.
14. Safety Results
The registry reports serious adverse events by randomized treatment arm as affected participants divided by participants at risk. These figures provide a direct arm-level summary of serious adverse events.
| Treatment arm | Serious adverse events | Affected / at risk |
|---|---|---|
| IFN Beta-1a 30 mcg | Serious adverse events | 194 / 922 |
| DAC HYP 150 mg | Serious adverse events | 221 / 919 |
The registry presents serious adverse events as affected/at-risk counts. The statistical interpretation should therefore preserve that distinction rather than treating the affected counts as percentages. The corresponding denominators are 922 and 919, respectively.
15. How to Interpret the Different Effect Measures
Rate ratio
Used for the primary recurrent-relapse outcome. It compares modeled event rates rather than the probability of experiencing at least one event.
Hazard ratio
Used for time-to-event endpoints. It compares modeled instantaneous event rates over follow-up.
Odds ratio
Used for the binary MSIS-29 worsening endpoint. It compares odds rather than probabilities.
Percent reduction
The registry uses several percentage-based expressions to communicate relative treatment effects. These should be interpreted according to the underlying measure from which they were derived.
A common statistical error is to treat all relative measures as interchangeable. They are not. The primary endpoint is a recurrent count/rate outcome; disability progression and relapse-free status have a time-to-event structure; and MSIS-29 worsening is binary. The appropriate effect measure follows from that endpoint structure.
16. Confidence Intervals and Statistical Uncertainty
| Endpoint | Estimate | 95% confidence interval | Interpretive point |
|---|---|---|---|
| Adjusted ARR | Rate ratio 0.550 | 0.469–0.645 | The interval remains below 1 for the rate-ratio scale. |
| New/newly enlarging T2 lesions | 54.4% reduction | 46.9–60.8% | The interval represents uncertainty around the reported percentage reduction. |
| Sustained disability progression | HR 0.84 | 0.66–1.07 | The interval includes 1 on the hazard-ratio scale. |
| Relapse-free at Week 144 | HR 0.59 | 0.50–0.69 | The interval remains below 1 on the hazard-ratio scale. |
| MSIS-29 worsening | OR 0.76 | 0.60–0.95 | The interval remains below 1 on the odds-ratio scale. |
The location of a confidence interval relative to its null value is informative, but the interval width is also important. Narrower intervals indicate greater statistical precision, while wider intervals indicate greater uncertainty. Confidence intervals do not describe the range of effects that individual participants experience.
17. P-Values and What They Do Not Tell You
The primary ARR analysis reports P < 0.0001. The MRI lesion analysis and relapse-free analysis also report P < 0.0001. The MSIS-29 analysis reports P = 0.0176, while the sustained disability progression hazard-ratio analysis reports P = 0.1575.
| Endpoint | P-value | What the P-value addresses |
|---|---|---|
| Adjusted ARR | < 0.0001 | Evidence against the specified null hypothesis for the primary rate comparison. |
| New/newly enlarging T2 lesions | < 0.0001 | Evidence against the specified null hypothesis for the MRI analysis. |
| Sustained disability progression | 0.1575 | Evidence against the specified null hypothesis for the reported Cox analysis. |
| Relapse-free at Week 144 | < 0.0001 | Evidence against the specified null hypothesis for the Cox analysis. |
| MSIS-29 worsening | 0.0176 | Evidence against the specified null hypothesis for the logistic regression analysis. |
A P-value is not the probability that the null hypothesis is true, nor is it a measure of clinical magnitude. The estimated effect and confidence interval must be considered together with the P-value.
18. Design Features That Matter Statistically
| Design feature | Statistical relevance |
|---|---|
| Randomization | Creates the framework for comparing outcomes between treatment assignments while reducing systematic allocation differences. |
| Parallel design | Participants remain in their assigned randomized groups rather than serving as their own crossover controls. |
| Quadruple masking | Masking can reduce the potential influence of treatment knowledge on trial conduct and outcome assessment. |
| Superiority hypothesis | The registry identifies superiority as the hypothesis type for the posted analyses. |
| Repeated relapses | Requires an analysis appropriate for count/rate data rather than a simple binary responder analysis. |
| Time-to-event outcomes | Require methods that accommodate differing follow-up and censoring, such as the reported Cox model. |
The endpoint structure is particularly important in DECIDE. A participant can experience multiple relapses, while sustained disability progression and relapse-free status have a time-to-event dimension. Applying one generic statistical method to all of these outcomes would discard important information about how the endpoints are generated.
19. What the Primary Rate Ratio Does — and Does Not — Mean
A rate ratio of 0.550 means that the estimated adjusted relapse rate for the first-listed treatment group was 0.550 times the rate for the reference group under the reported negative binomial model.
Because the registry reports the corresponding effect as a 45.0% reduction, the two quantities can be interpreted together: the modeled rate is 55.0% of the comparison rate, corresponding to a 45.0% relative reduction.
This does not mean that 45.0% of participants avoided relapse, that each participant experienced a 45.0% reduction, or that the absolute difference in the number of relapses was 45.0 percentage points.
The 95% CI of 0.469–0.645 quantifies uncertainty around the estimated rate ratio under the reported model and analysis framework. It provides more information about precision than the point estimate alone.
The primary P-value of < 0.0001 indicates the degree of evidence against the specified null hypothesis under the statistical test. It does not say that the treatment effect is "more than" any particular clinical threshold, and it does not quantify the amount of benefit.
20. Limitations
- Endpoint-specific populations: the registry identifies different analysis populations for several secondary outcomes, so results should not automatically be interpreted as though every endpoint used the same denominator.
- Model dependence: adjusted ARR is based on a negative binomial model, while disability progression and relapse-free status use Cox proportional-hazards models and MSIS-29 worsening uses logistic regression.
- Proportional-hazards assumption: Cox-model hazard ratios rely on the proportional-hazards framework. A single hazard ratio is a summary relative event-rate measure rather than a direct fixed-time risk ratio.
- Recurrent-event interpretation: annualized relapse rate is a rate outcome, so it should not be interpreted as though it were simply the proportion of participants who experienced at least one relapse.
- MRI assessment structure: the MRI lesion analysis is restricted to participants with baseline and at least one post-baseline MRI measurement and incorporates the logarithmic scan number as an offset.
- Alternative MS medications: for the primary ARR analysis, data after participants switched to alternative MS medications are excluded, which is part of the registry-defined analysis.
- Serious adverse-event summary: the registry reports affected/at-risk counts for serious adverse events rather than a formal comparative effect estimate in the information summarized here.
- Registry reporting: ClinicalTrials.gov provides the reported statistical analyses, but it does not provide every possible detail of a full statistical analysis plan in the summarized record.
21. Why This Trial Matters Statistically
DECIDE is useful as a teaching case because the trial applies different regression frameworks to different endpoint structures within the same randomized comparison. The primary outcome is a recurrent count/rate endpoint, while secondary outcomes include time-to-event and binary outcomes.
| Concept | How it appears in DECIDE |
|---|---|
| Randomization | Randomized allocation in a two-arm parallel phase 3 trial. |
| Blinding | Quadruple masking. |
| Count/rate analysis | Negative binomial regression for adjusted ARR. |
| Covariate adjustment | Primary ARR model adjusts for baseline relapse rate, prior IFN beta use, EDSS category, and age category. |
| Rate ratio | Primary effect measure for adjusted ARR. |
| Time-to-event analysis | Cox proportional-hazards models for sustained disability progression and relapse-free status. |
| Hazard ratio | Effect measure for time-to-event secondary outcomes. |
| Binary regression | Logistic regression for MSIS-29 physical impact worsening. |
| Odds ratio | Effect measure for the binary MSIS-29 endpoint. |
| Confidence intervals | 95% two-sided intervals accompany the reported effect estimates. |
| P-values | Reported for the posted formal analyses. |
The broader statistical lesson is that endpoint definition should drive the analysis. A recurrent relapse count is not statistically equivalent to a single disability event, and neither is equivalent to a binary patient-reported outcome. The DECIDE analyses illustrate how the estimand and effect measure change with the structure of the endpoint.
22. Related Tutorials
Learn more about the methods used in this trial:
23. Related Calculators
24. Sources
- ClinicalTrials.gov: NCT01064401.
- PubMed: PMID 37652990.
- PubMed: PMID 37072219.
- PubMed: PMID 28984179.
- PubMed: PMID 28485186.
- PubMed: PMID 28104250.
Continue through Clinical Biostats
Connect the endpoints and statistical methods used in this trial with deeper biostatistics tutorials, statistical calculators, and additional clinical trial analyses.
25. Record Summary
DECIDE provides a useful example of endpoint-specific clinical-trial statistics. Its primary adjusted annualized relapse rate was analyzed with negative binomial regression and reported as a rate ratio of 0.550, corresponding to a reported 45.0% reduction. Secondary analyses applied negative binomial regression to MRI lesion counts, Cox proportional-hazards models to sustained disability progression and relapse-free status, and logistic regression to MSIS-29 physical impact worsening. The results therefore illustrate the importance of matching the statistical model and effect measure to the structure of the clinical endpoint.
The most informative reading combines the point estimate, confidence interval, P-value, analysis population, endpoint definition, and model assumptions. In particular, the primary ARR result should be interpreted as a model-based comparison of relapse rates, the Cox results as relative time-to-event measures, and the MSIS-29 result as an odds ratio for a defined binary worsening outcome.