This page separates reported trial information from statistical interpretation. The ClinicalTrials.gov record identifies the primary endpoint and its calculation method, but the registry-reported statistical-analysis field contains no formal comparative analysis estimates, confidence intervals, or p-values.
1. Trial at a Glance
FREEDOMS II was a randomized, parallel, triple-masked phase 3 treatment trial in multiple sclerosis. The registry lists fingolimod and placebo as the interventions and identifies one registered primary endpoint: the Aggregate Annualized Relapse Rate (ARR) Estimate up to Month 24.
| Feature | FREEDOMS II |
|---|---|
| Trial name | FREEDOMS II |
| NCT identifier | NCT00355134 |
| Phase | Phase 3 |
| Status | COMPLETED |
| Therapeutic area | Neurology |
| Condition | Multiple Sclerosis |
| Brief title | Efficacy and Safety of Fingolimod (FTY720) in Patients With Relapsing-remitting Multiple Sclerosis |
| Allocation | RANDOMIZED |
| Design model | PARALLEL |
| Masking | TRIPLE |
| Primary purpose | TREATMENT |
| Enrollment | 1083.0 |
| Arms | 3 |
| Interventions | Fingolimod (drug); Placebo (drug) |
| Lead sponsor | Novartis |
| Sponsor type | INDUSTRY |
2. Clinical Question
The central statistical question was whether treatment with fingolimod could be evaluated against placebo in patients with multiple sclerosis using the prespecified aggregate annualized relapse rate through Month 24.
Population
The trial's brief title identifies patients with relapsing-remitting multiple sclerosis. The registered condition is Multiple Sclerosis.
Intervention
Fingolimod (FTY720), listed in the registry as a drug intervention.
Comparator
Placebo, listed in the registry as a drug intervention.
Primary question
How does the aggregate annualized relapse rate, as defined by the registry, behave through Month 24 in the randomized treatment comparison?
The registry extract does not provide a detailed eligibility table, baseline-characteristic table, or treatment-dose schedule beyond the fingolimod dose labels appearing in the serious-adverse-event data.
3. Trial Design
Random allocation is explicitly identified in the registry.
The registry identifies a parallel trial design rather than a crossover design.
The registry identifies the study as triple-masked.
The registered primary purpose is treatment.
Fingolimod 1.25 mg
- Serious adverse events: 53 affected / 370 at risk.
- The ClinicalTrials.gov record identifies this dose in the core safety population.
Fingolimod 0.5 mg
- Serious adverse events: 53 affected / 358 at risk.
- The ClinicalTrials.gov record identifies this dose in the core safety population.
Placebo
- Serious adverse events: 45 affected / 355 at risk.
- The ClinicalTrials.gov record identifies placebo as the comparator intervention.
Extension populations
- Fingolimod 1.25 mg: 27 affected / 308 at risk.
- Fingolimod 0.5 mg: 21 affected / 324 at risk.
4. Endpoints
| Endpoint | Time frame | Type | Registry definition / method |
|---|---|---|---|
| Aggregate Annualized Relapse Rate (ARR) Estimate up to Month 24 | 24 months | Count / rate | ARR is the average number of relapses in a year calculated by negative binomial regression as the sum of confirmed relapses of all patients in the group divided by the sum of the number of days on study of all patients in the group and multiplied by 365.25. |
Definition of relapse
The registry definition states that a relapse was defined as the appearance of a new neurological abnormality or worsening of previously stable or improving pre-existing neurological abnormality, separated by at least 30 days from onset of a preceding clinical demyelinating event. The abnormality must be present for at least 24 hours and occur in the absence of fever (\<37.5C) or known infection. A relapse must be confirmed by the Independent Evaluating Physician (examining neurologist). ARR estimates were calculated from a negative binomial regression model adjusted for treatment, pooled center, number of relapses in the previous 2 years prior to enrollment, and Baseline expanded disability status scale (EDSS).
5. What the Primary Endpoint Measures
Annualized relapse rate is a count/rate endpoint. Instead of simply asking how many relapses occurred, the registered calculation relates the total number of confirmed relapses to the amount of time that patients were actually under study.
The registry definition states that this annualization is calculated by negative binomial regression and uses the aggregate number of confirmed relapses and aggregate days on study.
The important statistical feature is the denominator. Patients do not necessarily contribute identical amounts of observation time. Expressing the event count relative to days on study creates a rate rather than a simple proportion.
The 365.25 multiplier expresses the resulting rate on an annual scale. Thus, the endpoint is designed to describe the expected relapse frequency per year rather than the percentage of patients who relapse.
Count
The numerator is based on confirmed relapses occurring among patients in the group.
Exposure time
The denominator uses the sum of the number of days on study, allowing the rate to account for observation time.
Annualization
Multiplication by 365.25 expresses the rate on a one-year scale.
Regression
The registry specifically identifies negative binomial regression as the method used to calculate the ARR.
6. Results and Analysis Status
Registry results status
The ClinicalTrials.gov record reports that results are posted and that 11 outcome measures are posted.
However, no formal statistical analyses were posted.
The registry data establish the primary endpoint and its calculation method, but the ClinicalTrials.gov record does not provide a formal comparative estimate for that endpoint.
| Item | What the ClinicalTrials.gov record establishes | What they do not establish |
|---|---|---|
| Primary endpoint | Aggregate Annualized Relapse Rate (ARR) Estimate up to Month 24 | A numerical treatment-effect estimate |
| Endpoint type | Count / rate | A reported confidence interval or p-value |
| Statistical method | Negative binomial regression is specified in the endpoint definition | Coefficient, rate ratio, standard error, confidence interval, or p-value |
| Results status | Results Posted: Yes; 11 outcome measures posted | A formal comparative analysis in the posted results |
What a formal analysis would normally report
For a count/rate endpoint of this type, a negative binomial model can account for the number of relapse events while incorporating the amount of observation time. A treatment comparison would ordinarily be presented through a model-based effect estimate, commonly expressed as a rate ratio or an equivalent treatment contrast, together with a measure of uncertainty and a hypothesis-test result when those quantities are part of the prespecified analysis.
None of those comparative numerical results are present in the posted results. They are therefore not reproduced or inferred here.
7. Statistical Methodology
Negative binomial regression
The registered primary endpoint explicitly states that ARR is calculated by negative binomial regression. This is a regression framework for count outcomes and is commonly useful when count data show greater variability than would be accommodated by a basic Poisson model.
For a relapse-count endpoint, the model conceptually connects the expected number of events to treatment and the amount of observation time. The exposure-time component is important because patients can contribute different numbers of days on study.
Here, Y represents the relapse count and exposure time represents the amount of observation contributing to the rate. The registry extract specifically identifies negative binomial regression for the ARR calculation, but it does not provide the complete model specification.
Why negative binomial rather than a simple arithmetic average?
A simple arithmetic average of individual relapse counts would not by itself account for unequal observation time. The registered ARR definition instead aggregates confirmed relapses and aggregates days on study, then expresses the rate annually through the 365.25 multiplier.
Negative binomial regression also provides a model-based framework for count data when the variability of the counts can exceed what a simple Poisson model assumes. That distinction matters because repeated clinical events such as relapses can be heterogeneous across patients.
Person-time and annualized rates
The registry's use of days on study is a form of person-time accounting. A patient contributing more observation time has more opportunity to experience a relapse, so the denominator should reflect that exposure.
An annualized relapse rate summarizes event frequency over observation time. It is not the same quantity as the percentage of patients who experience at least one relapse.
Randomization
The trial is identified as randomized. Randomization is important because, when properly implemented, it creates treatment groups whose expected baseline characteristics are comparable in repeated hypothetical trials. That provides the foundation for interpreting a treatment-group contrast causally rather than simply as an association.
Triple masking
The registry identifies the masking status as TRIPLE. Masking can reduce the opportunity for knowledge of treatment assignment to influence behavior, assessment, reporting, or other aspects of trial conduct. The ClinicalTrials.gov record does not specify which three participant or study roles were masked, so those roles are not assigned here.
Parallel design
The registry identifies the design model as PARALLEL. In a parallel design, participants remain associated with their randomized treatment group rather than serving as their own treatment-period comparator. That is conceptually different from a crossover design, and no crossover feature is reported in the ClinicalTrials.gov record.
8. Statistical Methods Explained
Why was a count/rate endpoint used?
The primary endpoint is based on the number of confirmed relapses occurring over a defined period of follow-up. A rate allows the analysis to account for the amount of time patients were actually observed rather than treating all participants as if they contributed exactly the same exposure time.
Why is ARR different from the percentage of patients who relapse?
A percentage answers a binary question: whether a patient experienced an event. ARR instead counts events and relates those events to observation time. A patient can experience more than one relapse, so the two measures summarize different aspects of the data.
Why does the formula use days on study?
Because the registry defines ARR using the sum of the number of days on study. This makes the denominator reflect the total observation time contributing to the relapse count. The final multiplication by 365.25 expresses that rate on an annual scale.
Why use negative binomial regression?
Negative binomial regression is a count-data model that can accommodate extra variability in event counts relative to a basic Poisson framework. The registry definition specifically names negative binomial regression as the method for calculating the registered ARR endpoint.
What does triple masking add to the design?
Triple masking is a design feature intended to reduce the possibility that knowledge of treatment assignment influences aspects of trial conduct or assessment. The registry identifies the masking level but does not identify the three masked roles.
Why does randomization matter for the statistical comparison?
Randomization provides the design basis for comparing treatment groups because treatment assignment is determined by the randomization process rather than by the patient's observed characteristics. Statistical adjustment can refine an analysis, but it does not replace the fundamental role of randomized allocation.
9. Safety Results
The ClinicalTrials.gov record reports serious adverse events by arm as affected patients over patients at risk. These data include both core and extension populations and should therefore be interpreted within those population labels.
| Population | Arm | Serious adverse events |
|---|---|---|
| Core | Fingolimod 1.25 mg | 53 / 370 |
| Core | Fingolimod 0.5 mg | 53 / 358 |
| Core | Placebo | 45 / 355 |
| Extension | Fingolimod 1.25 mg | 27 / 308 |
| Extension | Fingolimod 0.5 mg | 21 / 324 |
These are counts of affected patients relative to the corresponding number at risk, not event rates derived by the analysis on this page. The ClinicalTrials.gov record does not provide a formal statistical comparison of serious adverse events, so no comparative p-value or confidence interval is reported here.
10. Trial Timeline
Trial start
The ClinicalTrials.gov record lists 2006-06 as the study start.
Primary endpoint time frame
The registered primary endpoint is the Aggregate Annualized Relapse Rate estimate up to Month 24.
Primary completion
The ClinicalTrials.gov record lists 2011-06 as the primary completion date.
Current registry status in the ClinicalTrials.gov record
The trial is identified as COMPLETED.
11. Interpretation of the Annualized Relapse Rate
The registered ARR is an annualized event rate. It represents the average number of confirmed relapses per year after accounting for the total number of days that patients were on study.
ARR is not the percentage of patients who relapsed, not the probability that an individual patient will relapse, and not a time-to-first-relapse measure. It counts confirmed relapse events and relates them to observation time.
The registry specifies negative binomial regression. Consequently, the primary endpoint is not simply an unadjusted division of two totals; it is defined through a count-regression framework incorporating the accumulated observation time.
No formal statistical analyses were posted. Therefore, the ClinicalTrials.gov record does not support a numerical treatment effect, confidence interval, p-value, or formal statistical conclusion about the primary endpoint.
12. Important Limitations and Interpretation Issues
- No formal comparative analysis reported: The trial data state that results are posted, but no formal statistical analyses were posted. A numerical treatment comparison therefore cannot be reconstructed from the ClinicalTrials.gov record.
- Incomplete relapse definition in the registry-reported extract: The definition ends with “pre-existing neurological abnormality, separated by at least 30 days from onset of a preceding clinical demyelinating event. The abnormality must be present for at least 24 hours and occur in the absence of fever (\<37.5C) or known infection. A relapse must be confirmed by the Independent Evaluating Physician (examining neurologist). ARR estimates were calculated from a negative binomial regression model adjusted for treatment, pooled center, number of relapses in the previous 2 years prior to enrollment, and Baseline expanded disability status scale (EDSS).”, so the remainder is not reconstructed.
- Limited model specification: The registry identifies negative binomial regression but does not provide the full model specification, covariates, dispersion specification, or reported treatment-effect parameter in the registry-reported extract.
- Three-arm design with limited arm-level efficacy detail: The ClinicalTrials.gov record identifies 3 arms but do not provide arm-specific primary efficacy estimates.
- Safety populations differ: Serious adverse-event data are separately identified for core and extension populations. Those populations should not be pooled without an explicit analysis plan.
- Limited baseline information: The ClinicalTrials.gov record does not contain a baseline-characteristics table, so balance across randomized groups cannot be evaluated from this extract.
- Limited multiplicity information: The ClinicalTrials.gov record identifies one registered primary endpoint but do not provide a multiplicity strategy for the 11 posted outcome measures.
- Limited missing-data information: The ClinicalTrials.gov record does not specify a missing-data or imputation strategy for the primary endpoint.
- No subgroup analysis information: The ClinicalTrials.gov record does not provide subgroup-specific efficacy estimates or interaction analyses.
13. Design Features That Are Not Specified in the Supplied Data
| Design topic | What can be said from the ClinicalTrials.gov record |
|---|---|
| Non-inferiority margin | No non-inferiority margin is provided. |
| Crossover | No crossover design is reported; the design model is identified as PARALLEL. |
| Factorial design | No factorial structure is reported; the design model is identified as PARALLEL. |
| Multiplicity | No multiplicity-adjustment strategy is reported. |
| Interim analysis | No interim-analysis specification is reported. |
| Missing-data / imputation | No missing-data or imputation strategy is reported. |
| Stratification | No randomization stratification factors are reported. |
| Bayesian methods | No Bayesian method is identified. |
This distinction is important in statistical review. Absence of a method from the ClinicalTrials.gov record should not be converted into an assumption that the method was or was not used in the underlying trial.
14. Why This Trial Matters Statistically
FREEDOMS II is a useful teaching case because its registered primary endpoint illustrates a problem that occurs frequently in clinical-trial statistics: repeated clinical events must be analyzed as counts over exposure time, rather than as a simple binary outcome.
| Concept | How it appears in FREEDOMS II |
|---|---|
| Randomization | The allocation is identified as RANDOMIZED. |
| Parallel design | The design model is PARALLEL. |
| Blinding | The masking level is TRIPLE. |
| Count endpoint | The primary endpoint is classified as Count / rate. |
| Annualized rate | The primary endpoint is Aggregate Annualized Relapse Rate up to Month 24. |
| Person-time | The ARR definition uses the sum of the number of days on study. |
| Negative binomial regression | The registry explicitly specifies negative binomial regression for ARR calculation. |
| Repeated events | The endpoint counts confirmed relapses rather than reducing the outcome to whether a patient ever relapsed. |
| Safety populations | Serious adverse events are reported separately for core and extension populations. |
| Registry interpretation | Results are posted, but the posted statistical analyses contains no formal comparative analyses. |
15. A Deeper Statistical Reading of the Primary Endpoint
Rate construction is part of the estimand
One of the most important lessons is that the definition of an endpoint determines what a treatment comparison means. An annualized relapse rate is concerned with event frequency during observation. It is therefore different from asking whether treatment changes the probability of experiencing at least one relapse.
Exposure time prevents a misleading denominator
Suppose two groups had identical numbers of observed relapses but different total observation times. A raw event count would not distinguish those situations. The registered ARR definition does: the relapse count is divided by the total number of days on study and then annualized.
Negative binomial regression adds a model for the counts
Count data can exhibit substantial patient-to-patient variability. Negative binomial regression is designed for count outcomes with an additional dispersion component beyond the basic Poisson variance structure. In a clinical-trial setting, this can provide a more appropriate framework for recurrent event counts when that extra variability is present.
The analysis should remain aligned with the registered endpoint
Because the registry explicitly defines the primary endpoint as an aggregate annualized relapse rate and specifies negative binomial regression, replacing it with a different endpoint or a different primary analysis would change the statistical question. A statistical review should therefore preserve the registered endpoint wording.
Results require more than the endpoint definition
Knowing how an endpoint is defined does not reveal the treatment effect. To report an actual comparative result, the analysis output must provide the relevant numerical estimate and its uncertainty, and ideally the associated hypothesis-testing information where applicable. Those quantities are not present in the posted results.
16. What the Safety Data Do — and Do Not — Mean
The safety values are expressed as affected patients divided by patients at risk. For example, the registry-reported core data report 53/370 for fingolimod 1.25 mg, 53/358 for fingolimod 0.5 mg, and 45/355 for placebo.
Serious-adverse-event counts describe safety outcomes. They do not provide evidence about the registered ARR endpoint and should not be combined with efficacy measures into a single summary statistic.
The extension data are separately reported as 27/308 for fingolimod 1.25 mg and 21/324 for fingolimod 0.5 mg. These denominators should be retained as extension populations rather than treated as interchangeable with the core populations.
17. Clinical Interpretation vs Statistical Interpretation
Statistical interpretation
The registry specifies a randomized, parallel, triple-masked phase 3 trial with a count/rate primary endpoint and negative binomial regression for ARR. It does not supply a formal comparative estimate in the posted results.
Clinical interpretation
The trial addresses relapse frequency in patients with relapsing-remitting multiple sclerosis. The ClinicalTrials.gov record also provide serious-adverse-event counts for the core and extension populations.
The distinction matters because statistical methodology describes how evidence is quantified, whereas clinical interpretation concerns what that evidence means in the context of the disease and intervention. Without a registry-reported treatment-effect estimate, the latter cannot be reduced to a numerical efficacy conclusion here.
18. What a Complete Statistical Result Would Need
For an educational analysis of this primary endpoint, a complete comparative result would ideally identify:
- the treatment groups being compared;
- the observed or model-based relapse counts;
- the corresponding exposure time;
- the estimated annualized relapse rate or treatment contrast;
- the uncertainty associated with the estimate;
- the hypothesis test, if one was prespecified and reported;
- the analysis population used;
- the model specification and any covariates or offsets;
- the handling of missing or incomplete follow-up; and
- any multiplicity or interim-analysis framework that affected interpretation.
The ClinicalTrials.gov record establishes only part of this structure. They provide the registered endpoint, its time frame, its count/rate classification, its negative-binomial calculation method, and safety information, but not the full comparative statistical output.
19. Planned Analysis
Because no formal statistical analyses were posted, the appropriate educational approach is to describe the registered endpoint and the analysis framework rather than infer an outcome.
The registry defines ARR from confirmed relapse counts and total days on study, with annualization by 365.25.
For this type of endpoint, the negative binomial model provides a natural framework for recurrent count data and can incorporate observation time. The treatment comparison would normally be expressed through a model-based contrast between randomized groups, with uncertainty quantified from the fitted model.
20. Why Randomization and Blinding Matter
Randomization
Randomization is the central design mechanism for creating treatment groups whose expected prognostic characteristics are comparable. This is especially important when the endpoint is affected by many patient-level factors, because those factors need not be measured individually for randomization to provide its design-based protection against systematic treatment-assignment differences.
Blinding
The registry describes the trial as triple-masked. Masking can protect the integrity of outcome assessment and other study processes by limiting knowledge of treatment assignment. For an endpoint involving identification and confirmation of relapse events, maintaining consistent assessment procedures is particularly relevant to the reliability of the event count.
Why the two features work together
Randomization primarily addresses systematic differences created by treatment assignment, while masking addresses potential behavioral or assessment effects associated with knowledge of assignment. They address different sources of bias and are therefore complementary design features.
21. Related Tutorials
Learn more about the methods used in this trial:
22. Related Statistical Calculators
23. Sources
- ClinicalTrials.gov: FREEDOMS II — NCT00355134.
- Linked publication: PubMed PMID 36286605.
- Linked publication: PubMed PMID 27733070.
- Linked publication: PubMed PMID 26121423.
- Linked publication: PubMed PMID 25245812.
- Linked publication: PubMed PMID 24685276.
Continue through the Clinical Biostats statistical library
Explore related biostatistics tutorials, statistical calculators, and additional clinical-trial analyses.
24. Record Summary
FREEDOMS II provides a useful example of how a randomized clinical trial can define a primary outcome around recurrent clinical events and observation time. Its registered primary endpoint is the Aggregate Annualized Relapse Rate estimate up to Month 24, classified as a count/rate endpoint and calculated using negative binomial regression from confirmed relapse counts and total days on study. The trial is identified as randomized, parallel, and triple-masked, with 1083.0 enrolled and 3 arms.
The ClinicalTrials.gov record also provide serious-adverse-event counts for three core groups and two extension groups. Those safety populations are reported separately and should not be merged without an explicit analysis plan.
The central limitation for statistical interpretation is that, although results are identified as posted and 11 outcome measures are listed, the posted statistical analyses contains no formal comparative analyses.