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
GALA was a randomized, parallel-group, phase 3 trial in subjects with relapsing-remitting multiple sclerosis. The study compared glatiramer acetate 40 mg administered three times a week with placebo and evaluated confirmed relapses, annualized relapse rate, MRI lesion outcomes, and brain-volume change.
| Feature | GALA |
|---|---|
| Phase | Phase 3 |
| Condition | Relapsing Remitting Multiple Sclerosis |
| Design | Randomized, parallel-group, quadruple-masked |
| Allocation | Randomized |
| Primary purpose | Treatment |
| Enrollment | 1,404 |
| Primary endpoints | Total number of confirmed relapses during the placebo-controlled treatment period; annualized rate of confirmed relapses comparing early starters to delayed starters |
| Primary statistical method | Negative binomial regression |
| Results status | Completed; results posted |
| ClinicalTrials.gov | NCT01067521 |
| Lead sponsor | Teva Branded Pharmaceutical Products R&D, Inc. |
2. Clinical Question
The central statistical question was whether glatiramer acetate 40 mg administered three times a week reduced confirmed relapse activity compared with placebo during the placebo-controlled treatment period, and whether early initiation was associated with a lower annualized rate of confirmed relapses over the study period compared with delayed initiation.
Population
Subjects with relapsing-remitting multiple sclerosis enrolled in the phase 3 GALA study.
Intervention
Glatiramer acetate 40 mg administered three times a week.
Comparator
Placebo during the placebo-controlled treatment period, followed by glatiramer acetate in the delayed-start group.
Primary questions
Does early glatiramer acetate treatment reduce confirmed relapse activity during the placebo-controlled period and over the study period?
3. Trial Design
GA 40 mg / GA 40 mg
- Glatiramer acetate 40 mg during the placebo-controlled period.
- Glatiramer acetate 40 mg continued during the subsequent study period.
- Compared with delayed starters for the annualized relapse-rate analysis.
Placebo / GA 40 mg
- Placebo during the placebo-controlled period.
- Glatiramer acetate 40 mg during the subsequent study period.
- Provides the delayed-start comparison for the long-term annualized relapse-rate analysis.
The registry describes the study as randomized, parallel, and quadruple-masked, with treatment as the primary purpose. The trial started on June 22, 2010, and the primary completion date was May 8, 2012.
4. Endpoints
| Endpoint | Time frame | Registry definition / assessment |
|---|---|---|
| Total number of confirmed relapses during the placebo-controlled treatment period | Day 1 to 12 months | Relapses were monitored throughout the study. During the placebo-controlled period, two neurologists/physicians assessed subjects' general medical and neurological evaluations separately. A relapse required the appearance of one or more new neurological abnormalities or reappearance of one or more previously observed neurological abnormalities lasting at least 48 hours and immediately preceded by an improving neurological state of at least 30 days from onset of the previous relapse. Objective neurological changes included an increase of at least 0.5 in EDSS, an increase of one grade in the actual score of at least two of the seven functional systems, or an increase of two grades in the actual score of one functional system. |
| Annualized rate of confirmed relapses comparing early starters to delayed starters | Day 1 up to 6.5 years | Annualized relapse rate was calculated by dividing the cumulative number of confirmed relapses by the number of person-years of exposure to treatment. The analysis estimated an early-start versus delayed-start contrast from a baseline-adjusted negative binomial regression model with an offset based on the log of exposure to treatment. |
Secondary endpoints reported in the registry
The posted statistical analyses also included MRI lesion measures and brain-volume outcomes. These analyses used negative binomial regression, ANCOVA, or MMRM according to the endpoint.
| Secondary outcome | Time frame | Method | Effect measure |
|---|---|---|---|
| Cumulative number of new/enlarging T2 lesions during the placebo-controlled period | Baseline (Day -7), Month 6, Month 12 | Negative binomial regression | Risk ratio |
| Cumulative number of Gd-enhanced lesions on T1-weighted images during the placebo-controlled period | Baseline (Day -7), Month 6, Month 12 | Negative binomial regression | Risk ratio |
| Brain atrophy defined by percent change in normalized brain volume | Baseline (Day -7), Month 12 | ANCOVA | Mean difference |
| Number of new/enlarging T2 lesions | Baseline (Day -7), Months 6, 12, and 36 | Negative binomial regression | Risk ratio |
| Cumulative number of Gd-enhanced T1 lesions | Baseline (Day -7), Months 6, 12, and 36 | Negative binomial regression | Risk ratio |
| Brain atrophy defined by percent change in brain volume | Baseline (Day -7), Months 6, 12, and 36 | MMRM | Mean difference |
5. Statistical Methodology
Negative binomial regression for relapse counts
The primary relapse endpoints were analyzed using negative binomial regression. This is a count-regression model appropriate when the outcome is the number of events observed over a period of exposure and when the variability of the counts can exceed what a simple Poisson model would reasonably accommodate.
The registry describes an offset based on the log of exposure to treatment. This allows subjects with different amounts of observed treatment exposure to contribute to an annualized event-rate comparison.
For the placebo-controlled primary analysis, the baseline-adjusted model included baseline EDSS score, the log of the prior two-year number of relapses, baseline T2 lesion volume, baseline Gd-enhancing T1 activity status, and country or geographical region.
For the entire-study annualized relapse-rate analysis, the registry specifies treatment group, placebo-controlled baseline EDSS score, log of the number of relapses in the two years before placebo-controlled baseline, baseline T2 lesion volume, baseline Gd-enhancing T1 lesion activity, and country or geographical region.
Risk ratio as the reported effect measure
The registry reports the negative-binomial treatment contrast as a risk ratio. For these analyses, the estimate is most naturally interpreted as a relative comparison of the modeled relapse rate between the treatment groups, with the exposure offset accounting for person-time.
An RR of 0.656 corresponds to a modeled rate approximately 34.4% lower than the comparator, because 1 − 0.656 = 0.344. This is a relative model-based interpretation, not an absolute reduction in the number of relapses experienced by every subject.
ANCOVA for brain-volume change
The Month 12 brain-atrophy endpoint was analyzed using ANCOVA. ANCOVA compares an outcome between treatment groups while accounting for specified covariate information. The reported effect measure was the mean difference in final values.
MMRM for repeated brain-volume measurements
The brain-atrophy endpoint measured at Months 6, 12, and 36 was analyzed using a mixed model for repeated measures. MMRM is designed for longitudinal data in which each subject can contribute multiple measurements over time while accounting for within-subject correlation.
Rather than treating the Month 6, Month 12, and Month 36 observations as independent, an MMRM framework models the repeated measurements as longitudinal observations from the same participants.
Intention-to-treat analysis
The posted primary analyses were based on the intent-to-treat population. For a randomized comparison, ITT preserves the treatment assignment created by randomization and avoids redefining the primary efficacy comparison solely around treatment exposure or adherence.
6. Primary Results
Primary Endpoint 1: Confirmed relapses during the placebo-controlled period
Negative-binomial relapse-rate comparison
95% CI: 0.539–0.799 · P < 0.0001
Analysis population: Intent-to-Treat. Time frame: Day 1 to 12 months.
The analysis compared Early Start: GA 40 mg / GA 40 mg with Delayed Start: Placebo / GA 40 mg. The registry reports a baseline-adjusted negative binomial regression with an exposure offset. The covariates were baseline EDSS score, log of the prior two-year number of relapses, baseline T2 lesion volume, baseline Gd-enhancing T1 activity status, and country or geographical region.
The estimate of 0.656 means that the modeled confirmed-relapse rate for the early-start group was approximately 65.6% of the corresponding rate in the delayed-start group during the placebo-controlled period. Expressed as a relative difference, this corresponds to an estimated 34.4% lower modeled relapse rate because 1 − 0.656 = 0.344.
The 95% confidence interval of 0.539–0.799 describes statistical uncertainty around the estimated treatment contrast. It does not mean that 95% of individual patients would experience a relapse-rate reduction within that range, nor does it describe the range of biological effects in individual subjects.
The p-value of <0.0001 addresses the strength of evidence against the null hypothesis specified for the statistical comparison. It does not measure the magnitude of the treatment effect; the magnitude is described by the RR and its confidence interval.
Because this is a negative-binomial analysis of recurrent events, the estimate is a model-based comparison of event rates rather than a simple comparison of the proportion of subjects who experienced at least one relapse.
Primary Endpoint 2: Annualized confirmed relapse rate through 6.5 years
Early start versus delayed start
95% CI: 0.6995–0.9925 · P = 0.0409
Analysis population: ITT. Time frame: Day 1 up to 6.5 years.
The annualized relapse-rate analysis used a negative binomial regression model with an offset based on the log of exposure to treatment. The model included treatment group, placebo-controlled baseline EDSS score, log of the number of relapses in the two years before placebo-controlled baseline, baseline T2 lesion volume, baseline Gd-enhancing T1 lesion activity, and country or geographical region.
The estimate of 0.8332 means that the modeled annualized confirmed-relapse rate for the early-start strategy was approximately 83.32% of the corresponding rate for delayed start. As a simple relative interpretation, this corresponds to an estimated 16.68% lower modeled relapse rate because 1 − 0.8332 = 0.1668.
The 95% confidence interval of 0.6995–0.9925 is relatively close to 1.00 at its upper boundary. It therefore conveys more uncertainty about the magnitude of the longer-term treatment contrast than the much narrower separation from 1.00 seen in the placebo-controlled primary analysis.
The p-value of 0.0409 indicates the reported statistical evidence for the prespecified superiority comparison. It is not a measure of clinical importance and should not be interpreted as a probability that the treatment effect is real.
The two primary estimates should not be treated as interchangeable. The first is a placebo-controlled comparison over 12 months, whereas the second evaluates an early-start versus delayed-start strategy over up to 6.5 years after both groups ultimately received GA 40 mg.
7. Secondary MRI Results
New or enlarging T2 lesions during the placebo-controlled period
Cumulative number of new/enlarging T2 lesions
95% CI: 0.546–0.780 · P < 0.0001
Baseline (Day -7), Month 6, and Month 12; ITT population.
The analysis used negative binomial regression. When both Month 6 and Month 12 scans were missing, the subject was excluded. When the Month 12 scan was missing, Month 6 data were used with an offset of log(0.5). When the Month 6 scan was missing, Month 12 data were used with an offset of 0.
An RR of 0.653 represents a modeled lesion count approximately 65.3% of the comparator value, or approximately a 34.7% lower modeled lesion count relative to the comparator. The 95% CI of 0.546–0.780 indicates that the estimated relative reduction is subject to sampling uncertainty but remains below 1.00 throughout the reported interval.
The result concerns MRI lesion activity rather than clinical relapse directly. A reduction in one outcome should not be interpreted as proof that the numerical effect is identical for a different endpoint.
Gadolinium-enhanced T1 lesions during the placebo-controlled period
Cumulative number of Gd-enhanced T1 lesions
95% CI: 0.436–0.699 · P < 0.0001
Baseline (Day -7), Month 6, and Month 12; ITT population.
The estimated RR of 0.552 corresponds to a modeled lesion count approximately 55.2% of the comparator value, or approximately a 44.8% lower modeled count. The 95% CI of 0.436–0.699 quantifies uncertainty around that estimate.
The p-value of <0.0001 provides evidence against the null treatment comparison but does not indicate how clinically meaningful a particular MRI-lesion difference is for an individual subject.
8. Secondary Longitudinal MRI Results
New or enlarging T2 lesions
| Assessment time | Effect | 95% CI | P-value |
|---|---|---|---|
| Month 6 | RR 0.736 | 0.592–0.914 | 0.0056 |
| Month 12 | RR 0.633 | 0.524–0.765 | <0.0001 |
| Month 36 | RR 0.666 | 0.557–0.797 | <0.0001 |
These analyses used negative binomial regression in the ITT population of participants with MRIs at baseline and the designated timeframes. The registry reports the Month 6, Month 12, and Month 36 contrasts separately.
The estimates remain below 1.00 at all three reported time points. Numerically, the corresponding relative differences are approximately 26.4% lower at Month 6, 36.7% lower at Month 12, and 33.4% lower at Month 36. These are direct transformations of the reported RRs and should be understood as relative model-based differences in lesion counts.
The confidence intervals communicate the precision of each time-specific estimate. They do not establish that the treatment effect itself changed from one month to another; demonstrating a difference between time-specific treatment effects would require a formal comparison of those effects.
Gadolinium-enhanced T1 lesions
| Assessment time | Effect | 95% CI | P-value |
|---|---|---|---|
| Month 6 | RR 0.556 | 0.4–0.773 | <0.0001 |
| Month 12 | RR 0.514 | 0.388–0.679 | <0.0001 |
| Month 36 | RR 0.663 | 0.514–0.854 | 0.0015 |
The reported estimates correspond to approximately 44.4% lower modeled lesion counts at Month 6, 48.6% lower at Month 12, and 33.7% lower at Month 36. These calculations simply express the reported RRs relative to 1.00; they do not create new statistical estimates.
The Month 6 confidence interval begins at 0.4, while the Month 12 interval is 0.388–0.679 and the Month 36 interval is 0.514–0.854. The progressively different widths illustrate why the point estimate alone should not be used to compare the precision of these analyses.
9. Brain Atrophy and Brain-Volume Results
Month 12 brain-volume change
Percent change in normalized brain volume
95% CI: −0.154 to 0.033 · P = 0.2058
ANCOVA; baseline (Day -7) to Month 12; ITT population with SIENEX brain-volume estimates at both time points.
The reported mean difference of −0.061 is the adjusted difference in the final-value measure between the treatment groups as reported by the ANCOVA analysis. The negative sign indicates a lower estimated value for the first-listed group under the registry's effect coding.
The 95% CI of −0.154 to 0.033 crosses 0.00. The p-value of 0.2058 therefore does not provide strong statistical evidence against a zero mean difference under the reported superiority test. Importantly, this does not prove that the groups are identical; it indicates that the reported data are compatible with a range of effects that includes zero.
MMRM brain-volume results
| Assessment time | Mean difference | 95% CI | P-value |
|---|---|---|---|
| Month 6 | −0.084 | −0.166 to −0.003 | 0.0425 |
| Month 12 | −0.086 | −0.184 to 0.012 | 0.0844 |
| Month 36 | 0.017 | −0.91 to 0.225 | 0.8701 |
The MMRM analysis included participants with SIENEX brain scans at baseline and the designated timeframes. The registry reports the estimates at Months 6, 12, and 36 separately.
At Month 6, the estimated mean difference was −0.084, with a 95% CI of −0.166 to −0.003 and P = 0.0425. The confidence interval lies below zero at its upper boundary by only 0.003, so the estimate is relatively close to the boundary of the null value.
At Month 12, the estimate was −0.086, but the 95% CI of −0.184 to 0.012 included zero, with P = 0.0844. At Month 36, the estimate was 0.017, with a 95% CI of −0.91 to 0.225 and P = 0.8701.
These separate longitudinal estimates should not be treated as three independent confirmatory conclusions. The registry identifies them as repeated measurements within an MMRM framework, and the broader set of secondary analyses creates an additional multiplicity consideration.
10. Statistical Methods Explained
Why was negative binomial regression used for relapse counts?
The primary clinical outcome was a count of confirmed relapses rather than a simple binary endpoint. Negative binomial regression models event counts while accommodating differences in exposure time through an offset and allowing more variability in counts than a restrictive Poisson model. That structure matches the registry's description of annualized relapse-rate analysis.
What does the risk ratio of 0.656 mean?
The estimate of 0.656 means that the modeled confirmed-relapse rate in the early-start group was 0.656 times the corresponding rate in the delayed-start group during the placebo-controlled period. Equivalently, the modeled rate was approximately 34.4% lower. It does not mean that exactly 34.4% of subjects avoided a relapse or that every subject had the same relative reduction.
Why was an exposure offset included?
Subjects may contribute different amounts of observed treatment exposure. Using the log of exposure as an offset converts the count model into an analysis of event rates rather than raw event counts alone. This is especially important when the estimand is expressed as relapses per year.
Why adjust for baseline covariates?
The placebo-controlled relapse model adjusted for baseline EDSS, prior two-year relapse history, T2 lesion volume, Gd-enhancing T1 activity, and country or geographical region. Covariate adjustment can account for prognostic baseline information and improve precision while retaining the randomized treatment comparison.
What does an MMRM add to the brain-volume analysis?
Brain volume was assessed repeatedly at Months 6, 12, and 36. MMRM explicitly treats these observations as repeated measurements from the same subjects rather than unrelated observations. This allows the analysis to describe treatment-group differences over multiple assessment times while accounting for within-subject correlation.
Why is a confidence interval important when interpreting these results?
A point estimate such as RR 0.656 is only one estimate from the observed data. The 95% confidence interval provides an uncertainty range under the statistical model and sampling framework. A narrow interval indicates greater precision than a wide interval, while an interval that includes the null value requires a different interpretation from one entirely separated from it.
Why should p-values not be used as measures of effect size?
A p-value describes the statistical evidence against a specified null hypothesis under the model. It does not quantify the magnitude of a treatment effect. The magnitude is conveyed by the RR or mean difference, while the confidence interval provides information about its precision.
11. Analysis Populations and Missing MRI Data
| Endpoint | Analysis population | Registry handling |
|---|---|---|
| Confirmed relapses, placebo-controlled period | ITT | Baseline-adjusted negative binomial regression. |
| Annualized confirmed relapse rate | ITT | Negative binomial regression with log exposure offset. |
| T2 lesions, placebo-controlled period | ITT | If both Month 6 and Month 12 scans were missing, the subject was excluded. If Month 12 was missing, Month 6 was used with log(0.5) offset; if Month 6 was missing, Month 12 was used with offset 0. |
| Gd-enhanced T1 lesions, placebo-controlled period | ITT | If both Month 6 and Month 12 scans were missing, the subject was excluded. |
| Month 12 brain-volume change | ITT participants with SIENEX estimates at baseline and Month 12 | ANCOVA. |
| T2 lesions at Months 6, 12, and 36 | ITT participants with MRIs at baseline and designated timeframes | Negative binomial regression. |
| Gd-enhanced T1 lesions at Months 6, 12, and 36 | ITT participants with MRIs at baseline and designated timeframes | Negative binomial regression. |
| Brain-volume change at Months 6, 12, and 36 | ITT participants with SIENEX brain scans at baseline and designated timeframes | MMRM. |
The missing-data handling differs by endpoint. In particular, the placebo-controlled T2-lesion analysis specifies a defined use of the available Month 6 or Month 12 scan with an offset when only one of the two scans was missing. That is materially different from simply discarding every participant with one missing scan.
12. Safety Results
The registry reports serious adverse events separately by treatment sequence. These counts are presented as affected subjects over the corresponding subjects at risk.
| Treatment sequence | Serious adverse events | Affected / at risk |
|---|---|---|
| Delayed Start: GA 40 mg | Serious adverse events | 36 / 419 |
| Delayed Start: Placebo | Serious adverse events | 21 / 461 |
| Early Start: GA 40 mg / GA 40 mg | Serious adverse events | 117 / 943 |
These figures should be read as arm-specific safety counts rather than as an efficacy measure. The denominators are important because a count of affected subjects alone does not describe the frequency of an event.
Early-start sequence
117 of 943 subjects in the Early Start: GA 40 mg / GA 40 mg sequence were affected by serious adverse events.
Delayed-start sequence
The registry separately reports 36 of 419 subjects for delayed-start GA 40 mg and 21 of 461 subjects for delayed-start placebo.
13. Multiplicity and Interpretation of the Analysis Set
The registry contains multiple posted statistical analyses across primary and secondary endpoints. The primary relapse analyses are explicitly identified as superiority analyses with two-sided 95% confidence intervals.
The registry also contains an important limitation concerning the open-label period: analyses of Open-Label Period data are not adjusted for multiplicity and therefore do not control for type one statistical error.
This is particularly relevant when interpreting the secondary MRI and brain-volume analyses. A p-value such as 0.0425 for a single secondary time point should be considered in the context of the complete set of analyses rather than treated as an isolated definitive finding.
14. Primary Endpoint Interpretation in Context
| Feature | Placebo-controlled relapse endpoint | 6.5-year annualized relapse endpoint |
|---|---|---|
| Time frame | Day 1 to 12 months | Day 1 up to 6.5 years |
| Comparison | Early Start: GA 40 mg / GA 40 mg vs Delayed Start: Placebo / GA 40 mg | Early Start: GA 40 mg / GA 40 mg vs Delayed Start: Placebo / GA 40 mg |
| Method | Negative binomial regression | Negative binomial regression |
| Effect measure | Risk ratio | Risk ratio |
| Estimate | 0.656 | 0.8332 |
| 95% CI | 0.539–0.799 | 0.6995–0.9925 |
| P-value | <0.0001 | 0.0409 |
The difference between these estimates illustrates why treatment effects should always be interpreted with their estimand and observation period. The placebo-controlled estimate addresses a direct GA-versus-placebo comparison during the first 12 months. The 6.5-year estimate addresses an early-versus-delayed treatment strategy after the delayed group also receives GA 40 mg.
The longer-term estimate of 0.8332 should not be described as though it were simply a 6.5-year extension of the placebo-controlled comparison. Once both groups receive GA 40 mg, the contrast changes from "GA versus placebo" to "earlier versus delayed initiation." This distinction is central to the causal interpretation of the second primary endpoint.
Similarly, the numerical difference between 0.656 and 0.8332 does not by itself demonstrate that the treatment effect weakened over time. Establishing a change in treatment effect would require a formal statistical comparison of effects across periods, not simply comparing two point estimates.
15. Confidence Intervals: Reading the Primary Results
RR 0.656
The 95% CI of 0.539–0.799 remains below 1.00, indicating that the reported interval for the modeled treatment contrast does not include the null relative effect.
RR 0.8332
The 95% CI of 0.6995–0.9925 also remains below 1.00, but its upper boundary is close to the null value, conveying greater uncertainty about the size of the longer-term contrast.
A confidence interval should be interpreted as an interval generated by a statistical procedure with a specified coverage property under repeated sampling. It is not the probability that the true effect lies inside this particular interval, nor is it a prediction interval for individual patient outcomes.
16. Why This Trial Matters Statistically
GALA is a useful statistical case study because the trial connects a recurrent clinical event endpoint with longitudinal imaging outcomes and multiple modeling frameworks. The same randomized treatment comparison is examined through negative binomial regression, ANCOVA, and MMRM, each matched to a different structure of the outcome.
| Concept | How it appears in GALA |
|---|---|
| Randomization | Subjects were randomly allocated in a phase 3 parallel-group trial. |
| Quadruple masking | The registry identifies the study as quadruple-masked. |
| Intention-to-treat analysis | The primary and posted secondary efficacy analyses use ITT populations. |
| Negative binomial regression | Used for confirmed-relapse counts and MRI lesion counts. |
| Exposure offset | The annualized relapse-rate model used the log of treatment exposure as an offset. |
| Covariate adjustment | Primary relapse models adjusted for baseline disease characteristics and country/geographical region. |
| ANCOVA | Used for the Month 12 brain-volume endpoint. |
| MMRM | Used for repeated brain-volume measurements at Months 6, 12, and 36. |
| Risk ratio | Reported for the negative-binomial relapse and lesion analyses. |
| Mean difference | Reported for brain-volume analyses. |
| Multiplicity | The registry states that Open-Label Period analyses were not adjusted for multiplicity. |
17. Important Limitations and Interpretation Issues
- Different estimands: the first primary endpoint is a placebo-controlled 12-month comparison, while the second is an early-start versus delayed-start comparison extending up to 6.5 years.
- Count-model interpretation: negative binomial regression estimates relative event rates rather than the probability that an individual subject experiences a particular number of relapses.
- Exposure time: the annualized relapse-rate model uses an exposure offset, so the reported RR is tied to the rate-model framework rather than a raw comparison of cumulative counts.
- Secondary-endpoint multiplicity: multiple MRI and brain-volume analyses create a broader testing context. The registry specifically notes that Open-Label Period analyses were not adjusted for multiplicity.
- Missing imaging data: MRI analyses use endpoint-specific analysis populations and, for the placebo-controlled T2-lesion analysis, prespecified handling of missing Month 6 or Month 12 scans.
- Different populations across endpoints: although the analyses are generally ITT-based, imaging outcomes require available scans at specified time points.
- Statistical versus clinical effect: an RR below 1 quantifies a relative model-based difference but does not by itself describe the absolute number of relapses avoided or the clinical importance for an individual subject.
- Repeated measurements: longitudinal brain-volume results should be interpreted within the MMRM framework rather than as unrelated single-time-point comparisons.
18. What the Risk Ratio Does — and Does Not — Mean
An RR of 0.656 means that the modeled relapse rate for the early-start group was 0.656 times the corresponding rate in the delayed-start group during the placebo-controlled period.
It does not mean that 65.6% of subjects were relapse-free, that 34.4% of subjects were protected from relapse, or that every patient experienced a 34.4% reduction in their personal relapse risk.
The 95% CI of 0.539–0.799 provides information about the uncertainty around the estimated relative rate under the specified model. The interval is not a range of treatment effects that must contain the effect experienced by each patient.
The p-value of <0.0001 for the placebo-controlled relapse analysis provides evidence against the null hypothesis under the specified statistical model. It does not tell the reader whether the effect is small, moderate, or large. That question is addressed by the RR and its confidence interval.
19. Interpreting the Brain-Volume Results
The brain-volume endpoints illustrate an important difference between a statistically estimated treatment contrast and a binary conclusion about efficacy. The Month 12 ANCOVA estimate was −0.061 with a 95% CI of −0.154 to 0.033 and P = 0.2058. The corresponding MMRM Month 6 estimate was −0.084 with a 95% CI of −0.166 to −0.003 and P = 0.0425, while the Month 12 and Month 36 MMRM results had confidence intervals that included zero.
The correct statistical reading is not that a single p-value determines the entire brain-volume evidence base. Each estimate describes a particular model-based comparison at a particular assessment time, and the collection of secondary analyses must be considered together.
20. Statistical Analysis Workflow
This workflow highlights a general principle in clinical biostatistics: the statistical model should follow the structure of the endpoint. Recurrent relapse and lesion counts call for count-based methods; a single continuous brain-volume endpoint can be analyzed with ANCOVA; repeated continuous measurements can be analyzed longitudinally with MMRM.
21. Results Summary
| Endpoint | Method | Estimate | 95% CI | P-value |
|---|---|---|---|---|
| Confirmed relapses, Day 1 to 12 months | Negative binomial regression | RR 0.656 | 0.539–0.799 | <0.0001 |
| Annualized confirmed relapses, Day 1 to 6.5 years | Negative binomial regression | RR 0.8332 | 0.6995–0.9925 | 0.0409 |
| New/enlarging T2 lesions, PC period | Negative binomial regression | RR 0.653 | 0.546–0.780 | <0.0001 |
| Gd-enhanced T1 lesions, PC period | Negative binomial regression | RR 0.552 | 0.436–0.699 | <0.0001 |
| New/enlarging T2 lesions, Month 6 | Negative binomial regression | RR 0.736 | 0.592–0.914 | 0.0056 |
| New/enlarging T2 lesions, Month 12 | Negative binomial regression | RR 0.633 | 0.524–0.765 | <0.0001 |
| New/enlarging T2 lesions, Month 36 | Negative binomial regression | RR 0.666 | 0.557–0.797 | <0.0001 |
| Gd-enhanced T1 lesions, Month 6 | Negative binomial regression | RR 0.556 | 0.4–0.773 | <0.0001 |
| Gd-enhanced T1 lesions, Month 12 | Negative binomial regression | RR 0.514 | 0.388–0.679 | <0.0001 |
| Gd-enhanced T1 lesions, Month 36 | Negative binomial regression | RR 0.663 | 0.514–0.854 | 0.0015 |
| Brain-volume change, Month 12 | ANCOVA | Mean difference −0.061 | −0.154 to 0.033 | 0.2058 |
| Brain-volume change, Month 6 | MMRM | Mean difference −0.084 | −0.166 to −0.003 | 0.0425 |
| Brain-volume change, Month 12 | MMRM | Mean difference −0.086 | −0.184 to 0.012 | 0.0844 |
| Brain-volume change, Month 36 | MMRM | Mean difference 0.017 | −0.91 to 0.225 | 0.8701 |
The registry therefore reports a consistent set of negative-binomial estimates below 1.00 for relapse and MRI lesion outcomes, while the brain-volume analyses show different estimates across time points. Those endpoints answer different statistical questions and should not be collapsed into a single summary statistic.
22. Why This Trial Matters Statistically
GALA demonstrates how a clinical trial can require several complementary statistical methods because its endpoints have different mathematical structures. The primary clinical outcome is a recurrent event count, MRI outcomes are lesion counts, and brain-volume outcomes are continuous longitudinal measurements.
Recurrent events
Negative binomial regression models the number of confirmed relapses while accounting for treatment exposure.
Imaging counts
The same count-model framework is applied to new/enlarging T2 lesions and Gd-enhanced T1 lesions.
Continuous outcomes
ANCOVA provides the Month 12 brain-volume comparison, while MMRM handles repeated brain-volume measurements.
Multiplicity
The registry explicitly notes that Open-Label Period analyses were not adjusted for multiplicity, illustrating why nominal p-values require context.
23. Related Tutorials
Learn more about the methods used in this trial:
24. Related Calculators
25. Sources
- ClinicalTrials.gov: NCT01067521.
- PubMed: PMID 27503905.
Continue through the Clinical Biostats statistical library
Explore tutorials and statistical calculators that explain the methods used to design, analyze, and interpret randomized clinical trials.
26. Record Summary
GALA provides a detailed example of endpoint-specific clinical-trial methodology. The primary relapse outcomes were analyzed with baseline-adjusted negative binomial regression and an exposure offset, while MRI lesion outcomes used the same count-regression framework and brain-volume outcomes used ANCOVA or MMRM according to the measurement structure. The reported primary relapse estimates were RR 0.656 during the placebo-controlled period and RR 0.8332 for the annualized relapse-rate comparison through 6.5 years, each accompanied by a two-sided 95% confidence interval and superiority p-value.
The broader statistical interpretation requires keeping the estimands separate. The placebo-controlled result directly compares GA 40 mg with placebo over 12 months, whereas the longer-term analysis compares early versus delayed initiation after both treatment strategies include GA 40 mg. Secondary MRI results generally report RRs below 1.00, while brain-volume estimates vary across the ANCOVA and MMRM analyses. The registry also explicitly notes that Open-Label Period analyses were not adjusted for multiplicity, reinforcing the importance of interpreting individual secondary and later analyses within the full statistical context.