This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record. Reported estimates are kept separate from the interpretation that follows them.
1. Trial at a Glance
ADVANCE was a randomized, triple-masked, placebo-controlled phase 3 trial sponsored by Biogen. It asked whether peginterferon beta-1a, given every four weeks (Q4W) or every two weeks (Q2W), reduced relapse activity compared with placebo over the first year of treatment in people with relapsing multiple sclerosis.
| Feature | ADVANCE |
|---|---|
| Phase | Phase 3 |
| Condition | Relapsing multiple sclerosis |
| Design | Randomized, parallel-group, triple-masked, placebo-controlled |
| Arms | 3 (placebo; peginterferon beta-1a Q4W; peginterferon beta-1a Q2W) |
| Enrollment | 1516 |
| Primary endpoint | Annualized relapse rate (ARR) at 1 year |
| Primary analysis | Negative binomial regression, adjusted for baseline EDSS, age and relapse rate |
| Dates | Start 2009-06; primary completion 2012-10 |
| ClinicalTrials.gov | NCT00906399 |
| Sponsor | Biogen (industry) |
2. Clinical Question
The central question was whether peginterferon beta-1a, at either of two dosing frequencies, lowers the rate of confirmed relapses over one year relative to placebo in relapsing multiple sclerosis. Because relapses are recurrent events that a participant can experience more than once, the question is framed as a comparison of event rates, not of the proportion of participants with any event.
Population
Participants with relapsing multiple sclerosis who were randomized and received at least one dose of study treatment.
Intervention
Peginterferon beta-1a (BIIB017), administered every 4 weeks (Q4W) or every 2 weeks (Q2W).
Comparator
Placebo during year 1. In year 2, participants who had received placebo went on to receive peginterferon beta-1a.
Primary question
Does each peginterferon beta-1a schedule reduce the annualized relapse rate at 1 year compared with placebo?
3. Trial Design
Placebo
- Matching placebo during the controlled year
- Reference group for every efficacy comparison
- In year 2, followed by peginterferon beta-1a
Peginterferon beta-1a
- Q4W schedule (500 at risk in year 1)
- Q2W schedule (513 at risk in year 1)
- Each schedule compared separately against placebo
4. Endpoints
The registry lists one primary endpoint. Its full definition is reproduced below, followed by the secondary endpoints for which formal comparisons were posted.
| Role | Endpoint | Time frame | Definition / unit |
|---|---|---|---|
| Primary | Annualized Relapse Rate (ARR) at 1 Year | 1 Year | A relapse is defined as new or recurrent neurologic symptoms not associated with fever or infection, lasting for at least 24 hours, and accompanied by new objective neurologic findings. Only relapses confirmed by an independent neurology evaluation committee (INEC) are included. Data after participants switched to alternative MS medications are excluded. Unit: relapses per person-years. |
| Secondary | Number of New Or Newly Enlarging T2 Hyperintense Lesions at 1 Year | 1 Year | Count of lesions on MRI. Unit: lesions. |
| Secondary | Proportion of Participants Relapsed at 1 Year | Year 1 | Unit: proportion of participants; analysed as time to first relapse. |
| Secondary | Estimated Proportion of Participants With Sustained Disability Progression at 1 Year | 1 Year | Unit: proportion of participants; analysed as time to sustained progression. |
These four endpoints use three different statistical families. The ARR and the T2 lesion count are count outcomes analysed with negative binomial regression; relapse occurrence and disability progression are time-to-first-event outcomes analysed with a Cox proportional-hazards model. The choice of effect measure, and how to interpret it, follows directly from that distinction.
5. Analysis Populations
| Analysis | Population as described in the registry |
|---|---|
| ARR at 1 year | Intent-to-treat (ITT) population: participants who were randomized and received at least 1 dose of study treatment. Data after switching to alternative MS medications are excluded. |
| T2 lesions at 1 year | ITT population with at least 1 post-baseline assessment. Missing data before alternative MS medications, and visits after switching, imputed from previous visit data assuming a constant rate of lesion development, or from the group mean at the same visit. |
| Proportion relapsed | ITT population; participants without a relapse before switching to alternative MS medications or withdrawing were censored at the time of switch/withdrawal. |
| Sustained disability progression | ITT population; participants censored at withdrawal/switch if they withdrew or switched to alternative MS medication without a progression. |
The ITT definition used here is a modified ITT: it requires at least one dose of study treatment. In a blinded placebo-controlled trial, exclusion based on receiving no dose is unlikely to depend on assignment, so this restriction generally has little effect on the validity of the comparison. The more consequential choices concern what happens after a participant switches to another MS therapy, which is handled differently for each endpoint (exclusion of post-switch data, imputation, or censoring).
6. Results: Primary Endpoint — Annualized Relapse Rate at 1 Year
Both peginterferon beta-1a schedules were compared with placebo using negative binomial regression adjusted for baseline EDSS (< 4 versus ≥ 4), baseline relapse rate and age (< 40 versus ≥ 40 years). The effect measure is the rate ratio of relapses per person-year, active versus placebo; values below 1 favour peginterferon beta-1a. The registry posts the adjusted rate ratios, 95% two-sided confidence intervals and p-values for each comparison under a superiority hypothesis.
Peginterferon beta-1a Q2W vs placebo
Rate ratio for annualized relapse rate
95% CI: 0.500–0.831 · P = 0.0007
Negative binomial regression, ITT population, superiority
What it means. A rate ratio of 0.644 means that, after adjustment for baseline EDSS, age and prior relapse rate, the estimated relapse rate in the Q2W group was 64.4% of the rate in the placebo group, a 35.6% lower estimated relapse rate over the first year.
What it does not mean. It does not mean that 35.6% fewer participants relapsed, nor that each participant's own relapse rate fell by that amount. The ratio compares group-level event rates per unit of person-time; a reduction in relapse rate can come from fewer participants relapsing, fewer repeat relapses among those who do, or both.
Precision. The 95% CI of 0.500 to 0.831 is compatible with reductions in the relapse rate from about 17% to about 50%. The whole interval lies below 1, so the data are not consistent with no effect at conventional error levels, but the plausible magnitude of benefit still spans a wide range.
The p-value. P = 0.0007 quantifies how surprising the data would be if the true rate ratio were 1. It does not measure how large the effect is; that information is carried by the estimate and its interval.
Cautions. This is one of two primary comparisons against a shared placebo group, and the registry does not describe how type I error was controlled across them. Post-switch data were excluded, so participants contribute person-time only while on their assigned regimen or before switching. Negative binomial regression allows for overdispersion, which is typical of relapse counts, and yields a conditional (covariate-adjusted) rate ratio.
Peginterferon beta-1a Q4W vs placebo
Rate ratio for annualized relapse rate
95% CI: 0.565–0.930 · P = 0.0114
Negative binomial regression, ITT population, superiority
What it means. A rate ratio of 0.725 corresponds to a 27.5% lower estimated relapse rate with Q4W dosing than with placebo, after covariate adjustment.
What it does not mean. The point estimate for Q4W is closer to 1 than for Q2W, but that alone is not evidence that the two schedules differ. The two intervals overlap substantially, and a formal Q2W-versus-Q4W comparison is not among the posted analyses.
Precision. The 95% CI of 0.565 to 0.930 excludes 1, but its upper end corresponds to only a 7% lower relapse rate, while the lower end corresponds to a 43.5% lower rate. The data therefore support a reduction whose size is imprecisely estimated.
The p-value. P = 0.0114 is larger than for Q2W, reflecting a smaller estimated effect relative to its standard error. It should not be read as a measure of clinical importance.
Cautions. If a multiplicity procedure split the overall alpha across the two dose comparisons, or tested them in sequence, the threshold applied to this p-value may have been stricter than 0.05. The registry does not report the procedure, so readers should look to the primary publications for the prespecified testing strategy.
| Comparison vs placebo | Rate ratio | 95% CI | P-value | Implied change in relapse rate |
|---|---|---|---|---|
| Peginterferon beta-1a Q2W | 0.644 | 0.500–0.831 | 0.0007 | 35.6% lower |
| Peginterferon beta-1a Q4W | 0.725 | 0.565–0.930 | 0.0114 | 27.5% lower |
The registry reports the adjusted rate ratios but these summaries do not include the per-arm relapse rates, which would provide the absolute scale needed to translate a relative reduction into relapses avoided per participant-year. Without the baseline rate, a rate ratio says how much the rate changed in relative terms, not how many relapses that represents.
7. Secondary Endpoint Results
New or newly enlarging T2 hyperintense lesions at 1 year
Lesion counts were compared with negative binomial regression adjusted for the baseline number of T2 lesions. The effect measure is a lesion mean ratio: the ratio of the model-estimated mean number of new or newly enlarging lesions, active versus placebo.
| Comparison vs placebo | Lesion mean ratio | 95% CI | P-value | Implied change in mean lesion count |
|---|---|---|---|---|
| Peginterferon beta-1a Q2W | 0.33 | 0.27–0.40 | <0.0001 | 67% lower |
| Peginterferon beta-1a Q4W | 0.72 | 0.60–0.87 | 0.0008 | 28% lower |
The MRI endpoint shows the clearest difference between schedules. The Q2W interval (0.27–0.40) and the Q4W interval (0.60–0.87) do not overlap, which is suggestive of a genuine difference in effect on new lesion formation, although a direct Q2W-versus-Q4W test is not posted. Two features of this analysis deserve attention: it was restricted to participants with at least one post-baseline assessment, and missing values were imputed either by carrying forward a constant rate of lesion development from earlier visits or by the group mean at the same visit. Both are single-imputation approaches, which treat imputed values as if observed and can make confidence intervals somewhat too narrow.
Proportion of participants relapsed at 1 year
Although labelled a proportion, this endpoint was analysed as time to first relapse with a Cox proportional-hazards model. Participants who had not relapsed before switching to alternative MS medications or withdrawing were censored at that time.
| Comparison vs placebo | Hazard ratio | 95% CI | P-value | Implied change in hazard |
|---|---|---|---|---|
| Peginterferon beta-1a Q2W | 0.61 | 0.47–0.80 | 0.0003 | 39% lower |
| Peginterferon beta-1a Q4W | 0.74 | 0.57–0.95 | 0.0200 | 26% lower |
The hazard ratios describe the instantaneous rate of a first relapse among participants still relapse-free. They are close in size to the corresponding ARR rate ratios, which is what one would expect if the treatment reduces relapse occurrence broadly rather than only preventing repeat relapses. The two measures nevertheless answer different questions: the rate ratio uses all relapses, while the hazard ratio stops counting at the first.
Sustained disability progression at 1 year
Time to sustained disability progression was also analysed with a Cox model, censoring participants who withdrew or switched therapy without a progression.
| Comparison vs placebo | Hazard ratio | 95% CI | P-value | Implied change in hazard |
|---|---|---|---|---|
| Peginterferon beta-1a Q2W | 0.62 | 0.40–0.97 | 0.0383 | 38% lower |
| Peginterferon beta-1a Q4W | 0.62 | 0.40–0.97 | 0.0380 | 38% lower |
Both schedules have the same reported hazard ratio and interval. The intervals are wide, running from a 60% lower to a 3% lower hazard, because sustained progression is a relatively infrequent event over one year and the number of events drives the precision of a hazard ratio. With an upper limit this close to 1, the evidence for an effect on disability is weaker and less precise than for relapse outcomes, and it is more sensitive to how secondary endpoints were controlled for multiplicity.
8. Safety: Serious Adverse Events
The registry reports the number of participants with at least one serious adverse event, by group and study year.
| Period | Group | Affected / at risk |
|---|---|---|
| Year 1 | Placebo | 76/499 |
| Year 1 | Peginterferon beta-1a Q4W | 70/500 |
| Year 1 | Peginterferon beta-1a Q2W | 55/513 |
| Year 2 | Placebo followed by peginterferon beta-1a (first group) | 42/227 |
| Year 2 | Placebo followed by peginterferon beta-1a (second group) | 36/228 |
| Year 2 | Peginterferon beta-1a Q4W | 67/439 |
| Year 2 | Peginterferon beta-1a Q2W | 39/438 |
In year 1, the numbers of participants with serious adverse events were not higher in either active group than in the placebo group. These are raw counts without formal between-group testing, and serious adverse event frequencies were not a powered endpoint. Year-2 figures come from a period without a placebo comparator and with smaller numbers at risk, so they describe continued exposure rather than a randomized contrast. Counts of participants affected also say nothing about the type, severity or relatedness of individual events; that detail sits in the full adverse event tables.
9. Statistical Methodology
Negative binomial regression for relapse and lesion counts
Relapse counts and new lesion counts are non-negative integers whose variance is usually larger than their mean because some participants are much more active than others. A Poisson model would understate that variability and produce confidence intervals that are too narrow. The negative binomial model adds a dispersion parameter that absorbs this extra variation. For the ARR, each participant's follow-up time enters as an offset, so the model estimates events per person-year rather than raw counts.
where Yi is the relapse count, ti the person-time at risk and κ the dispersion parameter. The rate ratio is exp(β1).
Covariate adjustment
The ARR model adjusted for baseline EDSS (< 4 versus ≥ 4), age (< 40 versus ≥ 40 years) and baseline relapse rate (number of relapses in the 3 years before entry divided by 3). The lesion model adjusted for baseline number of T2 lesions. In a randomized trial, adjustment for strongly prognostic baseline variables does not remove confounding (randomization handles that) but improves precision. In non-linear models such as the negative binomial or Cox model, it also shifts the estimand toward a conditional effect.
Cox proportional-hazards model
Time to first relapse and time to sustained disability progression were analysed with Cox models, which estimate a hazard ratio without specifying the shape of the baseline hazard. The registry lists these outcomes as proportions at 1 year; such proportions are typically derived from Kaplan-Meier estimates, while the hazard ratio summarises the relative event rate over the whole year.
A single HR assumes the ratio of hazards is roughly constant over the year. It is a relative measure and is not the same as a ratio of the 1-year proportions.
Censoring and intercurrent events
Switching to an alternative MS medication is an intercurrent event that changes what is being measured. The trial handled it three ways: excluding post-switch data (ARR), imputing lesion counts (MRI), and censoring at switch or withdrawal (time-to-event endpoints). Censoring at switch is valid only if switching is unrelated to the underlying risk of the event after accounting for model covariates; if participants with more active disease switch sooner, censoring can bias the hazard ratio.
Missing data and imputation
For the T2 lesion endpoint, missing values were imputed from a participant's own earlier visits (assuming a constant rate of lesion development) or from the group mean at the same visit. These single-imputation rules are transparent but do not propagate imputation uncertainty. Multiple imputation, which generates several plausible datasets and combines them, is the usual way to reflect that uncertainty in the confidence interval.
Multiplicity
ADVANCE compared two active schedules against one placebo group for the primary endpoint and several secondary endpoints. Each additional comparison increases the chance of at least one false-positive finding. The registry does not describe the error-control strategy, so the posted p-values are presented as nominal values.
10. Statistical Methods Explained
Why was negative binomial regression used for the annualized relapse rate?
Relapses are counts accumulated over varying follow-up time, and participants differ widely in how often they relapse. The negative binomial model handles both: an offset for time converts counts into rates, and a dispersion parameter accommodates variance larger than the mean. Using a Poisson model instead would typically produce overly narrow intervals and overly small p-values.
What does a rate ratio of 0.644 mean in practice?
It means the adjusted relapse rate with Q2W dosing was estimated at 64.4% of the placebo rate, a 35.6% relative reduction. It is a relative measure: its absolute meaning depends on how often participants in the placebo group relapsed. The same ratio would translate into many more relapses avoided in a highly active population than in a quiet one.
How does the ARR rate ratio differ from the hazard ratio for first relapse?
The rate ratio counts every relapse over the year; the hazard ratio stops at the first relapse per participant. In ADVANCE the two measures are similar for each schedule (0.644 and 0.61 for Q2W; 0.725 and 0.74 for Q4W), which suggests the treatment effect is reflected both in whether participants relapse and in how often. They are not interchangeable, because they rely on different models and assumptions.
Why can a "proportion relapsed" endpoint be analysed with a Cox model?
Some participants stopped or switched before the end of year 1, so a simple proportion (relapsed divided by randomized) would treat their unobserved time incorrectly. A time-to-event analysis uses each participant's follow-up up to censoring, and the Cox model summarises the difference as a hazard ratio. Kaplan-Meier methods can then estimate the proportion relapsed by 1 year while accounting for censoring.
What does a lesion mean ratio of 0.33 tell us?
It indicates that the model-estimated mean number of new or newly enlarging T2 lesions in the Q2W group was about one-third of that in the placebo group, adjusted for baseline T2 lesion count. MRI lesions are measured objectively and occur more often than clinical relapses, so this endpoint typically gives tighter intervals, which is visible in the narrow 0.27–0.40 range.
Why do the two disability hazard ratios have identical intervals but different p-values?
Both comparisons are reported as HR 0.62 (95% CI 0.40–0.97), with p-values of 0.0383 and 0.0380. Estimates and interval bounds are rounded to two decimals, while the underlying unrounded values and standard errors differ slightly. Tiny differences like this are expected and carry no interpretive weight.
Does a smaller p-value for Q2W prove it is better than Q4W?
No. Each p-value tests a schedule against placebo, not against the other schedule. Showing that two doses differ requires a direct comparison or an interaction-type test. The non-overlapping lesion-ratio intervals are suggestive for MRI activity, but for the primary ARR endpoint the intervals overlap considerably.
11. Limitations
- One-year placebo control: the randomized comparison with placebo covers only year 1; year-2 data lack a concurrent control.
- Relative measures only: the posted analyses give rate ratios, lesion mean ratios and hazard ratios; per-arm relapse rates and 1-year proportions are needed to judge absolute benefit.
- Multiplicity: two active schedules and several secondary endpoints were compared against a shared placebo group, and the registry does not describe the error-control strategy.
- Handling of treatment switching: exclusion of post-switch data and censoring at switch assume switching is not informative about the outcome; that assumption cannot be verified from the data.
- Single imputation for MRI outcomes: carry-forward-type and group-mean imputation can understate uncertainty and may bias estimates if missingness is related to disease activity.
- Imprecise disability estimates: the sustained progression intervals reach 0.97, so the magnitude of any effect on disability over one year is poorly determined.
- Proportional hazards: the Cox hazard ratios assume a constant relative effect over the year; the registry does not report checks of this assumption.
- Safety counts: serious adverse events are reported as counts of affected participants without statistical comparison or event-level detail.
12. Why This Trial Matters Statistically
ADVANCE is a clear teaching example of how the type of outcome determines the statistical model, and how relative effect measures from different models relate to one another.
| Concept | How it appears in ADVANCE |
|---|---|
| Recurrent-event count | Annualized relapse rate analysed as relapses per person-year |
| Negative binomial regression | Primary ARR analysis and MRI lesion analysis, allowing for overdispersion |
| Rate ratio | 0.644 (Q2W) and 0.725 (Q4W) versus placebo |
| Covariate adjustment | Baseline EDSS, age, prior relapse rate; baseline T2 lesions for MRI |
| Cox model and hazard ratio | Time to first relapse and to sustained disability progression |
| Censoring | At switch to alternative MS medication or withdrawal |
| Missing data | Single imputation for lesion counts |
| Multiple active arms | Two dosing schedules compared against one placebo group |
| Confidence intervals | Precise for MRI outcomes, wide for disability progression |
Statistical interpretation
Both schedules produced primary-endpoint rate ratios below 1 with 95% intervals excluding 1. Secondary endpoints pointed in the same direction, with the most precise estimates for MRI lesions and the least precise for disability progression.
Clinical interpretation
The relative reductions in relapse rate need to be combined with the absolute relapse rate, MRI findings, disability outcomes and the full safety profile before judging the value of either schedule.
13. Related Tutorials
Learn more about the methods used in this trial:
14. Related Calculators
15. Sources
- ClinicalTrials.gov: NCT00906399 — Efficacy and Safety Study of Peginterferon Beta-1a in Participants With Relapsing Multiple Sclerosis.
- PubMed: PMID 24794721.
- PubMed: PMID 37652990.
- PubMed: PMID 33307998.
- PubMed: PMID 29104758.
- PubMed: PMID 28607743.
Continue exploring Clinical Biostats
Follow the methods used in ADVANCE into detailed tutorials, then try them with the statistical calculators.
16. Record Summary
ADVANCE compared two dosing schedules of peginterferon beta-1a with placebo over one year in relapsing multiple sclerosis. The primary analysis, a covariate-adjusted negative binomial regression of confirmed relapses, gave rate ratios of 0.644 (95% CI 0.500–0.831) for Q2W and 0.725 (95% CI 0.565–0.930) for Q4W. Secondary analyses of MRI lesions, time to first relapse and time to sustained disability progression pointed in the same direction, with precision varying considerably by endpoint. Reading the trial well means combining relative effects, interval width, the handling of treatment switching and missing data, and the multiplicity created by two active arms and several endpoints.