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
ASCLEPIOS II was a randomized, parallel-group, quadruple-masked phase 3 trial asking whether ofatumumab, given by subcutaneous injection, reduced the rate of confirmed relapses compared with the active oral comparator teriflunomide in patients with relapsing multiple sclerosis.
| Feature | ASCLEPIOS II |
|---|---|
| Phase | Phase 3 |
| Condition | Relapsing multiple sclerosis |
| Design | Randomized, parallel-group, active-controlled, quadruple-masked (double-dummy) |
| Arms | Ofatumumab 20 mg subcutaneous vs teriflunomide 14 mg oral |
| Enrollment | 955 |
| Primary endpoint | Annualized relapse rate (ARR), baseline up to 2.5 years |
| Primary analysis | Negative binomial regression with log link and log time-in-study offset; rate ratio |
| Status | Completed; results posted |
| Study code | COMB157G2302 (companion study for pooled analyses: COMB157G2301) |
| ClinicalTrials.gov | NCT02792231 |
| Sponsor | Novartis Pharmaceuticals (industry) |
2. Clinical Question
The central question was whether ofatumumab 20 mg could lower the annualized rate of confirmed relapses relative to teriflunomide 14 mg in relapsing multiple sclerosis. Because the comparator is an established active therapy rather than placebo, the trial estimates the incremental effect of switching mechanism of action, not the effect of treatment versus no treatment.
Population
Participants with relapsing multiple sclerosis enrolled in a phase 3 treatment trial.
Intervention
Ofatumumab 20 mg by subcutaneous injection, with teriflunomide-matching placebo capsules.
Comparator
Teriflunomide 14 mg capsules, with matching placebo ofatumumab subcutaneous injections.
Primary question
Is the annualized rate of confirmed relapses lower with ofatumumab than with teriflunomide over up to 2.5 years? (Superiority hypothesis.)
3. Trial Design
Ofatumumab arm
- Ofatumumab 20 mg subcutaneous injection
- Teriflunomide-matching placebo capsules
Teriflunomide arm
- Teriflunomide 14 mg capsules
- Matching placebo of ofatumumab subcutaneous injections
4. Endpoints
Primary endpoint
| Endpoint | Registry definition | Time frame |
|---|---|---|
| Annualized Relapse Rate (ARR) | The number of confirmed relapses in a year, calculated as the total number of relapses for all participants in the treatment group divided by the total participant-years of time in study. A confirmed MS relapse was one accompanied by a clinically relevant change in the EDSS performed by the Independent EDSS rater: an increase of at least 0.5 points on the EDSS score, or an increase of 1 point on two functional scores or 2 points on one functional score (excluding changes involving bowel/bladder or cerebral functional system), compared with the last EDSS rating that did not occur during a relapse. | Baseline up to 2.5 years |
Secondary endpoints with posted analyses
| Endpoint | Time frame | Analysis |
|---|---|---|
| 3-month Confirmed Disability Worsening (3mCDW) based on EDSS — pooled and study-level | Baseline, every 3 months up to 2.5 years | Cox regression; hazard ratio |
| 6-month Confirmed Disability Worsening (6mCDW) based on EDSS — pooled and study-level | Baseline, every 3 months up to 2.5 years | Cox regression; hazard ratio |
| 6-month Confirmed Disability Improvement (6mCDI) based on EDSS — pooled and study-level | Baseline, every 3 months up to 2.5 years | Cox regression; hazard ratio |
| 3mCDW and 6mCDW based on EDSS > 8 weeks after onset of treatment — pooled | Baseline up to 2.5 years | Cox regression; hazard ratio |
| Number of gadolinium-enhancing T1 lesions per MRI scan | Baseline, yearly up to 2.5 years | Negative binomial regression; rate ratio |
| Number of new or enlarging T2 lesions on MRI per year (annualized lesion rate) | Baseline, yearly up to 2.5 years | Negative binomial regression; rate ratio |
| Neurofilament light chain (NfL) concentration in serum | Month 3, 12 and 24 | Mixed model for repeated measures; geometric mean ratio |
| Annualized rate of brain volume loss (percent brain volume change from baseline) | Baseline, Months 12 and 24 | Random coefficient model; mean difference |
5. Statistical Methodology
Negative binomial regression for relapses
Relapses are counts: a participant may have none, one or several during follow-up, and participants are followed for different lengths of time. The registry describes the primary model as a negative binomial regression with log link fitted to the number of relapses, with the natural log of time-in-study as an offset to annualize the rate. The model was adjusted for treatment and region as factors, and for number of relapses in the previous year, baseline EDSS, baseline number of Gd-enhancing lesions and age at baseline as covariates.
The offset log(time) has a fixed coefficient of 1, so the model describes relapses per unit of time. The treatment rate ratio is exp(βtrt). The negative binomial distribution adds a dispersion parameter so that the variance can exceed the mean, which is typical when some patients relapse repeatedly and many do not relapse at all.
Cox proportional-hazards models for disability
Confirmed disability worsening and improvement are time-to-first-event outcomes assessed with EDSS every 3 months. The registry lists Cox regression, with the hazard ratio as the effect measure. Participants without a confirmed event are censored at their last assessment.
Mixed model for repeated measures for NfL
Serum NfL was measured at Months 3, 12 and 24 and analysed with a mixed model for repeated measures (MMRM). The effect is reported as a geometric mean ratio, which indicates that the analysis was carried out on the log scale and back-transformed. MMRM uses all available visits for each participant and, under a missing-at-random assumption, does not require complete data at every visit.
Random coefficient model for brain volume
Brain volume loss was analysed with a random coefficient model, in which each participant has their own intercept and slope over time. The treatment effect is expressed as a difference in the annualized rate (slope) of percent brain volume change.
Analysis population
Efficacy analyses used the full analysis set; pooled disability analyses used the full analysis set from the combined studies. A full analysis set analyses participants according to randomized assignment and is the usual implementation of the intention-to-treat principle in regulatory trials.
Pre-specified pooling with a companion study
The registry states that this study was not powered for some secondary endpoints as a stand-alone study, and that it was pre-specified in the protocol to combine its data with study COMB157G2301 to address those endpoints. The disability endpoints are therefore reported twice: as pooled estimates across both studies and as estimates from this study alone.
6. Primary Result: Annualized Relapse Rate
ARR rate ratio, ofatumumab 20 mg vs teriflunomide 14 mg
95% CI: 0.309–0.560 (two-sided) · P < 0.001
Negative binomial regression · Full analysis set · Superiority hypothesis
| Measure | Value |
|---|---|
| Effect measure | Rate ratio (OMB 20 mg / TER 14 mg) |
| Estimate | 0.416 |
| 95% confidence interval | 0.309 to 0.560 |
| P-value | <0.001 |
| Unit of the endpoint | Number of relapses in a year |
What the estimate means. A rate ratio of 0.416 means that, after adjustment for region, prior-year relapses, baseline EDSS, baseline Gd-enhancing lesions and age, the model-estimated relapse rate with ofatumumab was 41.6% of the rate with teriflunomide, a 58.4% lower estimated relapse rate.
What it does not mean. It does not mean that 58.4% of patients were spared a relapse, nor that each individual's relapse risk fell by that amount. A rate ratio compares event rates per unit of exposure time at the group level; it says nothing directly about the proportion of relapse-free participants, and it is a relative measure, so the absolute number of relapses avoided depends on the underlying rate in the comparator group.
Precision. The 95% confidence interval of 0.309 to 0.560 lies entirely below 1. Even its upper limit corresponds to a 44.0% lower estimated rate, so the data are inconsistent with a small or null difference under the model. The interval is fairly wide on the ratio scale, which is typical of count outcomes where many participants have zero events.
Why the p-value is not the effect size. P < 0.001 states that data this extreme would be very unlikely if the true rate ratio were 1. It does not measure how large the benefit is; the estimate and its interval carry that information.
Cautions. The estimate depends on the negative binomial variance assumption and on the offset correctly representing exposure. Participants who leave the study early contribute less time, and if early discontinuation is related to disease activity, the annualized rate can be biased. The comparator is active, so the effect is relative to teriflunomide, not to untreated disease.
7. Secondary Results: Confirmed Disability
Disability endpoints were pre-specified for pooled analysis. The table places the pooled estimate and the single-study estimate side by side for each endpoint.
| Endpoint | Data | Hazard ratio | 95% CI | P-value |
|---|---|---|---|---|
| 3mCDW | Pooled | 0.657 | 0.500–0.863 | 0.003 |
| 3mCDW | Study COMB157G2302 | 0.662 | 0.449–0.977 | 0.038 |
| 6mCDW | Pooled | 0.676 | 0.498–0.917 | 0.012 |
| 6mCDW | Study COMB157G2302 | 0.759 | 0.490–1.174 | 0.215 |
| 6mCDI | Pooled | 1.355 | 0.952–1.928 | 0.092 |
| 6mCDI | Study COMB157G2302 | 1.523 | 0.936–2.477 | 0.090 |
| 3mCDW > 8 weeks after treatment onset | Pooled | 0.641 | 0.486–0.847 | 0.002 |
| 6mCDW > 8 weeks after treatment onset | Pooled | 0.657 | 0.481–0.898 | 0.008 |
All pooled worsening estimates lie between 0.641 and 0.676, with confidence intervals entirely below 1. The single-study 3mCDW estimate (0.662) is almost identical to the pooled value (0.657) but with a wider interval, 0.449–0.977, whose upper limit sits close to 1. The single-study 6mCDW estimate (0.759) has an interval of 0.490–1.174 that includes 1. For improvement (6mCDI), a hazard ratio above 1 would favour ofatumumab, and both point estimates are above 1, but both intervals include 1.
What the estimate means. The pooled 3mCDW hazard ratio of 0.657 means that, under the Cox model and over the analysed follow-up, the estimated instantaneous rate of reaching a 3-month confirmed worsening in EDSS was 65.7% of that with teriflunomide, a 34.3% lower estimated hazard.
What it does not mean. It is not the proportion of patients who avoided disability worsening, and it is not a difference in cumulative incidence at 2.5 years. Hazard ratios summarise relative event rates over time; absolute proportions of participants with an event would be needed to judge absolute benefit.
Precision and the value of pooling. Comparing the pooled and single-study rows shows exactly why the protocol pre-specified pooling. Point estimates are similar, but the single-study intervals are wider because fewer disability events accrue in one trial. A single-study 6mCDW interval that crosses 1 (0.490–1.174) is an expression of limited information, not evidence that the effect is absent.
P-values. P = 0.038 for single-study 3mCDW and P = 0.215 for single-study 6mCDW do not mean that ofatumumab affects 3-month but not 6-month worsening. The two point estimates are close to each other; the difference in p-values mostly reflects differing numbers of events and interval width.
Cautions. A single hazard ratio assumes approximately proportional hazards across follow-up. Several disability endpoints were tested, so nominal p-values among secondary endpoints should be read with multiplicity in mind. The pooled analyses combine two trials, so their population is the full analysis set of both studies rather than this trial alone. The "> 8 weeks after onset of treatment" variants, whose estimates (0.641 and 0.657) are close to the main pooled values, act as a sensitivity check that early events, before the drug could plausibly act, are not driving the result.
8. Secondary Results: MRI Lesions
| Endpoint | Time point | Rate ratio | 95% CI | P-value |
|---|---|---|---|---|
| Gd-enhancing T1 lesions per scan | Baseline, yearly up to 2.5 years | 0.061 | 0.037–0.101 | <.001 |
| New or enlarging T2 lesions per year | Month 12 | 0.21 | 0.17–0.27 | <.001 |
| New or enlarging T2 lesions per year | Month 24 | 0.19 | 0.12–0.31 | <.001 |
| New or enlarging T2 lesions per year | End of Study | 0.15 | 0.13–0.19 | <.001 |
Both lesion outcomes are counts and were analysed with negative binomial regression, the same model family as the primary endpoint. A rate ratio of 0.061 for Gd-enhancing lesions corresponds to a 93.9% lower estimated number of enhancing lesions per scan. For new or enlarging T2 lesions, the estimated annualized lesion rate with ofatumumab was 21%, 19% and 15% of the teriflunomide rate at Month 12, Month 24 and End of Study, respectively.
9. Secondary Results: Serum NfL and Brain Volume
| Endpoint | Time point | Effect measure | Estimate | 95% CI | P-value |
|---|---|---|---|---|---|
| Serum NfL (pg/mL) | Month 3 | Geometric mean ratio | 0.89 | 0.85–0.93 | <.001 |
| Serum NfL (pg/mL) | Month 12 | Geometric mean ratio | 0.74 | 0.70–0.79 | <.001 |
| Serum NfL (pg/mL) | Month 24 | Geometric mean ratio | 0.76 | 0.71–0.81 | <.001 |
| Annualized brain volume loss (%) | Months 12 and 24 | Mean difference (net) | 0.07 | -0.02 to 0.15 | 0.128 |
Geometric mean NfL concentrations with ofatumumab were an estimated 11% lower at Month 3, 26% lower at Month 12 and 24% lower at Month 24 than with teriflunomide. The pattern of a smaller ratio at Month 3 and a larger, stable separation from Month 12 is consistent with a biomarker that responds over months rather than immediately.
For brain volume, the estimated net mean difference in annualized loss was 0.07 percentage points with a 95% confidence interval from -0.02 to 0.15 and P = 0.128. The interval includes zero, so this study does not distinguish the two treatments on this measure; it also does not demonstrate that they are equivalent, because equivalence would require a pre-specified margin that is not part of this analysis.
10. Safety: Serious Adverse Events
| Arm | Participants with a serious adverse event | Participants at risk |
|---|---|---|
| Ofatumumab 20 mg | 42 | 481 |
| Teriflunomide 14 mg | 37 | 474 |
Serious adverse events were reported in 42 of 481 participants in the ofatumumab arm and 37 of 474 in the teriflunomide arm. The registry does not report a formal statistical comparison of these counts. Safety counts of this kind are descriptive: the primary endpoint was relapse rate, and trials of this kind are not designed to detect differences in uncommon adverse events, and serious adverse events group together heterogeneous clinical events. Exposure time may also differ between participants, which crude counts do not take into account.
11. Multiplicity and Interpretation of Secondary Endpoints
ASCLEPIOS II reports one primary endpoint and a large set of secondary analyses: several disability definitions, each in pooled and single-study form, two MRI lesion outcomes at multiple time points, a biomarker at three visits and a brain-volume slope. The ClinicalTrials.gov record does not describe a testing hierarchy or multiplicity adjustment for these secondary endpoints.
| Analysis | Role | Interpretation |
|---|---|---|
| ARR | Primary endpoint | Confirmatory superiority comparison in this study |
| 3mCDW / 6mCDW / 6mCDI, pooled | Secondary, pre-specified pooling | Designed to answer the disability question with adequate information across two trials |
| 3mCDW / 6mCDW / 6mCDI, single study | Secondary, not individually powered | Supportive; wider intervals expected |
| MRI lesions, NfL, brain volume | Secondary | Mechanistic and imaging support; multiple time points increase the number of comparisons |
12. Statistical Methods Explained
Why was a negative binomial model used for relapses instead of Poisson regression?
A Poisson model assumes the variance of the count equals its mean. In multiple sclerosis, relapses cluster: many patients have none and some have several. That extra-Poisson variation (overdispersion) makes Poisson standard errors too small and confidence intervals too narrow. The negative binomial model adds a dispersion parameter, giving more honest uncertainty for the rate ratio of 0.416.
What does the log time-in-study offset do?
Participants were followed for up to 2.5 years, but not all for the same length of time. Including log(time) as an offset converts the model from "relapses per participant" to "relapses per participant-year", matching the registry's definition of ARR as total relapses divided by total participant-years. Without it, participants with longer follow-up would appear to relapse more simply because they were observed longer.
Why adjust for baseline covariates in a randomized trial?
Randomization makes the arms comparable on average, so adjustment is not needed to remove bias. Adjusting for strong predictors of relapse, such as relapses in the previous year, baseline EDSS, baseline Gd-enhancing lesions and age, explains part of the between-patient variation and usually yields a more precise estimate of the treatment effect. Adjustment for region reflects the multicentre structure.
Why were the disability endpoints pooled with another trial?
Confirmed disability worsening is a comparatively infrequent event over 2.5 years, so a single trial sized for relapses has limited power for it. The protocol pre-specified combining data with study COMB157G2301. Because the pooling was planned in advance rather than chosen after seeing results, it does not carry the usual risks of post hoc meta-analysis, although it still assumes that the two studies estimate a common treatment effect.
Why is a hazard ratio above 1 favourable for disability improvement?
For worsening, the event is harmful, so a hazard ratio below 1 favours ofatumumab. For confirmed improvement, the event is beneficial, so a higher hazard with ofatumumab (1.355 pooled, 1.523 single-study) would be favourable. In this trial both intervals include 1, so the direction is suggestive rather than established.
What does a geometric mean ratio of 0.74 for NfL mean?
NfL concentrations are right-skewed, so they are analysed on the log scale. A difference in log means, back-transformed, becomes a ratio of geometric means. A value of 0.74 at Month 12 means the typical (geometric mean) NfL concentration with ofatumumab was an estimated 74% of that with teriflunomide, or 26% lower. It is not a statement about arithmetic means or about every participant.
13. Limitations
- Stand-alone power: the registry states that this study was not powered for some secondary endpoints as a stand-alone study; single-study disability results should be read with that in mind.
- Pooled populations: pooled estimates combine this trial with COMB157G2301, so they describe the two-trial population rather than ASCLEPIOS II alone.
- Active comparator: all effects are relative to teriflunomide 14 mg. They cannot be translated into effects versus placebo or versus other therapies without additional assumptions.
- Variable follow-up: follow-up extended up to 2.5 years, and discontinuation related to disease activity or tolerability could affect annualized rates and censoring in time-to-event models.
- Proportional hazards: single hazard ratios for disability summarise the whole follow-up period and are less descriptive if the treatment effect changes over time.
- Multiplicity: numerous secondary analyses were reported with nominal p-values; the registry does not describe an adjustment strategy for them.
- Missing information in the registry: the ClinicalTrials.gov record does not report subgroup analyses or results beyond 2.5 years for this study, so heterogeneity of effect and durability beyond 2.5 years cannot be assessed from these results.
- Safety: serious adverse event counts are descriptive and not accompanied by a formal comparison.
14. Why This Trial Matters Statistically
ASCLEPIOS II is a clear teaching example of how a single randomized trial uses different models for different data types, and of how a programme of two trials can be designed so that endpoints too rare for one study are answered jointly.
| Concept | How it appears in ASCLEPIOS II |
|---|---|
| Count outcomes | Relapses and MRI lesions analysed with negative binomial regression |
| Offsets and exposure | Log time-in-study used to annualize relapse rates |
| Rate ratio | Primary effect measure (0.416) and MRI lesion effect measures |
| Covariate adjustment | Prior relapses, baseline EDSS, Gd lesions, age and region in the primary model |
| Time-to-event analysis | Confirmed disability worsening and improvement analysed with Cox regression |
| Hazard ratio | Relative rates of confirmed disability events |
| Pre-specified pooling | Disability endpoints combined with COMB157G2301 to gain information |
| Repeated measures | NfL at Months 3, 12 and 24 analysed with MMRM on the log scale |
| Random coefficient model | Participant-specific slopes for brain volume change |
| Double-dummy masking | Different routes of administration kept masked in both arms |
| Confidence intervals vs p-values | Similar pooled and single-study estimates with very different p-values |
15. Related Tutorials
Learn more about the methods used in this trial:
16. Related Calculators
17. Sources
- ClinicalTrials.gov: NCT02792231 — Efficacy and Safety of Ofatumumab Compared to Teriflunomide in Patients With Relapsing Multiple Sclerosis.
- PubMed: PMID 32757523.
- PubMed: PMID 35266417.
- PubMed: PMID 37606897.
- PubMed: PMID 39018512.
Continue exploring clinical trial statistics
Each trial page connects its endpoints and methods to deeper statistical tutorials and to calculators for working through similar analyses.
18. Record Summary
ASCLEPIOS II compared ofatumumab 20 mg with teriflunomide 14 mg in relapsing multiple sclerosis using a double-dummy, quadruple-masked design. Its primary result, an annualized relapse rate ratio of 0.416 (95% CI 0.309–0.560), comes from a covariate-adjusted negative binomial model with an exposure offset. Disability outcomes were answered mainly through pre-specified pooling with a companion trial, and MRI, NfL and brain-volume outcomes each used a model suited to their data type. The most useful reading combines relative effects, confidence intervals, the model behind each estimate and the design features, such as pooling and multiplicity, that shape how each result should be weighed.