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
RA-MOBILITY was a randomized, parallel, triple-masked phase 2/3 trial evaluating sarilumab on top of methotrexate in rheumatoid arthritis. The registry reports 1,675 enrolled participants, 10 arms, four registered primary endpoints, five posted outcome measures, and 13 posted statistical analyses.
| Feature | RA-MOBILITY |
|---|---|
| Trial name | RA-MOBILITY |
| Brief title | Evaluation of Sarilumab (SAR153191/REGN88) on Top of Methotrexate in Rheumatoid Arthritis Patients |
| Phase | Phase 2/3 |
| Status | Completed |
| Therapeutic area | Immunology & Rheumatology |
| Condition | Rheumatoid Arthritis |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | Triple |
| Primary purpose | Treatment |
| Enrollment | 1,675 |
| Lead sponsor | Sanofi |
| Study period | March 2010 to October 2013 for start and primary completion |
2. Clinical Question
The trial evaluated sarilumab added to methotrexate in participants with rheumatoid arthritis. The registered primary questions focused on whether different sarilumab dosing regimens produced higher ACR20 response rates than placebo, and whether selected sarilumab regimens improved physical function and limited structural damage relative to placebo.
Population
Participants with rheumatoid arthritis enrolled in the randomized RA-MOBILITY study and evaluated in Part A or the dose-selected Part B population.
Intervention
Sarilumab on top of methotrexate, with sarilumab regimens including 100 mg or 150 mg weekly in Part A, and 100 mg, 150 mg, or 200 mg every 2 weeks in the Part A dosing structure.
Comparator
Placebo for sarilumab, including weekly placebo in Part A and every-2-week placebo in Part B.
Primary question
Do sarilumab regimens improve clinical response, physical function, and structural outcomes compared with placebo when evaluated on top of methotrexate?
3. Trial Design
The registry describes RA-MOBILITY as randomized, parallel, triple-masked, and treatment-focused. The intervention list includes sarilumab, placebo for sarilumab, methotrexate, and folic acid. The statistical structure is notable because the trial used different methods for binary response endpoints, longitudinal physical-function measurements, and the structural-damage endpoint.
Dose-ranging comparisons
- SAR 100 mg weekly
- SAR 150 mg weekly
- SAR 100 mg every 2 weeks
- SAR 150 mg every 2 weeks
- SAR 200 mg every 2 weeks
- Placebo weekly
Dose-selected comparisons
- SAR 100 mg weekly, Cohort 1, non-selected
- SAR 150 mg weekly, Cohort 1, non-selected
- SAR 100 mg every 2 weeks, Cohort 1, non-selected
- SAR 150 mg every 2 weeks, selected
- SAR 200 mg every 2 weeks, selected
- Placebo every 2 weeks, selected
- Sarilumab rescue, selected cohort
The registry reports 10 total arms. The serious-adverse-event data identify the specific Part A and Part B cohorts represented in the safety reporting.
4. Endpoints
| Registered primary endpoint | Time frame | Endpoint type |
|---|---|---|
| Part A: Percentage of Participants Achieving American College of Rheumatology 20% (ACR20) Response at Week 12 | Baseline to Week 12 | Binary |
| Part B: Percentage of Participants Achieving ACR20 Response at Week 24 | Baseline to Week 24 | Binary |
| Part B: Change From Baseline in Health Assessment Question Disability Index (HAQ-DI) at Week 16 | Baseline, Week 16 | Continuous |
| Part B: Change From Baseline in Van Der Heijde Modified Total Sharp Score (mTSS) at Week 52 | Baseline, Week 52 | Continuous |
ACR20 at Week 12
ACR20 response was defined according to American College of Rheumatology guidelines as at least 20% improvement in tender joint count and swollen joint count together with at least 20% improvement in at least three of five measures: C-reactive protein, participant assessment of pain, participant global assessment of disease activity, physician global assessment of disease activity, and HAQ-DI. The Part A definition specifies last observation carried forward for missing data.
ACR20 at Week 24
For Part B, ACR20 responses were determined without imputation of missing post-baseline values. Data collected after treatment discontinuation or rescue were set to missing. Participants became non-responders for time points beyond the point at which they started rescue treatment or discontinued study treatment.
HAQ-DI at Week 16
HAQ-DI assessed difficulty with dress/groom, arising, eating, walking, reaching, gripping, hygiene, and common activities. The overall score ranges from 0, representing least difficulty, to 3, representing extreme difficulty. An increase indicates worsening physical function, while a decrease represents improvement. Data collected after treatment discontinuation were set to missing.
mTSS at Week 52
The Van Der Heijde modified Total Sharp Score evaluates structural damage using radiographic scores for erosions and joint-space narrowing. The total score ranges from 0, normal, to 448, the worst possible total score. An increase represents progression of structural damage. Missing data were imputed using linear extrapolation.
5. Statistical Methodology
The registry identifies three main statistical methods: the Cochran-Mantel-Haenszel test, mixed-effects modeling, and ANCOVA. These methods correspond to the different statistical structures of the registered endpoints.
| Endpoint / analysis | Statistical method | Effect measure | Analysis population |
|---|---|---|---|
| Part A ACR20 Week 12 | Cochran-Mantel-Haenszel test | Odds ratio | Part A ITT; all randomized participants |
| Part B ACR20 Week 24 | Cochran-Mantel-Haenszel test | Odds ratio | ITT population randomized after dose selection, cohort 2 |
| HAQ-DI Week 16 | Mixed-effects model / MMRM | LS mean difference | Cohort 2 ITT participants with available baseline and Week 16-or-earlier HAQ-DI |
| mTSS Week 52 | Rank ANCOVA | Not reported in the statistical-analysis record | Cohort 2 ITT participants with available baseline and Week 52-or-earlier mTSS |
Multiplicity control
Multiplicity was an explicit component of the registry-reported analysis. In Part A, pairwise response-rate comparisons between each sarilumab dose and placebo were performed using a two-sided Cochran-Mantel-Haenszel test, with multiplicity addressed using the Hommel procedure.
For Part B, the registry describes a Bonferroni correction for each dose together with a hierarchical testing procedure across the three co-primary and main secondary endpoints. The hierarchy continued only when the preceding endpoint was statistically significant at the 0.025 level. The HAQ-DI and mTSS analyses also identify hierarchical testing in their analysis notes.
Intention-to-treat analysis
The Part A ACR20 analyses used an ITT population defined as all randomized participants. The Part B ACR20 analyses used the ITT population of participants randomized after dose selection. This distinction matters because randomization establishes the comparison being tested, while the Part B population reflects the later dose-selected stage of the trial.
Missing-data handling
The endpoint definitions use different approaches to missing observations. Part A ACR20 used LOCF. Part B ACR20 did not impute missing post-baseline values and treated observations after discontinuation or rescue as missing, with participants becoming non-responders after those events. HAQ-DI observations after treatment discontinuation were set to missing. mTSS used linear extrapolation for missing data.
The statistical estimate is not independent of the analysis population and missing-data rules. Different endpoint definitions can therefore produce different inferential properties even when they are evaluated within the same randomized trial.
6. Part A ACR20 Results at Week 12
The Part A primary endpoint was the percentage of participants achieving an ACR20 response from baseline to Week 12. Five sarilumab-versus-placebo comparisons were posted in the primary statistical analyses.
| Comparison | Odds ratio | 95% CI | P-value |
|---|---|---|---|
| SAR 100 mg qw vs placebo qw | 1.99 | 0.85–4.64 | 0.1155 |
| SAR 150 mg qw vs placebo qw | 3.84 | 1.53–9.63 | 0.0041 |
| SAR 100 mg q2w vs placebo qw | 1.17 | 0.52–2.61 | 0.7119 |
| SAR 150 mg q2w vs placebo qw | 2.38 | 1.06–5.35 | 0.0363 |
| SAR 200 mg q2w vs placebo qw | 2.34 | 1.03–5.29 | 0.0426 |
All five comparisons used the Part A ITT population and the two-sided Cochran-Mantel-Haenszel test. The registry specifically states that pairwise comparisons were derived between each sarilumab dose and placebo and that multiplicity was addressed using the Hommel procedure.
The odds ratio compares the odds of achieving an ACR20 response between each sarilumab regimen and placebo. For example, an odds ratio of 3.84 for SAR 150 mg qw means that the estimated odds of ACR20 response were 3.84 times the corresponding odds under placebo in this analysis.
An odds ratio is not the same as a risk ratio or a percentage-point difference in response rates. Without the underlying response proportions, the odds ratio alone cannot be translated into an absolute increase in the probability of response.
The 95% confidence interval of 1.53–9.63 describes statistical uncertainty around the odds-ratio estimate. Its width indicates that the estimate is not known with arbitrary precision. The p-value of 0.0041 addresses evidence against the null hypothesis under the specified test; it does not measure the size or clinical importance of the treatment effect.
Because several dose comparisons were made, the Hommel procedure is important to interpretation. Individual nominal p-values should not be considered in isolation from the prespecified multiplicity adjustment.
Reading the Part A estimates together
The estimates vary across dosing comparisons. The 95% confidence intervals for SAR 150 mg qw, SAR 150 mg q2w, and SAR 200 mg q2w exclude 1.00, while the intervals for SAR 100 mg qw and SAR 100 mg q2w include 1.00. The registry's multiplicity procedure remains part of the formal interpretation of these multiple comparisons.
7. Part B ACR20 Results at Week 24
The Part B primary ACR20 endpoint was assessed from baseline to Week 24. The posted primary analyses compare the two selected sarilumab regimens with placebo every 2 weeks.
SAR 150 mg q2w vs placebo q2w
95% CI: 2.077–3.703 · P < 0.0001
SAR 200 mg q2w vs placebo q2w
95% CI: 2.957–5.344 · P < 0.0001
Both analyses used the Part B ITT population, defined as participants randomized after dose selection in cohort 2. The statistical method was a two-sided Cochran-Mantel-Haenszel test.
An odds ratio of 2.773 means that the estimated odds of ACR20 response were 2.773 times the odds under placebo for the SAR 150 mg q2w comparison. The corresponding estimate of 3.975 means that the estimated odds were 3.975 times those under placebo for SAR 200 mg q2w.
The confidence intervals are relatively concentrated around their respective estimates compared with the wider Part A intervals. For SAR 150 mg q2w, the 95% CI is 2.077–3.703; for SAR 200 mg q2w it is 2.957–5.344. These intervals quantify uncertainty in the odds-ratio estimates rather than variability among individual participants.
The p-values of <0.0001 provide evidence against the relevant null hypotheses under the specified analyses. They do not indicate that the probability of response is "greater than 99.99%", nor do they quantify clinical magnitude.
The Part B multiplicity strategy is especially important: the registry describes Bonferroni correction for each dose together with hierarchical testing across three co-primary and main secondary endpoints. The hierarchy continued only when the previous endpoint was statistically significant at the 0.025 level.
8. HAQ-DI Results at Week 16
The third registered primary endpoint evaluated change from baseline in HAQ-DI at Week 16. The registry used a mixed model for repeated measures, reported as a mixed-effects model, and expressed treatment differences as least-squares mean differences.
| Comparison | LS mean difference | 95% CI | P-value |
|---|---|---|---|
| SAR 150 mg q2w vs placebo q2w | -0.235 | -0.312 to -0.157 | <0.0001 |
| SAR 200 mg q2w vs placebo q2w | -0.258 | -0.336 to -0.181 | <0.0001 |
The analysis population was the cohort 2 ITT population restricted to participants with available HAQ-DI data at baseline and on or before Week 16. The registry states that testing followed the hierarchical procedure and was performed only when the previous endpoint was statistically significant.
Because lower HAQ-DI values represent less difficulty with daily activities, a negative least-squares mean difference indicates a lower estimated HAQ-DI value in the sarilumab group than in the placebo group. The estimate of -0.235 therefore represents an adjusted difference of 0.235 HAQ-DI units in favor of the sarilumab comparison under the model. The corresponding SAR 200 mg q2w estimate was -0.258.
The mixed-effects model uses information from the repeated measurements represented in the longitudinal analysis rather than reducing the entire trajectory to a single unadjusted comparison. The registry specifically identifies the method as MMRM.
The 95% CI of -0.312 to -0.157 quantifies uncertainty around the estimated LS mean difference for SAR 150 mg q2w. The p-value of <0.0001 indicates strong statistical evidence against the null hypothesis specified for the comparison, but it does not describe the magnitude of improvement or establish that a particular individual experienced the estimated average change.
Interpretation also depends on the analysis population and the handling of observations after treatment discontinuation, which were set to missing.
9. mTSS Results at Week 52
The fourth registered primary endpoint was change from baseline in Van Der Heijde modified Total Sharp Score at Week 52. The registry identifies a two-sided rank-based ANCOVA as the statistical method.
| Comparison | Method | P-value |
|---|---|---|
| SAR 150 mg q2w vs placebo q2w | Rank ANCOVA | <0.0001 |
| SAR 200 mg q2w vs placebo q2w | Rank ANCOVA | <0.0001 |
The analysis population was the cohort 2 ITT population with available mTSS data at baseline and on or before Week 52. The registry states that missing mTSS data were imputed using linear extrapolation and that the hierarchical testing procedure was used.
The mTSS endpoint measures structural damage on a scale from 0 to 448, with increasing scores representing progression. A rank-based ANCOVA was used rather than reporting a conventional mean difference in the posted statistical analysis.
The registry reports P < 0.0001 for both dose comparisons, but does not provide an effect estimate or confidence interval in these posted statistical-analysis records. Consequently, the p-values establish the reported inferential result without supplying a quantitative estimate of the magnitude of the between-group difference.
The distinction is important: a p-value answers an evidence-against-the-null question under the specified model and testing framework, whereas an effect estimate would describe the magnitude and direction of the treatment difference. Both are useful, but they answer different questions.
10. Secondary Endpoint Results
The registry posts two secondary statistical analyses, both evaluating the percentage of participants achieving a Major Clinical Response at Week 52.
| Comparison | Odds ratio | 95% CI | P-value |
|---|---|---|---|
| SAR 150 mg q2w vs placebo q2w | 4.661 | 2.451–8.863 | <0.0001 |
| SAR 200 mg q2w vs placebo q2w | 5.565 | 2.946–10.515 | <0.0001 |
Both analyses used the ITT population consisting of all participants randomized after dose selection in cohort 2 and the two-sided Cochran-Mantel-Haenszel test. The registry describes hierarchical testing for the SAR 150 mg q2w analysis.
The odds-ratio estimates of 4.661 and 5.565 describe substantially higher estimated odds of Major Clinical Response under the respective sarilumab regimens relative to placebo. They should not be read as 4.661-fold or 5.565-fold increases in the probability of response because odds and probabilities are different quantities.
The 95% confidence intervals exclude 1.00 for both comparisons. The intervals also show that the exact magnitude is uncertain: the SAR 150 mg q2w estimate is compatible with odds ratios from 2.451 to 8.863, while the SAR 200 mg q2w interval extends from 2.946 to 10.515.
The p-values are both <0.0001. Statistical significance should be considered within the trial's hierarchical testing structure rather than interpreted as a standalone measure of treatment magnitude.
11. Safety: Serious Adverse Events
The registry reports serious adverse events by arm as affected participants divided by participants at risk. These data span Part A and Part B cohorts and are presented below exactly as reported.
| Arm | Serious adverse events |
|---|---|
| Part A: SAR 100 mg qw | 3/50 |
| Part A: SAR 150 mg qw | 0/50 |
| Part A: SAR 100 mg q2w | 3/51 |
| Part A: SAR 150 mg q2w | 0/52 |
| Part A: SAR 200 mg q2w | 0/51 |
| Part A: Placebo qw | 2/51 |
| Part B: SAR 100 mg qw Cohort 1 (Non-selected) | 2/29 |
| Part B: SAR 150 mg qw Cohort 1 (Non-selected) | 2/26 |
| Part B: SAR 100 mg q2w Cohort 1 (Non-selected) | 0/29 |
| Part B: SAR 150 mg q2w Cohort 1 (Selected) | 35/370 |
| Part B: SAR 200 mg q2w Cohort 1 (Selected) | 46/369 |
| Part B: Placebo q2w Cohort 1 (Selected) | 21/259 |
| Part B: Sarilumab Rescue Cohort 1 (Selected) | 7/284 |
These counts should be interpreted as arm-specific affected participants over participants at risk. The denominators differ substantially between Part A and Part B, so direct comparison of raw counts across all rows would not be an appropriate safety comparison.
12. Statistical Methods Explained
Why was the Cochran-Mantel-Haenszel test used?
The ACR20 endpoints are binary outcomes: each participant is classified as a responder or non-responder. The Cochran-Mantel-Haenszel framework is designed for categorical treatment comparisons and can produce an adjusted comparison across relevant strata. In RA-MOBILITY, the registry explicitly identifies the two-sided Cochran-Mantel-Haenszel test for the ACR20 analyses.
What does an odds ratio of 3.975 mean?
An odds ratio of 3.975 means that the estimated odds of ACR20 response were 3.975 times the odds under placebo for the SAR 200 mg q2w versus placebo q2w comparison. It does not mean that 397.5% of participants responded or that the probability of response was 3.975 times higher. The conversion from odds to probability depends on the underlying baseline probability.
Because odds depend nonlinearly on probability, an odds ratio cannot be converted to an absolute risk difference without knowing the relevant baseline response probability.
Why was a mixed-effects model used for HAQ-DI?
HAQ-DI is a continuous measure observed longitudinally. A mixed-effects model for repeated measures is suited to settings where repeated observations from the same participant are statistically related. The RA-MOBILITY registry specifically identifies MMRM as the analysis method and reports treatment differences as least-squares mean differences.
What does a negative HAQ-DI difference mean?
HAQ-DI ranges from 0 to 3, with lower values indicating less difficulty. Therefore, a negative LS mean difference indicates a lower estimated HAQ-DI value in the sarilumab group relative to placebo. For SAR 150 mg q2w, the reported difference was -0.235; for SAR 200 mg q2w, it was -0.258.
Why was rank ANCOVA used for mTSS?
The registry reports a rank-based ANCOVA for the Week 52 mTSS endpoint. mTSS is a bounded structural-damage score, and a rank-based approach can analyze the ordering of observations without relying on the same distributional assumptions as a conventional mean-based analysis. The registry does not provide the underlying rank transformation details in the posted statistical-analysis record.
Why does multiplicity matter so much in this trial?
RA-MOBILITY evaluated multiple sarilumab doses and multiple primary endpoints. Testing many hypotheses independently can increase the chance of observing a statistically significant result by chance alone. The registry therefore reports explicit multiplicity procedures: Hommel adjustment for the Part A pairwise comparisons and Bonferroni correction plus hierarchical testing for Part B.
Why is the analysis population important?
The Part A ACR20 analysis used all randomized participants in the Part A ITT population. The Part B efficacy analyses instead used participants randomized after dose selection. The HAQ-DI and mTSS analyses additionally required relevant endpoint data. These distinctions affect which participants contribute to each estimate and therefore must be considered when comparing results across endpoints.
13. Multiplicity and Hierarchical Testing
Multiplicity is one of the central statistical features of RA-MOBILITY. The trial did not treat every reported p-value as though it arose from an isolated hypothesis test.
| Setting | Multiplicity approach reported | Statistical implication |
|---|---|---|
| Part A ACR20 | Hommel procedure | Addresses multiplicity across pairwise dose-versus-placebo comparisons. |
| Part B ACR20 | Bonferroni correction for each dose plus hierarchical testing | Controls the testing sequence across the co-primary and main secondary endpoints. |
| HAQ-DI | Hierarchical testing | Testing was performed only when the previous endpoint was statistically significant. |
| mTSS | Hierarchical testing | Testing was performed only when previous endpoints were statistically significant. |
The registry specifies a 0.025 significance level for continuation of the Part B hierarchical sequence. This means that the sequence itself is part of the inferential design rather than an administrative detail added after looking at results.
Multiplicity adjustment changes how evidence is judged across a family of hypotheses. A p-value that appears small in isolation does not automatically establish statistical significance for the corresponding endpoint when multiple comparisons are being considered. The relevant adjustment or hierarchy must be incorporated into the interpretation.
14. Missing Data and Estimand Considerations
RA-MOBILITY used different missing-data rules for different endpoints. This is statistically important because the definition of the analysis can influence the treatment effect estimate.
Part A ACR20
Missing data were imputed using Last Observation Carried Forward.
Part B ACR20
Missing post-baseline values were not imputed; data after discontinuation or rescue were set to missing, with participants treated as non-responders beyond those events.
HAQ-DI
Data collected after treatment discontinuation were set to missing.
mTSS
Missing data were imputed using linear extrapolation.
These approaches correspond to different endpoint structures and analysis assumptions. LOCF carries forward an earlier observation, whereas linear extrapolation uses an assumed trajectory to address missing structural-damage measurements. Neither should be interpreted as simply revealing an unobserved true value.
15. Intention-to-Treat and Analysis Populations
The ITT principle is central to the primary efficacy analyses. In Part A, the registry explicitly defines the ITT population as all randomized participants. For Part B ACR20, the ITT population includes participants randomized after dose selection in cohort 2.
| Endpoint | Population definition |
|---|---|
| Part A ACR20 Week 12 | Part A ITT population: all randomized participants |
| Part B ACR20 Week 24 | ITT population: all participants randomized after dose selection, cohort 2 |
| HAQ-DI Week 16 | Cohort 2 ITT participants with available baseline and on-or-before-Week-16 HAQ-DI data |
| mTSS Week 52 | Cohort 2 ITT participants with available baseline and on-or-before-Week-52 mTSS data |
| Major Clinical Response Week 52 | ITT population: all participants randomized after dose selection, cohort 2 |
The ITT framework preserves the treatment assignment generated by randomization, while endpoint-specific data requirements determine which participants contribute an observed measurement to a particular continuous analysis. This distinction is especially important when interpreting the HAQ-DI and mTSS models.
16. Confidence Intervals and P-values
Most of the registry-posted treatment effects include two-sided 95% confidence intervals. The intervals and p-values answer different statistical questions and should be read together.
Confidence interval
A 95% confidence interval quantifies statistical uncertainty around an estimated treatment effect under the specified analysis framework. It does not describe the range of effects that must occur in individual participants.
P-value
A p-value quantifies the compatibility of the observed data with a specified null hypothesis under the statistical test. It is not a measure of treatment-effect size.
For example, the SAR 150 mg q2w Part B ACR20 analysis has an odds ratio of 2.773 with a 95% CI of 2.077–3.703 and P < 0.0001. The odds ratio describes the estimated magnitude; the confidence interval describes uncertainty around that magnitude; and the p-value addresses the specified hypothesis test.
A very small p-value does not imply that the treatment effect is very large, just as a wide confidence interval does not imply that the null hypothesis is true. Magnitude, precision, and evidence against a null hypothesis are separate dimensions of an analysis.
17. Comparing Binary and Continuous Endpoints
RA-MOBILITY provides a useful example of why clinical trials use different statistical methods for different endpoint types.
| Endpoint | Data structure | Effect measure | Method |
|---|---|---|---|
| ACR20 Week 12 | Binary response | Odds ratio | Cochran-Mantel-Haenszel |
| ACR20 Week 24 | Binary response | Odds ratio | Cochran-Mantel-Haenszel |
| HAQ-DI Week 16 | Continuous longitudinal outcome | LS mean difference | Mixed-effects model / MMRM |
| mTSS Week 52 | Continuous structural score | Not reported | Rank ANCOVA |
| Major Clinical Response Week 52 | Binary response | Odds ratio | Cochran-Mantel-Haenszel |
This is more than a choice of statistical vocabulary. The effect measure determines what the reported number means. An odds ratio compares odds; an LS mean difference compares model-based adjusted means; and a rank-based ANCOVA tests treatment differences using an ordered representation of the continuous endpoint.
18. Clinical Biostats Synthesis of the Primary Results
The four registered primary endpoints address complementary aspects of rheumatoid arthritis response. ACR20 captures a categorical clinical response, HAQ-DI measures participant-reported physical function, and mTSS evaluates radiographic structural damage.
| Primary endpoint | Key posted result | Statistical interpretation |
|---|---|---|
| Part A ACR20 Week 12 | ORs 1.17 to 3.84 across five comparisons | Evidence and precision varied by sarilumab dose comparison. |
| Part B ACR20 Week 24 | OR 2.773 and 3.975; both P < 0.0001 | Both selected-dose comparisons had estimated odds of response above placebo with confidence intervals excluding 1.00. |
| HAQ-DI Week 16 | LS mean differences -0.235 and -0.258; both P < 0.0001 | Both selected-dose comparisons had lower model-estimated HAQ-DI values than placebo. |
| mTSS Week 52 | Both P < 0.0001 | The registry reports strong statistical evidence under rank ANCOVA but does not provide an effect estimate or confidence interval. |
The endpoint pattern illustrates why a clinical-trial conclusion should not be reduced to a single p-value. Clinical response, physical function, and structural damage are distinct outcomes, and each requires interpretation in the context of its scale, analysis method, missing-data rules, and multiplicity framework.
19. Important Limitations and Interpretation Issues
- Dose-ranging structure: Part A included multiple sarilumab dose comparisons with placebo, creating a multiplicity problem that required explicit adjustment.
- Part A versus Part B populations: Part B analyses used participants randomized after dose selection, so results from the two parts should not automatically be treated as though they arose from an identical analysis population.
- Different missing-data strategies: ACR20, HAQ-DI, and mTSS used different rules for missing observations. These assumptions are part of the endpoint analysis.
- mTSS effect magnitude: the registry posts p-values for the Week 52 rank ANCOVA analyses but does not provide an effect estimate or confidence interval in those statistical-analysis records.
- Odds ratios versus response probabilities: odds ratios are not probability ratios and cannot be translated into an absolute response difference without additional information.
- Hierarchical testing: later endpoints in the Part B sequence depended on statistical significance of preceding endpoints at the specified 0.025 level.
- Safety denominators: serious-adverse-event counts are reported across different cohorts and denominators, so raw event counts should not be compared without preserving the corresponding population structure.
- Interpretation of p-values: p-values quantify evidence against specified null hypotheses; they do not measure treatment-effect magnitude or individual patient benefit.
20. Why This Trial Matters Statistically
RA-MOBILITY is a useful teaching case because it combines dose-ranging comparisons, hierarchical confirmatory testing, binary clinical-response endpoints, longitudinal continuous outcomes, and a structural-damage endpoint within one randomized clinical-trial program.
| Concept | How it appears in RA-MOBILITY |
|---|---|
| Randomization | The trial is randomized with parallel allocation. |
| Blinding | The registry identifies triple masking. |
| ITT analysis | Primary analyses use specified ITT populations. |
| Cochran-Mantel-Haenszel test | Used for ACR20 and Major Clinical Response binary endpoints. |
| Odds ratio | Used to quantify binary treatment comparisons. |
| Mixed-effects model | Used for longitudinal HAQ-DI analysis. |
| ANCOVA | Rank ANCOVA used for Week 52 mTSS. |
| Multiplicity | Hommel, Bonferroni, and hierarchical procedures are reported. |
| Missing-data methods | LOCF, no imputation with specified responder rules, missing-after-discontinuation handling, and linear extrapolation are used across endpoints. |
| Multiple endpoint domains | Clinical response, physical function, and structural damage were all evaluated. |
21. Statistical Methods in Practical Terms
Binary endpoint
ACR20 converts a multidimensional clinical response definition into responder versus non-responder status, allowing treatment groups to be compared using categorical-data methods.
Continuous endpoint
HAQ-DI retains a numeric scale, allowing the analysis to estimate an adjusted difference in mean outcome rather than an odds ratio.
Repeated measurements
The MMRM framework accounts for the longitudinal nature of HAQ-DI observations and produces least-squares mean comparisons.
Rank-based analysis
Rank ANCOVA evaluates the mTSS endpoint using ordered information rather than relying solely on a conventional arithmetic-mean comparison.
22. What the Primary Results Do — and Do Not — Establish
The ACR20 odds ratios quantify relative odds of achieving the registered clinical-response definition. They do not directly quantify absolute response probability or improvement in every component of the ACR20 definition.
The negative LS mean differences indicate lower model-estimated HAQ-DI values under the selected sarilumab comparisons. They do not mean that every participant experienced exactly the reported reduction.
The reported P < 0.0001 results indicate evidence against the specified null hypotheses under the rank ANCOVA analyses. Because the registry does not provide an effect estimate or confidence interval for these analyses, the magnitude and precision of the structural-damage difference cannot be characterized from those posted analysis fields alone.
The statistical evidence needs to be interpreted through the trial's Hommel, Bonferroni, and hierarchical testing procedures. A collection of individual p-values is not equivalent to a collection of independent confirmatory tests.
23. Trial Timeline
Trial start
The registry lists March 2010 as the study start date.
Randomized phase 2/3 program
The study evaluated sarilumab dosing in rheumatoid arthritis using randomized parallel treatment groups and Part A and Part B analyses.
Primary completion
The registry lists October 2013 as the primary completion date.
Registry results posted
The ClinicalTrials.gov record reports posted outcome measures and statistical analyses, including 11 primary and 2 secondary statistical analyses.
24. Related Tutorials
Learn more about the methods used in this trial:
25. Related Calculators
26. Sources
- ClinicalTrials.gov: NCT01061736.
- PubMed: PMID 24297381.
- PubMed: PMID 25733246.
- PubMed: PMID 36413080.
- PubMed: PMID 36008838.
- PubMed: PMID 34519964.
Continue through the Clinical Biostats statistical library
Explore tutorials and statistical calculators that explain the methods used in randomized clinical trials, from categorical-data tests and odds ratios to longitudinal models and ANCOVA.
27. Record Summary
RA-MOBILITY provides a compact example of how a clinical trial can require several statistical frameworks at once. The randomized parallel design and ITT populations anchor the treatment comparisons, while the endpoint types determine the analysis methods: Cochran-Mantel-Haenszel testing for binary response outcomes, a mixed-effects repeated-measures model for HAQ-DI, and rank ANCOVA for mTSS.
The posted Part A ACR20 analyses produced odds ratios ranging from 1.17 to 3.84, with different levels of statistical evidence and precision across dose comparisons. In Part B, the ACR20 odds ratios were 2.773 and 3.975, while HAQ-DI LS mean differences were -0.235 and -0.258, all with P < 0.0001. The mTSS analyses also reported P < 0.0001 for both selected-dose comparisons, although the posted statistical-analysis records do not provide corresponding effect estimates or confidence intervals.
The statistical story is therefore broader than any single p-value. Dose comparisons, analysis populations, endpoint definitions, missing-data rules, effect measures, confidence intervals, and multiplicity procedures all contribute to the correct interpretation of the evidence.