This page separates reported trial results from statistical interpretation. The numerical results presented here are restricted to the ClinicalTrials.gov record.
1. Trial at a Glance
RUXCOVID-DEVENT was a randomized, double-blind, parallel phase 3 trial evaluating ruxolitinib in participants with COVID-19-associated ARDS who require mechanical ventilation. The registered primary endpoint was the percentage of participants who had died due to any cause from study start to Day 29.
| Feature | RUXCOVID-DEVENT |
|---|---|
| Phase | Phase 3 |
| Condition | COVID-19 |
| Study population | Participants with COVID-19-associated ARDS who require mechanical ventilation |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | Double |
| Primary purpose | Treatment |
| Enrollment | 211 |
| Interventions | Placebo; Ruxolitinib |
| Primary endpoint type | Binary |
| Primary endpoint analyses posted | 2 |
| Primary analyses with estimate + confidence interval | 2 |
| Trial status | Terminated |
2. Clinical Question
The registered clinical question was whether ruxolitinib 5 mg BID plus standard of care and ruxolitinib 15 mg BID plus standard of care could be evaluated against placebo plus standard of care for the 28-day mortality rate in participants with COVID-19-associated ARDS who require mechanical ventilation.
Population
Participants with COVID-19-associated ARDS who require mechanical ventilation.
Intervention
Ruxolitinib, evaluated at 15 mg BID and 5 mg BID, with standard of care.
Comparator
Placebo plus standard of care.
Primary question
How does the odds of death due to any cause from study start to Day 29 compare for each ruxolitinib dose plus standard of care versus placebo plus standard of care?
3. Trial Design
Ruxolitinib 15 mg BID + SoC
- Ruxolitinib 15 mg BID
- Standard of care therapy
- Primary comparison reported as an odds ratio versus placebo + SoC
Ruxolitinib 5 mg BID + SoC
- Ruxolitinib 5 mg BID
- Standard of care therapy
- Primary comparison reported as an odds ratio versus placebo + SoC
Placebo + SoC
- Placebo
- Standard of care therapy
- Reference group for the reported odds ratios
4. Endpoints
The registry lists one primary endpoint. It is a binary mortality endpoint evaluated from study start through Day 29.
| Endpoint | Registry definition | Time frame | Type |
|---|---|---|---|
| Percentage of Participants Who Have Died Due to Any Cause | To evaluate the 28-day mortality rate of ruxolitinib 5 mg BID + SoC therapy and ruxolitinib 15 mg BID + SoC compared with placebo + SoC therapy, in participants with COVID-19-associated ARDS who require mechanical ventilation. | Study start to Day 29 | Binary |
The distinction between the registry's endpoint label, its stated 28-day mortality objective, and the registered time frame of study start to Day 29 is important. The analysis presented below uses the endpoint wording and time frame exactly as reported in the registry rather than substituting a different definition.
5. Analysis Population and Statistical Framework
The primary analyses use the intention-to-treat (ITT) population. The registry defines the ITT population according to treatment assignment at the time of randomization regardless of the actual study drug a participant might take during participation in the study.
| Component | Registered analysis approach |
|---|---|
| Analysis population | ITT population |
| Endpoint | Percentage of participants who have died due to any cause |
| Endpoint type | Binary |
| Method | Mixed Models Analysis; normalized as a mixed-effects model |
| Model family | Longitudinal / mixed models |
| Effect measure | Odds ratio |
| Hypothesis type | Superiority |
| Fixed covariates | Treatment group and ARDS severity |
| Random effect | Investigational site |
The analysis text specifies a logistic regression mixed model that includes treatment group and ARDS severity as fixed covariates and investigational site as a random effect. Thus, the reported odds ratios are adjusted model-based comparisons rather than simple unadjusted ratios of observed event proportions.
For a binary endpoint, logistic regression models the probability of the event through the log-odds scale. The mixed-effects extension allows investigational site to contribute a random effect rather than treating all sites as having an identical baseline event propensity.
6. Statistical Methodology
Binary endpoint analysis
The primary endpoint is binary: each participant either has died due to any cause within the registered study period or has not. A logistic regression framework is therefore a natural model for the endpoint. The registry specifically reports a logistic regression mixed model.
Covariate adjustment
The model includes ARDS severity as a fixed covariate in addition to treatment group. Covariate adjustment can improve precision when an important baseline characteristic is associated with the outcome. It also means that the reported odds ratio should be interpreted as the treatment contrast from the fitted adjusted model, rather than as a purely unadjusted comparison.
Random effect for investigational site
Investigational site is included as a random effect. This allows the model to account for variation among sites without estimating a separate fixed treatment effect for every site. In a multicenter setting, this is a way of recognizing that participants treated at different sites may share site-level characteristics.
Odds ratio as the effect measure
The registry reports odds ratios rather than risk ratios or risk differences. An odds ratio compares the odds of the binary event between two groups:
An OR below 1 indicates lower estimated odds of the event in the treatment group relative to the reference group. It does not directly state the percentage-point difference in mortality or the relative risk.
Intention-to-treat analysis
The ITT definition reported by the registry assigns participants according to their randomized treatment assignment regardless of the actual study drug they might take. This preserves the treatment comparison created by randomization and avoids redefining treatment groups based on subsequent treatment exposure.
Superiority hypothesis
The registered hypothesis type is superiority. This means the analysis was structured to evaluate whether a ruxolitinib regimen differed from placebo plus standard of care in the specified direction of interest, rather than to establish equivalence or non-inferiority.
7. Primary Results
ClinicalTrials.gov reports two formal primary-endpoint analyses, one for each ruxolitinib dose versus placebo plus standard of care. Both use the same endpoint, ITT population, mixed-model framework, and covariate structure.
Ruxolitinib 15 mg BID + Standard of Care
Odds ratio for death due to any cause
95% CI: 0.201–1.028 · P = 0.0292
Two-sided confidence interval; superiority analysis in the ITT population.
| Primary endpoint | 15 mg BID + SoC vs placebo + SoC |
|---|---|
| Endpoint | Percentage of Participants Who Have Died Due to Any Cause |
| Time frame | Study start to Day 29 |
| Effect measure | Odds ratio |
| Estimate | 0.46 |
| 95% CI | 0.201–1.028 |
| P-value | 0.0292 |
| Model | Logistic regression mixed model |
| Adjustment | Treatment group and ARDS severity as fixed covariates; investigational site as a random effect |
The estimated odds ratio of 0.46 means that, in the fitted adjusted mixed-effects logistic model, the estimated odds of death for the ruxolitinib 15 mg BID + SoC group were 0.46 times the corresponding odds for placebo + SoC. Expressed as a simple relative interpretation of the model estimate, this corresponds to approximately 54% lower estimated odds of the event.
This does not mean that mortality was 54% lower as an absolute percentage, that 54% of participants benefited, or that an individual participant's probability of death was reduced by 54%. Odds and probabilities are related but are not interchangeable.
The 95% confidence interval of 0.201–1.028 describes uncertainty around the model-based odds-ratio estimate. Its width indicates that the treatment-effect estimate is not highly precise. In particular, the interval extends above 1, so the interval includes an odds ratio corresponding to higher estimated odds in the treatment group.
The P-value of 0.0292 is a measure of evidence against the specified null hypothesis under the model and testing framework. It is not a measure of the magnitude or clinical importance of the treatment effect. A p-value should therefore be read alongside the odds ratio and confidence interval rather than used as a substitute for them.
The analysis is based on the ITT population and adjusts for ARDS severity while accounting for investigational site as a random effect. Because the endpoint is binary and the model is logistic, proportional-hazards assumptions do not apply to this primary analysis. The registry supplies a two-sided 95% confidence interval; the ClinicalTrials.gov record does not describe a non-inferiority margin or a proportional-hazards analysis.
Ruxolitinib 5 mg BID + Standard of Care
Odds ratio for death due to any cause
95% CI: 0.171–1.023 · P = 0.0280
Two-sided confidence interval; superiority analysis in the ITT population.
| Primary endpoint | 5 mg BID + SoC vs placebo + SoC |
|---|---|
| Endpoint | Percentage of Participants Who Have Died Due to Any Cause |
| Time frame | Study start to Day 29 |
| Effect measure | Odds ratio |
| Estimate | 0.42 |
| 95% CI | 0.171–1.023 |
| P-value | 0.0280 |
| Model | Logistic regression mixed model |
| Adjustment | Treatment group and ARDS severity as fixed covariates; investigational site as a random effect |
The estimated odds ratio of 0.42 means that, under the fitted adjusted mixed-effects logistic model, the estimated odds of death for the ruxolitinib 5 mg BID + SoC group were 0.42 times the corresponding odds for placebo + SoC. As a direct interpretation of the model estimate, this is approximately 58% lower estimated odds of the event.
The estimate should not be translated directly into a 58% reduction in mortality probability. An odds ratio is a ratio of odds, not a ratio of probabilities, and the absolute effect depends on the underlying event probability.
The 95% confidence interval is 0.171–1.023. This interval indicates substantial uncertainty around the point estimate and extends slightly above 1. The interval therefore communicates less precision than the point estimate alone might suggest.
The P-value of 0.0280 quantifies the statistical evidence under the specified model and testing framework; it does not quantify the size of the treatment effect. The estimate, confidence interval, endpoint definition, and analysis population remain essential to interpretation.
As with the 15 mg BID analysis, the model uses treatment group and ARDS severity as fixed covariates and investigational site as a random effect. The ClinicalTrials.gov record identifies the hypothesis as superiority and the confidence intervals as two-sided. They do not provide a non-inferiority margin, a crossover analysis, a proportional-hazards assumption, or a Bayesian analysis for this primary endpoint.
8. Comparing the Two Primary Estimates
The two formal primary analyses provide separate comparisons of the two ruxolitinib doses with placebo plus standard of care. They should not be treated as a single combined treatment effect because the registry reports them as distinct treatment-group comparisons.
| Comparison | Odds ratio | 95% CI | P-value | Hypothesis |
|---|---|---|---|---|
| Ruxolitinib 15 mg BID + SoC vs placebo + SoC | 0.46 | 0.201–1.028 | 0.0292 | Superiority |
| Ruxolitinib 5 mg BID + SoC vs placebo + SoC | 0.42 | 0.171–1.023 | 0.0280 | Superiority |
Both point estimates are below 1. The 5 mg BID estimate is numerically lower than the 15 mg BID estimate, but the ClinicalTrials.gov record does not provide a formal statistical comparison between the two ruxolitinib doses. Therefore, the two point estimates should not be interpreted as establishing that one dose has a different treatment effect from the other.
9. Safety Results
The ClinicalTrials.gov record reports serious adverse events by treatment arm. The data provide affected participants and participants at risk for each arm, but do not provide a separate statistical comparison or confidence interval for serious adverse events.
| Arm | Serious adverse events | At risk |
|---|---|---|
| Ruxolitinib 15 mg BID | 23 | 77 |
| Ruxolitinib 5 mg BID | 21 | 87 |
| Placebo | 11 | 45 |
| Total | 55 | 209 |
The ClinicalTrials.gov record shows that serious adverse events were reported in all three arms. Because the ClinicalTrials.gov record does not include a formal safety-effect estimate, confidence interval, or p-value, the table should be read descriptively rather than as a statistical comparison of safety between treatment groups.
10. Statistical Methods Explained
Why was a mixed-effects model used for a binary endpoint?
The primary endpoint is binary, so logistic regression provides a natural framework for modeling it. The registry goes further by specifying a mixed model with investigational site as a random effect. This allows the analysis to acknowledge site-level variation while estimating the treatment comparison and adjusting for ARDS severity.
What does an odds ratio of 0.46 mean?
An odds ratio of 0.46 means that the modeled odds of death in the ruxolitinib 15 mg BID + SoC group are estimated at 46% of the odds in the placebo + SoC group. The corresponding simple relative interpretation is approximately 54% lower estimated odds. It is not equivalent to saying that the probability of death is 54% lower.
What does an odds ratio of 0.42 mean?
An odds ratio of 0.42 means that the modeled odds of death in the ruxolitinib 5 mg BID + SoC group are estimated at 42% of the odds in the placebo + SoC group. The simple relative interpretation is approximately 58% lower estimated odds. Again, the result does not directly provide an absolute mortality difference.
Why adjust for ARDS severity?
The registry's model includes ARDS severity as a fixed covariate. Adjustment incorporates information about disease severity into the model rather than relying solely on the treatment-group indicator. The resulting odds ratio is therefore an adjusted model estimate conditional on the covariates specified in the analysis.
Why include investigational site as a random effect?
Participants within the same investigational site may share site-specific characteristics. A random effect provides a way to model between-site variation while retaining a common treatment-effect framework. In this analysis, site is treated as a random rather than fixed effect.
Why use the ITT population?
Under the registry-reported definition, participants remain associated with their randomized assignment regardless of the actual study drug they might receive. This preserves the randomized comparison and avoids changing the treatment groups based on post-randomization treatment exposure.
Why does the confidence interval matter more than the point estimate alone?
The odds ratio is a single estimate, while the confidence interval communicates uncertainty around that estimate. For the 15 mg BID comparison, the interval is 0.201–1.028; for the 5 mg BID comparison, it is 0.171–1.023. In both cases, the interval is substantially wider than the point estimate alone would suggest and extends slightly above 1.
11. Understanding the P-values
The registry reports P = 0.0292 for the 15 mg BID comparison and P = 0.0280 for the 5 mg BID comparison. These are statistical quantities associated with the respective superiority analyses.
P-value ≠ effect size
A p-value describes statistical evidence under a specified null hypothesis and model. It does not say how large or clinically important the treatment effect is.
Effect estimate matters
The odds ratios of 0.46 and 0.42 describe the estimated relative effects on the odds scale. They should be interpreted together with their confidence intervals.
Precision matters
The intervals 0.201–1.028 and 0.171–1.023 show uncertainty that cannot be conveyed by the point estimates alone.
Testing framework matters
The ClinicalTrials.gov record identifies both analyses as superiority tests with two-sided 95% confidence intervals, but do not provide a multiplicity adjustment scheme.
12. Odds Ratios, Risks, and Probabilities
For a binary endpoint, three quantities are often confused: probability, odds, and odds ratio. They are mathematically related but have different interpretations.
| Quantity | Meaning |
|---|---|
| Probability | The proportion or modeled probability of participants experiencing the event. |
| Odds | Probability of the event divided by probability of not experiencing the event. |
| Odds ratio | The ratio of odds between two treatment groups. |
This distinction is particularly important when communicating an odds ratio to a clinical audience. An OR of 0.42 does not mean the event probability is 0.42 times the control probability in every setting. The numerical relationship between odds ratios and risk ratios depends on the baseline event probability.
The logistic model operates on the log-odds scale. Converting an odds ratio into an absolute probability difference requires a baseline risk or an appropriate model-based reference probability.
13. Covariate Adjustment and Randomization
Randomization is the foundation of the treatment comparison because treatment assignment is randomized rather than chosen according to prognosis. Covariate adjustment is then used within the prespecified model to account for ARDS severity while the mixed-effects structure accounts for investigational site.
| Model component | Role in the primary analysis |
|---|---|
| Treatment group | Estimates the treatment comparison of each ruxolitinib regimen with placebo + SoC. |
| ARDS severity | Fixed covariate included for adjustment. |
| Investigational site | Random effect accounting for site-level variation. |
| ITT assignment | Defines treatment groups according to randomization. |
| Logistic link | Models the binary mortality outcome through log-odds. |
Adjustment should not be confused with replacing randomization. The treatment comparison remains defined by randomized assignment in the ITT population; the regression model adds structure for the prespecified covariates and site effect.
14. What the Registry Data Do Not Establish
The registry-reported RUXCOVID-DEVENT data are sufficient to describe the primary endpoint analyses and the reported serious adverse events, but they do not support several analyses that might appear in a more extensive clinical-trial report.
- No median survival analysis: the ClinicalTrials.gov record does not report a time-to-event survival endpoint with median survival estimates.
- No hazard ratio analysis: the primary endpoint is binary and the reported effect measure is an odds ratio.
- No proportional-hazards analysis: a Cox model or other hazard-based analysis is not reported for the registry-reported primary endpoint.
- No non-inferiority margin: the hypothesis type is superiority, and no non-inferiority margin is reported.
- No crossover analysis: the ClinicalTrials.gov record does not describe crossover or treatment switching.
- No Bayesian analysis: no Bayesian method is reported in the statistical analyses posted on ClinicalTrials.gov.
- No formal subgroup results: the registry-reported primary analyses do not provide subgroup-specific effect estimates.
- No formal safety comparison: serious adverse events are reported by arm, but no safety-effect estimate or inferential test is reported.
15. Multiplicity and the Two Primary Comparisons
The ClinicalTrials.gov record identifies two primary-endpoint analyses: one comparing ruxolitinib 15 mg BID + SoC with placebo + SoC and one comparing ruxolitinib 5 mg BID + SoC with placebo + SoC. Both are labeled superiority analyses.
| Feature | Registry-supported interpretation |
|---|---|
| Number of primary analyses | 2 |
| Comparisons | 15 mg BID + SoC vs placebo + SoC; 5 mg BID + SoC vs placebo + SoC |
| Endpoint | Percentage of Participants Who Have Died Due to Any Cause |
| Hypothesis type | Superiority |
| Confidence intervals | 95%, two-sided |
| Multiplicity adjustment reported? | Not specified in the ClinicalTrials.gov record |
Because two primary treatment comparisons are reported, the interpretation of their p-values depends on the prespecified multiplicity strategy. The ClinicalTrials.gov record does not specify an alpha-allocation or multiplicity-adjustment procedure. Accordingly, this page reports the p-values exactly as posted without constructing an additional multiplicity correction.
16. Safety and Efficacy Are Different Statistical Questions
The primary efficacy endpoint asks about death due to any cause over the registered study period. Serious adverse events represent a separate safety domain. These endpoints should not be combined into a single numerical judgment without a prespecified benefit-risk framework.
Efficacy
The primary efficacy analysis reports adjusted odds ratios for death due to any cause from study start to Day 29.
Safety
Serious adverse events are reported descriptively as affected participants and participants at risk in each arm.
Different denominators
The serious-adverse-event table has 209 participants at risk, while overall enrollment is reported as 211.
Different inference
The efficacy results include estimates, confidence intervals, and p-values; the ClinicalTrials.gov record does not include a formal inferential comparison.
17. Limitations
- Registry-level reporting: the analysis is based on the ClinicalTrials.gov record rather than a complete protocol or statistical analysis plan.
- Limited primary endpoint detail: the registry identifies the endpoint and model but does not supply the underlying participant-level data needed to reconstruct the adjusted model.
- Confidence-interval width: both reported intervals are relatively broad around their point estimates, indicating meaningful uncertainty in the estimated treatment effects.
- Interval crossing 1: both 95% confidence intervals extend slightly above an odds ratio of 1, which is important when interpreting the precision of the estimates.
- Multiplicity information: two primary analyses are reported, but the ClinicalTrials.gov record does not specify how type I error was allocated between them.
- No absolute mortality estimates reported: the data provide odds ratios but do not provide the arm-specific mortality percentages required to calculate or display an absolute risk difference.
- No dose-to-dose comparison: the analyses posted on ClinicalTrials.gov compare each ruxolitinib dose with placebo rather than formally comparing the two ruxolitinib doses.
- Safety inference: serious adverse events are reported descriptively without a registry-reported statistical comparison.
- Generalizability: the study population is specifically defined as participants with COVID-19-associated ARDS who require mechanical ventilation, so the ClinicalTrials.gov record should not automatically be generalized to populations outside that clinical setting.
18. Why This Trial Matters Statistically
RUXCOVID-DEVENT is a useful teaching case because its primary analysis combines several fundamental ideas in clinical-trial statistics: randomized treatment assignment, a binary endpoint, intention-to-treat analysis, covariate adjustment, a mixed-effects model, a random site effect, odds-ratio estimation, confidence intervals, and superiority testing.
| Concept | How it appears in RUXCOVID-DEVENT |
|---|---|
| Randomization | Participants were assigned using a randomized parallel design. |
| Blinding | The trial was double-blind. |
| ITT analysis | The primary analyses use treatment assignment at randomization regardless of actual study drug received. |
| Binary endpoint | The primary outcome is the percentage of participants who have died due to any cause from study start to Day 29. |
| Logistic regression | The primary analysis uses a logistic regression mixed model. |
| Mixed-effects model | Investigational site is included as a random effect. |
| Covariate adjustment | ARDS severity is included as a fixed covariate. |
| Odds ratio | The treatment effect is reported on the odds scale. |
| Confidence interval | Each primary comparison has a two-sided 95% confidence interval. |
| P-value | Each primary comparison has a reported p-value associated with the superiority analysis. |
| Multiplicity | Two primary treatment comparisons are reported; the ClinicalTrials.gov record does not specify an adjustment strategy. |
| Safety analysis | Serious adverse events are reported descriptively by arm. |
19. Reading the Primary Results as a Statistician
The reported treatment contrast is the odds of death due to any cause from study start to Day 29 for each randomized ruxolitinib regimen relative to placebo plus standard of care, within the ITT framework and the specified mixed-effects logistic model.
The effect measure is an odds ratio. Values below 1 indicate lower estimated odds in the ruxolitinib group relative to placebo. This is different from a risk ratio and from an absolute risk difference.
The 15 mg BID estimate is 0.46 with a 95% CI of 0.201–1.028. The 5 mg BID estimate is 0.42 with a 95% CI of 0.171–1.023. The confidence intervals show that the point estimates should not be interpreted without acknowledging substantial statistical uncertainty.
The analysis is not an unadjusted 2×2 table alone. It uses a logistic regression mixed model with treatment group and ARDS severity as fixed covariates and investigational site as a random effect.
The p-values provide evidence under the specified testing framework, while the odds ratios describe the estimated relative effect and the confidence intervals describe uncertainty. None of these quantities alone supplies an absolute mortality difference.
20. A Compact Statistical Summary
| Element | RUXCOVID-DEVENT |
|---|---|
| Design | Randomized, double-blind, parallel, phase 3 |
| Enrollment | 211 |
| Primary endpoint | Percentage of Participants Who Have Died Due to Any Cause |
| Time frame | Study start to Day 29 |
| Endpoint type | Binary |
| Analysis population | ITT |
| Model | Logistic regression mixed model |
| Fixed covariates | Treatment group; ARDS severity |
| Random effect | Investigational site |
| Effect measure | Odds ratio |
| 15 mg BID estimate | OR 0.46; 95% CI 0.201–1.028; P = 0.0292 |
| 5 mg BID estimate | OR 0.42; 95% CI 0.171–1.023; P = 0.0280 |
| Safety information reported | Serious adverse events by arm |
21. Related Tutorials
Learn more about the methods used in this trial:
22. Related Statistical Calculators
23. Sources
- ClinicalTrials.gov: RUXCOVID-DEVENT, NCT04377620.
- Linked publication: PubMed record for PMID 36226977.
Continue through the Clinical Biostats statistical pathway
Use the trial's methods as a starting point for deeper study of randomized trials, binary endpoints, odds ratios, confidence intervals, covariate adjustment, and mixed-effects models.
24. Record Summary
RUXCOVID-DEVENT provides a focused example of statistical analysis for a randomized clinical trial with a binary primary endpoint. The primary analyses use the ITT population and a logistic regression mixed model incorporating treatment group and ARDS severity as fixed covariates and investigational site as a random effect. The reported effect measure is the odds ratio, with estimates of 0.46 for ruxolitinib 15 mg BID + SoC versus placebo + SoC and 0.42 for ruxolitinib 5 mg BID + SoC versus placebo + SoC. The corresponding two-sided 95% confidence intervals are 0.201–1.028 and 0.171–1.023, with reported p-values of 0.0292 and 0.0280, respectively.
The most important statistical lesson is that these numbers answer a specific model-based question. The odds ratios quantify relative treatment effects on the odds scale; the confidence intervals communicate uncertainty; and the p-values quantify statistical evidence under the specified testing framework. None should be interpreted in isolation or converted automatically into an absolute mortality difference. The ClinicalTrials.gov record is separate and descriptive, reporting serious adverse events by treatment arm without a formal inferential comparison.