← Clinical Trials
COVID-19 Associated Cytokine Storm Phase 3 Completed NCT04362137

RUXCOVID: Complete Statistical Analysis of Ruxolitinib in COVID-19 Associated Cytokine Storm

An independent statistical review of the randomized, double-blind phase 3 RUXCOVID trial evaluating ruxolitinib 5 mg versus placebo in patients with COVID-19 associated cytokine storm, with emphasis on the primary composite endpoint, secondary efficacy outcomes, time-to-event analyses, and interpretation of uncertainty.

Trial start: 2020-05-02  ·  Primary completion: 2020-10-17  ·  Enrollment: 432
Scope of this record

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

RUXCOVID was a randomized, double-blind, parallel-group phase 3 trial evaluating ruxolitinib 5 mg versus placebo in patients with COVID-19 associated cytokine storm. The registry reports 432 enrolled participants, two treatment arms, one primary binary endpoint, and statistical analyses using logistic regression, a Cox proportional-hazards model, and ANCOVA.

432
Enrolled
Phase 3
2
Arms
Parallel design
0.91
Primary OR
95% CI 0.48–1.73
0.769
Primary P-value
Two-sided
FeatureRUXCOVID
Trial nameRUXCOVID
Brief titleStudy to Assess the Efficacy and Safety of Ruxolitinib in Patients With COVID-19 Associated Cytokine Storm
PhasePhase 3
ConditionCytokine Storm (Covid-19)
DesignRandomized, double-blind, parallel-group
Primary purposeTreatment
InterventionsRuxolitinib 5 mg and placebo
Enrollment432
Primary endpoint typeBinary
Hypothesis typeSuperiority
Results postedYes
Statistical analyses posted17
Lead sponsorNovartis Pharmaceuticals

2. Clinical Question

The primary statistical question was whether ruxolitinib 5 mg differed from placebo with respect to the proportion of patients who died, developed respiratory failure requiring mechanical ventilation, or required intensive care unit care during Day 1 through Day 29.

Population

Patients enrolled in the phase 3 RUXCOVID trial with COVID-19 associated cytokine storm.

Intervention

Ruxolitinib 5 mg.

Comparator

Placebo.

Primary question

Does ruxolitinib 5 mg produce a different proportion of patients experiencing the registered composite clinical outcome over Day 1 through Day 29?

3. Trial Design

01
Randomize432 enrolled
02
Two armsRuxolitinib 5 mg vs placebo
03
Double-blindParallel-group design
04
AssessClinical outcomes
05
AnalyzeBinary and time-to-event endpoints
ARM A · RUXOLITINIB 5 MG

Ruxolitinib

  • Ruxolitinib 5 mg
  • Randomized treatment assignment
  • Double-blind trial design
  • Compared with placebo
ARM B · PLACEBO

Placebo

  • Placebo
  • Randomized treatment assignment
  • Double-blind trial design
  • Compared with ruxolitinib 5 mg
Blinding is statistically relevant. The registry identifies the study as double-blind. Blinding can reduce the influence of treatment knowledge on participant behavior, outcome assessment, and other trial processes. It does not, however, remove uncertainty caused by sampling variation, missing assessments, or the assumptions of the statistical model.

4. Endpoints

EndpointTime frameTypeAnalysis reported
Proportion of Patients Who Die, Develop Respiratory Failure [Require Mechanical Ventilation] or Require Intensive Care Unit (ICU) Care Day 1 - Day 29 Binary Logistic regression
Percentage of Patients With at Least Two-point Improvement From Baseline in Clinical Status Baseline, Day 15, Day 29 Binary Logistic regression
Percentage of Patients With at Least One-point Improvement From Baseline in Clinical Status Baseline, Day 15, Day 29 Binary Logistic regression
Percentage of Patients With at Least One-point Deterioration From Baseline in Clinical Status Baseline, Day 15, Day 29 Binary Logistic regression
Time to Improvement in Clinical Status 29 days Time-to-event Cox proportional-hazards model
Mean Change From Baseline in the Clinical Status Baseline, Day 15, Day 29 Continuous ANCOVA
Mortality Rate Day 15, Day 29 Binary Logistic regression
Proportion of Patients Requiring Mechanical Ventilation Day 1 - Day 29 Binary Logistic regression
Duration of Hospitalization 29 days Time-to-event Cox proportional-hazards model
Time to Hospital Discharge or to a NEWS2 Score of ≤2 29 days Time-to-event Cox proportional-hazards model
Proportion of Patients With no Oxygen Therapy Day 15, Day 29 Binary Logistic regression
Primary endpoint definition: Efficacy is measured by a composite endpoint of proportion of patients who die, develop respiratory failure [require mechanical ventilation], or require intensive care unit [ICU] care for the treatment of COVID-19. Analyses are cumulative, thus analysis on Day 29 includes all events till that day. The posted primary analysis excludes randomized participants who had developed respiratory failure and/or required ICU at randomization.

5. Statistical Methodology

The registry reports three main statistical method families for RUXCOVID: logistic regression for binary outcomes, a Cox proportional-hazards model for time-to-event outcomes, and ANCOVA for the continuous clinical-status change endpoint.

Logistic regression for binary outcomes

The primary endpoint is binary: a participant either experiences the specified composite outcome during the registered time frame or does not. Logistic regression models the odds of the binary outcome and expresses the treatment comparison as an odds ratio.

Odds ratio
OR = odds in the ruxolitinib 5 mg group / odds in the placebo group

An OR of 1 represents equal odds under the model. An OR below 1 indicates lower estimated odds in the ruxolitinib 5 mg group; an OR above 1 indicates higher estimated odds.

Cox proportional-hazards model

The registry uses a proportional-hazards model for time to improvement in clinical status, duration of hospitalization, and time to hospital discharge or to a NEWS2 score of ≤2. These analyses use a hazard ratio as the effect measure.

Hazard ratio
HR = estimated instantaneous event rate in ruxolitinib 5 mg / placebo

For the posted time-to-event analyses, the registry states that a hazard ratio greater than 1 favors the ruxolitinib 5 mg arm. A hazard ratio is not a probability, a risk ratio, or a difference in median time.

ANCOVA for change in clinical status

ANCOVA was used for the continuous endpoint Mean Change From Baseline in the Clinical Status. The registry reports the effect as a least-squares mean, normalized here as a mean difference between treatment groups.

Mean-difference interpretation
Estimated difference = LS mean change, ruxolitinib 5 mg − placebo

A value near zero indicates little estimated difference in mean change between the groups under the fitted model. The sign must be interpreted in the context of how the clinical-status scale is coded.

Intention-to-treat principle

For mortality, the registry explicitly states that the analysis population includes all randomized participants, including those who did not receive any dose, according to the intent-to-treat principle, while excluding patients lost to follow-up. This preserves treatment assignment as the basis for the randomized comparison.

Competing-risk framework

For the time to improvement in clinical status, duration of hospitalization, and time to hospital discharge or to a NEWS2 score of ≤2, the registry notes that the between-group comparison used a competing-risk framework. This matters because a participant may experience a competing event that prevents the event of primary interest from occurring in the same way.

6. Results: Primary Endpoint

The primary analysis compared ruxolitinib 5 mg with placebo for the registered composite endpoint over Day 1 through Day 29. The analysis population consisted of randomized participants excluding those who developed respiratory failure and/or required ICU care at randomization.

Odds ratio for the primary composite endpoint

0.91

95% CI: 0.48–1.73   ·   P = 0.769

Two-sided superiority analysis; logistic regression

Clinical Biostats interpretation

The estimated odds ratio of 0.91 means that the modeled odds of experiencing the composite endpoint were estimated to be approximately 9% lower with ruxolitinib 5 mg than with placebo. This is a relative comparison of odds, not a statement that 9% fewer patients experienced the outcome and not a 9-percentage-point absolute risk reduction.

The 95% confidence interval of 0.48–1.73 is fairly wide relative to the point estimate and includes 1.00. Thus, the reported estimate is compatible with lower or higher odds under the statistical uncertainty represented by the interval. The interval describes uncertainty around the estimated treatment effect; it does not describe the range of individual patient outcomes.

The P-value of 0.769 is a measure of how compatible the observed data are with the null hypothesis under the specified statistical framework. It does not measure the size, clinical importance, or probability of the treatment effect. A large P-value is not evidence that the treatment effects are exactly equal.

The analysis is also specific to the stated analysis population and cumulative Day 1 through Day 29 composite endpoint. Because the endpoint combines death, respiratory failure requiring mechanical ventilation, and ICU care, the overall odds ratio does not identify which component, if any, drives the observed estimate.

7. Secondary Binary Outcomes

The registry reports several secondary binary analyses. Most use logistic regression and odds ratios, with assessments at Day 15 and Day 29 where specified. These endpoints should be read individually rather than treating their nominal P-values as if they were a single prespecified confirmatory test.

OutcomeTimeOR95% CIP-value
At least two-point improvement in clinical status Day 15 0.89 0.55–1.46 0.647
At least two-point improvement in clinical status Day 29 1.00 0.52–1.92 0.997
At least one-point improvement in clinical status Day 15 0.98 0.51–1.87 0.946
At least one-point improvement in clinical status Day 29 0.79 0.35–1.79 0.573
At least one-point deterioration in clinical status Day 15 0.75 0.31–1.83 0.532
At least one-point deterioration in clinical status Day 29 1.18 0.40–3.49 0.764
Mortality rate Day 15 0.94 0.20–5.57 0.944
Mortality rate Day 29 1.21 0.35–5.11 0.775
Requiring mechanical ventilation Day 1 - Day 29 0.99 0.45–2.21 0.987
No oxygen therapy Day 15 0.61 0.23–1.63 0.325
No oxygen therapy Day 29 1.22 0.25–5.40 —
Direction of the odds ratio: The registry's analysis notes generally state that an OR < 1 favors ruxolitinib 5 mg for these comparisons, while the two clinical-status improvement analyses explicitly state that an OR > 1 favors ruxolitinib 5 mg. The direction therefore needs to be read together with the endpoint-specific registry analysis note rather than inferred solely from the numerical value.

8. Time-to-Event Results

Three secondary endpoints were analyzed using a proportional-hazards model with a competing-risk framework. These analyses answer questions about the relative rate at which an event occurs over time rather than simply whether an event occurred by a fixed date.

EndpointHR95% CIP-valueRegistry direction
Time to Improvement in Clinical Status 1.11 0.90–1.37 0.330 HR > 1 favors ruxolitinib 5 mg
Duration of Hospitalization 1.04 0.84–1.28 0.738 HR > 1 favors ruxolitinib 5 mg
Time to Hospital Discharge or to a NEWS2 Score of ≤2 1.02 0.84–1.23 0.869 HR > 1 favors ruxolitinib 5 mg

Time to Improvement in Clinical Status

Hazard ratio

1.11

95% CI: 0.90–1.37   ·   P = 0.330

The estimate of 1.11 corresponds to an approximately 11% higher estimated instantaneous rate of improvement in the ruxolitinib 5 mg group under the fitted model. The confidence interval spans both values below and above 1, so the estimate has substantial uncertainty. The P-value does not quantify the size of the estimated difference.

Duration of Hospitalization

Hazard ratio

1.04

95% CI: 0.84–1.28   ·   P = 0.738

The registry uses a competing-risk framework for this analysis and states that an HR greater than 1 favors ruxolitinib 5 mg. An estimate of 1.04 is therefore close to the no-difference value of 1.00 on the reported hazard-ratio scale.

Time to Hospital Discharge or to a NEWS2 Score of ≤2

Hazard ratio

1.02

95% CI: 0.84–1.23   ·   P = 0.869

The estimated hazard ratio is close to 1.00, with a confidence interval that includes 1.00. The analysis therefore does not establish a precise directional difference from the registry's reported estimate and uncertainty interval.

9. Continuous Clinical-Status Outcome

The registry also reports the mean change from baseline in clinical status using ANCOVA. Two analyses correspond to the two specified post-baseline time points.

Time pointEffect measureMean difference95% CIP-value
Day 15 LS mean -0.03 -0.31–0.25 0.831
Day 29 LS mean 0.08 -0.23–0.38 0.624

Day 15

ANCOVA mean difference

-0.03

95% CI: -0.31–0.25   ·   P = 0.831

The estimated between-group difference in least-squares mean change was -0.03. The confidence interval extends from -0.31 to 0.25, crossing zero. The statistical interpretation therefore depends on the scale's direction as well as the model-based estimate; the ClinicalTrials.gov record does not provide the scale coding needed to translate the sign into a clinical-status direction.

Day 29

ANCOVA mean difference

0.08

95% CI: -0.23–0.38   ·   P = 0.624

At Day 29, the estimated mean difference was 0.08, with a 95% confidence interval from -0.23 to 0.38. Because the interval includes zero, the data represented by this analysis are compatible with differences in either direction around the point estimate.

10. Statistical Methods Explained

Why was logistic regression used for the primary endpoint?

The primary endpoint is binary: the composite outcome either occurred or did not occur during the registered Day 1 through Day 29 period. Logistic regression is designed for this structure and produces an odds ratio comparing the modeled odds between the randomized treatment groups.

What does an odds ratio of 0.91 mean?

An OR of 0.91 means the estimated odds in the ruxolitinib 5 mg group were about 9% lower than the estimated odds in the placebo group. It does not mean the probability of the event was 9% lower. Odds and probabilities are related but are not interchangeable, particularly when outcomes are not rare.

Why is the confidence interval more informative than the point estimate alone?

The point estimate is only one estimate of the treatment contrast. The 95% confidence interval of 0.48–1.73 for the primary endpoint shows how uncertain that estimate is within the stated statistical framework. A point estimate of 0.91 by itself could give a misleading impression of precision.

What does the P-value of 0.769 tell us?

The P-value describes the compatibility of the observed data with the null hypothesis under the specified model and testing framework. It is not the probability that the null hypothesis is true, and it is not a measure of treatment magnitude. Effect size and uncertainty should be read from the odds ratio and confidence interval.

Why was a Cox proportional-hazards model used?

The registry classifies time to improvement in clinical status, duration of hospitalization, and time to hospital discharge or to a NEWS2 score of ≤2 as time-to-event outcomes. A Cox model allows treatment groups to be compared through a hazard ratio while accounting for the timing of events and censoring.

What does a hazard ratio of 1.11 mean?

For time to improvement in clinical status, the registry reports an HR of 1.11. Because the registry states that HR values above 1 favor ruxolitinib 5 mg for this analysis, the estimate corresponds to an approximately 11% higher modeled instantaneous rate of improvement. It does not mean that patients improve 11% faster in a simple arithmetic sense, nor does it imply an 11-percentage-point increase in the probability of improvement.

Why does the competing-risk framework matter?

The registry states that the between-group comparisons for the reported time-to-event outcomes used a competing-risk framework. When another event can prevent the event of interest from occurring in the same way, ordinary time-to-event reasoning can require modification. The competing-risk framework explicitly recognizes that structure rather than treating every participant as if all possible event pathways were interchangeable.

11. Secondary Outcome Interpretation

Clinical-status improvement

The two-point improvement estimates were 0.89 at Day 15 and 1.00 at Day 29. Their confidence intervals were 0.55–1.46 and 0.52–1.92, respectively.

Clinical-status deterioration

The one-point deterioration estimates were 0.75 at Day 15 and 1.18 at Day 29, with confidence intervals of 0.31–1.83 and 0.40–3.49.

Mortality

The reported odds ratios were 0.94 at Day 15 and 1.21 at Day 29. Both confidence intervals are wide: 0.20–5.57 and 0.35–5.11.

Mechanical ventilation

The reported odds ratio was 0.99, with a 95% CI of 0.45–2.21 and a P-value of 0.987.

Several of the secondary binary estimates have confidence intervals spanning relatively broad ranges. This is especially important when interpreting rare outcomes such as mortality, where the odds-ratio scale can become statistically unstable and imprecise. The point estimate should therefore not be separated from its confidence interval.

12. Safety Results

The registry data provide serious adverse events by randomized treatment arm as affected participants over the stated at-risk counts.

Safety measureRuxolitinib 5 mgPlacebo
Serious adverse events 31/281 15/143

These counts should be read as reported rather than converted into percentages. The denominators differ from the overall enrollment figure of 432, so the serious-adverse-event counts should not be interpreted as though all 432 enrolled participants formed the denominator for this particular safety summary.

Safety versus efficacy: serious adverse events are a separate dimension of trial evidence. They should not be combined mathematically with the primary efficacy odds ratio to create a single overall treatment score. The reported safety counts describe affected participants and participants at risk; they do not by themselves establish causality or severity beyond the registry's classification as serious adverse events.

13. Analysis Populations

Endpoint / analysisAnalysis population
Primary composite endpoint Randomized participants excluding those who developed respiratory failure and/or required ICU at randomization.
Clinical-status improvement and deterioration Randomized participants with a valid assessment of clinical status at baseline.
Time to improvement in clinical status Randomized participants with a valid assessment of clinical status at baseline.
Mean change from baseline in clinical status Randomized participants with a valid assessment for the outcome measure.
Mortality rate All randomized participants including those who did not receive any dose, as per intent-to-treat principle, and excluding patients lost to follow-up.
Mechanical ventilation, hospitalization, discharge/NEWS2, oxygen therapy Randomized participants with a valid assessment for the outcome measure.

This distinction is important. The randomized population is the foundation of an intention-to-treat comparison, but individual endpoint analyses can have different valid-assessment requirements. Consequently, an effect estimate for one endpoint should not automatically be assumed to have exactly the same analysis population as every other endpoint.

14. Multiplicity and Multiple Secondary Endpoints

The registry reports one primary endpoint and multiple secondary analyses. The ClinicalTrials.gov record identifies the hypothesis type as superiority but do not provide an alpha-spending scheme, multiplicity adjustment procedure, or endpoint hierarchy for the secondary analyses.

FeatureWhat the ClinicalTrials.gov record establishes
Primary endpointOne registered binary composite endpoint
Primary hypothesis typeSuperiority
Secondary analysesMultiple binary, continuous, and time-to-event outcomes
Statistical methodsLogistic regression, Cox proportional-hazards model, ANCOVA
Multiplicity adjustmentNot reported in the ClinicalTrials.gov record
Alpha-spending procedureNot reported in the ClinicalTrials.gov record

Accordingly, the secondary P-values should be interpreted as the P-values attached to their individual reported analyses. Without a documented multiplicity strategy in the ClinicalTrials.gov record, they should not automatically be interpreted as establishing independent confirmatory findings across the entire collection of secondary endpoints.

15. Confidence Intervals and Precision

The RUXCOVID results illustrate why confidence intervals are essential when interpreting effect estimates.

OutcomeEstimate95% CINull value
Primary composite endpointOR 0.910.48–1.731.00
Two-point improvement, Day 15OR 0.890.55–1.461.00
Two-point improvement, Day 29OR 1.000.52–1.921.00
Time to improvementHR 1.110.90–1.371.00
Mean change, Day 15-0.03-0.31–0.250
Mean change, Day 290.08-0.23–0.380

For odds ratios and hazard ratios, the conventional null value is 1. For a mean difference, the null value is 0. Several of the reported intervals include their corresponding null values, which is consistent with the reported two-sided P-values being above conventional significance thresholds. More importantly, the widths of the intervals show that a point estimate should not be treated as a precise measurement of the underlying treatment effect.

16. What the Odds Ratio Does — and Does Not — Mean

Statistical interpretation

The primary odds ratio of 0.91 means that the modeled odds of the composite endpoint were approximately 9% lower in the ruxolitinib 5 mg group than in the placebo group.

It does not mean that the probability of the composite endpoint was exactly 9% lower, that 9% of participants benefited, or that every patient experienced the same proportional change.

Why the confidence interval matters

The 95% CI of 0.48–1.73 indicates considerable uncertainty around the primary odds-ratio estimate. Because it includes 1.00, the data represented by the analysis are compatible with odds lower than, close to, or higher than the placebo group under the specified model.

Why the P-value is not an effect-size measure

The P-value of 0.769 should not be interpreted as a 76.9% probability that there is no treatment effect. Nor does it indicate that the treatment effect is small. Effect magnitude is described by the estimate; statistical precision is described by the confidence interval; the P-value addresses compatibility with a null hypothesis under the specified testing framework.

17. Reading the Time-to-Event Analyses

Time-to-event endpoints contain more information than a simple yes/no assessment because they incorporate the timing of events. A participant who experiences an event early and one who experiences it later are not treated as identical observations when the analysis properly uses event times.

The RUXCOVID registry analyses use Cox proportional-hazards models and report hazard ratios. The proportional-hazards assumption is therefore an important model consideration. If the relative hazard changes materially over time, a single hazard ratio can become a less complete summary of the treatment contrast.

Hazard ratio

A relative measure of the instantaneous event rate under the fitted Cox model. It is not the same as a risk ratio or absolute time difference.

Censoring

Time-to-event analyses can use information up to the point at which a participant's event status is no longer observed, subject to the assumptions of the analysis.

Competing risks

The registry explicitly identifies a competing-risk framework for the reported time-to-event comparisons.

Kaplan-Meier

Kaplan-Meier estimation is a standard descriptive method for time-to-event distributions. The registry-reported RUXCOVID method fields specifically identify Cox proportional-hazards modeling rather than reporting a Kaplan-Meier analysis.

18. Missing Data and Assessment Requirements

The ClinicalTrials.gov record does not specify a detailed imputation strategy, such as multiple imputation, last observation carried forward, or a prespecified missing-not-at-random sensitivity analysis.

What the registry does specify is that several analyses require participants to have a valid assessment for the relevant outcome, while the mortality analysis explicitly uses an intention-to-treat population and excludes patients lost to follow-up. These distinctions matter because excluding participants with unavailable outcome assessments can change the population contributing information to an analysis.

Interpretation caution: the absence of a detailed imputation method in the ClinicalTrials.gov record should not be interpreted as evidence that no missing-data procedures existed in the underlying statistical analysis plan. It means only that the ClinicalTrials.gov record does not provide that detail.

19. Randomization and Blinding

Allocation
Randomized
Participants were assigned to one of two parallel treatment groups.
Masking
Double
The registry classifies the trial as double-blind.
Purpose
Treatment
The primary purpose is registered as treatment.
Hypothesis
Superiority
The posted statistical analyses are identified as superiority analyses.

Randomization is the key design feature supporting a causal comparison between treatment assignments, provided the randomized comparison is analyzed appropriately. Double blinding provides an additional design safeguard against differential behavior or assessment based on knowledge of assignment.

20. Trial Timeline

2020-05-02

Trial start

The registry lists May 2, 2020 as the study start date.

2020-10-17

Primary completion

The registry lists October 17, 2020 as the primary completion date.

Completed

Registry status

The RUXCOVID study is listed as completed, with results posted on ClinicalTrials.gov.

21. Statistical Profile of the Reported Results

Analysis familyRUXCOVID useReported effect measure
Logistic regression Primary composite and multiple binary secondary outcomes Odds ratio
Cox proportional-hazards model Time to improvement, duration of hospitalization, discharge/NEWS2 endpoint Hazard ratio
ANCOVA Mean change from baseline in clinical status Mean difference / LS mean

This combination of methods is coherent with the endpoint structures reported in the registry: binary outcomes are modeled as categorical responses, time-to-event outcomes use a survival-analysis model, and the continuous change-from-baseline endpoint uses a linear-model approach.

22. Important Limitations and Interpretation Issues

23. Why This Trial Matters Statistically

RUXCOVID is a useful teaching case because it combines three major clinical-trial analysis frameworks in one randomized phase 3 study: logistic regression for binary outcomes, Cox proportional-hazards modeling for time-to-event outcomes, and ANCOVA for continuous change from baseline.

ConceptHow it appears in RUXCOVID
RandomizationRandomized, parallel-group phase 3 design
BlindingDouble-blind study
Binary endpointPrimary composite endpoint over Day 1 - Day 29
Logistic regressionPrimary and multiple secondary binary analyses
Odds ratioPrimary effect measure and secondary binary effect measure
Confidence intervalReported around the odds ratios, hazard ratios, and mean differences
P-valueReported for the primary and secondary analyses
Time-to-event analysisTime to improvement, hospitalization, and discharge/NEWS2 outcomes
Cox modelUsed for reported time-to-event analyses
Hazard ratioEffect measure for time-to-event outcomes
Competing risksUsed for the reported between-group time-to-event comparisons
ANCOVAMean change from baseline in clinical status
Intention-to-treatExplicitly identified for mortality analysis
Safety analysisSerious adverse events reported by treatment arm

24. Clinical Interpretation vs Statistical Interpretation

Statistical interpretation

The primary logistic-regression analysis produced an odds ratio of 0.91 with a 95% CI of 0.48–1.73 and a P-value of 0.769. The interval includes the null value of 1.00, and the estimate should be interpreted together with its uncertainty.

Endpoint interpretation

The secondary analyses address several distinct dimensions of clinical status, including improvement, deterioration, mortality, mechanical ventilation, hospitalization, discharge/NEWS2 status, oxygen therapy, and continuous change in clinical status.

Model interpretation

Odds ratios, hazard ratios, and ANCOVA mean differences describe different mathematical contrasts. They should not be placed on a common scale or interpreted as interchangeable measures of treatment effect.

Safety interpretation

The serious-adverse-event counts of 31/281 and 15/143 provide a separate safety description and should be interpreted independently of the efficacy estimates.

25. Statistical Methods Summary

QuestionMethodEffect measure
Did the primary binary composite outcome differ between treatment groups? Logistic regression OR 0.91 (95% CI 0.48–1.73)
Did the odds of binary clinical outcomes differ? Logistic regression Odds ratios
Did the rate of reaching a time-to-event endpoint differ? Cox proportional-hazards model Hazard ratios
Did mean clinical status change differ? ANCOVA Mean difference / LS mean

26. Related Tutorials

Learn more about the methods used in this trial:

27. Related Calculators

28. Sources

Continue through Clinical Biostats

Use the related tutorials and statistical calculators to explore the regression, survival-analysis, confidence-interval, and clinical-trial concepts represented in RUXCOVID.

29. Record Summary

RUXCOVID provides a compact example of how different clinical-trial endpoint structures require different statistical models. The primary composite endpoint is binary and was analyzed with logistic regression, producing an odds ratio of 0.91 with a 95% CI of 0.48–1.73 and a P-value of 0.769. Secondary outcomes extend the analysis to binary clinical-status measures, mortality, mechanical ventilation, oxygen therapy, continuous change from baseline, and several time-to-event endpoints analyzed with Cox proportional-hazards models using a competing-risk framework.

The most important statistical lesson is that the estimates cannot be interpreted in isolation. Odds ratios describe binary-outcome odds, hazard ratios describe modeled instantaneous event rates, and ANCOVA estimates describe differences in adjusted mean change. Confidence intervals quantify uncertainty around each estimate, while P-values address compatibility with a null hypothesis rather than treatment magnitude. Analysis populations also differ by endpoint, making it important to identify exactly which participants contributed to each reported comparison.

Clinical Biostats methodology: A trial-results page should not merely repeat reported estimates. The goal is to reconstruct the statistical story of the trial while clearly distinguishing the registry's reported analyses from the educational interpretation of effect measures, uncertainty, analysis populations, model assumptions, and endpoint structure.