← Clinical Trials
COVID-19-induced Pneumonia Phase 3 Randomized NCT04362813

CAN-COVID: Complete Statistical Analysis of Canakinumab in COVID-19-induced Pneumonia

An independent statistical review of the randomized, double-blind phase 3 CAN-COVID trial evaluating canakinumab versus placebo in participants with cytokine release syndrome in COVID-19-induced pneumonia, with emphasis on its binary primary endpoint and logistic-regression analysis.

Trial status: COMPLETED  ·  Enrollment: 454  ·  Primary completion: 2020-09-16
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. Trial-specific numerical results on this page are limited to the ClinicalTrials.gov record.

1. Trial at a Glance

CAN-COVID was a randomized, double-blind, parallel-group phase 3 trial evaluating canakinumab versus placebo in participants with cytokine release syndrome in COVID-19-induced pneumonia. The registry reports an enrollment of 454 participants, two treatment arms, one registered primary endpoint, and formal statistical analyses using logistic regression.

454
Enrolled
Phase 3
2
Treatment arms
Canakinumab vs placebo
1.39
Primary OR
95% CI 0.76–2.54
0.2874
Primary p-value
Two-sided
FeatureCAN-COVID
Trial nameCAN-COVID
NCT identifierNCT04362813
PhasePhase 3
ConditionCytokine Release Syndrome (CRS) in Patients With COVID-19-induced Pneumonia
AllocationRandomized
Design modelParallel
MaskingDouble
Primary purposeTreatment
Enrollment454
Lead sponsorNovartis Pharmaceuticals
Sponsor typeIndustry
Trial statusCompleted
Start2020-04-30
Primary completion2020-09-16

2. Clinical Question

The central statistical question was whether participants randomized to canakinumab had higher odds than participants randomized to placebo of surviving without requiring invasive mechanical ventilation from Day 3 to Day 29, using the registry's prespecified binary primary endpoint.

Population

Participants with cytokine release syndrome in COVID-19-induced pneumonia.

Intervention

Canakinumab.

Comparator

Placebo.

Primary question

Does canakinumab increase the odds of surviving without requiring invasive mechanical ventilation from Day 3 to Day 29?

3. Trial Design

01
Randomize454 participants enrolled
02
Two armsCanakinumab vs placebo
03
Double blindMasked treatment assignment
04
Assess9-point ordinal scale
05
AnalyzeLogistic regression
INTERVENTION

Canakinumab

  • Drug intervention
  • Randomized treatment assignment
  • Included in the primary comparison against placebo
COMPARATOR

Placebo

  • Drug comparator
  • Randomized treatment assignment
  • Reference group for the reported odds ratio
Why the design matters statistically. Randomization creates the framework for comparing treatment groups without relying on observational adjustment alone. Double masking is intended to reduce the potential influence of treatment knowledge on participant behavior, clinical management, assessment, and other aspects of trial conduct.

4. Endpoint and Assessment Framework

EndpointTime frameTypeRegistry definition
Participants Who Survived Without Requiring Invasive Mechanical Ventilation From Day 3 to Day 29, Primary Analysis Day 3 to Day 29 Binary Number of responders who survived without requiring invasive mechanical ventilation from Day 3 to Day 29. An early dropout without requiring invasive mechanical ventilation is considered as a responder if discharged from hospital with 9-point ordinal scale ≤1 or with last 9-point ordinal scale on or after Day 15 better than baseline.

The registry describes the primary endpoint as a binary outcome. In practical terms, each participant contributes to a responder/non-responder classification according to the endpoint definition rather than contributing a continuous measurement.

Primary endpoint structure
Responder = survived without requiring invasive mechanical ventilation from Day 3 to Day 29

The registry's endpoint definition also specifies how certain early dropouts are classified, using hospital discharge status or the participant's last available 9-point ordinal scale assessment.

5. Analysis Population and Covariate Adjustment

The primary analysis population was defined as randomized participants with at least one assessment of the 9-point ordinal scale between Day 3 and Day 29. The registry specifies that the ordinal scale ranges from 0, representing uninfected, to 8, representing death.

Analysis componentRegistry specification
PopulationRandomized participants with at least one 9-point ordinal scale assessment between Day 3 and Day 29
Groups comparedCanakinumab vs placebo
Regression methodLogistic regression
Effect measureOdds ratio
Hypothesis typeSuperiority
Adjustment variablesTreatment, region (North America vs Europe), and baseline 9-point ordinal scale (≤4, ≥5)

This distinction between the randomized population and the specific analysis population is important. The registry does not define the primary analysis simply as all 454 enrolled participants. Instead, the posted primary analysis uses randomized participants who had at least one qualifying ordinal-scale assessment during the specified Day 3 to Day 29 period.

6. Primary Results

The registry reports a formal statistical analysis for the primary binary endpoint using logistic regression adjusted for treatment, region, and baseline 9-point ordinal scale.

Odds ratio for the primary endpoint

1.39

95% CI: 0.76–2.54   ·   P = 0.2874

Two-sided superiority analysis; canakinumab versus placebo.

Primary endpointCanakinumab vs placebo
Effect measureOdds ratio
Estimate1.39
95% confidence interval0.76–2.54
P-value0.2874
ModelLogistic regression
AdjustmentTreatment, region, and baseline 9-point ordinal scale
Clinical Biostats interpretation

An odds ratio of 1.39 means that the estimated odds of meeting the primary responder definition were 1.39 times as high in the canakinumab group as in the placebo group, based on the fitted logistic-regression model. Expressed as a simple relative comparison of odds, 1.39 corresponds to estimated odds that are 39% higher in the canakinumab group.

The odds ratio does not mean that 39% more participants survived, that the probability of response was 39% higher, or that an individual participant's probability of avoiding invasive mechanical ventilation increased by 39%. Odds and probabilities are different quantities, particularly when the event is not rare.

The 95% confidence interval of 0.76–2.54 describes uncertainty around the estimated odds ratio under the statistical model and sampling framework. Because the interval includes 1, the data are compatible with both a lower and a higher odds of the primary outcome under the range represented by this interval.

The p-value of 0.2874 is a measure of compatibility between the observed data and the null hypothesis used for the statistical test. It is not a measure of effect size, clinical importance, or the probability that the treatment works. A p-value also does not tell us the probability that the null hypothesis is true.

The interpretation is further conditioned on the analysis population and the prespecified definition of the binary endpoint. Because logistic regression was adjusted for region and baseline ordinal-scale category, the reported odds ratio is a model-based adjusted estimate rather than a simple unadjusted ratio of two observed response proportions.

7. Why Logistic Regression Was Used

The primary endpoint is binary: each participant is classified according to whether the specified responder definition was met. Logistic regression is a standard method for modeling a binary outcome while simultaneously estimating the association between treatment and the outcome and adjusting for prespecified covariates.

Conceptual logistic model
logit[P(Y = 1)] = β0 + β1Treatment + β2Region + β3Baseline ordinal category

The treatment coefficient is transformed to an odds ratio. The actual fitted coefficients are not reported in the ClinicalTrials.gov record; the registry reports the adjusted odds ratio of 1.39.

For this trial, the adjustment variables were treatment, region, and baseline 9-point ordinal scale category. This structure allows the treatment comparison to account for the specified baseline and regional factors rather than relying only on a crude two-group comparison.

8. What an Odds Ratio of 1.39 Means

An odds ratio compares odds, not probabilities. If the odds of response in the placebo group were represented by a value \(O\), an odds ratio of 1.39 would correspond to estimated odds of \(1.39O\) in the canakinumab group, under the fitted model.

What it says

The fitted model estimated higher odds of meeting the primary responder definition with canakinumab than with placebo, with an odds ratio of 1.39.

What it does not say

It does not directly state the difference in responder probabilities between the two groups.

Why the CI matters

The 95% CI of 0.76–2.54 spans 1, so the estimate is uncertain enough to include both directions of an odds-ratio comparison around the null value.

Why the p-value matters differently

The p-value of 0.2874 addresses the statistical test; it should not be used as a substitute for the effect estimate or its confidence interval.

9. Secondary Endpoint Result: COVID-19-related Death

The registry also reports a formal analysis for the secondary endpoint COVID-19-related Death After Study Treatment, assessed at 29 days.

Odds ratio for COVID-19-related death

0.67

95% CI: 0.30–1.50   ·   P = 0.3303

Two-sided superiority analysis; canakinumab versus placebo.

Secondary endpointReported analysis
OutcomeCOVID-19-related Death After Study Treatment
Time frame29 days
Effect measureOdds ratio
Estimate0.67
95% confidence interval0.30–1.50
P-value0.3303
ModelLogistic regression
AdjustmentTreatment, region (North America vs Europe), and baseline 9-point ordinal scale (≤4, ≥5)
Clinical Biostats interpretation

An odds ratio of 0.67 corresponds to estimated odds of COVID-19-related death that were 33% lower in the canakinumab group than in the placebo group, under the fitted logistic-regression model. This is a relative comparison of odds, not a 33-percentage-point reduction in mortality probability.

The 95% CI of 0.30–1.50 is wide relative to the point estimate and includes 1. The interval therefore includes values compatible with lower odds as well as values compatible with higher odds under the model.

The p-value of 0.3303 does not measure the magnitude of the estimated effect. It describes the statistical evidence against the null hypothesis in the specified analysis and should be considered together with the odds ratio and confidence interval.

The analysis population was defined according to the intent-to-treat principle, including all randomized participants who did not receive any dose, with an additional registry-specified rule excluding early dropouts if their last available 9-point ordinal scale was greater than 1, including non-available cases as described in the registry analysis record.

10. Primary and Secondary Analysis Side by Side

FeaturePrimary endpointSecondary endpoint
Outcome Survived without requiring invasive mechanical ventilation from Day 3 to Day 29 COVID-19-related death after study treatment
Time frame Day 3 to Day 29 29 days
Endpoint type Binary Binary
Method Logistic regression Logistic regression
Effect measure Odds ratio Odds ratio
Estimate 1.39 0.67
95% CI 0.76–2.54 0.30–1.50
P-value 0.2874 0.3303

The two estimates point in different directions numerically, but their confidence intervals are broad and both include the null odds ratio of 1. The registry's reported analyses therefore need to be read as estimates with uncertainty rather than as definitive evidence based on the point estimates alone.

11. Intention-to-Treat Analysis

The secondary mortality analysis explicitly states that the analysis population included all randomized participants including those who did not receive any dose, as per intent-to-treat principle, subject to the registry's specified early-dropout rule.

The intention-to-treat principle preserves the connection between the analysis and the original randomized treatment assignment. This is particularly important in randomized trials because post-randomization treatment changes, discontinuation, or nonreceipt of treatment can otherwise create differences between the groups that were not generated by the original randomization.

Primary versus secondary population: the ClinicalTrials.gov record does not describe the primary analysis population simply as "all randomized participants." The primary analysis was restricted to randomized participants with at least one qualifying 9-point ordinal-scale assessment between Day 3 and Day 29. The secondary death analysis explicitly invokes the intent-to-treat principle with its own registry-specified early-dropout rule.

12. The Role of the 9-point Ordinal Scale

The primary analysis uses the 9-point ordinal scale in two related ways: the primary endpoint definition depends on ordinal-scale assessments, and baseline ordinal-scale category is included as an adjustment variable in the logistic regression.

Use of the ordinal scaleRole in the analysis
Primary endpointHelps determine whether an early dropout meets the registry's responder definition.
Baseline adjustmentBaseline scale is categorized as ≤4 versus ≥5 and included in the logistic-regression adjustment.
Analysis populationParticipants needed at least one ordinal-scale assessment between Day 3 and Day 29 for the primary analysis.
Scale range reported by registry0 = uninfected; 8 = death.

This is an example of why endpoint definition and statistical model cannot be separated. The outcome is binary, but its classification incorporates information from an ordinal clinical scale, and the baseline value of that same scale is included as a covariate.

13. Confidence Intervals and Statistical Uncertainty

A confidence interval is more informative than a point estimate alone because it shows the range of parameter values that remain compatible with the data under the specified inferential framework.

Primary OR: 1.39

The point estimate is above 1, corresponding to higher estimated odds with canakinumab.

Primary 95% CI: 0.76–2.54

The interval crosses 1, so the null odds ratio is within the reported uncertainty interval.

Secondary OR: 0.67

The point estimate is below 1, corresponding to lower estimated odds of COVID-19-related death.

Secondary 95% CI: 0.30–1.50

The interval also crosses 1, so the uncertainty includes the null odds ratio.

It is important not to interpret the width of a confidence interval as a probability statement about where the true effect lies for this particular trial. Instead, the interval quantifies uncertainty associated with the estimated parameter under the statistical framework used to construct it.

14. P-values: What They Tell Us and What They Do Not

The primary analysis reported a two-sided p-value of 0.2874. The secondary COVID-19-related death analysis reported a two-sided p-value of 0.3303.

Null-value framework
Odds ratio = 1  →  no difference in odds between treatment groups

The p-value evaluates evidence against the null hypothesis within the specified model and testing framework. It does not measure the size of the observed odds ratio and does not provide the probability that the null hypothesis is true.

A small p-value would not automatically imply a large or clinically important treatment effect, and a larger p-value does not prove that treatment groups are identical. The confidence interval and effect estimate provide essential complementary information.

15. Statistical Methods Explained

Why was logistic regression used?

Because the primary endpoint is binary. Logistic regression is designed for outcomes that can be represented as two categories and allows treatment effects to be estimated while adjusting for specified covariates. Here, the registry reports adjustment for region and baseline 9-point ordinal scale in addition to treatment.

What does an odds ratio of 1.39 mean?

It means that the model estimated the odds of meeting the primary responder definition to be 1.39 times as high with canakinumab as with placebo. It does not mean that the probability of response was 39 percentage points higher.

Why is an odds ratio of 1 the null value?

An odds ratio of 1 means that the estimated odds are the same in the two groups. Values above 1 indicate higher odds in the numerator group, while values below 1 indicate lower odds.

Why adjust for baseline ordinal scale?

The registry specifies baseline 9-point ordinal scale category as an adjustment factor, with categories ≤4 and ≥5. Including this prespecified baseline factor can account for differences in baseline clinical status represented by the ordinal scale when estimating the treatment association.

Why does the confidence interval matter more than the point estimate alone?

The point estimate is only one estimate of the treatment association. The confidence interval shows how uncertain that estimate is under the model and data. For the primary analysis, the interval of 0.76–2.54 includes 1, indicating substantial uncertainty about the direction and magnitude of the odds ratio.

Why is the p-value not a measure of effect size?

The p-value addresses the statistical evidence against a null hypothesis; it does not quantify how large the treatment effect is. Effect size is represented here by the odds ratio, while the confidence interval describes uncertainty around that estimate.

Why does the analysis population matter?

The statistical interpretation applies to the participants included in the specified analysis population. For the primary endpoint, this means randomized participants with at least one qualifying ordinal-scale assessment between Day 3 and Day 29. A treatment estimate cannot automatically be generalized to participants who were not represented in the analysis.

16. Blinding and Randomization as Statistical Protection

Two design features are especially important for interpreting the reported regression estimates: randomization and double masking.

Randomization
Random assignment provides the foundation for comparing the treatment groups and reduces systematic allocation differences that could otherwise confound the treatment comparison.
Double masking
Double masking reduces the opportunity for knowledge of treatment assignment to influence participant behavior, treatment delivery, assessment, or other trial processes.
Parallel design
The registry identifies the study as a parallel-group design, so participants remain associated with their randomized treatment comparison rather than sequentially receiving both randomized interventions.
Superiority hypothesis
The posted analyses are identified as superiority analyses. The question is therefore whether the treatment groups differ, rather than whether one treatment is merely not worse than another within a non-inferiority margin.

17. Safety Results

The ClinicalTrials.gov record reports serious adverse events by treatment arm as the number affected divided by the number at risk.

Safety measureCanakinumabPlacebo
Serious adverse events47/22553/223
Serious adverse events: affected / at risk
Canakinumab
47/225
Placebo
53/223

The ClinicalTrials.gov record reports serious adverse events by arm but do not provide a formal statistical comparison for this safety measure. The counts should therefore be treated as descriptive safety information rather than as a reported hypothesis test.

Safety denominator matters. The registry reports serious adverse events as affected participants divided by participants at risk: 47/225 for canakinumab and 53/223 for placebo. The denominators are part of the reported information and should not be replaced by the overall enrollment of 454.

18. What the Registry Results Do Not Establish

The posted statistical analyses are informative, but several conclusions would require evidence that is not contained in the ClinicalTrials.gov record.

19. Limitations

20. Why This Trial Matters Statistically

CAN-COVID is a useful teaching example because it demonstrates how a randomized clinical trial can use a binary clinical endpoint rather than a continuous measurement or a conventional time-to-event endpoint, and then use logistic regression to adjust the treatment comparison for prespecified baseline and regional factors.

ConceptHow it appears in CAN-COVID
RandomizationParticipants were randomized to canakinumab or placebo.
BlindingThe trial was double masked.
Parallel designThe registry identifies a parallel-group design.
Binary endpointThe primary outcome classifies participants according to survival without requiring invasive mechanical ventilation from Day 3 to Day 29.
Logistic regressionThe primary and reported secondary analyses used logistic regression.
Odds ratioThe treatment effect was summarized using an odds ratio.
Confidence intervalBoth reported effect estimates include two-sided 95% confidence intervals.
P-valueBoth formal analyses report two-sided p-values under a superiority hypothesis.
Covariate adjustmentThe logistic model was adjusted for treatment, region, and baseline 9-point ordinal scale.
Intention-to-treatThe secondary COVID-19-related death analysis explicitly invokes the intent-to-treat principle.

21. Statistical Interpretation of the Primary Result

Effect estimate

The primary odds ratio of 1.39 is above the null value of 1. On the fitted model scale, the estimated odds of meeting the primary responder definition were higher with canakinumab than with placebo.

Precision

The corresponding 95% CI of 0.76–2.54 is relatively broad and crosses the null value. The point estimate therefore should not be interpreted without the uncertainty interval.

Statistical evidence

The reported two-sided p-value of 0.2874 is not a measure of the size of the observed association. It should be considered alongside the odds ratio, confidence interval, endpoint definition, and analysis population.

Model context

The estimate is an adjusted odds ratio from logistic regression incorporating treatment, region, and baseline 9-point ordinal scale category. It is therefore not simply the ratio of two raw response percentages.

Clinical meaning

The statistical result describes the observed randomized-group comparison for the specified endpoint. It does not by itself establish how an individual participant would respond or establish a treatment recommendation.

22. Primary vs Secondary Evidence

Primary endpoint

Survival without requiring invasive mechanical ventilation from Day 3 to Day 29. Adjusted OR 1.39; 95% CI 0.76–2.54; P = 0.2874.

Secondary endpoint

COVID-19-related death after study treatment at 29 days. Adjusted OR 0.67; 95% CI 0.30–1.50; P = 0.3303.

The two estimates should not be collapsed into a single summary statistic. They represent different binary outcomes and, importantly, have different analysis-population specifications in the ClinicalTrials.gov record.

23. Trial Timeline

2020-04-30

Trial start

The registry lists 2020-04-30 as the study start date.

2020-09-16

Primary completion

The registry lists 2020-09-16 as the primary completion date.

Completed

Registry status

the ClinicalTrials.gov record identifies CAN-COVID as completed and reports results on ClinicalTrials.gov.

24. What Would Normally Be Reported for a Binary Endpoint?

For a binary clinical endpoint, a complete statistical presentation often benefits from showing both the underlying event proportions and the model-based effect estimate. The registry analysis reports the adjusted odds ratio, confidence interval, and p-value, but the ClinicalTrials.gov record does not give the treatment-specific responder counts or responder percentages for the primary endpoint.

ComponentAvailable in the ClinicalTrials.gov recordInterpretive role
Binary endpoint definitionYesDefines exactly what counts as a responder.
Analysis populationYesDefines which randomized participants enter the primary analysis.
Regression methodYesSpecifies the statistical model.
Adjusted odds ratioYes: 1.39Summarizes the modeled treatment association.
95% CIYes: 0.76–2.54Quantifies uncertainty around the estimated odds ratio.
P-valueYes: 0.2874Reports the statistical test result.
Raw responder count by armNot reportedWould show the observed number of responders in each group.
Raw responder percentage by armNot reportedWould provide an absolute probability-scale description.

This distinction is important because an adjusted odds ratio and an absolute response rate answer different questions. The odds ratio describes the modeled relative association on the odds scale, whereas response percentages would describe the observed absolute frequency of the binary outcome.

25. Statistical Methods Summary

MethodRole in CAN-COVID
RandomizationCreates the randomized treatment comparison between canakinumab and placebo.
Double maskingReduces potential influence of treatment knowledge on trial conduct and assessment.
Binary endpoint analysisPrimary endpoint is analyzed as a responder/non-responder outcome.
Logistic regressionModels the probability of the binary outcome while adjusting for specified covariates.
Odds ratioPrimary effect measure; compares modeled odds between canakinumab and placebo.
95% confidence intervalQuantifies uncertainty around the odds-ratio estimate.
Two-sided testingBoth registry-reported formal analyses report two-sided confidence intervals and p-values.
Superiority testingBoth formal analyses are identified as superiority analyses.
Intent-to-treat principleExplicitly identified for the secondary COVID-19-related death analysis.

26. Related Tutorials

Learn more about the methods used in this trial:

27. Related Calculators

28. Sources

Continue through Clinical Biostats

Build from the statistical concepts in this trial with focused tutorials, statistical calculators, and additional clinical-trial analyses.

29. Record Summary

CAN-COVID provides a clear example of a randomized phase 3 trial in which a clinically meaningful binary endpoint was analyzed using logistic regression. The primary endpoint was survival without requiring invasive mechanical ventilation from Day 3 to Day 29, with early-dropout rules incorporating hospital discharge and the 9-point ordinal scale. The reported adjusted odds ratio was 1.39, with a two-sided 95% confidence interval of 0.76–2.54 and a p-value of 0.2874. The registry also reports a secondary analysis of COVID-19-related death at 29 days, with an adjusted odds ratio of 0.67, 95% confidence interval 0.30–1.50, and p-value 0.3303.

The most useful statistical reading of these results is not to focus on any single number. The odds ratio describes the estimated treatment association, the confidence interval describes uncertainty around that estimate, and the p-value describes the statistical test under the specified model. Together with the randomized, double-blind design, the endpoint definitions, the analysis populations, and the covariate adjustment, these elements form the statistical framework for interpreting the posted CAN-COVID results.

Clinical Biostats methodology: A trial-results page should distinguish reported evidence from statistical interpretation. For CAN-COVID, the registry supplies a binary primary endpoint, logistic-regression analyses, adjusted odds ratios, confidence intervals, p-values, analysis-population definitions, and descriptive serious-adverse-event counts. The interpretation above stays within those the ClinicalTrials.gov record rather than inferring unreported response rates, subgroup effects, survival curves, or additional statistical analyses.