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
CANTOS (Cardiovascular Risk Reduction Study, Reduction in Recurrent Major CV Disease Events) was a randomized, parallel-group, triple-masked phase 3 trial that asked whether canakinumab, at any of three doses, lowered the hazard of a first major adverse cardiovascular event (MACE) compared with placebo in participants with atherosclerosis.
| Feature | CANTOS |
|---|---|
| Registry title | Cardiovascular Risk Reduction Study (Reduction in Recurrent Major CV Disease Events) |
| Phase | Phase 3 |
| Condition | Atherosclerosis |
| Design | Randomized, parallel-group, triple-masked, treatment purpose |
| Arms | 4: canakinumab 50 mg, 150 mg, 300 mg, and placebo |
| Interventions listed | Canakinumab (drug); placebo (drug); standard of care (drug) |
| Enrollment | 10066 |
| Primary endpoints | Time to first adjudicated MACE (core phase); two substudy endpoints (carotid plaque burden; insulin secretion rate) |
| Dates | Start 2011-04-11; primary completion 2017-03-28 |
| Status | Completed; results posted |
| ClinicalTrials.gov | NCT01327846 |
| Sponsor | Novartis Pharmaceuticals (industry) |
2. Clinical Question
The central question was whether adding canakinumab to the care of people with established atherosclerosis reduced the rate of recurrent major cardiovascular events. Because the trial randomized three separate doses against a single placebo group, it was simultaneously a superiority trial and an exploration of the dose–response relationship.
Population
Participants with atherosclerosis, as registered on ClinicalTrials.gov. Efficacy was analysed in the Full Analysis Set.
Intervention
Canakinumab at one of three doses: 50 mg (Group III), 150 mg (Group II) or 300 mg (Group I).
Comparator
Placebo (Group IV), under triple masking.
Primary question
Is the hazard of a first adjudication-committee-confirmed MACE lower with each canakinumab dose than with placebo?
3. Trial Design
The registry identifies the arms through the hypothesis statements attached to each analysis: Group I is the canakinumab 300 mg group, Group II the 150 mg group, Group III the 50 mg group and Group IV the placebo group. The numbers below are the participants at risk in the serious adverse event tables; the four at-risk counts sum to the enrollment of 10066.
Canakinumab 300 mg
- Highest dose studied
- Tested against placebo in hypothesis H11
Canakinumab 150 mg
- Middle dose
- Tested against placebo in hypothesis H21
- Dose given to all patients in the Extension phase
Canakinumab 50 mg
- Lowest dose studied
- Tested against placebo in hypothesis H31
Placebo
- Shared control group for all three dose comparisons
- Largest of the four randomized groups
4. Analysis Population
| Analysis population | Definition in the registry | Role |
|---|---|---|
| Full Analysis Set (FAS) | All randomized patients except for those who were misrandomized or from sites with serious GCP violations | All posted efficacy analyses |
| Safety (at risk) | Participants at risk as listed in the serious adverse event tables for each group | Serious adverse event reporting |
The FAS is close to, but not identical with, a pure intention-to-treat population. Excluding misrandomized patients and sites with serious Good Clinical Practice violations is a prespecified, assignment-independent exclusion; because the decision does not depend on treatment received or on outcomes, it generally preserves the protection that randomization provides. It is best described as a modified intention-to-treat analysis.
5. Endpoints
Primary endpoints
| Endpoint | Registry definition | Time frame |
|---|---|---|
| Core phase first CEC-confirmed MACE and its components | Time to occurrence of CEC (cardiovascular clinical events adjudication committee) confirmed MACE, a composite of CEC-confirmed CV death, CEC-confirmed non-fatal MI, or CEC-confirmed non-fatal stroke. Patients with the CEC-adjudicated reason for death of "Unknown" were counted as CV death. | From randomization, to end of treatment plus 30 days, up to approximately 6 years |
| Substudy 1 (core phase) | Change from baseline in carotid plaque burden in the bifurcation region of the index carotid artery | 24 months |
| Substudy 2 (core phase) | Change from baseline of the insulin secretion rate (ISR) relative to glucose 0–30 min, defined as Φ30 = AUCISR 0-30 / AUCGluc 0-30, averaged across the Year 3, 4, 5 visits | From randomization up to approximately 6 years |
Secondary endpoints with posted analyses
| Endpoint | Time frame |
|---|---|
| Core phase CEC-confirmed CV death, non-fatal MI, non-fatal stroke, or hospitalization for unstable angina requiring unplanned revascularization | From randomization, to end of treatment plus 30 days, up to approximately 6 years |
| Core phase new-onset type 2 diabetes among patients with pre-diabetes at randomization | From randomization up to approximately 6 years |
| Core phase all-cause mortality, non-fatal MI, or non-fatal stroke | From randomization, to end of treatment plus 30 days, up to approximately 6 years |
| Core phase all-cause mortality | From randomization, to end of treatment plus 30 days, up to approximately 6 years |
Why adjudication and the "unknown death" rule matter
Every event in the primary composite had to be confirmed by an independent clinical events committee. Central adjudication reduces misclassification, which in a time-to-event analysis tends to dilute the hazard ratio toward 1. Counting deaths of unknown cause as cardiovascular deaths is a conservative, prespecified convention: it avoids the bias that would arise if unexplained deaths were simply dropped, at the cost of including some non-cardiovascular deaths in the endpoint. Because the rule applies identically in every arm and is decided under masking, it should not favour either treatment.
6. Statistical Methodology
Time-to-first-event framework
MACE was analysed as a time-to-first-event endpoint. Each participant contributes follow-up from randomization until the first qualifying event or until censoring at the end of the observation window (end of treatment plus 30 days). Participants who had not experienced a MACE by then are right-censored: they contribute information about the period during which they were known to be event-free.
Log-rank test for the primary hypotheses
For each of the three dose comparisons (H11, H21, H31), the registry lists a log-rank test as the statistical method. The log-rank test compares the observed number of events in each group with the number expected if the hazard were the same in both groups, accumulated across all event times. Its hypotheses were written in a one-directional form, for example for the 150 mg dose:
H1: hazard of first MACE (150 mg) < hazard of first MACE (placebo)
Framing the null hypothesis as "greater than or equal to" makes the test one-directional: only evidence that canakinumab lowers the hazard counts against the null.
Cox proportional-hazards model
Hazard ratios and 95% confidence intervals were estimated with a Cox proportional-hazards model. The registry states that a hazard ratio below 1 favours canakinumab. The same Cox framework was used for the individual components (CV death, MI and stroke) and for the secondary endpoints.
The baseline hazard h0(t) is left unspecified; only the ratio between groups is estimated. That ratio is assumed constant over follow-up.
Multiplicity
Three doses compared with one placebo group, plus several components and secondary endpoints, create many opportunities for a chance finding. The registry states that, beyond the primary endpoint, all p-values for the canakinumab 50 mg and 300 mg groups are to be considered nominal only. In practice this means that the formal, error-controlled chain of inference beyond MACE runs through the 150 mg comparison, and that p-values for the other two doses on secondary endpoints are descriptive. The registry does not post the full alpha allocation or testing sequence.
Kaplan-Meier estimation
Time-to-event endpoints of this kind are normally summarised with Kaplan-Meier curves, which estimate the probability of remaining event-free over time while accounting for censoring. The ClinicalTrials.gov record does not report Kaplan-Meier estimates, cumulative incidence at fixed time points, or median event times for CANTOS, so the results below are expressed as hazard ratios.
7. Primary Results: First Adjudicated MACE
All three canakinumab doses produced point estimates of the hazard ratio below 1. The 150 mg comparison had the smallest p-value, and its 95% confidence interval excluded 1.
Canakinumab 150 mg vs placebo (H21)
95% CI: 0.74–0.98 · log-rank p = 0.0241
Full Analysis Set · Group II vs Group IV
| Comparison | Hypothesis | Hazard ratio | 95% CI (two-sided) | Log-rank p |
|---|---|---|---|---|
| 300 mg vs placebo (Group I vs IV) | H11 | 0.86 | 0.75–0.99 | 0.0648 |
| 150 mg vs placebo (Group II vs IV) | H21 | 0.85 | 0.74–0.98 | 0.0241 |
| 50 mg vs placebo (Group III vs IV) | H31 | 0.93 | 0.80–1.07 | 0.1895 |
What the estimate means. An HR of 0.85 for 150 mg means that, under the Cox model and averaged over the follow-up period, the estimated hazard of a first adjudicated MACE was 15% lower with canakinumab 150 mg than with placebo. For 300 mg (HR 0.86) the estimated hazard was 14% lower, and for 50 mg (HR 0.93) 7% lower.
What it does not mean. It does not mean that 15% fewer participants had an event, that each individual's risk fell by 15%, or that 15% of participants benefited. A hazard ratio is a relative, instantaneous-rate measure; it says nothing directly about absolute risk, and the registry does not report absolute event rates or numbers needed to treat alongside these analyses.
Precision. The 150 mg interval (0.74–0.98) is compatible with anything from a 26% to a 2% lower hazard. The upper limit sits close to 1, so the data are consistent with a modest effect as well as a moderate one. For 50 mg, the interval (0.80–1.07) includes 1 and also includes a small increase in hazard.
Why the p-value is not the effect size. The 300 mg and 150 mg estimates (0.86 and 0.85) are almost identical, yet their p-values differ (0.0648 and 0.0241). A p-value mixes the size of an effect with the amount of information and with the testing procedure; it is not a measure of how large or clinically important the effect is. Two nearly equal estimates can fall on different sides of a significance threshold.
An apparent inconsistency for 300 mg. The 300 mg interval (0.75–0.99) excludes 1, but the posted p-value is 0.0648. The interval comes from the Cox model while the p-value comes from the log-rank test, and the registered hypotheses are one-directional. The p-values may also reflect the trial's prespecified multiplicity framework. The registry does not explain the difference, so the 300 mg result is best described as borderline rather than clearly positive or negative.
Cautions. The single HR assumes the hazard ratio is roughly constant over up to approximately 6 years of follow-up; the registry does not report a check of this assumption. Three comparisons share one placebo group, so some form of multiplicity control applies to the primary family. The analysis population is the FAS, not every randomized participant.
Dose–response pattern
The three point estimates do not fall in a strictly graded order: 50 mg gave 0.93, 150 mg 0.85 and 300 mg 0.86. Doubling the dose from 150 mg to 300 mg did not produce a further reduction in the estimated hazard, and the confidence intervals for all three doses overlap substantially. Overlapping intervals do not prove that the doses are equivalent, and a formal test of trend across doses is not posted on ClinicalTrials.gov. The descriptive pattern is compatible with an effect that plateaus somewhere between the lower and middle doses, but it is also compatible with chance variation around a common effect.
8. Primary Results: MACE Components
The registry also posts Cox-model hazard ratios for each component of MACE. These component analyses describe where the composite effect came from; they are not independently powered tests.
| Component | 300 mg vs placebo | 150 mg vs placebo | 50 mg vs placebo |
|---|---|---|---|
| CV death | 0.94 (0.77–1.16), p = 0.572 | 0.90 (0.73–1.10), p = 0.296 | 0.91 (0.73–1.12), p = 0.369 |
| MI (fatal and non-fatal) | 0.84 (0.69–1.01), p = 0.067 | 0.76 (0.63–0.92), p = 0.006 | 0.94 (0.78–1.14), p = 0.542 |
| MI (non-fatal) | 0.83 (0.68–1.00) | 0.76 (0.62–0.92) | 0.94 (0.78–1.14) |
| Stroke (fatal and non-fatal) | 0.80 (0.56–1.12), p = 0.190 | 0.98 (0.71–1.35), p = 0.912 | 1.03 (0.74–1.43), p = 0.871 |
| Stroke (non-fatal) | 0.80 (0.57–1.13) | 0.99 (0.72–1.37) | 1.04 (0.75–1.45) |
Hazard ratio (95% two-sided CI). No statistical test or p-value is posted for the non-fatal MI and non-fatal stroke analyses.
What the components show. For 150 mg, the component with the clearest separation from placebo was myocardial infarction: HR 0.76 (0.63–0.92) for fatal and non-fatal MI, corresponding to a 24% lower estimated hazard. The CV death estimate (0.90) and the stroke estimates (0.98 and 0.99) were closer to 1 with intervals that include 1. Much of the composite signal therefore appears to come from MI.
What it does not show. A nonsignificant component is not evidence that the treatment has no effect on that component. The stroke intervals are wide (for 300 mg, 0.56–1.12), reflecting relatively few events; they are compatible with meaningful benefit, no effect, or modest harm. Likewise, the apparently lower stroke estimate for 300 mg than for 150 mg should not be read as a dose-specific effect without a formal comparison.
Precision and p-values. Component intervals are wider than the composite interval because each component has fewer events. The p-value of 0.006 for MI at 150 mg is small, but it is one of many component tests, and the component results are descriptive of the primary composite rather than separate confirmatory claims.
Cautions. In a first-event composite, a participant who first has a non-fatal MI and later dies of CV causes counts only once for MACE, but may count in the separate CV death analysis. Component and composite analyses therefore do not add up. Non-cardiovascular death acts as a competing event for the components, which the cause-specific Cox model handles by censoring.
9. Primary Substudy Endpoints
The registry lists two substudy endpoints as primary and marks results as posted for both: change from baseline in carotid plaque burden at 24 months (Substudy 1), and change from baseline in the early-phase insulin secretion rate relative to glucose, Φ30, averaged across the Year 3, 4 and 5 visits (Substudy 2). No formal statistical analyses for either substudy endpoint are posted to ClinicalTrials.gov.
Both are continuous change-from-baseline outcomes. Endpoints of this type are typically analysed with an analysis of covariance (ANCOVA) or a mixed model for repeated measures, adjusting for the baseline value, and summarised as a least-squares mean difference between each dose and placebo with a 95% confidence interval. For Φ30, which is a ratio of two areas under the curve and is often right-skewed, analysis on the log scale with results expressed as a ratio of geometric means is a common alternative. Because these substudies enrolled only a subset of the trial population, their precision would be considerably lower than that of the main MACE analysis.
10. Secondary Endpoint Results
MACE or hospitalization for unstable angina requiring unplanned revascularization
| Analysis | 300 mg vs placebo | 150 mg vs placebo | 50 mg vs placebo |
|---|---|---|---|
| Expanded composite (log-rank) | 0.82 (0.72–0.94), p = 0.0648 | 0.83 (0.73–0.95), p = 0.0241 | 0.90 (0.79–1.03), p = 0.1895 |
| Unstable angina component (Cox) | 0.58 (0.39–0.86), p = 0.007 | 0.64 (0.44–0.94), p = 0.022 | 0.71 (0.48–1.05), p = 0.086 |
Adding unstable-angina hospitalizations with unplanned revascularization to MACE produced hazard ratios of 0.82, 0.83 and 0.90, slightly further from 1 than the primary MACE estimates. The p-values posted for the log-rank analyses of this expanded composite (0.0648, 0.0241 and 0.1895) are identical to those posted for the primary MACE comparisons, even though the hazard ratios differ. Identical p-values for different endpoints are unusual, and readers who need the exact test results for this endpoint should consult the full publications. Under the registry's multiplicity statement, the 300 mg and 50 mg p-values here are nominal.
New-onset type 2 diabetes among patients with pre-diabetes
| Comparison | Hazard ratio (95% CI) | Log-rank p |
|---|---|---|
| 300 mg vs placebo | 1.01 (0.83–1.23) | 0.8456 |
| 150 mg vs placebo | 1.06 (0.87–1.29) | 0.8456 |
| 50 mg vs placebo | 0.98 (0.80–1.20) | 0.6541 |
All three estimates are close to 1, and all three intervals span 1 comfortably. This is a situation in which the confidence interval is far more informative than the p-value: for 150 mg the data are compatible with anything from a 13% lower to a 29% higher hazard of progressing to type 2 diabetes. The result is best described as showing no clear difference, not as demonstrating that canakinumab has no effect on diabetes onset; a trial designed to show the absence of an effect would need a prespecified equivalence or non-inferiority margin, which this endpoint did not have. The same p-value of 0.8456 is posted for both the 300 mg and 150 mg comparisons.
All-cause mortality, non-fatal MI, or non-fatal stroke
| Comparison | Hazard ratio (95% CI) | Cox p |
|---|---|---|
| 300 mg vs placebo | 0.87 (0.77–0.99) | 0.028 |
| 150 mg vs placebo | 0.85 (0.75–0.96) | 0.011 |
| 50 mg vs placebo | 0.94 (0.83–1.07) | 0.377 |
Replacing CV death with all-cause death gives a composite that does not depend on adjudicated cause of death. The estimates (0.87, 0.85, 0.94) closely track the primary MACE results, which suggests the primary finding is not an artefact of how deaths were classified.
All-cause mortality
| Comparison | Hazard ratio (95% CI) | Cox p |
|---|---|---|
| 300 mg vs placebo | 0.93 (0.79–1.10) | 0.406 |
| 150 mg vs placebo | 0.92 (0.78–1.09) | 0.329 |
| 50 mg vs placebo | 0.96 (0.81–1.13) | 0.597 |
All-cause mortality estimates are slightly below 1 for every dose, but every interval includes 1. The trial was not designed or powered to detect a mortality difference, so these results neither establish nor exclude an effect on survival.
11. Safety: Serious Adverse Events
| Group | Participants with ≥1 serious adverse event | Participants at risk |
|---|---|---|
| Group I (canakinumab 300 mg) | 759 | 2263 |
| Group II (canakinumab 150 mg) | 743 | 2285 |
| Group III (canakinumab 50 mg) | 686 | 2170 |
| Group IV (placebo) | 1099 | 3348 |
| Group V | 1216 | 5777 |
Because the placebo group is larger than each dose group, the raw counts of affected participants are not directly comparable; any comparison must be made on proportions of participants at risk. Serious adverse events are recorded regardless of causality, so a proportion in a treatment group reflects both treatment effects and the background illness of the population. The registry posts no formal statistical comparison of serious adverse events between groups. Group V is a separately reported group with 5777 participants at risk; it is not one of the four randomized core-phase comparisons, and the registry notes that all patients in the Extension phase received canakinumab 150 mg.
12. Statistical Methods Explained
Why was a log-rank test used for the primary endpoint rather than a comparison of proportions?
Follow-up in CANTOS lasted up to approximately 6 years and differed between participants. Comparing the proportion with MACE would ignore when events happened and would treat a participant censored early the same as one followed for years without an event. The log-rank test uses every participant's follow-up time and handles censoring directly, which makes it the standard test for a time-to-first-event endpoint in a randomized trial.
What does a hazard ratio of 0.85 mean for the 150 mg dose?
It means that at any given time during follow-up, among participants who had not yet had a MACE, the estimated rate of a first MACE in the 150 mg group was 0.85 times the rate in the placebo group, a 15% relative reduction in the estimated hazard. It is a relative measure; without the underlying event rates, it cannot be translated into how many events were prevented.
How can the 300 mg interval exclude 1 when its p-value is 0.0648?
The confidence interval comes from the Cox model and the p-value from the log-rank test, and the registered hypotheses were one-directional. The two procedures are closely related but are not guaranteed to agree exactly, particularly when a result is near the boundary; a multiplicity-adjusted testing framework could also account for the difference. The registry does not explain it. The practical lesson is that when an interval barely excludes 1 and a p-value is just above 0.05, the evidence is borderline under either summary.
Why are the 50 mg and 300 mg p-values for secondary endpoints called nominal?
A nominal p-value is calculated as if the test stood alone, without accounting for the other tests performed. With three doses and many endpoints, the probability that at least one test crosses 0.05 by chance grows quickly. The registry states that, beyond the primary endpoint, p-values for the 50 mg and 300 mg groups are nominal only, so results such as p = 0.007 for unstable angina at 300 mg should be treated as hypothesis-generating rather than confirmatory.
Why do the three dose comparisons share a placebo group, and does it matter?
Sharing a placebo group is efficient: fewer participants receive placebo than if three separate two-arm trials were run. The cost is that the three comparisons are positively correlated. A chance fluctuation in the placebo group affects every comparison, so three hazard ratios below 1 are less independent confirmation than they might appear.
What does the proportional-hazards assumption mean here?
The Cox model summarises the treatment effect with one number that is assumed to hold at every point in follow-up. If the effect of canakinumab were larger early and smaller later, or took time to emerge, the single hazard ratio would be a weighted average of a changing effect. The registry does not report diagnostics for this assumption, so the HR is best interpreted as an average relative effect over the observed follow-up.
13. Limitations
- Relative measures only: the posted analyses are hazard ratios. Without absolute event rates at fixed times, the clinical magnitude of the effect cannot be judged from the registry alone.
- Multiplicity: three doses, several components and several secondary endpoints were tested. P-values for the 50 mg and 300 mg groups beyond the primary endpoint are nominal, and the full alpha allocation is not posted.
- Borderline primary results: the 150 mg interval upper limit (0.98) and the 300 mg interval upper limit (0.99) lie close to 1, and the 300 mg log-rank p-value (0.0648) is above 0.05.
- Non-monotonic dose response: the 300 mg estimate was not lower than the 150 mg estimate, and no formal trend test is posted.
- Component imprecision: stroke and CV death estimates have wide intervals and cannot establish or exclude an effect on those outcomes.
- Analysis population: the FAS excludes misrandomized patients and sites with serious GCP violations, so it is a modified rather than a strict intention-to-treat population.
- Proportional hazards: no assessment of the proportional-hazards assumption is reported in the registry.
- Extension phase: all patients received canakinumab 150 mg in the Extension phase, so no randomized comparison is possible for that period.
- Reporting irregularities: some posted p-values are identical across different endpoints or comparisons, and several component analyses carry no test statistic; the registry entries should be read together with the full publications.
- Substudies: no formal statistical analyses of the two substudy endpoints are posted to ClinicalTrials.gov.
14. Why This Trial Matters Statistically
CANTOS is a useful teaching case because a single large trial illustrates several recurring questions in cardiovascular outcome trials: how to read borderline hazard ratios, how multiplicity changes the meaning of a p-value, and how multi-arm designs with a shared control complicate interpretation.
| Concept | How it appears in CANTOS |
|---|---|
| Multi-arm design | Three canakinumab doses compared with one shared placebo group |
| Time-to-event endpoint | Time to first adjudicated MACE, with follow-up up to approximately 6 years |
| Log-rank test | Primary test for each dose-versus-placebo hypothesis (H11, H21, H31) |
| Cox model and hazard ratio | Relative effect estimates for MACE, its components and secondary endpoints |
| Confidence intervals | Show that the 150 mg and 300 mg effects are compatible with small as well as moderate reductions |
| P-values vs effect size | Nearly identical HRs (0.85 and 0.86) with different p-values (0.0241 and 0.0648) |
| Multiplicity | Nominal p-values for 50 mg and 300 mg beyond the primary endpoint |
| Composite endpoints | MACE driven mainly by MI, with imprecise estimates for stroke and CV death |
| Analysis population | Full Analysis Set as a modified intention-to-treat population |
| Dose response | Estimates that do not fall in strict dose order |
15. Related Tutorials
Learn more about the methods used in this trial:
16. Related Calculators
17. Sources
- ClinicalTrials.gov: NCT01327846 — Cardiovascular Risk Reduction Study (Reduction in Recurrent Major CV Disease Events).
- PubMed: PMID 42141513.
- PubMed: PMID 38701435.
- PubMed: PMID 37934948.
- PubMed: PMID 35943897.
- PubMed: PMID 35385050.
Keep learning with Clinical Biostats
Explore the statistical methods behind this trial in more depth, or work through them with our calculators.
18. Record Summary
CANTOS randomized 10066 participants with atherosclerosis to one of three canakinumab doses or placebo and followed them for up to approximately 6 years for a first adjudicated major adverse cardiovascular event. The 150 mg dose gave a hazard ratio of 0.85 (95% CI 0.74–0.98; p = 0.0241), the 300 mg dose 0.86 (0.75–0.99; p = 0.0648) and the 50 mg dose 0.93 (0.80–1.07; p = 0.1895). The signal was concentrated in myocardial infarction, with imprecise estimates for stroke and CV death and no clear difference in all-cause mortality or new-onset diabetes. The most useful reading of the trial combines these relative effects with their confidence intervals, the multiplicity structure of a multi-arm design, and an awareness that absolute risk reductions are not reported in the registry.