← Clinical Trials
Transcatheter Aortic Valve Replacement Phase 3 Randomized NCT02556203

GALILEO: Complete Statistical Analysis of Rivaroxaban After Transcatheter Aortic Valve Replacement

An independent statistical analysis of the GALILEO phase 3 trial comparing a rivaroxaban-based antithrombotic strategy with an antiplatelet-based strategy after transcatheter aortic valve replacement, with emphasis on time-to-event methods, non-inferiority testing, hazard ratios, confidence intervals, and bleeding outcomes.

Phase 3  ·  Randomized parallel design  ·  Enrollment 1653  ·  Status: Terminated
Scope of this record

This page separates reported trial results from statistical interpretation. Numerical results and trial characteristics are restricted to the ClinicalTrials.gov data posted on ClinicalTrials.gov for NCT02556203. The registry reports results for both efficacy and bleeding time-to-event endpoints, including separate analyses of the primary death or thromboembolic-event endpoint.

Registry note: 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

GALILEO was a randomized, phase 3, parallel-group, open-label trial evaluating a rivaroxaban-based antithrombotic strategy versus an antiplatelet-based strategy after transcatheter aortic valve replacement.

1653
Enrollment
Randomized trial
2
Arms
Parallel design
1.35
DTE HR
95% CI 1.01–1.81
1.50
PBE HR
95% CI 0.95–2.37
FeatureGALILEO
Trial nameGALILEO
PhasePhase 3
ConditionTranscatheter Aortic Valve Replacement
Therapeutic areaCardiology
DesignRandomized, parallel-group
MaskingNone
Primary purposePrevention
Enrollment1653.0
Arms2
StatusTerminated
Start2015-12-16
Primary completion2018-11-27
Lead sponsorBayer
Sponsor typeIndustry
ClinicalTrials.govNCT02556203

2. Clinical Question

The statistical question was whether a rivaroxaban-based antithrombotic strategy after transcatheter aortic valve replacement could be compared with an antiplatelet-based strategy with respect to death or first thromboembolic event, while also evaluating primary bleeding events and other clinical outcomes.

Population

Participants undergoing transcatheter aortic valve replacement, as represented by the registered condition and trial population.

Intervention

A rivaroxaban-based strategy involving rivaroxaban (Xarelto, BAY59-7939), with the registered intervention set also including acetylsalicylic acid (ASA), clopidogrel, and vitamin K antagonist components.

Comparator

An antiplatelet-based strategy.

Primary question

How did the rivaroxaban-based strategy compare with the antiplatelet-based strategy for death or first thromboembolic event and primary bleeding event?

3. Trial Design

01
Randomize1653 participants
02
Two strategiesRivaroxaban-based vs antiplatelet-based
03
Follow-upTime-to-event outcomes
04
AdjudicationClinical event definitions
05
AnalysisLog-rank and hazard ratios
Allocation
Randomized.
Model
Parallel.
Masking
None.
Primary purpose
Prevention.
RIVAROXABAN-BASED STRATEGY

Rivaroxaban strategy

  • Rivaroxaban (Xarelto, BAY59-7939)
  • Registered intervention set includes acetylsalicylic acid (ASA)
  • Registered intervention set includes clopidogrel
  • Registered intervention set includes vitamin K antagonist (VKA)
ANTIPLATELET STRATEGY

Antiplatelet strategy

  • Antiplatelet-based treatment strategy
  • The registry identifies ASA and clopidogrel among the trial interventions
Important design boundary: the ClinicalTrials.gov record identifies the two randomized strategies and lists the individual interventions, but it does not provide enough detail here to reconstruct a complete treatment schedule, treatment duration, or allocation ratio. Those details are therefore not inferred on this page.

4. Analysis Populations

The registry provides two important analysis populations for the primary time-to-event endpoint. The distinction matters because the non-inferiority analysis and the later superiority analysis use different populations and time frames.

Analysis populationDefinition / role
Safety analysis set (SAF)Included all randomized subjects who had been exposed to study drug at least once; 1608 subjects overall.
Full analysis set (FAS)Included all randomized subjects and results presented according to randomized treatment arm; 1644 subjects overall.
Why this distinction matters: the 14-month DTE non-inferiority analysis was based on the safety analysis set, whereas the 16-month DTE analysis and the primary bleeding analysis used the full analysis set. These are not interchangeable analysis populations, so their estimates should be presented as separate analyses rather than silently combined.

5. Endpoints

EndpointRegistered definitionTime frameType
Death or First Thromboembolic Event (DTE) Death or first adjudicated thromboembolic event (DTE), defined as composite of all-cause death, any stroke, myocardial infarction (MI), symptomatic valve thrombosis, pulmonary embolism (PE), deep vein thrombosis (DVT), and non-central nervous system (CNS) systemic embolism. Through study completion, on average 14 months Binary / time-to-event in posted analysis
Death or First Thromboembolic Event (DTE) Death or first adjudicated thromboembolic event (DTE), defined as composite of all-cause death, any stroke, myocardial infarction (MI), symptomatic valve thrombosis, pulmonary embolism (PE), deep vein thrombosis (DVT), and non-central nervous system (CNS) systemic embolism. Through study completion, on average 16 months Binary / time-to-event in posted analysis
Primary Bleeding Event (PBE) PBE is defined according to VARC (Valve Academic Research Consortium) definitions as the adjudicated composite of: Life-threatening, disabling or major bleeding. Through study completion, on average 16 months Binary / time-to-event in posted analysis

The registry therefore contains three posted primary-endpoint analyses: two analyses of DTE at different average follow-up times and one analysis of primary bleeding events. The statistical interpretation should preserve that distinction.

6. Statistical Methodology

Time-to-event analysis

The posted analyses treat the clinical outcomes as time-to-event endpoints. This is important because participants can experience an event at different times, while others may remain event-free through their available follow-up. A time-to-event analysis uses both the occurrence of events and the amount of observed follow-up rather than reducing every participant to a simple yes/no outcome without regard to timing.

Core quantity
HR = estimated hazard in rivaroxaban-based strategy / estimated hazard in antiplatelet-based strategy

An HR above 1 indicates a higher estimated hazard in the rivaroxaban arm compared with the antiplatelet arm, consistent with the registry's analysis notes.

Log-rank testing

The 16-month DTE analysis and the primary bleeding analysis were reported using a log-rank test. The log-rank procedure compares the observed and expected pattern of events between treatment groups across follow-up and is commonly paired with Kaplan-Meier estimation and Cox modeling for time-to-event endpoints.

Cox proportional-hazards modeling

The registry states that the reported hazard-ratio estimates for the posted analyses are based on a Cox proportional hazards model. The analysis notes also explicitly state that HRs greater than 1 indicate a higher hazard rate in the rivaroxaban arm compared with the antiplatelet arm.

Interpretation of the hazard ratio
HR > 1  →  higher estimated hazard in the rivaroxaban arm

The HR is a relative time-to-event measure. It is not an absolute risk difference, a risk ratio at a particular time point, or the percentage of participants who experience an event.

Intention-to-treat analysis

The registry identifies intention-to-treat analysis as an analysis concept for the full analysis set. Under the intention-to-treat principle, randomized participants are analyzed according to randomized treatment assignment. This preserves the comparison created by randomization and is particularly important when interpreting treatment effects in a randomized trial.

Non-inferiority analysis

The 14-month DTE analysis was explicitly identified as a non-inferiority analysis. The registry defines the null hypothesis as H0: HR(t) ≥ 1.20 for all time points t ≥ 0, corresponding to the statement that the hazard for the primary efficacy endpoint in the rivaroxaban-based treatment group is more than 20% larger than that in the antiplatelet-based control group.

Non-inferiority logic in this analysis
H0: HR(t) ≥ 1.20   vs   non-inferiority when the upper confidence bound is below 1.20

The relevant comparison is therefore between the one-sided upper confidence limit and the prespecified 1.20 threshold, rather than simply asking whether a conventional two-sided p-value is below 0.05.

7. Primary Results: Death or First Thromboembolic Event

14-month non-inferiority analysis

The first posted primary DTE analysis evaluated the endpoint through study completion, on average 14 months, in the safety analysis set. The registry reports a one-sided 97.5% confidence interval and a non-inferiority hypothesis.

Hazard ratio for death or first thromboembolic event

1.2089302

One-sided 97.5% CI upper limit: 1.7040824

Analysis population: Safety analysis set, 1608 subjects overall

Hypothesis: Non-inferiority

Clinical Biostats interpretation

The estimated hazard ratio of 1.2089302 means that the estimated hazard of the composite DTE endpoint was higher in the rivaroxaban-based strategy than in the antiplatelet-based strategy under the reported time-to-event model. Expressed descriptively, the point estimate is approximately 20.89302% above a hazard ratio of 1.

The estimate does not mean that 20.89302% more participants experienced the endpoint, nor does it represent an absolute increase in event probability. A hazard ratio compares event rates over time within the fitted survival-analysis framework.

The one-sided 97.5% confidence limit extends to 1.7040824. Because the upper confidence limit is above the registry's non-inferiority threshold of 1.20, this confidence interval does not establish non-inferiority against that stated margin.

The confidence interval also communicates substantial uncertainty around the point estimate. The non-inferiority conclusion depends specifically on the prespecified margin and one-sided confidence-bound logic; it should not be replaced by an ordinary two-sided significance test.

16-month analysis of death or first thromboembolic event

A second posted primary analysis evaluated the same DTE endpoint through study completion, on average 16 months, using the full analysis set. Unlike the preceding analysis, this was reported as a superiority analysis using a two-sided log-rank test.

Hazard ratio for death or first thromboembolic event

1.35

95% CI: 1.01–1.81   ·   P = 0.04223

Analysis population: Full analysis set, 1644 subjects overall

Method: Log-rank test; HR based on Cox proportional hazards model

Clinical Biostats interpretation

The HR of 1.35 indicates a 35% higher estimated hazard for the DTE composite in the rivaroxaban-based strategy relative to the antiplatelet-based strategy, under the reported Cox model.

It does not mean that 35% of participants experienced DTE, or that an individual participant's probability of an event was necessarily 35% higher. The hazard ratio is a relative time-to-event measure rather than an absolute risk measure.

The 95% confidence interval of 1.01–1.81 indicates uncertainty around the estimated hazard ratio. Its lower bound is just above 1, while the upper bound permits a materially larger relative hazard than the point estimate.

The reported P = 0.04223 is a measure of evidence against the stated null hypothesis under the specified two-sided log-rank framework. It is not a measure of effect size. The magnitude of the effect is conveyed by the HR, while its precision is conveyed by the confidence interval.

Because the registry identifies this as a superiority analysis rather than the preceding non-inferiority analysis, the two results answer different statistical questions and should not be interpreted as though they were one analysis with one hypothesis.

Educational note: a Kaplan-Meier curve is not reconstructed here from summary statistics. A valid curve requires the underlying event and censoring information or sufficiently detailed source data.

8. Primary Result: Primary Bleeding Event

The third posted primary analysis evaluated the registered primary bleeding event through study completion, on average 16 months, in the full analysis set.

Hazard ratio for primary bleeding event

1.50

95% CI: 0.95–2.37   ·   P = 0.07745

Analysis population: Full analysis set, 1644 subjects overall

Method: Log-rank test; HR based on Cox proportional hazards model

Clinical Biostats interpretation

The HR of 1.50 indicates a 50% higher estimated hazard of the registered primary bleeding event in the rivaroxaban-based strategy compared with the antiplatelet-based strategy under the reported model.

This does not mean that 50% of participants experienced a primary bleeding event, nor does it mean that the absolute probability of bleeding increased by 50 percentage points.

The 95% confidence interval of 0.95–2.37 crosses 1.00. Thus, the interval includes values compatible with a lower hazard as well as values compatible with a substantially higher hazard in the rivaroxaban arm.

The reported P = 0.07745 is not a measure of the size or clinical importance of the observed HR. It describes the evidence against the relevant null hypothesis under the reported two-sided log-rank framework. The confidence interval is essential for understanding the precision and range of compatible relative effects.

The registry labels this analysis as descriptive and identifies the hypothesis type as other / not stated. It therefore should not be retroactively assigned a different confirmatory hypothesis solely because the reported p-value can be compared with conventional thresholds.

9. Secondary Endpoint Results

The registry also reports five secondary time-to-event analyses. Each used the full analysis set, intention-to-treat analysis concepts, a two-sided log-rank test, and hazard ratios based on Cox proportional hazards models.

Secondary endpointHR95% CIP-valueHypothesis
Number of Participants With Net-clinical Benefit 1.39 1.08–1.80 = 0.01156 Superiority
Number of Participants With Cardiovascular Death or Thromboembolic Event 1.22 0.89–1.69 = 0.21595 Superiority
Number of Participants With TIMI Major / Minor Bleeds 1.78 1.08–2.94 = 0.02216 Other / not stated
Number of Participants With ISTH Major Bleeds 1.66 1.05–2.62 = 0.02702 Other / not stated
Number of Participants With Composite Bleeding Endpoint of BARC 2, 3, or 5 Bleeds 1.84 1.41–2.41 = 0.00001 Other / not stated

How to read these secondary analyses

All five reported HRs are above 1, so each point estimate corresponds to a higher estimated hazard in the rivaroxaban arm under the registry's stated interpretation of HRs above 1. The confidence intervals quantify uncertainty around those estimates.

The secondary analyses should nevertheless be distinguished from the primary DTE and PBE analyses. Several are explicitly labelled descriptive or have the hypothesis type "Other / not stated." A nominal p-value does not, by itself, establish that a secondary endpoint was part of a multiplicity-controlled confirmatory testing sequence.

Multiplicity caution: the ClinicalTrials.gov record reports multiple primary and secondary endpoint analyses but does not provide a complete multiplicity-adjustment strategy in the ClinicalTrials.gov record. The individual p-values should therefore be interpreted in the context of their stated hypothesis types rather than treated as a single collection of independent confirmatory tests.

10. Secondary Results in Detail

Net-clinical benefit

Reported estimate
HR 1.39  ·  95% CI 1.08–1.80  ·  P = 0.01156

The estimated hazard for the registered net-clinical-benefit endpoint was higher in the rivaroxaban arm under the reported Cox model. The 95% confidence interval lies above 1, although the estimate should be interpreted according to the endpoint definition and the registry's stated superiority framework.

Cardiovascular death or thromboembolic event

Reported estimate
HR 1.22  ·  95% CI 0.89–1.69  ·  P = 0.21595

The point estimate is above 1, but the 95% confidence interval includes 1. The interval therefore permits both a lower and a higher hazard relative to the antiplatelet strategy under the reported model.

TIMI major / minor bleeds

Reported estimate
HR 1.78  ·  95% CI 1.08–2.94  ·  P = 0.02216

The point estimate corresponds to a 78% higher estimated hazard of the registered TIMI major/minor bleeding endpoint in the rivaroxaban arm. The confidence interval indicates substantial uncertainty around the magnitude of the relative effect.

ISTH major bleeds

Reported estimate
HR 1.66  ·  95% CI 1.05–2.62  ·  P = 0.02702

The estimated hazard of the registered ISTH major-bleeding endpoint was higher in the rivaroxaban arm. The 95% confidence interval is above 1 but remains broad enough that the precise magnitude of the relative difference is uncertain.

BARC 2, 3, or 5 bleeding

Reported estimate
HR 1.84  ·  95% CI 1.41–2.41  ·  P = 0.00001

The estimated hazard of the composite BARC 2, 3, or 5 bleeding endpoint was 84% higher in the rivaroxaban arm under the reported Cox model. The confidence interval remains entirely above 1 and quantifies uncertainty around the estimated relative hazard.

11. Safety Results

The ClinicalTrials.gov record reports serious adverse events by randomized arm in terms of affected participants and participants at risk.

Safety measureRivaroxaban armAntiplatelet arm
Serious adverse events296 / 801282 / 807
Serious adverse events by arm — affected / at risk
Rivaroxaban
296 / 801
Antiplatelet
282 / 807

The registry's serious-adverse-event figures should be distinguished from the time-to-event bleeding analyses. A count of participants affected by a serious adverse event and a Cox-model hazard ratio for a bleeding endpoint answer different statistical questions.

Safety interpretation: 296/801 and 282/807 are affected/at-risk counts as reported in the registry by the registry. They should not be converted into an inferred comparative risk measure on this page because the task data does not provide the full safety analysis framework needed for such a comparison.

12. Non-Inferiority and Superiority: Two Different Questions

GALILEO is particularly useful statistically because the ClinicalTrials.gov record contains both a non-inferiority analysis and a later superiority analysis of the same broad DTE endpoint. These analyses should not be conflated.

Feature14-month DTE analysis16-month DTE analysis
Analysis populationSafety analysis set, 1608 subjects overallFull analysis set, 1644 subjects overall
Time frameThrough study completion, on average 14 monthsThrough study completion, on average 16 months
EstimateHR 1.2089302HR 1.35
Confidence intervalOne-sided 97.5%; upper 1.704082495% two-sided: 1.01–1.81
Hypothesis typeNon-inferioritySuperiority
Key statistical questionIs the upper confidence bound below the 1.20 non-inferiority margin?Does the time-to-event comparison provide evidence against the superiority null under the reported log-rank framework?

The non-inferiority margin is a substantive part of the statistical hypothesis. A point estimate close to 1 does not establish non-inferiority if the upper confidence bound remains above the permitted margin. Conversely, a result can have a confidence interval that excludes 1 while still requiring separate consideration of whether a non-inferiority or superiority framework was prespecified.

13. Statistical Methods Explained

Why use a time-to-event analysis rather than simply comparing event counts?

Participants can have different amounts of observed follow-up, and the timing of an event contains information. Time-to-event methods incorporate the timing of events and censoring. This is why the GALILEO analyses are reported with hazard ratios and log-rank tests rather than only raw event proportions.

What does an HR of 1.35 mean?

An HR of 1.35 means the estimated hazard in the rivaroxaban arm was 1.35 times the estimated hazard in the antiplatelet arm under the reported Cox model. Equivalently, it is a 35% higher estimated hazard. It does not mean that the event probability was 35 percentage points higher.

Why was a one-sided confidence interval used for the non-inferiority analysis?

Non-inferiority testing is directional. The relevant concern is whether the experimental strategy could be worse than the comparator by more than a prespecified amount. Here, the registry specifies an upper hazard-ratio margin of 1.20, so the upper confidence bound is compared directly with that threshold.

Why does the p-value not measure the size of the effect?

A p-value summarizes evidence against a null hypothesis under a specified statistical model and testing procedure. It does not tell us how large the treatment effect is. In GALILEO, the HR describes relative magnitude while the confidence interval describes precision; the p-value provides a separate measure of statistical evidence under the stated testing framework.

Why does the analysis population matter?

The 14-month DTE non-inferiority analysis used the safety analysis set, while the 16-month DTE and PBE analyses used the full analysis set. Because these populations are defined differently, their estimates answer questions about different sets of participants. Comparing them as though they came from exactly the same analysis population would obscure an important part of the statistical design.

What does an HR of 1.84 for BARC 2, 3, or 5 bleeding mean?

Under the reported Cox model, the estimated hazard of that composite bleeding endpoint was 1.84 times the hazard in the antiplatelet arm. The registry specifically states that HRs above 1 indicate a higher hazard in the rivaroxaban arm. The estimate does not mean that 84% of patients bled or that absolute bleeding risk increased by 84 percentage points.

Why is the confidence interval important even when the p-value is below 0.05?

The p-value and confidence interval provide different information. For example, the BARC analysis has an HR of 1.84 with a 95% CI of 1.41–2.41. The interval communicates that the precise magnitude of the relative hazard remains uncertain even though the entire reported interval is above 1.

14. Kaplan-Meier Estimation and Censoring

The registered analyses are time-to-event analyses, making Kaplan-Meier estimation a natural descriptive framework for these data. Kaplan-Meier methods estimate the probability of remaining event-free over time while allowing participants to be censored when their event status is no longer observed.

Conceptual Kaplan-Meier form
S(t) = ∏ti ≤ t (1 − di/ni)

Here, di is the number of events at time ti and ni is the number at risk immediately before that time.

The ClinicalTrials.gov record does not provide the individual event and censoring times required to reconstruct the GALILEO Kaplan-Meier curves. Accordingly, no numerical survival curve is fabricated from the summary hazard ratios and confidence intervals.

15. Cox Proportional-Hazards Interpretation

The registry states that the reported hazard-ratio estimates are based on Cox proportional hazards models. The standard Cox interpretation concerns the relative hazard between treatment groups under the model.

Conceptual model
h(t | X) = h0(t) exp(βX)

For a binary treatment indicator, the exponentiated treatment coefficient gives the hazard ratio. The model's proportional-hazards interpretation assumes that the relative hazard is represented appropriately by a common hazard ratio over the relevant follow-up.

Model caution: a single Cox hazard ratio can be difficult to interpret if the proportional-hazards assumption does not adequately describe the underlying event processes. The ClinicalTrials.gov record does not report a formal proportional-hazards diagnostic, so this page does not claim that the assumption was empirically verified.

16. Endpoint Architecture and Multiplicity

The registry identifies three registered primary endpoints: two records for the DTE endpoint at different average follow-up times and one primary bleeding endpoint. It also reports five secondary statistical analyses.

Endpoint / analysisRoleStatistical framework
DTE, average 14 monthsPrimaryNon-inferiority; one-sided confidence interval
DTE, average 16 monthsPrimarySuperiority; two-sided log-rank test; Cox HR
Primary bleeding event, average 16 monthsPrimaryTwo-sided log-rank test; Cox HR; descriptive / other hypothesis
Net-clinical benefitSecondarySuperiority; two-sided log-rank test; Cox HR
Cardiovascular death or thromboembolic eventSecondarySuperiority; two-sided log-rank test; Cox HR
TIMI major / minor bleedsSecondaryDescriptive; two-sided log-rank test; Cox HR
ISTH major bleedsSecondaryDescriptive; two-sided log-rank test; Cox HR
BARC 2, 3, or 5 bleedsSecondaryDescriptive; two-sided log-rank test; Cox HR

Multiplicity is important whenever multiple endpoints or hypotheses are examined. The ClinicalTrials.gov record identifies the hypothesis types and reports the individual p-values, but it does not provide a complete multiplicity-adjustment procedure. The page therefore avoids treating all reported p-values as though they belonged to one explicitly defined familywise error-control procedure.

17. What the Confidence Intervals Say

DTE at 16 months

The HR of 1.35 has a 95% CI of 1.01–1.81. The interval is entirely above 1, but its width shows that the exact relative hazard is not known with high precision.

Primary bleeding

The HR of 1.50 has a 95% CI of 0.95–2.37. The interval crosses 1, so the reported data are compatible with a range that includes no relative hazard difference as well as higher hazard.

Non-inferiority DTE

The one-sided 97.5% upper limit is 1.7040824. Because the non-inferiority margin is 1.20, the upper confidence bound is not below the margin.

BARC bleeding

The HR is 1.84 with a 95% CI of 1.41–2.41, showing that the entire reported confidence interval lies above 1 while still leaving uncertainty about the precise magnitude.

18. Understanding the Difference Between Relative and Absolute Effects

Hazard ratios are relative measures. An HR of 1.50 does not provide the absolute probability that a participant will experience the event. To obtain an absolute risk difference at a particular time, one would need the corresponding estimated survival or cumulative-incidence quantities for both treatment groups.

This distinction is especially important when interpreting composite endpoints. DTE includes all-cause death, any stroke, myocardial infarction, symptomatic valve thrombosis, pulmonary embolism, deep vein thrombosis, and non-CNS systemic embolism. A single HR summarizes the relative time-to-first-event comparison for that composite rather than describing each component separately.

Composite-endpoint caution: the DTE HR should not be interpreted as though it were a hazard ratio for death alone or for any single thromboembolic component. The registry defines DTE as the first occurrence of the specified composite endpoint.

19. Comparing the Primary Efficacy and Bleeding Signals

EndpointHR95% CIP-valueInterpretive direction
DTE, 16 months1.351.01–1.81= 0.04223Higher estimated hazard in rivaroxaban arm
Primary bleeding event1.500.95–2.37= 0.07745Higher point-estimated hazard in rivaroxaban arm
Net-clinical benefit1.391.08–1.80= 0.01156Higher estimated hazard in rivaroxaban arm
TIMI major / minor bleeds1.781.08–2.94= 0.02216Higher estimated hazard in rivaroxaban arm
ISTH major bleeds1.661.05–2.62= 0.02702Higher estimated hazard in rivaroxaban arm
BARC 2, 3, or 5 bleeds1.841.41–2.41= 0.00001Higher estimated hazard in rivaroxaban arm

The statistical pattern is notable because the point estimates for every posted time-to-event comparison in the ClinicalTrials.gov record is above 1. This means that, under the registry's stated interpretation, every reported point estimate corresponds to a higher estimated hazard in the rivaroxaban arm. The size and precision of those estimates vary, and their hypothesis classifications differ.

This table should not be reduced to a single p-value-based conclusion. The DTE non-inferiority analysis uses a 1.20 margin; the 16-month DTE analysis is labelled superiority; PBE is labelled descriptive; and the secondary endpoints include both superiority and other/not-stated hypothesis types. The statistical question changes with the prespecified hypothesis.

20. Trial Timeline

2015-12-16

Trial start

The GALILEO trial began on December 16, 2015 according to the registry data.

2018-11-27

Primary completion

The registry lists November 27, 2018 as the primary completion date.

Registry status

Terminated

The ClinicalTrials.gov record lists the study status as terminated.

21. Important Limitations and Interpretation Issues

22. Why This Trial Matters Statistically

GALILEO is a useful teaching case because it illustrates how the interpretation of a randomized trial can change depending on the statistical question being asked. The same broad clinical endpoint appears in both a non-inferiority analysis and a later superiority analysis, with different follow-up times and different analysis populations.

ConceptHow it appears in GALILEO
RandomizationRandomized phase 3 parallel-group design with 1653 participants enrolled.
Time-to-event endpointsDTE and primary bleeding event were analyzed as time-to-event outcomes.
Hazard ratioUsed as the principal relative effect measure for the posted analyses.
Log-rank testReported for the 16-month DTE analysis and primary bleeding analysis and for the secondary analyses.
Cox modelRegistry notes state that HR estimates were based on Cox proportional hazards models.
Intention-to-treat principleIdentified as an analysis concept for the full analysis set.
Non-inferiorityDTE analysis used a one-sided 97.5% confidence limit and a 1.20 hazard-ratio margin.
SuperiorityThe 16-month DTE analysis was labelled a superiority analysis with a two-sided log-rank test.
Confidence intervalsUsed to communicate uncertainty around hazard-ratio estimates and to evaluate the non-inferiority margin.
Composite endpointsDTE combines death and several adjudicated thromboembolic events.
MultiplicityMultiple primary and secondary endpoint analyses require attention to the stated hypothesis types and error-control framework.
Safety analysisSerious adverse events are reported separately by arm, with 296/801 and 282/807 affected/at risk.

23. Clinical Interpretation vs Statistical Interpretation

Statistical interpretation

The posted DTE analyses report HRs above 1, with the 14-month analysis using a non-inferiority margin of 1.20 and the 16-month analysis reporting HR 1.35 with a 95% CI of 1.01–1.81. The primary bleeding analysis reports HR 1.50 with a 95% CI of 0.95–2.37.

Clinical interpretation

The statistical results describe comparative time-to-event outcomes between the two randomized strategies. They should be interpreted alongside the registered endpoint definitions and the separately reported serious-adverse-event counts rather than collapsed into one numerical measure.

The most important interpretive point is that an HR is not a complete description of clinical benefit or harm. The endpoint definition, analysis population, follow-up, confidence interval, hypothesis type, and safety evidence all contribute to the statistical story.

24. Sources

Continue with the statistical methods behind GALILEO

Explore the underlying concepts used to understand randomized time-to-event trials, non-inferiority hypotheses, hazard ratios, confidence intervals, and log-rank testing.

25. Related Tutorials

Learn more about the methods used in this trial:

26. Related Calculators

27. Record Summary

GALILEO provides a detailed example of how statistical interpretation depends on the exact endpoint, analysis population, follow-up period, and hypothesis. The registry reports a 14-month non-inferiority analysis of death or first thromboembolic event with HR 1.2089302 and a one-sided 97.5% upper confidence limit of 1.7040824 against a 1.20 margin. A separate 16-month full-analysis-set DTE analysis reports HR 1.35 with 95% CI 1.01–1.81 and P = 0.04223. The primary bleeding analysis reports HR 1.50 with 95% CI 0.95–2.37 and P = 0.07745.

The secondary analyses consistently report point estimates above 1, including HR 1.39 for net-clinical benefit, 1.22 for cardiovascular death or thromboembolic event, 1.78 for TIMI major/minor bleeding, 1.66 for ISTH major bleeding, and 1.84 for BARC 2, 3, or 5 bleeding. These estimates must be interpreted according to their stated hypothesis types rather than as a uniform set of confirmatory tests.

Clinical Biostats methodology: A trial-results page should not merely repeat reported numbers. The goal is to reconstruct the statistical structure of the trial, explain what each estimate means, preserve the distinction between non-inferiority and superiority questions, and clearly identify the assumptions and limitations that affect interpretation.