← Clinical Trials
Acute Ischaemic Stroke Transient Ischaemic Attack Phase 3 Completed NCT03354429

THALES: Complete Statistical Analysis of Ticagrelor in Acute Ischaemic Stroke and Transient Ischaemic Attack

An independent statistical review of the randomized phase 3 THALES trial evaluating ticagrelor versus placebo in participants with acute ischaemic stroke or transient ischaemic attack, with emphasis on the registered primary endpoint, time-to-event analysis, logistic regression, and bleeding outcomes.

Trial: THALES  ·  NCT03354429  ·  Phase 3  ·  Enrollment 11,016
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

THALES was a completed, randomized, parallel-group, quadruple-masked phase 3 trial in acute ischaemic stroke and transient ischaemic attack. The registry reports 11,016 enrolled participants, two intervention arms, seven posted statistical analyses, and a primary analysis comparing ticagrelor with placebo for subsequent stroke or death from randomisation through visit 3.

11,016
Enrolled
Participants
2
Arms
Ticagrelor vs placebo
0.83
Primary HR
95% CI 0.71–0.96
0.015
Primary P-value
Two-sided
FeatureTHALES
Trial nameTHALES
NCT identifierNCT03354429
Therapeutic areaNeurology
ConditionsAcute Ischaemic Stroke; Transient Ischaemic Attack
PhasePhase 3
StatusCompleted
AllocationRandomized
Design modelParallel
MaskingQuadruple
Primary purposeTreatment
Enrollment11,016
Lead sponsorAstraZeneca
Sponsor typeIndustry
StartJanuary 22, 2018
Primary completionDecember 13, 2019

2. Clinical Question

The central statistical question was whether ticagrelor differed from placebo with respect to the registered composite endpoint of subsequent stroke or death among participants with acute ischaemic stroke or transient ischaemic attack, measured from randomisation through visit 3.

Population

Participants with acute ischaemic stroke or transient ischaemic attack enrolled in the phase 3 THALES trial.

Intervention

Ticagrelor.

Comparator

Placebo.

Primary question

How does ticagrelor compare with placebo for the composite of subsequent stroke or death from randomisation through visit 3?

3. Trial Design

01
Randomize11,016 participants
02
Parallel armsTicagrelor vs placebo
03
Quadruple maskingMasked trial design
04
Follow-upDay 1 through visit 3
05
AnalysisStroke, death, function, bleeding
ARM 1 · TICAGRELOR

Ticagrelor

  • Intervention type: drug
  • Randomized treatment arm
  • Compared with placebo in the posted analyses
ARM 2 · PLACEBO

Placebo

  • Intervention type: drug
  • Randomized comparator arm
  • Reference group in the posted effect measures
Why the design matters statistically. Randomization establishes the treatment comparison before outcome data are observed, while the parallel-group structure permits direct comparison of the randomized groups. Quadruple masking is a design feature intended to reduce the potential influence of knowledge of treatment assignment on trial conduct and assessment.

4. Endpoints

The registry lists one primary endpoint and seven posted outcome measures/statistical analyses. The primary endpoint is registered as binary, while its posted statistical analysis uses a time-to-event formulation and a Cox proportional-hazards model.

EndpointRegistry definition / time frameAnalysis
Composite of Subsequent Stroke or Death Participants with subsequent stroke or death; from randomisation (day 1) to visit 3 (day 30-34) Cox proportional-hazards model; hazard ratio
Ischaemic Stroke From randomisation (day 1) to visit 3 (day 30-34) Cox proportional-hazards model; hazard ratio
Number of Participants With Modified Rankin Scale (mRS) Score >1 at Visit 3 Visit 3 (day 30-34) Logistic regression; odds ratio
Bleeding Event That Fulfils Serious Adverse Event Criteria and is Categorised as GUSTO Severe From randomisation (day 1) to visit 3 (day 30-34) Cox proportional-hazards model; hazard ratio
ICH or Fatal Bleeding Event From randomisation (day 1) to visit 3 (day 30-34) Cox proportional-hazards model; hazard ratio
Bleeding Event That Fulfils Serious Adverse Event Criteria and is Categorised as GUSTO Moderate/Severe From randomisation (day 1) to visit 3 (day 30-34) Cox proportional-hazards model; hazard ratio
Premature Permanent Discontinuation of IP Due to Bleeding From randomisation (day 1) to visit 3 (day 30-34) Cox proportional-hazards model; hazard ratio

5. Statistical Methodology

Primary analysis population

The primary endpoint analysis used the full analysis set including all randomised patients. The analysis text also identifies the intention-to-treat principle. This is important because the treatment comparison remains anchored to the randomized assignment rather than being redefined according to subsequent treatment exposure.

Cox proportional-hazards model

The registry reports a Cox regression analysis for the primary composite and for the other time-to-event endpoints. The effect measure is the hazard ratio, with placebo as the reference group.

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

The hazard ratio compares the modeled instantaneous event rates between treatment groups. In the THALES analyses, an HR below 1 indicates a lower estimated hazard for ticagrelor relative to placebo under the fitted model.

Logistic regression

The mRS >1 endpoint at visit 3 was analysed with logistic regression. The posted analysis specifies covariate adjustment for NIHSS score and history of stroke (yes/no). Patients with missing mRS score or missing covariates, NIHSS score, and history of stroke were addressed in the analysis-population wording in the registry record.

Conceptual logistic model
logit[P(Y=1)] = α + β1Treatment + β2NIHSS + β3Stroke history

The exponentiated treatment coefficient can be interpreted as an odds ratio comparing ticagrelor with placebo after adjustment for the listed covariates.

Effect measures

Two principal effect measures appear in the posted analyses: hazard ratios for time-to-event endpoints and an odds ratio for the binary mRS endpoint. These measures should not be treated as interchangeable.

MeasureUsed in THALES forWhat it compares
Hazard ratioPrimary composite, ischaemic stroke, bleeding endpoints, discontinuation due to bleedingRelative modeled event hazard over time
Odds ratiomRS score >1 at visit 3Relative odds of the binary outcome

6. Results: Primary Endpoint

Composite of Subsequent Stroke or Death

The primary analysis compared ticagrelor with placebo in the full analysis set including all randomised patients. The registered time frame was from randomisation (day 1) to visit 3 (day 30-34), and the posted analysis used a Cox proportional-hazards model.

Hazard ratio for subsequent stroke or death

0.83

95% CI: 0.71–0.96   ·   P = 0.015   ·   Two-sided

Reference group: placebo

Clinical Biostats interpretation

An HR of 0.83 means that, under the fitted Cox model, the estimated instantaneous hazard of the composite event was 0.83 times that in the placebo group. Expressed as a simple relative interpretation, this corresponds to a 17% lower estimated hazard for ticagrelor relative to placebo.

The HR does not mean that 17% of participants avoided stroke or death, that every participant experienced a 17% reduction in risk, or that the absolute probability of the event was reduced by 17 percentage points. Hazard is a time-dependent rate concept, not an individual probability.

The 95% CI of 0.71–0.96 describes the statistical uncertainty around the estimated hazard ratio under the analysis framework. It is narrower than an interval extending substantially above 1, but it does not describe the range of individual treatment effects across participants.

The P = 0.015 value addresses evidence against the null hypothesis within the prespecified superiority framework; it does not measure the size or clinical importance of the treatment effect. Effect size and uncertainty are better conveyed jointly by the HR and its confidence interval.

Because this is a Cox analysis, interpretation also depends on the model's proportional-hazards assumption. A single HR summarizes the relative hazard over the analysis period and can be incomplete if the relative hazards change materially over time. The analysis population is also important: this result comes from the full analysis set including all randomised patients.

7. Secondary Endpoint Results

Ischaemic Stroke

The secondary endpoint of ischaemic stroke was analysed from randomisation (day 1) to visit 3 (day 30-34) using a Cox proportional-hazards model in the full analysis set including all randomised patients.

Hazard ratio for ischaemic stroke

0.79

95% CI: 0.68–0.93   ·   P = 0.004   ·   Two-sided

Reference group: placebo

Clinical Biostats interpretation

The HR of 0.79 corresponds to a 21% lower estimated hazard of ischaemic stroke for ticagrelor relative to placebo under the Cox model. This is a relative hazard interpretation, not a statement that 21% of participants were protected from stroke or that absolute stroke risk fell by 21 percentage points.

The 95% CI of 0.68–0.93 quantifies uncertainty around the estimated relative hazard. The P-value of 0.004 is evidence evaluated against the relevant null hypothesis; it is not a measure of effect magnitude. As with the primary analysis, the interpretation assumes that the Cox model provides an appropriate representation of the time-to-event relationship.

Number of Participants With Modified Rankin Scale (mRS) Score >1 at Visit 3

This binary endpoint was measured at visit 3 (day 30-34). The posted analysis used logistic regression with NIHSS score and history of stroke (yes/no) included as covariates. The analysis population was the full analysis set including all randomised patients, with the registry specifying exclusions related to missing mRS score or missing covariate information.

Adjusted odds ratio for mRS score >1

0.98

95% CI: 0.89–1.07   ·   P = 0.613   ·   Two-sided

Reference group: placebo

Clinical Biostats interpretation

An adjusted OR of 0.98 indicates that the modeled odds of having an mRS score >1 at visit 3 were estimated to be 0.98 times the corresponding odds in the placebo group after adjustment for NIHSS score and history of stroke.

The OR is close to 1, and the 95% CI of 0.89–1.07 spans 1. The P-value of 0.613 does not provide evidence of a statistically detectable difference under the reported superiority analysis. This should not be translated into proof that the treatment groups are identical; the confidence interval expresses the uncertainty around the estimated comparison.

An odds ratio is also not a risk ratio. If the outcome probability is not small, the numerical value of an OR can differ materially from the corresponding risk ratio. The covariate adjustment is another important part of the result: this is an adjusted estimate rather than an unadjusted comparison of proportions.

8. Bleeding and Treatment Discontinuation Results

The registry contains four additional prespecified analyses involving bleeding or premature permanent discontinuation of investigational product due to bleeding. Each used the full analysis set including all randomised patients and a Cox proportional-hazards model, with placebo as the reference group.

EndpointHazard ratio95% CIP-value
Bleeding Event That Fulfils Serious Adverse Event Criteria and is Categorised as GUSTO Severe 3.99 1.74–9.14 0.001
ICH or Fatal Bleeding Event 3.66 1.48–9.02 0.005
Bleeding Event That Fulfils Serious Adverse Event Criteria and is Categorised as GUSTO Moderate/Severe 3.27 1.67–6.43 <0.001
Premature Permanent Discontinuation of IP Due to Bleeding 4.80 3.28–7.02 <0.001

How to interpret these hazard ratios

Each estimate is above 1, meaning the estimated hazard was higher in the ticagrelor group than in the placebo group under the corresponding Cox model. For example, an HR of 3.99 means an estimated instantaneous hazard approximately 3.99 times that of placebo; it does not mean that 399% of participants experienced the event or that the absolute probability was 3.99 times higher.

Clinical Biostats interpretation

The confidence intervals are important because the bleeding endpoints are event-specific estimates rather than measurements of all adverse outcomes. The intervals for all four posted bleeding-related analyses lie above 1, while their P-values are reported as 0.001, 0.005, <0.001, and <0.001, respectively.

These results also illustrate why a clinical trial cannot be reduced to one efficacy statistic. The primary composite and ischaemic-stroke analyses have HRs below 1, while the reported bleeding analyses have HRs above 1. Statistical interpretation therefore requires examining the endpoint definitions, effect measures, confidence intervals, and time frame separately.

9. Safety Results

The registry data provide serious adverse event counts by randomized treatment arm in terms of affected participants and participants at risk.

Safety measureTicagrelorPlacebo
Serious adverse events 571 / 5,523 affected / at risk 609 / 5,493 affected / at risk
Serious adverse events · affected participants
Ticagrelor
571
Placebo
609

The serious-adverse-event figures should be kept distinct from the specific bleeding endpoints above. The bleeding analyses concern narrowly defined outcomes and use time-to-event models, whereas the safety summary here reports affected participants and participants at risk for serious adverse events.

Denominator matters. The safety data are reported as affected participants over participants at risk: 571/5,523 for ticagrelor and 609/5,493 for placebo. These denominators should not be replaced with a different analysis population or used to infer an unreported hazard ratio.

10. Statistical Methods Explained

Why was a Cox proportional-hazards model used?

The primary composite and several secondary or prespecified safety endpoints are time-to-event outcomes. A Cox model uses both whether an event occurred and when it occurred, while accommodating censoring. That makes it suitable for comparing event hazards over a defined follow-up period rather than reducing the analysis to a single binary count.

What does the primary hazard ratio of 0.83 mean?

Because placebo is the reference group, an HR of 0.83 indicates that the modeled instantaneous hazard of subsequent stroke or death was estimated at 83% of the placebo hazard. Equivalently, 1 − 0.83 = 0.17, or a 17% lower estimated hazard. This is not the same as a 17-percentage-point reduction in event probability.

Why is the confidence interval important?

The estimate 0.83 is a single point estimate. Its 95% CI of 0.71–0.96 shows the uncertainty around that estimate under the statistical model. A confidence interval is therefore more informative than the point estimate alone: it communicates both the estimated effect and the range of values compatible with the sampling uncertainty represented by the analysis.

Why doesn't the P-value measure effect size?

The P-value evaluates the evidence against a null hypothesis within the specified statistical framework. It is affected by both the observed data and the amount of information available. Two studies can have similar effect estimates but different P-values, or different effect estimates with similar P-values. The HR or OR and its confidence interval are therefore necessary to describe the magnitude and precision of the estimated treatment effect.

Why was logistic regression used for the mRS endpoint?

The registry describes the mRS >1 endpoint as binary at visit 3, making logistic regression a natural model for the probability of belonging to one of the two outcome categories. The posted analysis additionally adjusted for NIHSS score and history of stroke, so the resulting OR of 0.98 is an adjusted comparison rather than a simple unadjusted treatment-group ratio.

Why are the bleeding hazard ratios interpreted differently from the primary hazard ratio?

The mathematical interpretation of the HR is the same, but the endpoint differs. An HR below 1 for the primary composite indicates a lower modeled hazard for ticagrelor relative to placebo, whereas an HR above 1 for a bleeding endpoint indicates a higher modeled hazard. Statistical interpretation must therefore begin with the precise endpoint definition before assigning clinical meaning to the direction of the estimate.

11. Intention-to-Treat Analysis

The primary endpoint analysis used the full analysis set including all randomised patients, and the registry explicitly identifies intention-to-treat analysis as a concept in the analysis text. This is a central feature of randomized trial inference.

Preserves randomization

Analysing participants according to randomized assignment maintains the original treatment comparison created by randomization.

Handles post-randomization events conceptually

Participants do not cease to belong to their randomized group simply because subsequent events or treatment changes occur.

Primary analysis population

The posted primary analysis specifies the full analysis set including all randomised patients.

Not the same as every possible analysis

Safety and other endpoint-specific analyses can have different data-handling requirements, so the analysis population must always be checked for the endpoint being interpreted.

12. Covariate Adjustment for the mRS Endpoint

The logistic regression analysis for the number of participants with mRS score >1 at visit 3 included NIHSS score and history of stroke (yes/no) as covariates.

Adjusted treatment comparison
Treatment effect = OR adjusted for NIHSS score + history of stroke

The purpose of covariate adjustment is to estimate the treatment association while accounting for prespecified or analysis-specified prognostic variables included in the model. The resulting OR should therefore not be interpreted as an unadjusted odds ratio.

For educational purposes, this distinction is important because adjustment changes the question being answered. An unadjusted comparison asks how the outcome differs between randomized groups without incorporating additional predictors into the model. The posted THALES mRS analysis instead reports a model-based treatment comparison after incorporating NIHSS and history of stroke.

13. Time-to-Event Analysis and Censoring

The primary composite and the other Cox-model endpoints were measured from randomisation (day 1) to visit 3 (day 30-34). Time-to-event analysis uses the timing of the event rather than simply classifying each participant as an event or non-event without regard to when the event occurred.

Conceptual survival function
S(t) = P(T > t)

The survival function represents the probability of remaining free of the specified event beyond time t. Cox regression models the relative hazard rather than directly modeling the survival probability itself.

Censoring is important because not every participant necessarily contributes an observed event during the complete analysis period. A censored observation contributes information up to the censoring time under the assumptions required by the survival-analysis framework. The ClinicalTrials.gov record does not provide detailed censoring counts or censoring patterns, so no additional quantitative characterization is made here.

14. Hazard Ratios Versus Odds Ratios

FeatureHazard ratioOdds ratio
THALES endpoint example Composite of subsequent stroke or death mRS score >1 at visit 3
Outcome structure Time-to-event Binary at a defined visit
Model Cox proportional-hazards model Logistic regression
Reference group Placebo Placebo
Interpretation Relative modeled instantaneous hazard Relative odds of the binary outcome

Keeping these effect measures separate prevents a common statistical mistake: treating every ratio below 1 as if it represented the same type of relative risk. The primary HR of 0.83 and the mRS OR of 0.98 answer different statistical questions because they arise from different outcome structures and models.

15. Multiple Endpoints and Interpretation

The registry reports seven statistical analyses: one primary analysis, one secondary ischaemic-stroke analysis, one secondary mRS analysis, and four additional prespecified bleeding or discontinuation analyses.

Endpoint roleNumber represented in posted analysesPrincipal method
Primary1Cox proportional-hazards model
Secondary2Cox proportional-hazards model; logistic regression
Other prespecified4Cox proportional-hazards model
Total statistical analyses posted7Cox proportional-hazards model and logistic regression

The primary endpoint has a reported superiority analysis with a two-sided confidence interval and P-value. The other posted analyses should be interpreted according to their own endpoint roles and statistical context. The ClinicalTrials.gov record does not specify a multiplicity-adjustment procedure, alpha allocation, or hierarchical testing strategy for these seven posted analyses.

Multiplicity caution: Seven posted analyses do not automatically constitute seven independent confirmatory hypotheses, nor does statistical significance for one endpoint by itself establish a global treatment conclusion across all endpoints. The registry-reported THALES registry data do not provide enough detail to reconstruct an endpoint hierarchy or multiplicity-adjustment procedure beyond the reported hypothesis type and individual analysis results.

16. What the Primary Result Does — and Does Not — Mean

Relative effect

The primary HR of 0.83 represents a modeled relative hazard comparison of ticagrelor versus placebo for subsequent stroke or death. A value below 1 indicates a lower estimated hazard in the ticagrelor group under the Cox model.

Not an absolute risk difference

The HR does not tell us the absolute number or percentage of participants who avoided the composite endpoint. An absolute treatment effect requires event probabilities, cumulative incidence estimates, or another absolute-risk measure.

Confidence interval

The 95% CI of 0.71–0.96 quantifies uncertainty around the estimated HR. It does not describe the range of responses that individual participants would experience.

P-value

The P-value of 0.015 is evidence against the relevant null hypothesis in the reported superiority analysis. It should not be interpreted as the probability that the null hypothesis is true, nor as a measure of how large the treatment effect is.

17. Comparing Efficacy and Bleeding Signals

One of the most instructive statistical features of THALES is that the direction of the estimated effect depends on the endpoint being studied.

EndpointEffectDirection relative to placebo
Composite of Subsequent Stroke or DeathHR 0.83 (95% CI 0.71–0.96)Lower estimated hazard
Ischaemic StrokeHR 0.79 (95% CI 0.68–0.93)Lower estimated hazard
mRS Score >1 at Visit 3OR 0.98 (95% CI 0.89–1.07)Odds estimate close to 1
GUSTO Severe bleedingHR 3.99 (95% CI 1.74–9.14)Higher estimated hazard
ICH or Fatal BleedingHR 3.66 (95% CI 1.48–9.02)Higher estimated hazard
GUSTO Moderate/Severe bleedingHR 3.27 (95% CI 1.67–6.43)Higher estimated hazard
Permanent discontinuation due to bleedingHR 4.80 (95% CI 3.28–7.02)Higher estimated hazard

This is a useful reminder that treatment effects are always endpoint-specific. A drug can have estimates in opposite directions for different outcomes without any statistical contradiction. The appropriate interpretation is to examine the complete set of prespecified outcomes, their definitions, analysis populations, effect measures, uncertainty, and statistical context.

18. Trial Timeline

January 22, 2018

Trial start

The THALES trial began according to the ClinicalTrials.gov record.

December 13, 2019

Primary completion

The ClinicalTrials.gov record identifies December 13, 2019 as the primary completion date.

Completed

Results posted

The registry record is marked completed and contains seven posted outcome measures and seven statistical analyses.

19. Limitations

20. Why This Trial Matters Statistically

THALES is a useful teaching case because it combines randomized treatment comparison, quadruple masking, time-to-event analysis, binary-outcome modeling, covariate adjustment, and endpoint-specific safety analysis in a single phase 3 trial.

ConceptHow it appears in THALES
RandomizationThe trial uses randomized allocation with two parallel treatment arms.
BlindingThe registry describes the trial as quadruple masked.
Intention-to-treat analysisThe primary analysis uses the full analysis set including all randomised patients.
Time-to-event endpointsThe primary composite, ischaemic stroke, bleeding endpoints, and discontinuation endpoint are analysed with Cox regression.
Cox proportional-hazards modelUsed for the primary and multiple prespecified time-to-event analyses.
Hazard ratioThe principal effect measure for the time-to-event outcomes.
Logistic regressionUsed for mRS score >1 at visit 3.
Odds ratioUsed for the adjusted binary mRS endpoint.
Covariate adjustmentNIHSS score and history of stroke were included in the mRS logistic regression.
Confidence intervalsAll posted estimates with confidence intervals use two-sided 95% intervals.
Endpoint-specific safetyFour additional posted analyses address bleeding or permanent discontinuation due to bleeding.

21. Related Tutorials

Learn more about the methods used in this trial:

22. Related Calculators

23. Sources

Continue through the Clinical Biostats statistical pathway

Use the trial's endpoints and methods as a practical starting point for deeper study of survival analysis, regression, confidence intervals, and randomized clinical-trial methodology.

24. Record Summary

THALES provides a compact example of how several statistical methods operate together in a randomized phase 3 trial. The primary endpoint, a composite of subsequent stroke or death from randomisation (day 1) to visit 3 (day 30-34), was analysed with a Cox proportional-hazards model in the full analysis set including all randomised patients. The reported HR was 0.83 with a 95% CI of 0.71–0.96 and P = 0.015. A secondary ischaemic-stroke analysis produced an HR of 0.79, while the adjusted mRS analysis used logistic regression and produced an OR of 0.98.

The additional prespecified analyses demonstrate why endpoint-specific interpretation matters. The reported bleeding hazard ratios were above 1, including HRs of 3.99, 3.66, 3.27, and 4.80 for the specified bleeding and discontinuation outcomes. The serious-adverse-event summary also reports 571/5,523 affected/at risk for ticagrelor and 609/5,493 for placebo.

Clinical Biostats methodology: The statistical story of a trial is more informative than a single headline result. THALES illustrates the importance of matching the statistical method to the endpoint, distinguishing hazard ratios from odds ratios, reporting confidence intervals alongside point estimates, preserving the randomized analysis population, understanding covariate adjustment, and interpreting efficacy and safety endpoints separately.