← Clinical Trials
Type 2 Diabetes Mellitus Phase 3 Time-to-Event Analysis NCT01144338

EXSCEL: Complete Statistical Analysis of Exenatide Once Weekly in Type 2 Diabetes Mellitus

An independent statistical review of the randomized phase 3 EXSCEL trial evaluating exenatide once weekly versus placebo in patients with type 2 diabetes mellitus, with emphasis on its MACE endpoint, Cox proportional-hazards analyses, superiority question, and non-inferiority framework.

Trial: EXSCEL  ·  Phase 3  ·  Enrollment: 14752  ·  Status: Completed
Scope of this record

This page separates reported trial results from statistical interpretation. Numerical trial results on this page are restricted to the information contained in the ClinicalTrials.gov trial data posted on ClinicalTrials.gov for EXSCEL and the statistical analyses posted there.

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

EXSCEL was a randomized, parallel, triple-masked phase 3 trial evaluating exenatide once weekly against placebo in patients with type 2 diabetes mellitus. The primary analyses used Cox proportional-hazards models in the intention-to-treat population. The same MACE composite served as both the primary efficacy and primary safety outcome, but the two analyses addressed different statistical hypotheses: superiority for efficacy and non-inferiority for safety.

14752
Enrollment
Randomized trial
2
Arms
Parallel design
0.91
Primary HR
95% CI 0.832–1.004
1.3
NI Margin
HR scale
FeatureEXSCEL
Trial nameExenatide Study of Cardiovascular Event Lowering Trial (EXSCEL)
PhasePhase 3
ConditionType 2 Diabetes Mellitus
DesignRandomized, parallel, triple-masked
AllocationRandomized
Primary purposeTreatment
Enrollment14752
Arms2
InterventionsExenatide Once Weekly; Placebo
Lead sponsorAstraZeneca
Sponsor typeIndustry
Study period2010-06-18 to 2017-04-21 primary completion
ClinicalTrials.govNCT01144338

2. Clinical Question

The central statistical question was whether treatment with exenatide once weekly, compared with placebo, changed the time to first occurrence of the composite cardiovascular endpoint defined as cardiovascular death, nonfatal myocardial infarction, or nonfatal stroke.

Population

Participants enrolled in the phase 3 EXSCEL trial with type 2 diabetes mellitus.

Intervention

Exenatide Once Weekly.

Comparator

Placebo.

Primary question

For the primary efficacy analysis, is the hazard of first MACE lower with exenatide once weekly than with placebo? For the primary safety analysis, is exenatide once weekly non-inferior to placebo with respect to MACE?

3. Trial Design

01
Randomize14752 participants
02
AssignExenatide once weekly or placebo
03
FollowStudy period
04
ObserveTime to first MACE
05
AnalyzeCox proportional-hazards model
ACTIVE ARM

Exenatide Once Weekly

  • Exenatide Once Weekly
  • Randomized parallel-group assignment
  • Primary analyses compared this group with placebo
CONTROL ARM

Placebo

  • Placebo
  • Randomized parallel-group assignment
  • Reference group for the primary Cox analyses

The registry identifies EXSCEL as randomized, with a parallel design and triple masking. The primary purpose is recorded as treatment. These design features establish the basic framework for interpreting the between-group time-to-event comparisons.

4. Trial Timeline

2010-06-18

Study start

The registry records 2010-06-18 as the study start date.

2017-04-21

Primary completion

The registry records 2017-04-21 as the primary completion date.

Completed

Registry status

The trial is recorded as completed, with results posted for the registered outcome measures.

5. Primary Endpoints

EndpointRegistry definitionTime frameHypothesis
Primary Efficacy Outcome MACE Events The primary efficacy outcome variable is defined as the composite endpoint of cardiovascular death, nonfatal MI, or nonfatal stroke. The number of participants who had an event is reported in the results. The primary efficacy endpoint is the same as the primary safety endpoint, and the statistical analysis tests the superiority of exenatide against the placebo. Time to first event. Information collected during study period (anticipated to be up to 7.5 years). Superiority
Primary Safety Outcome MACE Events The primary safety outcome variable is defined as the composite endpoint of cardiovascular death, nonfatal MI, or nonfatal stroke. The number of participants who had an event is reported in the results. The primary safety endpoint is the same as the primary efficacy endpoints, and the statistical analysis tests the non-inferiority of exenatide against placebo. Time to first event. Information collected during study period (anticipated to be up to 7.5 years). Non-inferiority

The important statistical feature is that the endpoint itself does not change between the two primary analyses. What changes is the question being asked. The efficacy analysis asks whether exenatide once weekly has a lower hazard than placebo. The safety analysis asks whether the hazard is sufficiently below the prespecified non-inferiority boundary of HR=1.3.

6. Analysis Population

Intention-to-treat population

The primary analyses use the ITT population: all patients consented and randomized in the study without a major GCP violation. The ITT population is analyzed as randomized.

ITT analysis → participants remain in their randomized treatment group for the primary comparison

This preserves the treatment comparison generated by randomization and avoids redefining treatment groups according to subsequent exposure or treatment behavior.

For a time-to-event endpoint, the ITT principle is particularly important because participants can contribute follow-up information even when their observed treatment course is not identical to the originally assigned intervention. The registry explicitly identifies the ITT population as the analysis population for all eight posted statistical analyses.

7. Statistical Methodology

Cox proportional-hazards model

The statistical method reported for the primary efficacy and safety analyses is regression, Cox, normalized here as a Cox proportional-hazards model. This is a survival-analysis model designed for outcomes defined by time until an event.

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

For a two-group comparison, the exponentiated treatment coefficient corresponds to a hazard ratio. An HR below 1 indicates a lower estimated instantaneous event rate in the exenatide group relative to placebo, under the fitted model.

Hazard ratio

The effect measure reported for every posted statistical analysis is the hazard ratio (HR). For EXSCEL, the registry compares placebo with exenatide once weekly and reports the HR as the treatment effect measure.

How to read the primary hazard ratio

HR 0.91

The estimated hazard ratio is below 1.

An HR of 0.91 corresponds to an estimated hazard approximately 9% lower in the exenatide once weekly group than in the placebo group, expressed on the hazard-ratio scale.

This is a relative time-to-event measure. It is not a statement that 9% of participants avoided an event, that each participant experienced exactly a 9% reduction in risk, or that the absolute difference in event probability was 9 percentage points.

Confidence intervals

The primary analyses report two-sided 95% confidence intervals. The interval quantifies uncertainty around the estimated hazard ratio under the model and statistical sampling framework.

Primary 95% confidence interval
0.832 ≤ HR ≤ 1.004

Because the interval extends across HR=1, it includes values corresponding to no relative hazard difference as well as values below 1. For the non-inferiority analysis, however, the relevant comparison is with the prespecified upper margin of HR=1.3.

Time-to-event interpretation

The endpoint is defined as time to first event. Participants are therefore not analyzed solely according to whether an event eventually occurred. Their observed follow-up time and event or censoring status contribute to the survival analysis.

This distinction explains why the Cox model is appropriate to the registered endpoint: the analysis incorporates the timing of the first MACE event rather than reducing the study to a simple binary proportion without regard to follow-up time.

8. Primary Result: Efficacy — MACE

The primary efficacy analysis evaluates the composite endpoint of cardiovascular death, nonfatal MI, or nonfatal stroke using a Cox proportional-hazards model in the ITT population. The hypothesis is a superiority test of exenatide once weekly over placebo:

Registered superiority hypothesis
H0: HR ≥ 1    vs.    H1: HR < 1

Primary efficacy MACE hazard ratio

0.91

95% CI: 0.832–1.004   ·   P = 0.061

Two-sided 95% confidence interval; ITT population; Cox proportional-hazards model.

FeaturePrimary efficacy analysis
OutcomePrimary Efficacy Outcome MACE Events
DefinitionComposite of cardiovascular death, nonfatal MI, or nonfatal stroke
Time frameTime to first event; information collected during study period (anticipated to be up to 7.5 years)
PopulationITT
ComparisonPlacebo vs Exenatide Once Weekly
MethodCox proportional-hazards model
Effect measureHazard ratio
Estimate0.91
95% CI0.832–1.004
P-value0.061
HypothesisSuperiority
Clinical Biostats interpretation

The HR of 0.91 estimates a lower hazard of first MACE with exenatide once weekly relative to placebo under the fitted Cox model. Expressed simply, the estimate is about 9% below the reference value of 1.

The estimate does not mean that the probability of MACE was exactly 9% lower, nor does it establish that every patient experienced a similar reduction. A hazard ratio is a model-based relative measure of event occurrence over time.

The two-sided 95% CI of 0.832–1.004 indicates uncertainty around the estimated HR. Because the interval includes HR=1, the interval is compatible with both a lower hazard and a value very close to no relative hazard difference.

The P-value of 0.061 addresses the statistical evidence against the specified superiority null hypothesis; it does not measure the magnitude or clinical importance of the HR. The p-value should therefore be read together with the HR and confidence interval rather than as a standalone measure of effect size.

The analysis is also a time-to-event analysis, so interpretation depends on the censoring structure and the assumptions underlying the Cox model, including the proportional-hazards assumption. The registry does not provide additional model diagnostics in the ClinicalTrials.gov record.

9. Primary Result: Safety — MACE Non-Inferiority

The primary safety analysis uses the same composite MACE endpoint and the same Cox proportional-hazards methodology, but changes the hypothesis from superiority to non-inferiority. The registry specifies a non-inferiority margin of HR=1.3.

Registered non-inferiority hypothesis
H0: HR ≥ 1.3    vs.    H1: HR < 1.3

Primary safety MACE hazard ratio

0.91

95% CI: 0.832–1.004   ·   P < 0.001

Two-sided 95% confidence interval; ITT population; Cox proportional-hazards model; non-inferiority margin HR=1.3.

FeaturePrimary safety analysis
OutcomePrimary Safety Outcome MACE Events
DefinitionComposite of cardiovascular death, nonfatal MI, or nonfatal stroke
Time frameTime to first event; information collected during study period (anticipated to be up to 7.5 years)
PopulationITT
ComparisonPlacebo vs Exenatide Once Weekly
MethodCox proportional-hazards model
Effect measureHazard ratio
Estimate0.91
95% CI0.832–1.004
P-value< 0.001
HypothesisNon-inferiority
Non-inferiority marginHR=1.3
Clinical Biostats interpretation

The estimated HR is again 0.91, but the statistical question is different. For non-inferiority, the relevant concern is whether the hazard could be at or above the prespecified unacceptable boundary of 1.3.

The upper confidence-limit of 1.004 is below the non-inferiority margin of 1.3. Thus, the reported confidence interval lies entirely on the non-inferiority side of the prespecified margin.

The P-value of < 0.001 is tied to the non-inferiority hypothesis, not the superiority hypothesis. It therefore should not be interpreted as evidence that the superiority test was positive. The same estimated HR can produce a different inferential conclusion because the null hypothesis and decision boundary are different.

This is one of the most important statistical lessons from EXSCEL: non-inferiority and superiority are not interchangeable tests. An estimate below 1 may be compatible with non-inferiority even when the corresponding superiority analysis does not meet its specified significance threshold.

As with the efficacy analysis, the interpretation relies on a Cox model and time-to-event data. Non-inferiority conclusions also depend on the prespecified margin being clinically and statistically appropriate and on preservation of the randomized comparison. The ClinicalTrials.gov record provides the margin but do not provide further justification for how that margin was selected.

10. Why the Same HR Can Produce Different P-values

At first glance, the primary efficacy and safety results may appear contradictory because both use an HR of 0.91 and the same confidence interval of 0.832–1.004, yet their reported p-values are different.

Superiority question

The efficacy analysis tests whether HR is below 1. The reported P-value is 0.061, with the 95% CI extending above 1.

Non-inferiority question

The safety analysis tests whether HR is below 1.3. The entire 95% CI is below 1.3, and the reported P-value is < 0.001.

The statistical reference point therefore matters. A hypothesis test is not determined solely by the observed estimate. It also depends on the null hypothesis being tested.

Two different reference values
Superiority: compare HR with 1    |    Non-inferiority: compare HR with 1.3

The same estimated effect can be insufficiently separated from 1 for superiority while being clearly separated from the non-inferiority boundary of 1.3.

11. Secondary Endpoint Results

Six secondary efficacy outcomes have statistical analyses posted in the ClinicalTrials.gov record. All use the ITT population, compare placebo with exenatide once weekly, and use Cox proportional-hazards models with hazard ratios and two-sided 95% confidence intervals.

Secondary endpointHR95% CIP-value
All-Cause Mortality0.860.77–0.970.016
CV Death0.880.76–1.020.096
MI0.970.85–1.100.622
Stroke0.850.70–1.030.095
Hospitalization for ACS1.050.94–1.180.402
Hospitalization for HF0.940.78–1.130.485

All-Cause Mortality

Hazard ratio

0.86

95% CI: 0.77–0.97   ·   P = 0.016

The HR of 0.86 is below 1, corresponding to an estimated hazard approximately 14% lower with exenatide once weekly than placebo under the fitted Cox model. The 95% CI of 0.77–0.97 remains below 1.

CV Death

Hazard ratio

0.88

95% CI: 0.76–1.02   ·   P = 0.096

The point estimate is below 1, but the two-sided 95% CI extends above 1. The result therefore carries more uncertainty about the direction of the relative hazard than a confidence interval entirely below 1 would.

MI

Hazard ratio

0.97

95% CI: 0.85–1.10   ·   P = 0.622

The point estimate is close to 1, and the confidence interval includes values below and above 1. The estimate should be understood as the observed model-based relative hazard estimate rather than as evidence that all participants experienced the same relative effect.

Stroke

Hazard ratio

0.85

95% CI: 0.70–1.03   ·   P = 0.095

The point estimate corresponds to an approximately 15% lower estimated hazard, but the confidence interval extends from 0.70 to 1.03 and therefore includes HR=1.

Hospitalization for ACS

Hazard ratio

1.05

95% CI: 0.94–1.18   ·   P = 0.402

The point estimate is above 1, corresponding to a higher estimated hazard in the exenatide once weekly group under the fitted model. The confidence interval includes 1 and extends on both sides of the null value.

Hospitalization for HF

Hazard ratio

0.94

95% CI: 0.78–1.13   ·   P = 0.485

The point estimate is below 1, but the confidence interval spans 1. The estimate therefore should not be interpreted without acknowledging the uncertainty represented by the confidence interval.

12. Secondary Results in Context

The secondary analyses illustrate why clinical-trial interpretation should not be reduced to counting how many p-values fall below a threshold. The estimates range from 0.85 for stroke to 1.05 for hospitalization for ACS, and the corresponding confidence intervals vary in width and relationship to HR=1.

PatternWhat it tells the reader
HR below 1The estimated hazard is lower in the exenatide once weekly group under the Cox model.
HR above 1The estimated hazard is higher in the exenatide once weekly group under the Cox model.
CI includes 1The confidence interval is compatible with no relative hazard difference.
CI entirely below 1The confidence interval excludes HR=1 on the lower side.
P-valueMeasures evidence against the specified null hypothesis; it is not an effect-size measure.

The six secondary analyses are also distinct from the primary safety non-inferiority test. Their registry entries specify superiority hypotheses rather than the HR=1.3 non-inferiority boundary used for the primary safety analysis.

13. Safety Results

The ClinicalTrials.gov record reports serious adverse events by randomized arm. The registry limitation states that only serious adverse events are collected in this study. Therefore, these figures should not be interpreted as a complete summary of all adverse events.

Safety measureAffectedAt risk
Placebo12227372
Exenatide Once Weekly12347344

Placebo

Serious adverse events were reported for 1222 of 7372 participants at risk.

Exenatide Once Weekly

Serious adverse events were reported for 1234 of 7344 participants at risk.

Safety-data limitation: The registry specifically states that only serious adverse events are collected in this study. The reported serious-adverse-event counts therefore should not be presented as if they represented all adverse events, treatment-emergent adverse events, or a complete safety profile.

14. Statistical Methods Explained

Why was a Cox proportional-hazards model used?

The registered primary endpoints are defined as time to first event. A Cox proportional-hazards model is designed for this type of survival outcome because it uses information about both whether an event occurred and when it occurred, while accommodating censored observations.

What does an HR of 0.91 mean?

An HR of 0.91 means that the fitted model estimates the instantaneous event hazard in the exenatide once weekly group to be 0.91 times the corresponding hazard in the placebo group. This is approximately a 9% lower estimated hazard on the relative hazard scale. It does not mean that the absolute probability of an event was reduced by 9 percentage points.

Why can the efficacy and safety analyses have different p-values with the same HR?

Because they test different hypotheses. The superiority analysis evaluates HR relative to 1, while the non-inferiority analysis evaluates HR relative to the prespecified margin of 1.3. The estimate and confidence interval can therefore provide different evidence relative to those two reference points.

Why is the non-inferiority margin important?

A non-inferiority trial does not simply ask whether the treatment is "not statistically different" from placebo. It asks whether the treatment effect is sufficiently far from an unacceptable loss of efficacy or safety, as represented here by an HR margin of 1.3. The relevant question is whether the uncertainty around the treatment effect remains below that boundary.

Why does the 95% CI matter?

The confidence interval communicates the precision of the HR estimate. For the primary analysis, the interval is 0.832–1.004. Relative to the superiority null of 1, that interval includes the null. Relative to the non-inferiority margin of 1.3, the entire interval lies below the margin.

Why does the p-value not measure effect size?

A p-value describes the statistical evidence against a specified null hypothesis. It does not tell the reader whether an observed HR is large or small in clinical terms. The HR itself describes the estimated relative effect, while the confidence interval describes uncertainty around that estimate.

What does intention-to-treat mean here?

The registry defines the ITT population as all patients consented and randomized in the study without a major GCP violation and states that the ITT population is analyzed as randomized. The key idea is that the treatment comparison follows the randomized assignment rather than being redefined according to subsequent treatment behavior.

15. Confidence Intervals and the Two Primary Questions

The EXSCEL primary analyses provide a particularly clear demonstration of why confidence intervals should be interpreted against the correct statistical reference value.

ReferencePrimary efficacyPrimary safety
Estimated HR0.910.91
95% CI0.832–1.0040.832–1.004
Reference value / margin11.3
QuestionIs HR below 1?Is HR below 1.3?
Reported P-value0.061< 0.001
Hypothesis typeSuperiorityNon-inferiority

The table demonstrates an important general principle: statistical significance has meaning only relative to a specified null hypothesis. The same confidence interval cannot be interpreted without first identifying the question being tested.

16. Understanding the MACE Composite Endpoint

The primary MACE endpoint is a composite of three event types:

Cardiovascular death

One component of the registered composite endpoint.

Nonfatal MI

One component of the registered composite endpoint.

Nonfatal stroke

One component of the registered composite endpoint.

Time to first event

The analysis records the first qualifying component event for the registered endpoint.

A composite endpoint can improve statistical efficiency by combining multiple clinically relevant event types into one prespecified outcome. However, the resulting HR describes the composite endpoint as a whole. It should not automatically be interpreted as showing that every individual component has the same treatment effect.

EXSCEL's the ClinicalTrials.gov record separately report CV death, MI, and stroke as secondary endpoints. Those analyses provide component-specific estimates, but they remain separate secondary analyses rather than interchangeable replacements for the primary composite endpoint.

17. Primary Efficacy vs Primary Safety: A Statistical Distinction

One of the most instructive features of EXSCEL is that the primary efficacy endpoint and primary safety endpoint are identical in definition, yet they have different inferential roles.

Primary efficacy
Superiority
H0: HR ≥ 1 versus H1: HR < 1

Reported HR 0.91; 95% CI 0.832–1.004; P=0.061.
Primary safety
Non-inferiority
H0: HR ≥ 1.3 versus H1: HR < 1.3

Reported HR 0.91; 95% CI 0.832–1.004; P<0.001.

This is not a contradiction. It is a consequence of asking two different questions of the same estimated treatment effect.

18. What the Hazard Ratio Does — and Does Not — Mean

Statistical interpretation

The primary HR of 0.91 indicates a lower estimated instantaneous hazard of first MACE in the exenatide once weekly group than in the placebo group under the fitted Cox model.

It does not mean that 9% of participants avoided MACE, that every participant had a 9% reduction in risk, or that the absolute probability of MACE differed by exactly 9 percentage points.

Why the confidence interval matters

The 95% CI of 0.832–1.004 describes uncertainty around the estimated HR. It does not describe the range of effects that individual participants experienced.

Why the reference value matters

For superiority, HR=1 is the null reference. For the EXSCEL non-inferiority analysis, HR=1.3 is the prespecified margin. Consequently, the same interval can include 1 while remaining entirely below 1.3.

19. Reading the Secondary Hazard Ratios

The secondary results demonstrate several common patterns in time-to-event analysis.

Reported secondary hazard ratios
All-Cause Mortality
0.86
CV Death
0.88
MI
0.97
Stroke
0.85
Hospitalization for ACS
1.05
Hospitalization for HF
0.94

The visual is intended only as a conceptual display of the reported point estimates. It should not be interpreted as a forest plot because the confidence intervals are not plotted here. The numerical interpretation should always rely on the reported HRs and confidence intervals in the table.

20. Statistical Interpretation of P-values

EXSCEL provides a useful demonstration of why p-values should be tied explicitly to their hypotheses.

AnalysisEstimateP-valueQuestion tested
Primary efficacy MACEHR 0.910.061Superiority: HR < 1
Primary safety MACEHR 0.91< 0.001Non-inferiority: HR < 1.3
All-Cause MortalityHR 0.860.016Superiority: HR < 1
CV DeathHR 0.880.096Superiority: HR < 1
MIHR 0.970.622Superiority: HR < 1
StrokeHR 0.850.095Superiority: HR < 1
Hospitalization for ACSHR 1.050.402Superiority: HR < 1
Hospitalization for HFHR 0.940.485Superiority: HR < 1

A p-value should therefore never be detached from its corresponding hypothesis. In EXSCEL, the difference between 0.061 and <0.001 for the two primary analyses is explained by the different statistical questions rather than by different estimated HRs.

21. Limitations and Interpretation Issues

22. What the Registry Data Do Not Establish

A careful statistical analysis should distinguish what is directly reported from what would require additional trial documents or analyses.

TopicWhat can be stated from the ClinicalTrials.gov record
Primary analysis methodCox proportional-hazards model.
Primary effect measureHazard ratio.
Analysis populationITT, analyzed as randomized.
Non-inferiority marginHR=1.3.
Stratification factorsNot provided in the ClinicalTrials.gov record.
Interim analysis planNot provided in the ClinicalTrials.gov record.
Multiplicity strategyNot provided in the ClinicalTrials.gov record.
Missing-data or imputation strategyNot provided in the ClinicalTrials.gov record.
Bayesian methodsNot reported in the ClinicalTrials.gov record.
CrossoverNot reported in the ClinicalTrials.gov record.

This distinction matters because the absence of a reported methodological detail in the ClinicalTrials.gov record should not be replaced with assumptions about how the trial must have been conducted.

23. Why This Trial Matters Statistically

EXSCEL is a useful teaching example because a single randomized clinical trial contains several important ideas in survival analysis and confirmatory trial design.

ConceptHow it appears in EXSCEL
RandomizationThe trial uses randomized allocation in a parallel design.
BlindingThe registry identifies the study as triple-masked.
Intention-to-treat analysisThe primary analyses use an ITT population analyzed as randomized.
Time-to-event endpointMACE is defined as time to first event.
Cox proportional-hazards modelThe registry reports Cox regression for all eight posted statistical analyses.
Hazard ratioHR is the reported effect measure for primary and secondary analyses.
Confidence intervalPrimary analyses report two-sided 95% CIs.
Superiority testingThe primary efficacy analysis tests HR below 1.
Non-inferiority testingThe primary safety analysis tests HR below the margin of 1.3.
Composite endpointMACE combines cardiovascular death, nonfatal MI, and nonfatal stroke.

The most important statistical lesson is that the analysis question determines the interpretation of the estimate. EXSCEL uses the same MACE definition and the same estimated HR for its primary efficacy and safety analyses, yet the inferential framework changes from superiority to non-inferiority.

24. Primary Analysis as a Worked Statistical Example

Step 1 · Identify the endpoint

The primary endpoint is time to first MACE, where MACE is cardiovascular death, nonfatal MI, or nonfatal stroke.

Step 2 · Identify the population

The analysis uses the ITT population and analyzes participants as randomized.

Step 3 · Identify the model

The registry reports Cox proportional-hazards regression.

Step 4 · Identify the effect measure

The effect measure is a hazard ratio, with an estimate of 0.91.

Step 5 · Identify the uncertainty

The two-sided 95% CI is 0.832–1.004.

Step 6 · Identify the hypothesis

For efficacy, the superiority null is HR ≥ 1. For safety, the non-inferiority null is HR ≥ 1.3.

This sequence is broadly useful when reading a clinical-trial statistical analysis: first identify what happened, then who was analyzed, then how it was modeled, then what effect measure was reported, and finally what hypothesis that effect was tested against.

25. Related Tutorials

Learn more about the methods used in this trial:

26. Related Calculators

27. Sources

Continue through the Clinical Biostats statistical pathway

Move from the trial's endpoints and statistical methods to deeper tutorials and practical statistical tools.

28. Record Summary

EXSCEL provides a particularly useful example of how a randomized clinical trial can use the same time-to-event endpoint for different statistical purposes. The trial enrolled 14752 participants in a randomized, parallel, triple-masked phase 3 design comparing exenatide once weekly with placebo. The primary endpoint was time to first MACE, defined as cardiovascular death, nonfatal MI, or nonfatal stroke.

The primary efficacy analysis used a Cox proportional-hazards model in the ITT population and tested superiority. It reported an HR of 0.91, with a two-sided 95% CI of 0.832–1.004 and a P-value of 0.061. The primary safety analysis used the same endpoint, analysis population, model, HR, and confidence interval but tested non-inferiority against an HR margin of 1.3, with a reported P-value of < 0.001.

The secondary analyses extend the same Cox-model framework to all-cause mortality, CV death, MI, stroke, hospitalization for ACS, and hospitalization for HF. Their HR estimates range from 0.85 to 1.05, with confidence intervals that vary in their precision and relationship to HR=1.

The central statistical lesson is therefore not simply the numerical value of the primary HR. It is the distinction between effect estimation and hypothesis testing. The HR describes the estimated relative hazard, the confidence interval describes uncertainty around that estimate, and the p-value must be interpreted in relation to the particular null hypothesis. EXSCEL makes this distinction especially clear because the same primary HR and confidence interval are evaluated once against HR=1 for superiority and once against HR=1.3 for non-inferiority.

Clinical Biostats methodology: A trial-results page should reconstruct the statistical story rather than simply repeat a registry summary. For EXSCEL, that story centers on randomized comparison, ITT analysis, time-to-event methodology, Cox regression, hazard ratios, confidence intervals, composite endpoints, and the distinction between superiority and non-inferiority hypotheses.