← Clinical Trials
Differentiated Thyroid Cancer Phase 3 Time-to-Event Analysis NCT01321554

SELECT: Complete Statistical Analysis of Lenvatinib in Differentiated Thyroid Cancer

An independent statistical review of the randomized phase 3 SELECT trial of lenvatinib versus placebo in 131I-refractory differentiated thyroid cancer, focusing on progression-free survival, the intention-to-treat analysis population, hazard-ratio estimation, confidence intervals, and safety.

Trial status: Completed  ·  Enrollment: 392  ·  Sponsor: Eisai Inc.
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. Numerical trial facts on this page are limited to the ClinicalTrials.gov record.

1. Trial at a Glance

SELECT was a completed, multicenter, randomized, double-blind, placebo-controlled phase 3 trial evaluating lenvatinib in participants with 131I-refractory differentiated thyroid cancer. The ClinicalTrials.gov record identifies progression-free survival as the single registered primary endpoint and report a formal hazard-ratio analysis comparing lenvatinib 24 mg with placebo.

392
Enrolled
Total enrollment
4
Arms
Parallel design
0.21
PFS HR
Lenvatinib 24 mg vs placebo
99%
Two-Sided CI
0.14–0.31
FeatureSELECT
Trial nameSELECT
Brief titleA Multicenter, Randomized, Double-Blind, Placebo-Controlled, Trial of Lenvatinib (E7080) in 131I-Refractory Differentiated Thyroid Cancer (DTC)
PhasePhase 3
ConditionThyroid Cancer
AllocationRandomized
Design modelParallel
MaskingTriple
Primary purposeTreatment
Enrollment392
Registered primary endpoints1
Primary endpoint typeTime-to-event
Results postedYes
Statistical analyses posted1
Effect measureHazard ratio; reported as Cox Proportional Hazard
Hypothesis typeSuperiority
ClinicalTrials.govNCT01321554
Lead sponsorEisai Inc.

2. Clinical Question

The primary statistical question was whether lenvatinib 24 mg produced a different time-to-progression-or-death profile from placebo in participants with 131I-refractory differentiated thyroid cancer, under a prespecified superiority framework.

Population

Participants with 131I-refractory differentiated thyroid cancer, as described by the trial's brief title and registered condition.

Intervention

Lenvatinib 24 mg in the Randomization Phase comparison reported in the statistical analysis.

Comparator

Placebo in the Randomization Phase comparison reported in the statistical analysis.

Primary question

Does lenvatinib improve progression-free survival relative to placebo?

3. Trial Design

01
Randomize392 enrolled
02
Parallel arms4 registered arms
03
BlindingTriple masking
04
Assess PFSProgression or death
05
CompareHazard ratio
Allocation
Randomized allocation in a parallel-group design.
Masking
The registry classifies the trial as triple-masked and placebo-controlled.
Primary purpose
Treatment.
Hypothesis
Superiority of the intervention over placebo for the registered primary endpoint.
RANDOMIZATION PHASE

Lenvatinib 24 mg

  • Randomization Phase treatment group.
  • Included in the posted primary PFS comparison.
  • Serious adverse events: 171 affected of 261 at risk.
RANDOMIZATION PHASE

Placebo

  • Randomization Phase comparator group.
  • Included in the posted primary PFS comparison.
  • Serious adverse events: 31 affected of 131 at risk.
Four-arm design, but one reported primary comparison: the registry profile lists 4 arms. The single posted statistical analysis in the ClinicalTrials.gov record compares Randomization Phase lenvatinib 24 mg with Randomization Phase placebo. The ClinicalTrials.gov record also contain two OOL, Treatment Period safety groups: lenvatinib 24 mg and lenvatinib 20 mg. No additional efficacy comparison is posted on ClinicalTrials.gov for the other registered arms.

4. Trial Timing and Registry Status

2011-03-17

Trial start

The registry profile gives March 17, 2011 as the study start date.

2013-11-15

Primary completion

The registry profile gives November 15, 2013 as the primary completion date.

Completed

Current registry status in the ClinicalTrials.gov record

The trial is classified as completed, with results posted on ClinicalTrials.gov.

5. Primary Endpoint

EndpointRegistered definition / time frameStatistical information posted
Progression Free Survival (PFS) Date of randomization to the date of disease progression or death (whichever occurred first), assessed up to data cutoff. Cox Proportional Hazard; hazard ratio; full analysis set / Intent-to-Treat Analysis Set.
Endpoint structure: PFS is a time-to-event endpoint. Each participant contributes follow-up from randomization until the event defined by the endpoint or an appropriate censoring point. The registry's primary endpoint time frame explicitly combines disease progression and death as the event, with whichever occurs first.

6. Analysis Population

The posted primary analysis used the full analysis set (Intent-to-Treat Analysis Set), which included all randomized participants.

PopulationDefinition in registry-reported analysisRole
Full analysis set / ITT Included all randomized participants. Primary PFS efficacy analysis.

This distinction is statistically important. An intention-to-treat analysis preserves the randomized comparison by retaining participants according to their randomized assignment rather than redefining treatment groups based on subsequent treatment exposure. For a randomized superiority comparison, that protects the causal interpretation provided by randomization, subject to the usual assumptions and limitations of the trial design and endpoint analysis.

7. Primary PFS Result

The ClinicalTrials.gov statistical analysis compares Randomization Phase lenvatinib 24 mg with Randomization Phase placebo for progression-free survival. The effect measure is reported as a Cox proportional-hazard measure and normalized as a hazard ratio.

Hazard ratio for progression or death

0.21

99% two-sided CI: 0.14–0.31

Superiority hypothesis; analysis population: full analysis set / Intent-to-Treat Analysis Set.

EndpointLenvatinib 24 mgPlaceboEffect estimate
Progression Free Survival Randomization Phase Randomization Phase HR 0.21
99% two-sided CI 0.14–0.31
Clinical Biostats interpretation

What the estimate means: A hazard ratio of 0.21 means that, under the fitted time-to-event model, the estimated instantaneous rate of the PFS event was approximately 21% as high in the lenvatinib 24 mg group as in the placebo group over the analyzed follow-up. Equivalently, this corresponds to an estimated 79% lower instantaneous hazard for the lenvatinib group relative to placebo.

What it does not mean: HR 0.21 does not mean that 21% of patients experienced progression, that 79% of patients were progression-free, or that every individual patient's probability of progression was reduced by exactly 79%. A hazard ratio is a relative model-based measure of event rates over time.

What the confidence interval says: The 99% two-sided confidence interval of 0.14–0.31 quantifies uncertainty around the estimated hazard ratio under the statistical model and sampling framework. It indicates that the point estimate is not the only plausible value; the interval provides a range of parameter values compatible with the specified confidence procedure.

Why the p-value is not discussed as a number: The ClinicalTrials.gov record does not report a p-value. It therefore would be inappropriate to insert one from another source or infer one from the confidence interval. In general, a p-value addresses evidence against a null hypothesis; it does not measure the magnitude or clinical importance of an effect.

Important modeling caution: The reported effect measure is based on a Cox proportional-hazards framework. Interpretation of a single hazard ratio is most straightforward when the proportional-hazards representation is reasonable over the relevant follow-up. The ClinicalTrials.gov record does not report a formal proportional-hazards diagnostic, so this page does not make such a diagnostic claim.

Censoring matters: PFS is a time-to-event endpoint and can involve censored observations. The hazard-ratio estimate therefore incorporates both observed events and the available follow-up information rather than simply comparing proportions with progression.

Educational note: a Kaplan-Meier curve is not reconstructed here because the ClinicalTrials.gov record provides the hazard-ratio summary but do not provide the underlying event and censoring times required to construct an original curve.

8. Statistical Methodology

Time-to-event analysis

PFS is fundamentally different from a simple binary endpoint because the analysis uses both whether an event occurred and the timing of that event. Participants who have not experienced progression or death by the relevant observation point can contribute censored follow-up information.

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

The survival function represents the probability that the event time T exceeds time t. For PFS, the event is the first qualifying disease progression or death.

Kaplan-Meier estimation

For a time-to-event endpoint such as PFS, Kaplan-Meier estimation is a standard nonparametric way to describe the event-time distribution while accounting for right censoring. It produces an estimated probability of remaining event-free over time.

Kaplan-Meier estimator
Ŝ(t) = ∏ti ≤ t (1 − di/ni)

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

Cox proportional-hazards model

The posted effect measure is explicitly reported as a Cox Proportional Hazard and normalized as a hazard ratio. The Cox model expresses the hazard for an individual or treatment group relative to a baseline hazard through a multiplicative treatment effect.

Cox model
h(t | X) = h0(t) exp(βX)

For a binary treatment indicator, exp(β) corresponds to the hazard ratio comparing the treatment group with the reference group under the fitted model.

Intention-to-treat analysis

The registry-reported analysis specifies that the full analysis set / Intent-to-Treat Analysis Set included all randomized participants. This is important because treatment assignment is determined before post-randomization outcomes occur. Keeping participants associated with their randomized assignment helps preserve the comparability created by randomization.

Superiority testing

The hypothesis type is registered as superiority. In a superiority framework, the statistical question is whether the treatment effect differs from the comparator in the favorable direction under the prespecified hypothesis-testing framework. The ClinicalTrials.gov record is an estimated hazard ratio of 0.21, with a 99% two-sided confidence interval of 0.14–0.31.

Blinded independent imaging review

The registry-reported PFS definition states that progression was determined by blinded independent imaging review using RECIST 1.1 during the double-blind treatment period. Blinding of outcome assessment can reduce the opportunity for knowledge of treatment assignment to influence the classification of radiographic progression.

9. Statistical Methods Explained

Why is PFS analyzed as a time-to-event endpoint?

Because PFS incorporates the time from randomization until progression or death, a participant who remains event-free for longer contributes different information from one who experiences an event early. Time-to-event methods also allow censored participants to contribute their available follow-up without pretending that an unobserved event occurred at a particular time.

What does a hazard ratio of 0.21 mean?

Under the fitted Cox model, the estimated instantaneous event rate for the lenvatinib 24 mg group was about 21% of the corresponding rate for placebo. The complement, 1 − 0.21 = 0.79, gives the directly derived statement that the estimated hazard was about 79% lower. Neither statement is an absolute probability of remaining progression-free.

Why is the confidence interval 99% rather than 95%?

The registry-reported analysis explicitly reports a 99% two-sided confidence interval. A confidence interval describes uncertainty around the estimated parameter under its specified confidence procedure. The choice of confidence level changes the width of the interval; it does not change the point estimate itself.

Why is a p-value not enough to describe the treatment effect?

A p-value is designed to quantify evidence against a specified null hypothesis under a statistical model. It does not describe how large an effect is. Here, the effect magnitude is conveyed by the hazard ratio of 0.21, while the 99% confidence interval of 0.14–0.31 conveys precision. The ClinicalTrials.gov record does not provide a numerical p-value.

Why does the ITT population matter?

Analyzing all randomized participants according to the randomized comparison maintains the treatment groups created by randomization. This reduces the risk that post-randomization treatment behavior selectively changes the groups being compared. It does not eliminate missing-data, censoring, adherence, or other sources of potential bias.

Why does blinding matter for PFS?

The endpoint depends on identifying disease progression from imaging. The registry-reported definition states that progression was determined by blinded independent imaging review. Blinding can reduce the possibility that knowledge of treatment assignment influences assessment of whether radiographic criteria for progression have been met.

What is the difference between a hazard ratio and a risk ratio?

A hazard ratio compares modeled instantaneous event rates over time. A risk ratio compares cumulative probabilities of an event over a specified period. They are therefore not interchangeable. A hazard ratio of 0.21 should not be rewritten as a 0.21 risk ratio without an appropriate time-specific risk analysis.

10. Confidence Interval: Reading the 99% CI

The posted hazard-ratio estimate is 0.21, while the 99% two-sided confidence interval extends from 0.14 to 0.31. This interval is important because the point estimate alone can give a false impression of certainty.

Reported hazard-ratio interval
Lower bound
0.14
Estimate
0.21
Upper bound
0.31

All values in the reported interval are below 1.00. For a hazard-ratio comparison in which values below 1 favor lenvatinib relative to placebo, the entire registry-reported 99% interval is on the treatment-favoring side of the null value. That is a descriptive interpretation of the reported interval; it should not be converted into an unreported p-value.

Precision versus clinical magnitude: a relatively narrow confidence interval can indicate greater statistical precision, but precision and clinical importance are different concepts. The hazard ratio describes a relative time-to-event effect; it does not by itself describe absolute differences in months, survival probabilities, or individual patient benefit.

11. Safety Results

The ClinicalTrials.gov record reports serious adverse events by treatment group and risk set. These figures are distinct from the primary PFS analysis and should be interpreted as safety-event counts rather than efficacy outcomes.

Period / groupSerious adverse events affectedAt risk
Randomization Phase: Lenvatinib 24 mg171261
Randomization Phase: Placebo31131
OOL, Treatment Period: Lenvatinib 24 mg6285
OOL, Treatment Period: Lenvatinib 20 mg1630

The two Randomization Phase rows correspond to the same groups used in the posted primary PFS comparison. The OOL, Treatment Period rows are reported separately in the ClinicalTrials.gov record and should not be combined with the Randomization Phase safety figures as though they represented the same analysis population.

Do not infer efficacy from safety counts: the serious-adverse-event figures describe safety events among specified risk sets. They are not adjusted treatment-effect estimates, and they should not be compared using the PFS hazard ratio.

12. Safety and Efficacy Use Different Statistical Questions

Efficacy question

PFS asks when the first qualifying progression or death occurred after randomization. The posted analysis uses a time-to-event hazard ratio.

Safety question

Serious adverse-event reporting describes affected participants within specified treatment-period risk sets.

Why populations matter

The efficacy analysis explicitly uses all randomized participants. Safety rows are presented with their own affected and at-risk counts.

Why the measures differ

A hazard ratio and an affected/at-risk count answer different questions and should not be treated as interchangeable measures of treatment effect.

13. What Is and Is Not Reported in the Supplied Analysis

The ClinicalTrials.gov record contains one formal statistical analysis: the primary PFS comparison between Randomization Phase lenvatinib 24 mg and placebo. Four outcome measures are reported in the registry profile, but the ClinicalTrials.gov record does not provide detailed statistical analyses for additional endpoints.

ItemSupplied information
Primary endpointProgression Free Survival (PFS)
Primary endpoint typeTime-to-event
Formal statistical analysisYes
Number of statistical analyses reported1
Analysis populationFull analysis set / Intent-to-Treat Analysis Set; all randomized participants
ComparisonRandomization Phase: Lenvatinib 24 mg vs Randomization Phase: Placebo
Effect measureHazard ratio; reported as Cox Proportional Hazard
Estimate0.21
Confidence interval99% two-sided: 0.14–0.31
p-valueNot reported in the registry-reported statistical analysis
Secondary efficacy estimatesNot reported in the statistical analysis

This distinction is deliberate. A complete statistical analysis should not fill gaps in a registry record with numbers from memory or from a different source when the requested data rules restrict the numerical evidence to the ClinicalTrials.gov record.

14. Design Topics Not Supported by the Supplied Data

Several statistical design features commonly discussed for phase 3 trials are not described in the registry-reported SELECT data. They therefore cannot be characterized here as features of this trial.

Design topicWhat can be stated from the ClinicalTrials.gov record
Non-inferiority marginNot reported; the registered hypothesis type is superiority.
CrossoverNot reported.
Factorial designNot reported; the design model is parallel.
Multiplicity strategyNot reported.
Interim analysisNot reported.
Missing-data or imputation strategyNot reported.
Stratification factorsNot reported.
Bayesian methodsNot reported.
Why this matters: absence of a detail from the ClinicalTrials.gov recordset is not evidence that the underlying trial protocol or statistical analysis plan lacked that feature. It means only that the ClinicalTrials.gov record does not support a factual description of it.

15. Interpreting the Primary Result as a Statistical Story

The SELECT primary analysis can be understood as a sequence of linked statistical decisions rather than as a single number.

01
RandomizationCreates treatment groups
02
Follow-upMeasures time to event
03
PFS eventProgression or death
04
Cox modelEstimates relative hazard
05
CIQuantifies uncertainty

First, randomization establishes the comparison framework. Second, PFS records the timing of the first qualifying event. Third, time-to-event methods accommodate censoring. Fourth, the Cox model summarizes the relative event rate through a hazard ratio. Finally, the confidence interval communicates uncertainty around that estimate. The resulting HR of 0.21 is therefore best understood as the output of the entire analysis framework rather than as a simple percentage difference between two groups.

A useful mental model

Randomization answers: who is being compared?

PFS answers: what event and time are being measured?

Kaplan-Meier answers: how does the event-free distribution evolve over time?

Cox regression answers: what relative hazard summarizes the treatment comparison?

The confidence interval answers: how uncertain is the estimated hazard ratio?

16. Limitations and Interpretation Issues

17. Why This Trial Matters Statistically

SELECT is a useful teaching case because the registry analysis illustrates a classic randomized time-to-event framework while also showing why a statistical result must be read in context.

ConceptHow it appears in SELECT
RandomizationThe trial uses randomized allocation.
BlindingThe registry classifies the trial as triple-masked and placebo-controlled.
Parallel designThe design model is parallel.
ITT analysisThe full analysis set included all randomized participants.
Time-to-event endpointPFS measures time from randomization to progression or death.
RECIST assessmentThe registry-reported PFS definition uses RECIST 1.1 and blinded independent imaging review.
Cox modelThe posted effect measure is reported as Cox Proportional Hazard.
Hazard ratioThe reported PFS estimate is HR 0.21.
Confidence intervalThe estimate has a 99% two-sided CI of 0.14–0.31.
SuperiorityThe registered hypothesis type is superiority.
Safety analysisSerious adverse events are reported by treatment period and risk set.

18. Statistical Concepts in This Trial

Learn more about the methods used in this trial:

19. Related Statistical Calculators

20. Sources

Continue with the underlying statistical methods

Explore the concepts that connect randomized clinical-trial design with survival analysis, hazard ratios, confidence intervals, and intention-to-treat analysis.

21. Record Summary

SELECT provides a clear example of how a randomized phase 3 trial can translate a time-to-event endpoint into a model-based treatment-effect estimate. The registry-reported primary analysis uses the full analysis set / Intent-to-Treat Analysis Set, includes all randomized participants, compares Randomization Phase lenvatinib 24 mg with placebo, and reports a Cox proportional-hazard effect measure. The resulting hazard ratio is 0.21, with a 99% two-sided confidence interval of 0.14–0.31.

The most important statistical lesson is that the hazard ratio should not be read in isolation. The endpoint definition establishes what event is being measured; the ITT population establishes who contributes to the randomized comparison; the time-to-event framework accounts for event timing and censoring; the Cox model supplies the relative hazard estimate; and the confidence interval communicates uncertainty. The ClinicalTrials.gov record does not report a numerical p-value, so none is added here.

Clinical Biostats methodology: A trial-results page should separate reported evidence from statistical interpretation. Where the ClinicalTrials.gov record provides an estimate and confidence interval, those quantities are interpreted directly. Where a methodological detail or additional result is not reported, it is not reconstructed from memory or from an unrelated source.