← Clinical Trials
Breast Cancer Phase 3 Completed NCT01419197

TH3RESA: Complete Statistical Analysis of Trastuzumab Emtansine in HER2-Positive Breast Cancer

An independent statistical analysis of the randomized phase 3 TH3RESA trial comparing trastuzumab emtansine with treatment of physician's choice in participants with HER2-positive breast cancer who had received at least two prior regimens of HER2-directed therapy.

Trial start: 2011-09  ·  Primary completion: 2013-02  ·  Enrollment: 602
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

TH3RESA was a randomized, parallel-group, open-label phase 3 trial evaluating trastuzumab emtansine versus treatment of physician's choice in participants with HER2-positive breast cancer who had received at least two prior regimens of HER2-directed therapy. The registry reports 602 enrolled participants, two arms, two co-primary time-to-event endpoints, and nine posted statistical analyses.

602
Enrolled
Randomized phase 3 trial
2
Arms
Parallel design
0.528
Primary PFS HR
95% CI 0.422–0.661
0.552
Primary OS HR
95% CI 0.369–0.826
FeatureTH3RESA
PhasePhase 3
ConditionBreast Cancer
PopulationParticipants with HER2-positive breast cancer who had received at least two prior regimens of HER2-directed therapy
DesignRandomized, parallel, open-label
AllocationRandomized
Primary purposeTreatment
InterventionsTrastuzumab emtansine; treatment of physician's choice
Primary endpointsProgression-free survival and overall survival
Primary endpoint typeTime-to-event
Results postedYes
Statistical analyses posted9
Lead sponsorHoffmann-La Roche

2. Clinical Question

The central question was whether trastuzumab emtansine produced a superior time-to-event outcome compared with treatment of physician's choice in participants with HER2-positive breast cancer who had already received at least two prior regimens of HER2-directed therapy.

Population

Participants with HER2-positive breast cancer who had received at least two prior regimens of HER2-directed therapy.

Intervention

Trastuzumab emtansine.

Comparator

Treatment of physician's choice.

Primary question

Does trastuzumab emtansine improve progression-free survival and overall survival relative to treatment of physician's choice?

3. Trial Design

01
Randomize602 participants
02
2 ArmsTrastuzumab emtansine vs TPC
03
FollowTime-to-event endpoints
04
AssessPFS / OS / response / survival
05
AnalyzeStratified methods
INTERVENTION

Trastuzumab Emtansine

  • Trastuzumab emtansine
  • Compared with treatment of physician's choice
  • Primary efficacy analysis used the randomized population
COMPARATOR

Treatment of Physician's Choice

  • Treatment of physician's choice
  • Compared directly with trastuzumab emtansine
  • Primary efficacy analysis used the randomized population
Allocation
Randomized.
Masking
None.
Design model
Parallel.
Primary purpose
Treatment.

4. Randomization, Stratification, and Analysis Populations

The registry reports that the primary progression-free survival and overall survival analyses were conducted in the randomized population. Participants were included in the treatment group to which they were randomized. Both primary time-to-event analyses were stratified by three prespecified factors.

Analysis featureRegistry description
Primary efficacy populationRandomized population: all participants who were randomized to the study.
Treatment assignmentParticipants were included in the treatment group to which they were randomized.
World regionUnited States, Western Europe, or Other
Number of prior regimens≤ 3 or > 3, excluding single-agent hormones, for treatment of metastatic or unresectable locally advanced/recurrent disease
Visceral diseaseAny visceral disease versus no visceral disease

Stratification is important because the comparison was not treated as though every participant came from one homogeneous population. The survival analysis preserved the stratified structure of the randomized design, while the Cox regression registry-reported the hazard-ratio estimate.

5. Endpoints

EndpointRegistered definitionTime frame
Progression-free Survival Time from randomization to the first documented disease progression by investigator assessment using Response Evaluation Criteria In Solid Tumors (RECIST) v1.1 or death from any cause, whichever occurred first. Progression-free survival was a co-primary endpoint. Baseline to the clinical cut-off date of 11 Feb 2013 (up to 2 years)
Overall Survival Time from randomization to death from any cause. Overall survival was a co-primary endpoint. Baseline to the clinical cut-off date of 11 Feb 2013 (up to 2 years)

The two primary endpoints are both time-to-event outcomes. That distinction drives much of the statistical methodology: patients can contribute information even when they have not experienced the event by the analysis cutoff, because their follow-up can be censored.

6. Statistical Methodology

Log-rank testing for time-to-event outcomes

The registry reports the log-rank test for both co-primary endpoints. The log-rank framework compares the survival experience of the randomized groups across follow-up rather than reducing the entire endpoint to a single binary outcome at one fixed time.

Hazard-ratio estimation with Cox regression

For both primary endpoints, the hazard ratio was estimated by Cox regression. The registry specifies stratification by world region, number of prior regimens, and presence of visceral disease.

Conceptual interpretation
HR = estimated hazard in trastuzumab emtansine group ÷ estimated hazard in treatment-of-physician's-choice group

A hazard ratio below 1 indicates a lower estimated instantaneous event rate in the trastuzumab emtansine group under the fitted time-to-event model. It is not a percentage of patients who experience the event and is not the same quantity as an absolute risk difference.

Cochran-Mantel-Haenszel analysis

The secondary objective response endpoint used a Cochran-Mantel-Haenszel test with stratified analysis. This approach compares categorical outcomes while accounting for the same reported stratification factors.

Wald / z-tests for survival percentages

The registry reports Wald / z-tests for the differences in 6-month and 1-year survival percentages. These analyses provide fixed-time contrasts rather than a single global comparison of the complete time-to-event distributions.

Two-sided confidence intervals

All posted statistical analyses report two-sided 95% confidence intervals. The interval quantifies uncertainty around the estimated effect under the corresponding statistical framework; it does not describe the range of effects that individual participants experienced.

7. Primary Results: Progression-free Survival

The registry reports a formal primary analysis of progression-free survival in the randomized population, comparing trastuzumab emtansine with treatment of physician's choice. The analysis used a stratified log-rank test, with the hazard ratio estimated by Cox regression.

Progression-free survival hazard ratio

0.528

95% CI: 0.422–0.661   ·   P < 0.0001

Two-sided 95% confidence interval; superiority hypothesis.

Primary endpointComparisonMethodEstimate95% CIP-value
Progression-free Survival Trastuzumab emtansine vs Treatment of Physician's Choice Stratified log-rank; Cox regression for HR HR 0.528 0.422–0.661 <0.0001
Clinical Biostats interpretation

The estimated hazard ratio of 0.528 means that, under the fitted stratified time-to-event model, the estimated instantaneous rate of progression or death in the trastuzumab emtansine group was approximately 52.8% of that in the treatment-of-physician's-choice group. Equivalently, 1 − 0.528 = 0.472, so the estimate corresponds to an approximately 47.2% lower estimated hazard.

The hazard ratio does not mean that 47.2% of participants avoided progression, nor does it mean that every participant experienced exactly a 47.2% reduction in risk. It is a relative time-to-event measure derived from a statistical model.

The 95% CI of 0.422–0.661 describes the statistical uncertainty surrounding the estimated hazard ratio. Because the interval lies below 1, the interval is consistent with a lower estimated hazard in the trastuzumab emtansine group under this analysis.

The P < 0.0001 value addresses evidence against the null hypothesis under the specified test. It does not measure the size of the treatment effect, and it should not be interpreted as the probability that the treatment effect is real.

The analysis was stratified by world region, number of prior regimens, and presence of visceral disease. The Cox-model interpretation also depends on the model's assumptions, including the proportional-hazards framework. Censoring and the definition of progression therefore remain important parts of the interpretation.

8. Primary Results: Overall Survival

Overall survival was the second co-primary endpoint. The registry defines it as the time from randomization to death from any cause and reports a stratified log-rank analysis with the hazard ratio estimated by Cox regression.

Overall survival hazard ratio

0.552

95% CI: 0.369–0.826   ·   P = 0.0034

Two-sided 95% confidence interval; superiority hypothesis.

Primary endpointComparisonMethodEstimate95% CIP-value
Overall Survival Trastuzumab emtansine vs Treatment of Physician's Choice Stratified log-rank; Cox regression for HR HR 0.552 0.369–0.826 0.0034
Clinical Biostats interpretation

The estimated overall-survival hazard ratio of 0.552 corresponds to an estimated instantaneous death rate that was approximately 55.2% of that in the treatment-of-physician's-choice group under the fitted model. The complementary calculation, 1 − 0.552 = 0.448, corresponds to an approximately 44.8% lower estimated hazard.

This does not mean that 44.8% of participants were saved, that 44.8% were cured, or that each individual participant experienced the same reduction in mortality risk. A hazard ratio summarizes a relative time-to-event comparison.

The 95% CI of 0.369–0.826 quantifies uncertainty around the estimated hazard ratio. It does not describe the range of survival times for individual participants or the probability that the true hazard ratio falls inside the interval.

The P = 0.0034 value measures the compatibility of the observed data with the null hypothesis under the specified testing framework. It is not an effect-size measure. A small P-value and a clinically meaningful effect are related concepts but are not interchangeable.

As with the PFS analysis, interpretation depends on the stratified design, censoring, and the assumptions underlying the Cox model. The analysis was designed to test superiority, and the confidence interval is two-sided.

9. Secondary Results: Objective Response

The registry reports a secondary endpoint measuring the percentage of participants with an objective response. Only participants with measurable disease at baseline were included in this analysis, and participants remained in their randomized treatment groups.

Difference in response percentage

22.7 percentage points

95% CI: 16.2–29.2   ·   P < 0.0001

Cochran-Mantel-Haenszel analysis; two-sided 95% confidence interval.

Secondary endpointMethodEffect measureEstimate95% CIP-value
Percentage of Participants With an Objective Response Cochran-Mantel-Haenszel test Difference in Response Percentage 22.7 16.2–29.2 <0.0001

The estimate is a difference in response percentage, not a hazard ratio. Its interpretation is therefore different from the primary survival analyses. A value of 22.7 indicates a 22.7-percentage-point difference in the reported response percentages between the randomized groups under the stratified analysis.

10. Secondary Results: 6-Month and 1-Year Survival

The registry separates the fixed-time survival comparisons into 6-month and 1-year analyses. Both use a Wald / z-test and report a difference in survival percentage.

Time pointMethodEffect measureEstimate95% CIP-value
6-month survival Wald / z-test Difference in Survival Percentage 12.6 5.03–20.09 0.0011
1-year survival Wald / z-test Difference in Survival Percentage 11.7 -5.41–28.75 0.1805

These estimates illustrate why fixed-time survival percentages and hazard ratios should not be treated as interchangeable. The 6-month comparison has a confidence interval entirely above zero, whereas the 1-year comparison has a confidence interval that includes zero.

Clinical Biostats interpretation

The 6-month survival estimate of 12.6 is a difference in survival percentage between the randomized groups. Its 95% CI of 5.03–20.09 describes uncertainty around that fixed-time difference.

The 1-year survival estimate of 11.7 is also a difference in survival percentage, but its 95% CI of -5.41–28.75 includes zero. The corresponding P = 0.1805 indicates that this particular fixed-time comparison did not provide the same level of statistical evidence against a zero difference under the reported test.

Neither result overturns or replaces the co-primary time-to-event analyses. A fixed-time comparison examines survival at one selected time point, whereas a log-rank/Cox analysis uses information across follow-up.

11. Secondary Results: Time to Pain Symptom Progression

Time to pain symptom progression was analyzed in the randomized population among participants with a baseline pain score and at least one post-baseline pain score. The registry reports a stratified log-rank analysis with a Cox-model hazard ratio.

Time to pain symptom progression

HR 1.115

95% CI: 0.819–1.517   ·   P = 0.4952

Two-sided 95% confidence interval; superiority hypothesis.

EndpointMethodEffect measureEstimate95% CIP-value
Time to Pain Symptom Progression Stratified log-rank; Cox regression framework Hazard Ratio 1.115 0.819–1.517 0.4952

An HR of 1.115 is above 1, so the point estimate corresponds to a higher estimated instantaneous event rate in the trastuzumab emtansine group for this endpoint. However, the 95% CI of 0.819–1.517 includes 1, and the reported P-value is 0.4952. This secondary endpoint therefore does not provide evidence of a clear difference under the reported analysis.

12. Secondary Results: Final Overall Survival Analysis

The registry also posts a later secondary analysis labeled Overall Survival (Final Analysis), using a clinical cut-off date of 13 Feb 2015 (up to 4 years). This is distinct from the earlier co-primary overall-survival analysis with a 11 Feb 2013 cutoff.

Final overall survival hazard ratio

0.677

95% CI: 0.539–0.850   ·   P = 0.0007

Stratified log-rank analysis with Cox regression for the hazard ratio.

AnalysisTime frameMethodEstimate95% CIP-value
Overall Survival (Final Analysis) Baseline to the clinical cut-off date of 13 Feb 2015 (up to 4 years) Stratified log-rank; Cox regression for HR HR 0.677 0.539–0.850 0.0007

The final-analysis hazard ratio of 0.677 is closer to 1 than the earlier overall-survival estimate of 0.552. That does not mean that the two estimates contradict one another: they correspond to different clinical cut-off dates and therefore different accumulated follow-up and event information.

13. Secondary Results: Final 6-Month and 1-Year Survival

The registry reports two additional fixed-time survival analyses at the final-analysis cutoff of 13 Feb 2015. Both use Wald / z-tests and differences in survival percentage.

Time pointTime frameEstimate95% CIP-value
6-month survival Baseline to the clinical cut-off date of 13 Feb 2015 (up to 4 years) 12.4 5.67–19.14 0.0003
1-year survival Baseline to the clinical cut-off date of 13 Feb 2015 (up to 4 years) 11.0 2.58–19.33 0.0104

Both final-analysis fixed-time estimates have confidence intervals above zero. They should nevertheless be interpreted as fixed-time differences, not as substitutes for the hazard-ratio analysis of the complete time-to-event experience.

14. Safety Results

The registry provides serious adverse-event counts by treatment group. It also reports a third category, participants receiving trastuzumab emtansine after treatment of physician's choice. That post-TPC category should not be combined with the randomized trastuzumab emtansine arm when describing the randomized comparison.

GroupSerious adverse events affectedAt risk
Trastuzumab Emtansine 102 403
Treatment of Physician's Choice (TPC) 41 184
Trastuzumab Emtansine - Post TPC Treatment 19 94

The serious-adverse-event data are descriptive counts rather than a reported formal hypothesis test in the statistical analyses posted on ClinicalTrials.gov. They therefore should not be converted into an efficacy-style treatment-effect claim without an appropriate safety analysis framework.

Population distinction: the post-TPC trastuzumab emtansine group is reported separately in the registry data. It is not the same population as the randomized trastuzumab emtansine group and should not be added to it when interpreting the randomized trial comparison.

15. Statistical Methods Explained

Why was a log-rank test used for the primary endpoints?

Progression-free survival and overall survival are time-to-event outcomes. Participants may have different follow-up durations, and some may not have experienced the event by the clinical cut-off. The log-rank test is designed to compare the event-time distributions of two groups while accounting for this censoring structure.

What does a hazard ratio of 0.528 mean?

A hazard ratio of 0.528 means that the estimated instantaneous event rate in the trastuzumab emtansine group was 52.8% of the comparator group's estimated rate under the fitted model. The complementary value, 47.2%, is the corresponding relative reduction in estimated hazard. It is not an absolute reduction in the percentage of participants who progress or die.

Why was the analysis stratified?

The registry identifies world region, number of prior regimens, and presence of visceral disease as stratification factors. Stratification allows the time-to-event comparison to account for these factors rather than treating the trial population as completely unstructured.

What does the 95% confidence interval tell us?

A 95% confidence interval describes statistical uncertainty around the estimated effect under the relevant model and sampling framework. For the PFS HR, the interval is 0.422–0.661; for the primary OS HR, it is 0.369–0.826. The interval is not a prediction interval for individual participants.

Why is a P-value not an effect-size measure?

A P-value evaluates the compatibility of the observed data with a specified null hypothesis under the statistical test. It depends on the effect estimate, variability, sample information, and analysis structure. The magnitude of the treatment effect is communicated by the hazard ratio or difference estimate and its confidence interval, not by the P-value alone.

Why are the 6-month and 1-year analyses different from the hazard-ratio analysis?

The fixed-time analyses estimate a difference in survival percentage at a particular time point and use a Wald / z-test. The hazard-ratio analysis summarizes the relative time-to-event experience over the analyzed follow-up. A treatment comparison can therefore look different at a selected time point than it does when the complete event-time information is considered.

16. Understanding the Primary Hazard Ratios

PFS

The PFS estimate was HR 0.528 with a 95% CI of 0.422–0.661. The point estimate is below 1, corresponding to a lower estimated hazard of progression or death in the trastuzumab emtansine group.

OS

The primary OS estimate was HR 0.552 with a 95% CI of 0.369–0.826. The point estimate is below 1, corresponding to a lower estimated hazard of death in the trastuzumab emtansine group.

Why both measures matter

PFS and OS answer different questions. PFS incorporates documented disease progression or death, whichever occurs first, whereas OS is defined solely by death from any cause. Reporting both co-primary endpoints provides two distinct perspectives on treatment effect.

17. Comparing the Primary and Final OS Analyses

AnalysisClinical cut-offHazard ratio95% CIP-value
Overall Survival — primary analysis 11 Feb 2013 0.552 0.369–0.826 0.0034
Overall Survival — final analysis 13 Feb 2015 0.677 0.539–0.850 0.0007

The two estimates should be read as analyses of different information sets. The earlier analysis uses the 11 Feb 2013 clinical cut-off, while the final analysis uses the 13 Feb 2015 cut-off. A later estimate can change as additional events accumulate and the risk sets evolve.

This is an important general lesson in longitudinal clinical-trial analysis: a hazard ratio is not a permanent property of a treatment. It is an estimate derived from a specified analysis population, follow-up period, event definition, censoring structure, and statistical model.

18. Time-to-Event Analysis in TH3RESA

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

The Kaplan-Meier estimator updates the estimated probability of remaining event-free at each observed event time, using the number of participants at risk immediately before that time and the number experiencing the event.

Although the ClinicalTrials.gov record reports formal log-rank and Cox analyses, they do not provide the underlying event-by-event risk-set information needed to reconstruct a Kaplan-Meier curve from the ClinicalTrials.gov record alone. A valid curve reconstruction would require those underlying event and censoring times or an appropriately detailed source.

Censoring is part of the analysis

For time-to-event endpoints, a participant who has not experienced the endpoint by the analysis cut-off does not simply disappear from the dataset. Their observed follow-up contributes information up to the time at which their outcome becomes censored. This is one reason time-to-event methods are preferable to treating every participant as though everyone had identical follow-up.

The proportional-hazards issue

The Cox model expresses the treatment effect through a hazard ratio. Interpreting one hazard ratio as a concise summary assumes that the model provides an appropriate representation of the relative hazard over time. If hazards change substantially relative to one another, a single HR can conceal important temporal features of the treatment effect.

19. Statistical Interpretation of the Secondary Endpoints

EndpointEstimate95% CIP-valueStatistical reading
Objective response 22.7 16.2–29.2 <0.0001 Estimated response-percentage difference above zero
6-month survival 12.6 5.03–20.09 0.0011 Estimated survival-percentage difference above zero
1-year survival 11.7 -5.41–28.75 0.1805 Confidence interval includes zero
Time to pain symptom progression HR 1.115 0.819–1.517 0.4952 Confidence interval includes 1
Final overall survival HR 0.677 0.539–0.850 0.0007 Estimated death hazard below comparator
Final 6-month survival 12.4 5.67–19.14 0.0003 Estimated survival-percentage difference above zero
Final 1-year survival 11.0 2.58–19.33 0.0104 Estimated survival-percentage difference above zero

These analyses use different effect measures, so they should not be ranked by their P-values or compared numerically as though they were estimates of the same quantity. A hazard ratio, a percentage-point difference, and a fixed-time survival difference answer different statistical questions.

20. Multiplicity and the Co-Primary Endpoints

The registry identifies progression-free survival and overall survival as the two co-primary endpoints and labels their hypotheses as superiority. Because both endpoints are primary, the interpretation of the trial's confirmatory evidence depends on the prespecified testing framework governing those endpoints.

The ClinicalTrials.gov record does not specify an alpha-allocation or multiplicity-adjustment procedure between the two co-primary endpoints. Accordingly, this page does not infer one from the reported P-values.

Important distinction: the presence of two co-primary endpoints means that the statistical interpretation should follow the trial's prespecified multiplicity framework. The ClinicalTrials.gov record identifies the endpoints, methods, estimates, confidence intervals, and P-values, but do not provide enough information to reconstruct an unreported alpha-allocation rule.

21. Stratified Analysis and the Cochran-Mantel-Haenszel Framework

TH3RESA provides a useful example of how stratification can be used across different endpoint types. The survival analyses were stratified by world region, prior-regimen count, and visceral disease. The objective-response analysis also used a stratified Cochran-Mantel-Haenszel test based on those factors.

Stratification factorCategoriesWhere used in registry-reported analyses
World region United States; Western Europe; Other Primary and selected secondary stratified analyses
Number of prior regimens ≤ 3; > 3 Primary and selected secondary stratified analyses
Visceral disease Any visceral disease; no visceral disease Primary and selected secondary stratified analyses

Stratification does not mean that treatment effects are separately estimated and then simply averaged. Rather, the statistical procedure accounts for the strata while producing an overall treatment comparison. This can improve alignment between the analysis and the randomized design.

22. Planned Analysis vs Reported Analysis

For TH3RESA, the registry reports results for both co-primary endpoints and seven additional statistical analyses. Therefore, this page focuses on the reported analyses rather than constructing a hypothetical planned-analysis section.

Where the registry identifies a formal method, the interpretation above uses that method directly: log-rank testing and Cox regression for the time-to-event analyses, Cochran-Mantel-Haenszel testing for objective response, and Wald / z-testing for fixed-time survival percentages.

23. Important Limitations and Interpretation Issues

24. Why This Trial Matters Statistically

TH3RESA is a useful teaching case because its registry results bring together several core methods used in randomized oncology trials: randomized treatment assignment, stratified time-to-event analysis, Cox hazard ratios, log-rank testing, fixed-time survival comparisons, and stratified categorical analysis.

ConceptHow it appears in TH3RESA
RandomizationParticipants were randomized to trastuzumab emtansine or treatment of physician's choice.
Time-to-event endpointsProgression-free survival and overall survival were co-primary endpoints.
Log-rank testUsed for the primary PFS and OS analyses and selected secondary time-to-event analyses.
Hazard ratioUsed to quantify the relative treatment effect for PFS, OS, time to pain symptom progression, and final OS.
Cox regressionUsed to estimate the primary PFS and OS hazard ratios and the final OS hazard ratio.
Stratified analysisAccounts for world region, number of prior regimens, and visceral disease in specified analyses.
Cochran-Mantel-Haenszel testUsed for the objective-response comparison.
Wald / z-testUsed for 6-month and 1-year survival percentage differences.
Confidence intervalsTwo-sided 95% intervals accompany the reported treatment-effect estimates.
Multiple analysis time pointsThe registry contains a primary OS analysis with a 11 Feb 2013 cutoff and a final OS analysis with a 13 Feb 2015 cutoff.

25. Clinical Interpretation vs Statistical Interpretation

Statistical interpretation

The two co-primary time-to-event analyses report hazard ratios below 1, with two-sided 95% confidence intervals below 1 and reported P-values of <0.0001 for PFS and 0.0034 for the primary OS analysis.

Endpoint-specific interpretation

The secondary results are heterogeneous by endpoint and effect measure. Objective response and several fixed-time survival comparisons report positive percentage differences, while time to pain symptom progression has an HR of 1.115 with a confidence interval including 1.

This distinction matters because a clinical-trial evidence set is not summarized adequately by a single P-value. Treatment effects need to be understood in relation to their endpoint definitions, effect measures, uncertainty intervals, analysis populations, follow-up period, and statistical method.

26. Trial Timeline

2011-09 · Trial start

Study initiated

The registry identifies September 2011 as the trial start date.

2013-02 · Primary completion

Primary study completion

The registry identifies February 2013 as the primary completion period.

11 Feb 2013 · Primary cutoff

Primary PFS and OS analysis cutoff

The co-primary endpoints were analyzed from baseline to the clinical cut-off date of 11 Feb 2013, up to 2 years.

13 Feb 2015 · Final OS cutoff

Final overall-survival analysis

The registry reports a later overall-survival analysis from baseline to the clinical cut-off date of 13 Feb 2015, up to 4 years.

27. Related Tutorials

Learn more about the methods used in this trial:

28. Related Calculators

29. Sources

Continue through the Clinical Biostats statistical pathway

Connect this trial's endpoints and methods to deeper statistical tutorials and practical analysis tools.

30. Record Summary

TH3RESA provides a compact example of several important clinical-trial statistical principles. The trial randomized 602 participants in a parallel, open-label phase 3 design and specified progression-free survival and overall survival as co-primary time-to-event endpoints. The primary analyses used stratified log-rank tests, with hazard ratios estimated by Cox regression. The reported PFS hazard ratio was 0.528 (95% CI 0.422–0.661; P < 0.0001), while the primary OS hazard ratio was 0.552 (95% CI 0.369–0.826; P = 0.0034).

The registry also reports secondary analyses using three different statistical structures: Cochran-Mantel-Haenszel testing for objective response, Wald / z-tests for fixed-time survival percentages, and stratified log-rank/Cox methods for time to pain symptom progression and the final overall-survival analysis. The resulting estimates should be interpreted according to their specific effect measures rather than reduced to a single numerical measure of "benefit."

The final overall-survival analysis, using the 13 Feb 2015 clinical cut-off, reported an HR of 0.677 (95% CI 0.539–0.850; P = 0.0007). Comparing this estimate with the earlier primary OS analysis illustrates an important statistical principle: treatment-effect estimates can change as follow-up and event information accumulate. The analysis cutoff, endpoint definition, analysis population, stratification, censoring, and model all form part of the statistical meaning of a reported estimate.

Clinical Biostats methodology: A trial-results page should distinguish reported evidence from statistical interpretation. For TH3RESA, the most informative reading combines the randomized design, endpoint definitions, stratified analysis methods, hazard ratios, confidence intervals, fixed-time survival differences, secondary endpoint results, and the limitations of interpreting each measure.