This page separates reported trial results from statistical interpretation. Trial-specific numerical results and design facts are restricted to the ClinicalTrials.gov record. Where the registry does not provide a particular statistical detail, this page does not infer it from external publications.
1. Trial at a Glance
RECOURSE was a completed phase 3 randomized, parallel-group trial evaluating TAS-102 versus placebo in patients with metastatic colorectal cancer refractory to standard chemotherapies. The registered primary endpoint was overall survival, a time-to-event outcome analyzed using a stratified log-rank test in the intention-to-treat population.
| Feature | RECOURSE |
|---|---|
| Trial name | RECOURSE |
| Brief title | Study of TAS-102 in Patients With Metastatic Colorectal Cancer Refractory to Standard Chemotherapies |
| Phase | Phase 3 |
| Status | Completed |
| Condition | Colorectal Cancer |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | Quadruple |
| Primary purpose | Treatment |
| Enrollment | 800 |
| Interventions | TAS-102; Placebo |
| Primary endpoint | Overall Survival |
| Primary endpoint type | Time-to-event |
| Statistical method | Stratified log-rank test |
| Effect measure | Hazard ratio |
| Hypothesis type | Superiority |
| Lead sponsor | Taiho Oncology, Inc. |
| Sponsor type | Industry |
| ClinicalTrials.gov identifier | NCT01607957 |
2. Clinical Question
The statistical question is whether assignment to TAS-102 is associated with a different overall-survival experience than assignment to placebo in patients with metastatic colorectal cancer refractory to standard chemotherapies.
Population
Patients with metastatic colorectal cancer refractory to standard chemotherapies, as stated in the trial's brief title.
Intervention
TAS-102.
Comparator
Placebo.
Primary question
Under a superiority framework, does the overall-survival distribution differ between the randomized TAS-102 and placebo groups?
This framing is important because the endpoint is not a simple binary outcome measured at one fixed date. Overall survival records the time from randomization to death, with participants who remain alive at the specified cutoff handled as censored observations in the posted analysis.
3. Trial Design
TAS-102
- Intervention type: drug
- Evaluated against placebo
- Included in the randomized parallel-group comparison
Placebo
- Comparator type: drug
- Evaluated against TAS-102
- Included in the randomized parallel-group comparison
The registry identifies the study as randomized, parallel, quadruple-masked, and treatment-oriented. Those design features establish the basic framework for interpreting the reported comparison. Randomization creates the principal basis for comparing groups, while masking is intended to reduce the potential for knowledge of assigned treatment to influence trial conduct or assessment.
4. Trial Timeline
Trial start
The registered study start date was 2012-06-17.
Primary completion
The registered primary completion date was 2014-01-31.
Registry status
The trial is listed as completed, with results posted on ClinicalTrials.gov.
The dates describe the registry's study timeline. They should not be confused with the event-analysis cutoff, which is described separately in the posted overall-survival analysis.
5. Endpoints
| Endpoint | Registered definition / time frame | Role |
|---|---|---|
| Overall Survival | Overall survival was defined as the time from the date of randomization to the date of death for participants. If a participant discontinued study medication for reasons other than radiologic disease progression, the participant was followed for tumor response until radiologic disease progression or initiation of new anticancer therapy. Time frame: Every 8 weeks, up to 12 months after the last participant was randomized or until the target number of events (deaths) was met, whichever was later. (Overall survival data was collected till 24 Jan 2014 which was date of observation of the 571st death) | Primary |
| Progression-free Survival | Time frame: Every 8 weeks, up to 12 months after the last participant was randomized or until the date of the investigator-assessed radiological disease progression or death due to any cause,whichever was later. (Progression free survival cutoff: 31 Jan 2014) | Secondary |
Only one primary endpoint is registered: overall survival. Progression-free survival is represented among the posted secondary analyses. Both are time-to-event outcomes, so their interpretation depends on the timing of events and censoring rather than simply on whether an event occurred.
6. Analysis Populations and Censoring
The posted primary overall-survival analysis was performed in the intention-to-treat (ITT) population. This means the statistical comparison retained participants according to their randomized treatment assignment rather than redefining treatment groups based on later treatment exposure.
| Analysis feature | Registry-reported approach |
|---|---|
| Primary efficacy population | Intention-to-treat population |
| Groups compared | TAS-102 vs Placebo |
| Primary endpoint | Overall Survival |
| Primary method | Stratified log-rank test |
| Effect measure | Hazard Ratio |
| Survivors at cutoff | Censored on the cutoff date post consent |
The censoring rule matters because a participant who is alive when follow-up ends has not reported an observed death time. Instead, that participant contributes survival information up to the cutoff. The analysis therefore uses both events and the amount of observed follow-up.
For an individual who dies during follow-up, the event time is observed. For an individual who remains alive at the analysis cutoff, the observed follow-up ends at censoring.
The registry does not provide additional details in the ClinicalTrials.gov record about missing-data imputation, specific censoring rules beyond the stated cutoff rule, or the handling of other potential deviations from the assigned treatment. Those issues therefore should not be reconstructed from assumptions.
7. Primary Result: Overall Survival
The registered primary endpoint was analyzed by a stratified log-rank test in the ITT population, comparing TAS-102 with placebo. The reported effect measure was a hazard ratio under a superiority hypothesis.
Hazard ratio for death
95% CI: 0.58–0.81 · P < 0.0001
Analysis population: ITT · Method: stratified log-rank test
| Primary endpoint | TAS-102 vs Placebo |
|---|---|
| Outcome | Overall Survival |
| Effect measure | Hazard Ratio |
| Estimate | 0.68 |
| 95% confidence interval | 0.58–0.81 |
| P-value | <0.0001 |
| Hypothesis | Superiority |
| Analysis population | ITT |
| Statistical method | Stratified log-rank test |
The reported HR of 0.68 means that the estimated hazard of death associated with TAS-102 was 0.68 times the corresponding hazard in the placebo group under the time-to-event comparison. Expressed as a simple relative-hazard interpretation, this corresponds to an estimated 32% lower hazard for TAS-102 relative to placebo.
The HR does not mean that 32% of patients benefited, that 32% fewer patients died, or that each individual patient experienced exactly a 32% reduction in risk. A hazard ratio is a relative time-to-event measure, not an absolute survival probability.
The 95% CI of 0.58–0.81 describes statistical uncertainty around the estimated hazard ratio under the analysis framework. It does not describe the range of outcomes that individual patients might experience, and it does not mean that 95% of individual treatment effects fall within the interval.
The P-value <0.0001 addresses the statistical evidence against the null hypothesis under the specified testing framework. It does not measure the size of the treatment effect. Effect size is conveyed by the HR, while the confidence interval provides information about precision.
Because this is a time-to-event analysis, interpretation also depends on censoring and on the assumptions underlying the hazard-ratio representation. The ClinicalTrials.gov record identifies the stratified log-rank test and hazard ratio but do not report the specific model used to estimate the HR or whether a proportional-hazards diagnostic was performed. Those details should therefore not be inferred.
Relative interpretation of the estimate
For a treatment-versus-control comparison, an HR below 1 indicates a lower estimated instantaneous event rate in the treatment group relative to the comparator, within the statistical framework used for the analysis.
The confidence interval is entirely below 1.00: its lower bound is 0.58 and its upper bound is 0.81. Thus the interval remains on the side of the hazard-ratio scale corresponding to a lower estimated hazard for TAS-102. The P-value provides a separate measure of compatibility with the null hypothesis and should not be substituted for the effect estimate itself.
8. Secondary Result: Progression-free Survival
Progression-free survival was posted as a secondary time-to-event endpoint. The registry reports the same broad analytical framework used for the primary endpoint: ITT analysis, comparison of TAS-102 with placebo, and a stratified log-rank test.
Hazard ratio for progression-free survival
95% CI: 0.41–0.57 · P < 0.0001
Analysis population: ITT · Method: stratified log-rank test
| Secondary endpoint | TAS-102 vs Placebo |
|---|---|
| Outcome | Progression-free Survival |
| Effect measure | Hazard Ratio |
| Estimate | 0.48 |
| 95% confidence interval | 0.41–0.57 |
| P-value | <0.0001 |
| Hypothesis | Superiority |
| Analysis population | ITT |
| Statistical method | Stratified log-rank test |
The reported HR of 0.48 means that the estimated instantaneous rate of the progression-free-survival event was 0.48 times the corresponding rate in the placebo group under the reported time-to-event analysis. As a simple relative interpretation, this corresponds to an estimated 52% lower hazard for TAS-102.
That statement is about the event hazard, not a statement that 52% of patients avoided progression or death. The HR should not be converted directly into an absolute percentage of patients benefiting.
The 95% CI of 0.41–0.57 describes uncertainty around the estimated hazard ratio. Its width provides information about precision, while its position relative to 1.00 indicates the direction of the estimated relative treatment effect.
The P-value <0.0001 indicates strong statistical evidence against the null hypothesis under the posted testing framework. It does not indicate that the treatment effect is “48% significant,” nor does it quantify clinical importance.
Progression-free survival is also a time-to-event endpoint, so censoring and the precise definition of the event are central to interpretation. The ClinicalTrials.gov record does not provide the complete endpoint definition after the words “investigator-assessed,” so no additional event-definition details are added here.
Comparing the two hazard ratios
| Endpoint | HR | 95% CI | P-value | Role |
|---|---|---|---|---|
| Overall Survival | 0.68 | 0.58–0.81 | <0.0001 | Primary |
| Progression-free Survival | 0.48 | 0.41–0.57 | <0.0001 | Secondary |
The two estimates answer related but different questions. Overall survival concerns death, whereas progression-free survival concerns the earlier time-to-event endpoint registered in the trial. The numerical difference between 0.68 and 0.48 should not be treated as proof that one endpoint shows a particular magnitude of biological benefit relative to the other. Different endpoints have different event processes and statistical properties.
9. Safety Results
The ClinicalTrials.gov record includes serious adverse events by randomized arm as affected participants over participants at risk. These data provide an arm-level safety summary, but they do not provide the complete adverse-event profile.
| Safety measure | Affected | At risk | Proportion from reported counts |
|---|---|---|---|
| TAS-102 serious adverse events | 158 | 533 | 158/533 |
| Placebo serious adverse events | 89 | 265 | 89/265 |
The ClinicalTrials.gov record reports the counts as TAS-102: 158/533 and Placebo: 89/265. The denominators sum to 798, whereas total enrollment is 800. The source data do not explain that difference, so this page does not infer why the safety denominators differ from the overall enrollment.
It is also important not to turn the serious-adverse-event counts into a broad statement about the entire safety profile. The ClinicalTrials.gov recordset contains no complete table of adverse-event grades, individual adverse-event terms, treatment discontinuations, or other safety categories.
10. Statistical Methodology
Time-to-event analysis
Both posted efficacy analyses concern time-to-event outcomes. This distinguishes RECOURSE from a trial whose primary endpoint is simply the proportion of patients with an outcome at one prespecified visit.
For a time-to-event endpoint, two participants can both remain event-free at the end of their observed follow-up, yet contribute different amounts of information if they have been followed for different lengths of time. Censoring allows such participants to contribute their observed follow-up without pretending that an unobserved event time is known.
The analysis uses the sequence of observed events and the participants at risk at those event times rather than reducing every participant to a single yes/no outcome.
Stratified log-rank test
The registry identifies the stratified log-rank test as the statistical method for both the primary overall-survival analysis and the secondary progression-free-survival analysis.
The log-rank framework compares the observed pattern of events between treatment groups over follow-up. A stratified version performs the comparison while accounting for prespecified strata. The ClinicalTrials.gov record does not identify the specific stratification factors, so this page does not name or reconstruct them.
Hazard ratio
The registry reports the hazard ratio as the effect measure for both posted efficacy analyses. The hazard ratio is a relative measure of event hazard over time.
An HR of 1 would represent equal hazards under the model or comparison being used. An HR below 1 favors a lower event hazard for the treatment group; an HR above 1 corresponds to a higher event hazard.
The registry data do not identify the specific regression model used to obtain the posted hazard ratios. A Cox proportional-hazards model is a common way to estimate hazard ratios in clinical-trial survival analysis, but it should not be attributed to RECOURSE without supporting trial-specific documentation. Accordingly, this page describes the reported HR without assigning an unreported model.
Intention-to-treat analysis
The primary overall-survival analysis was performed in the ITT population, and the secondary progression-free-survival analysis was also performed in the ITT population. This preserves the randomized treatment assignment as the basis of the efficacy comparison.
The value of ITT is especially important in a randomized trial because removing participants after randomization can change the comparability created by randomization. ITT does not mean that every participant necessarily receives every planned treatment exposure; it means that the efficacy comparison remains anchored to randomized assignment.
Superiority testing
The registered hypothesis type is superiority. Thus the analytical question is whether the evidence supports a difference in the specified direction rather than whether TAS-102 falls within a prespecified non-inferiority margin.
There is no non-inferiority margin in the ClinicalTrials.gov record, so non-inferiority logic is not applied to this trial.
11. Confidence Intervals and Statistical Precision
A confidence interval adds information that a point estimate and P-value cannot provide alone. For RECOURSE, the primary overall-survival estimate is 0.68, while the 95% confidence interval extends from 0.58 to 0.81.
Point estimate
The HR of 0.68 is the single reported estimate of the relative treatment effect for overall survival.
Interval estimate
The 95% CI of 0.58–0.81 expresses statistical uncertainty around that point estimate under the analysis framework.
P-value
The P-value <0.0001 addresses evidence against the null hypothesis; it is not a measure of effect magnitude.
Clinical meaning
Neither the HR nor its confidence interval directly gives an absolute probability of survival for an individual patient.
The same distinction applies to progression-free survival: the HR is 0.48, while the 95% CI is 0.41–0.57. The interval gives a range of statistical uncertainty around the estimate rather than a distribution of individual patient outcomes.
12. Statistical Methods Explained
Why was an ITT analysis used for overall survival?
The registry explicitly states that the overall-survival analysis was performed in the ITT population. An ITT framework retains participants according to randomized assignment, preserving the treatment comparison generated by randomization. This is particularly useful for an efficacy question because treatment discontinuation or other post-randomization events should not automatically redefine the groups being compared.
Why use a stratified log-rank test?
The registry reports a stratified log-rank test for both overall survival and progression-free survival. A log-rank test is designed for comparing time-to-event experience between groups. Stratification allows the comparison to account for the trial's prespecified strata. Because the ClinicalTrials.gov record does not identify those strata, the statistical principle can be explained without assigning particular variables to the strata.
What does an overall-survival HR of 0.68 mean?
An HR of 0.68 means that the estimated death hazard for TAS-102 relative to placebo was 0.68 under the reported analysis. As a simple relative interpretation, that is a 32% lower estimated hazard. It does not mean a 32% absolute improvement in survival probability, nor does it mean that exactly 32% of participants benefited.
Why is the confidence interval important?
The 95% CI of 0.58–0.81 shows the statistical uncertainty surrounding the reported HR. A narrow interval generally provides more precision than a wide interval, although precision and clinical importance are different concepts. The confidence interval should therefore be read alongside the point estimate and the endpoint itself.
Why does the P-value not measure effect size?
The P-value is a measure of evidence against a null hypothesis under a specified statistical framework. It is influenced by both the observed data and the amount of information available. The magnitude of the treatment effect is described by the HR, while its statistical uncertainty is described by the confidence interval.
What does censoring mean in this trial?
For the primary overall-survival analysis, the registry states that participants who were alive as of the overall-survival cutoff date were censored on the cutoff date post consent. Censoring means that the exact death time was not observed within the analysis period, but the participant still contributes information through the observed follow-up.
Why should the PFS HR not be treated as another primary endpoint?
The registry identifies overall survival as the sole primary endpoint and progression-free survival as a secondary endpoint. Although both analyses report a hazard ratio and P-value, their roles in the trial hierarchy are different. A secondary endpoint should therefore not be relabeled as a co-primary endpoint simply because its statistical result is available.
13. Interpreting the Primary Endpoint Without Overinterpreting It
The reported overall-survival HR was 0.68, with a 95% CI of 0.58–0.81 and a two-sided P-value of <0.0001. The analysis used the ITT population and a stratified log-rank test, under a superiority hypothesis.
The result does not provide a median overall-survival time, a specific absolute survival probability, or an individual patient's probability of benefit. None of those quantities is contained in the ClinicalTrials.gov record, so they are not substituted for the reported hazard ratio.
A hazard ratio summarizes relative time-to-event differences. Absolute survival probabilities or median survival times answer different questions. Because those additional absolute measures are not reported in the ClinicalTrials.gov record for this page, the interpretation remains deliberately centered on the reported HR, confidence interval, P-value, and analysis method.
14. Understanding the Relationship Between OS and PFS
RECOURSE reports both overall survival and progression-free survival, but the two endpoints should not be conflated.
| Dimension | Overall Survival | Progression-free Survival |
|---|---|---|
| Role | Primary endpoint | Secondary endpoint |
| Endpoint type | Time-to-event | Time-to-event |
| Reported HR | 0.68 | 0.48 |
| 95% CI | 0.58–0.81 | 0.41–0.57 |
| P-value | <0.0001 | <0.0001 |
| Analysis population | ITT | ITT |
| Method | Stratified log-rank test | Stratified log-rank test |
Overall survival is anchored to death, while progression-free survival is an earlier disease-related time-to-event endpoint. An earlier endpoint can show a different relative effect from overall survival because its event process occurs at different times and may incorporate different types of events.
The difference between the two estimates should therefore be described rather than mathematically combined. The ClinicalTrials.gov record supports the statement that both reported HRs are below 1 and both have P-values below 0.0001; they do not support a claim that the numerical difference between the HRs represents a particular biological quantity.
15. Safety and Efficacy Are Different Statistical Questions
The efficacy analyses and serious-adverse-event data illustrate why a clinical trial cannot be reduced to a single statistic.
| Evidence domain | Reported RECOURSE information | Statistical interpretation |
|---|---|---|
| Overall survival | HR 0.68; 95% CI 0.58–0.81; P <0.0001 | Formal time-to-event efficacy comparison |
| Progression-free survival | HR 0.48; 95% CI 0.41–0.57; P <0.0001 | Formal secondary time-to-event comparison |
| Serious adverse events | 158/533 vs 89/265 | Descriptive arm-level safety counts in the ClinicalTrials.gov record |
The survival endpoints have formal reported statistical analyses. The serious-adverse-event ClinicalTrials.gov record do not include a statistical comparison. Therefore, it would be inappropriate to place all three rows on the same inferential scale.
This distinction is also important because a treatment can affect efficacy and safety through different mechanisms and with different event definitions. A rigorous trial analysis keeps each endpoint tied to its prespecified definition and statistical framework.
16. What the Registry Does Not Report in the Supplied Data
A strong statistical analysis also identifies the boundary between documented information and information that would require additional trial documentation.
- Median overall survival: not reported in the ClinicalTrials.gov record.
- Median progression-free survival: not reported in the ClinicalTrials.gov record.
- Absolute survival probabilities: not reported in the ClinicalTrials.gov record.
- Kaplan-Meier curve coordinates: not reported.
- Baseline characteristics: not reported.
- Subgroup analyses: not reported.
- Specific stratification factors: not reported.
- Cox-model specification: not reported.
- Proportional-hazards diagnostics: not reported.
- Multiplicity strategy: not reported.
- Interim-analysis plan: not reported.
- Missing-data or imputation strategy: not reported.
- Non-inferiority margin: not applicable to the registry-reported superiority hypothesis and not reported.
- Bayesian methods: not reported.
- Crossover analysis: not reported.
17. Limitations
- Limited registry detail: the ClinicalTrials.gov record provides the primary and secondary statistical results but not the complete statistical analysis plan.
- Incomplete endpoint time-frame text: the registry-reported primary and secondary time-frame fields terminate mid-entry, so additional wording is not inferred.
- Hazard-ratio interpretation: the registry identifies the hazard ratio but does not specify the model used to obtain it in the ClinicalTrials.gov record.
- Proportional-hazards uncertainty: no proportional-hazards diagnostic is reported in the ClinicalTrials.gov record. A single hazard ratio should therefore be understood as the reported summary measure rather than assumed to describe every point in time identically.
- No absolute survival estimates: the ClinicalTrials.gov record does not include median survival or time-specific survival probabilities, limiting absolute interpretation.
- No subgroup results: the ClinicalTrials.gov record does not support conclusions about consistency of the treatment effect across patient subgroups.
- Safety denominator difference: serious adverse events are reported as 158/533 and 89/265, while total enrollment is 800. The ClinicalTrials.gov record does not explain the difference between the safety denominators and overall enrollment.
- Multiplicity information: the registry data do not provide a multiplicity-adjustment framework. The secondary PFS result should therefore be interpreted according to its registered role rather than treated as another primary endpoint.
- Missing-data methods: no imputation strategy is reported. For time-to-event analyses, censoring is explicitly described for participants alive at the overall-survival cutoff, but broader missing-data procedures cannot be reconstructed.
18. Why This Trial Matters Statistically
RECOURSE is a useful teaching example because the trial's main results illustrate several fundamental ideas in clinical-trial statistics without requiring the analysis to be reduced to a single P-value.
| Concept | How it appears in RECOURSE |
|---|---|
| Randomization | The trial used randomized allocation in a parallel-group design. |
| Blinding | The registry identifies quadruple masking. |
| ITT analysis | The primary and secondary posted efficacy analyses were performed in the ITT population. |
| Time-to-event endpoints | Overall survival was the primary endpoint and progression-free survival was a secondary endpoint. |
| Stratified log-rank testing | Used for both posted efficacy analyses. |
| Hazard ratio | Reported as the effect measure for OS and PFS. |
| Confidence interval | 95% two-sided intervals quantify uncertainty around both HR estimates. |
| P-value | Both posted efficacy analyses report P <0.0001. |
| Censoring | Alive participants at the OS cutoff were censored on the cutoff date post consent. |
| Superiority hypothesis | The registered hypothesis type is superiority rather than non-inferiority. |
From a statistical-learning perspective, the most important lesson is that the hazard ratio is only one part of the analysis. The endpoint definition, randomized population, censoring rule, statistical test, confidence interval, and hypothesis type all determine how the number should be interpreted.
19. A Practical Reading Framework for the RECOURSE Results
Step 1 · Identify the endpoint
Overall survival is the primary endpoint. Progression-free survival is secondary.
Step 2 · Identify the population
The posted efficacy analyses use the intention-to-treat population.
Step 3 · Identify the effect measure
The registry reports hazard ratios rather than risk ratios or odds ratios.
Step 4 · Read the interval
The 95% CI indicates the statistical precision of the reported HR.
Step 5 · Read the P-value separately
The P-value describes evidence against the null hypothesis, not effect magnitude.
Step 6 · Check the analysis method
The registry reports a stratified log-rank test for both efficacy endpoints.
This framework prevents several common errors: treating an HR as an absolute risk reduction, treating a P-value as an effect-size measure, assuming an unreported Cox model, or treating a secondary endpoint as though it were a co-primary endpoint.
20. Overall Statistical Interpretation
Primary statistical result
95% CI 0.58–0.81 · P <0.0001
Overall survival was analyzed in the ITT population using a stratified log-rank test under a superiority hypothesis.
The primary analysis reports a hazard ratio below 1, with a 95% confidence interval from 0.58 to 0.81 and a two-sided P-value below 0.0001. Within the reported statistical framework, the estimate corresponds to a lower estimated hazard of death for TAS-102 than for placebo.
The secondary progression-free-survival analysis reports an HR of 0.48, with a 95% CI of 0.41–0.57 and P <0.0001. This provides a second time-to-event result in the same general direction, but its registered role is secondary rather than primary.
The ClinicalTrials.gov record does not include median survival, absolute survival probabilities, subgroup effects, detailed stratification factors, or the full statistical analysis plan. The appropriate interpretation therefore centers on what is actually reported: randomized allocation, quadruple masking, ITT efficacy analysis, time-to-event endpoints, stratified log-rank testing, hazard ratios, confidence intervals, P-values, and the stated censoring rule.
21. Statistical Concepts in This Trial
Learn more about the methods used in this trial:
22. Related Statistical Calculators
23. Sources
- ClinicalTrials.gov: RECOURSE, NCT01607957.
- Linked publication: PubMed PMID 31912801.
- Linked publication: PubMed PMID 29274618.
- Linked publication: PubMed PMID 27443414.
- Linked publication: PubMed PMID 25970050.
Continue through the Clinical Biostats statistical pathway
Explore the statistical concepts behind randomized trials, survival analysis, confidence intervals, hazard ratios, and time-to-event endpoints.
24. Record Summary
RECOURSE provides a clear example of a randomized phase 3 time-to-event analysis. The trial enrolled 800 participants, used randomized parallel-group allocation with quadruple masking, and compared TAS-102 with placebo. Overall survival was the sole registered primary endpoint, analyzed in the ITT population using a stratified log-rank test and reported with a hazard ratio of 0.68, a 95% CI of 0.58–0.81, and P <0.0001. Progression-free survival was a secondary endpoint with a reported HR of 0.48, 95% CI 0.41–0.57, and P <0.0001.
The statistical interpretation depends on keeping several distinctions intact: a hazard ratio is not an absolute risk difference; a confidence interval is not a range of individual patient effects; a P-value is not a measure of effect size; and a secondary endpoint is not automatically a primary endpoint. The ClinicalTrials.gov record also show why registry-based analysis should remain disciplined about what is known and what is not: detailed subgroup results, absolute survival measures, stratification factors, model specifications, multiplicity procedures, and missing-data methods are not reported and therefore are not reconstructed here.