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
AXIS was a randomized, parallel, open-label phase 3 trial in kidney neoplasms. The trial compared axitinib (AG-013736) with sorafenib in 723 participants, with progression-free survival as the registered primary endpoint and overall survival and objective response rate reported as secondary outcomes.
| Feature | AXIS |
|---|---|
| Trial name | AXIS |
| Brief title | Axitinib (AG 013736) As Second Line Therapy For Metastatic Renal Cell Cancer |
| Phase | Phase 3 |
| Status | Completed |
| Condition | Kidney Neoplasms |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | None |
| Primary purpose | Treatment |
| Enrollment | 723 |
| Interventions | Axitinib (AG-013736); Sorafenib |
| Primary endpoint | Progression-Free Survival (PFS) |
| Primary endpoint type | Time-to-event |
| Results posted | Yes |
| Outcome measures posted | 14 |
| Statistical analyses posted | 3 |
2. Clinical Question
The central statistical question was whether axitinib could produce a different time-to-progression-or-death profile from sorafenib in the trial population, with the primary hypothesis specified as superiority.
Population
The trial's brief title describes patients receiving second-line therapy for metastatic renal cell cancer. The registry condition is Kidney Neoplasms.
Intervention
Axitinib (AG-013736), with the reported arm-specific safety denominator identified in the registry data as 359 participants receiving axitinib 5 mg.
Comparator
Sorafenib, with the reported arm-specific safety denominator identified in the registry data as 355 participants receiving sorafenib 400 mg.
Primary question
Does axitinib produce superior progression-free survival compared with sorafenib?
3. Trial Design
Axitinib 5 mg
- Intervention: Axitinib (AG-013736)
- Safety denominator reported: 359
- Primary comparison: progression-free survival versus sorafenib
Sorafenib 400 mg
- Intervention: Sorafenib
- Safety denominator reported: 355
- Primary comparison: progression-free survival versus axitinib
The registry identifies the study as randomized and parallel, which is important for the causal interpretation of the between-arm comparison. Randomization creates the design framework for comparing outcomes according to assigned treatment. The absence of masking is a separate design feature: the registry identifies the trial as having no masking, so treatment assignment was not blinded.
4. Endpoints
| Endpoint | Role | Time frame | Type |
|---|---|---|---|
| Progression-Free Survival (PFS) | Primary | From initiation of treatment up to follow-up period (up to 3 years) | Time-to-event |
| Overall Survival (OS) | Secondary | From initiation of treatment up to follow-up period (up to 3 years) | Time-to-event |
| Objective Response Rate (ORR) | Secondary | From initiation of treatment up to follow-up period (up to 3 years) | Binary |
Primary endpoint: Progression-Free Survival
The registry defines PFS as the time in months from start of study treatment to the first documentation of objective tumor progression of disease or death due to any cause, whichever occurs first. Disease progression was assessed using Response Evaluation Criteria in Solid Tumors (RECIST) version 1.0.
Secondary endpoint: Overall Survival
Overall survival was defined as a time-to-event outcome measured from initiation of treatment through the follow-up period of up to 3 years. The posted analysis compared axitinib 5 mg with sorafenib 400 mg using a log-rank test and reported a hazard ratio.
Secondary endpoint: Objective Response Rate
Objective response rate was a binary endpoint expressed as the percentage of participants. The posted analysis compared axitinib 5 mg with sorafenib 400 mg using a Cochran-Mantel-Haenszel test and reported a risk ratio.
5. Analysis Populations and Stratification
| Feature | Registry-supported description |
|---|---|
| Primary efficacy population | Full analysis set (FAS) |
| Assignment for analysis | According to initial randomization |
| Primary PFS comparison | Axitinib 5 mg vs Sorafenib 400 mg |
| OS analysis | Stratified analysis; the analysis note identifies stratification by ECOG performance status and prior treatment |
| ORR analysis | Stratified analysis; the analysis note identifies stratification by ECOG performance status and prior treatment |
Stratification is important because it allows the treatment comparison to account for prespecified factors that may be associated with outcome. In the posted OS and ORR analyses, the registry specifically identifies ECOG performance status and prior treatment as stratification factors. This is different from simply comparing two unadjusted percentages or two unadjusted survival curves.
6. Statistical Methodology
Log-rank test for time-to-event endpoints
The primary PFS analysis and the secondary OS analysis used the log-rank test. A log-rank test compares the event experience of two groups across follow-up while accounting for the timing of events and right-censoring.
The test evaluates whether the observed pattern of events over time is compatible with the null hypothesis of no difference between the treatment groups.
Cochran-Mantel-Haenszel test for response
The ORR analysis used the Cochran-Mantel-Haenszel test. This is appropriate for a categorical endpoint when the comparison is conducted across strata. In this trial's posted analysis, the strata were based on ECOG performance status and prior treatment.
The reported risk ratio summarizes the relative probability of response, while the Cochran-Mantel-Haenszel framework accounts for the specified stratification structure.
Hazard ratio
The posted PFS and OS analyses reported hazard ratios. A hazard ratio is a relative measure of the instantaneous event rate between treatment groups under a time-to-event analysis. A value below 1 indicates a lower estimated hazard in the axitinib group relative to the sorafenib group.
The hazard ratio is not the same as a relative risk, a risk difference, or the percentage of participants who experience the event.
Confidence intervals
The primary PFS estimate was accompanied by a two-sided 95% confidence interval. Confidence intervals quantify statistical uncertainty around an estimated effect under the specified analysis framework. They should not be interpreted as a range containing the effects for individual patients.
Superiority hypothesis
The registry identifies the posted analyses as superiority analyses. That means the inferential question is whether the evidence supports a difference favoring the experimental treatment rather than whether axitinib merely satisfies a non-inferiority margin. No non-inferiority margin is reported in the ClinicalTrials.gov record.
7. Results: Progression-Free Survival
PFS was the single registered primary endpoint. The posted analysis compared axitinib 5 mg with sorafenib 400 mg using a log-rank test in the full analysis set.
Primary endpoint: PFS hazard ratio
95% CI: 0.544–0.812 · P < 0.0001
Two-sided 95% confidence interval · Superiority hypothesis
| Primary endpoint | Axitinib 5 mg vs Sorafenib 400 mg |
|---|---|
| Outcome | Progression-Free Survival (PFS) |
| Analysis | Log-rank test |
| Effect measure | Hazard ratio |
| Estimate | 0.665 |
| 95% CI | 0.544–0.812 |
| P-value | <0.0001 |
| Hypothesis | Superiority |
| Analysis population | Full analysis set |
The estimated PFS hazard ratio of 0.665 means that, within the time-to-event model used for the analysis, the estimated instantaneous rate of progression or death in the axitinib group was approximately 66.5% of that in the sorafenib group. Expressed as a relative hazard comparison, this corresponds to an estimated 33.5% lower hazard because 1 − 0.665 = 0.335.
This does not mean that 33.5% of participants avoided progression, that 33.5% of participants benefited, or that each participant experienced exactly a 33.5% reduction in risk. The hazard ratio is a population-level relative measure derived from a time-to-event analysis.
The two-sided 95% confidence interval of 0.544–0.812 describes uncertainty around the estimated hazard ratio. It indicates that the estimate is not being presented as a single known quantity. The interval remains below 1, which is consistent with the reported superiority hypothesis test.
The P-value < 0.0001 addresses the statistical evidence against the null hypothesis under the specified test. It does not measure the size of the treatment effect, the probability that axitinib works, or the clinical importance of the result. Effect size and uncertainty are conveyed by the hazard ratio and its confidence interval.
Because PFS is a time-to-event endpoint, interpretation also depends on censoring and on the assumptions underlying the chosen survival-analysis framework. The ClinicalTrials.gov record does not report a formal assessment of the proportional-hazards assumption, so this page does not make a separate claim about that assumption.
8. Results: Overall Survival
Overall survival was reported as a secondary time-to-event endpoint. The posted analysis used a log-rank test comparing axitinib 5 mg with sorafenib 400 mg and reported a hazard ratio with a two-sided 95% confidence interval.
Secondary endpoint: OS hazard ratio
95% CI: 0.800–1.174 · P = 0.3744
Log-rank analysis · Stratified by ECOG performance status and prior treatment
| Secondary endpoint | Axitinib 5 mg vs Sorafenib 400 mg |
|---|---|
| Outcome | Overall Survival (OS) |
| Analysis | Log-rank test |
| Effect measure | Hazard ratio |
| Estimate | 0.969 |
| 95% CI | 0.800–1.174 |
| P-value | 0.3744 |
| Hypothesis | Superiority |
| Stratification | ECOG performance status and prior treatment |
| Analysis population | Full analysis set |
The OS hazard ratio of 0.969 is close to 1. In the reported time-to-event analysis, the estimated instantaneous rate of death in the axitinib group was approximately 96.9% of that in the sorafenib group. Expressed mechanically as a relative hazard difference, 1 − 0.969 corresponds to approximately a 3.1% lower estimated hazard for axitinib, but that arithmetic should not be interpreted as a clinically meaningful effect estimate on its own.
The confidence interval of 0.800–1.174 includes 1. That means the data are compatible with a range of relative hazard values on both sides of the no-difference value under the specified analysis framework.
The P-value of 0.3744 does not measure the magnitude of the observed hazard ratio. It measures the evidence against the specified null hypothesis under the reported testing procedure. A relatively large P-value should not be translated into a statement that the two treatments are proven identical; rather, the reported analysis does not provide statistically strong evidence of superiority for the OS comparison.
The analysis note states that the P-value was obtained from a 1-sided log-rank test of treatment stratified by ECOG performance status and prior treatment, while the reported confidence interval is two-sided at 95%. This distinction matters: the tail specification for a hypothesis test and the confidence level for an interval are separate features of statistical inference.
9. Results: Objective Response Rate
Objective response rate was a secondary binary endpoint. The posted analysis used the Cochran-Mantel-Haenszel test, with the treatment comparison stratified by ECOG performance status and prior treatment.
Secondary endpoint: response risk ratio
95% CI: 1.408–3.003 · P = 0.0001
Two-sided 95% confidence interval · Cochran-Mantel-Haenszel analysis
| Secondary endpoint | Axitinib 5 mg vs Sorafenib 400 mg |
|---|---|
| Outcome | Objective Response Rate (ORR) |
| Endpoint type | Binary |
| Analysis | Cochran-Mantel-Haenszel test |
| Effect measure | Risk ratio / relative risk |
| Estimate | 2.056 |
| 95% CI | 1.408–3.003 |
| P-value | 0.0001 |
| Hypothesis | Superiority |
| Stratification | ECOG performance status and prior treatment |
| Analysis population | Full analysis set |
The reported risk ratio of 2.056 means that the probability of objective response in the axitinib group was estimated to be about 2.056 times the corresponding probability in the sorafenib group, under the reported stratified analysis framework.
This is different from saying that 2.056 times as many patients responded, or that axitinib caused a response in a particular proportion of individuals. A risk ratio requires the underlying response probabilities to determine the absolute difference between groups.
The 95% confidence interval of 1.408–3.003 quantifies uncertainty around the estimated relative response probability. The entire interval is above 1, so the reported interval is consistent with a higher response probability for axitinib relative to sorafenib under this analysis.
The P-value of 0.0001 addresses the evidence against the null hypothesis of no treatment difference in the specified stratified test. It is not a measure of how large or clinically important the response difference is.
The registry analysis note states that the P-value was obtained from a 1-sided Cochran-Mantel-Haenszel test stratified by ECOG performance status and prior treatment. The ClinicalTrials.gov record does not provide the arm-specific ORR percentages themselves, so this page does not reconstruct absolute response rates from the risk ratio.
10. Safety Results
The ClinicalTrials.gov record reports serious adverse events by treatment arm as the number affected divided by the number at risk.
| Safety measure | Axitinib 5 mg | Sorafenib 400 mg |
|---|---|---|
| Serious adverse events | 146/359 | 127/355 |
The ClinicalTrials.gov record is limited to serious adverse events by arm. The page therefore does not infer rates for other adverse-event categories, severity grades, treatment discontinuations, or individual adverse-event terms.
11. Statistical Methods Explained
Why was a log-rank test used for PFS?
PFS is a time-to-event endpoint because participants can experience progression or death at different times, while some participants may remain event-free at the end of their observed follow-up. A log-rank test uses the ordering and timing of events rather than reducing every participant to a simple yes/no outcome.
What does a hazard ratio of 0.665 mean?
A hazard ratio of 0.665 means that the estimated instantaneous event rate for progression or death in the axitinib group was 66.5% of that in the sorafenib group under the reported analysis. The complementary quantity, 1 − 0.665, is 0.335, or a 33.5% lower estimated hazard. This is a relative time-to-event measure, not an absolute probability.
Why is the PFS confidence interval important?
The interval of 0.544–0.812 gives more information than the point estimate alone. It describes the statistical uncertainty around the estimated hazard ratio. Because it does not cross 1, the reported interval is consistent with a treatment effect in the direction of lower hazard for axitinib.
How is the ORR risk ratio different from the PFS hazard ratio?
The ORR risk ratio compares probabilities of a binary outcome: response versus no response. The PFS hazard ratio compares event rates over time while accounting for censoring. They therefore answer different statistical questions and cannot be interpreted interchangeably.
Why was the Cochran-Mantel-Haenszel test used for ORR?
The posted ORR analysis used the Cochran-Mantel-Haenszel test and identified stratification by ECOG performance status and prior treatment. The method provides a way to compare treatment groups while incorporating the specified strata rather than treating all participants as belonging to one undifferentiated population.
Why does the OS P-value need to be distinguished from the OS effect estimate?
The OS hazard ratio of 0.969 describes the estimated relative hazard, while the P-value of 0.3744 describes evidence against the null hypothesis under the reported testing procedure. A P-value is therefore not a substitute for the effect estimate or its confidence interval.
Why does the analysis population matter?
The posted analyses use the full analysis set, with participants analyzed according to initial randomization. This preserves the randomized treatment comparison as the basis for efficacy inference. It also means the treatment comparison is not simply a comparison of participants who remained on treatment for the same length of time.
12. Understanding the Primary PFS Result
The primary PFS result is best understood by considering three related quantities together: the hazard ratio, the confidence interval, and the P-value.
Effect size
HR 0.665 is the reported relative time-to-event estimate. It indicates a lower estimated hazard of progression or death for axitinib under the model used.
Precision
The 95% CI of 0.544–0.812 shows the statistical uncertainty surrounding the point estimate.
Evidence against the null
The reported P-value is <0.0001. This concerns statistical evidence under the specified test rather than effect magnitude.
Clinical meaning
The hazard ratio alone does not describe median PFS, absolute event probabilities, or the experience of any particular participant.
This distinction is especially useful when interpreting clinical-trial results. A very small P-value can occur with a modest effect in a sufficiently informative dataset, while a large effect estimate can be imprecise in a small dataset. The AXIS primary analysis provides the effect estimate and its confidence interval directly, so both should be considered rather than relying on the P-value alone.
13. Understanding the Secondary OS Result
The OS result illustrates why a clinical trial can have different statistical conclusions across different endpoints. The reported PFS hazard ratio was 0.665, whereas the OS hazard ratio was 0.969 with a 95% confidence interval of 0.800–1.174.
| Endpoint | Effect measure | Estimate | 95% CI | P-value |
|---|---|---|---|---|
| PFS | Hazard ratio | 0.665 | 0.544–0.812 | <0.0001 |
| OS | Hazard ratio | 0.969 | 0.800–1.174 | 0.3744 |
| ORR | Risk ratio | 2.056 | 1.408–3.003 | 0.0001 |
These are not interchangeable outcomes. PFS measures time to progression or death, OS measures time to death, and ORR measures a binary response outcome. Differences between their estimates do not by themselves indicate an inconsistency in statistical methodology. They reflect the fact that each endpoint captures a different aspect of the treatment comparison.
14. One-Sided Tests and Two-Sided Confidence Intervals
The registry analysis notes contain an important inferential detail. The OS P-value was obtained from a 1-sided log-rank test, and the ORR P-value was obtained from a 1-sided Cochran-Mantel-Haenszel test. At the same time, the posted confidence intervals for these analyses are identified as two-sided 95% intervals.
The directionality of a hypothesis test and the coverage level of a confidence interval are separate aspects of statistical reporting. They should not be treated as interchangeable descriptions.
This is particularly important when reading a trial registry because a reader might otherwise assume that every P-value and confidence interval uses the same tail convention. Here, the registry-reported analysis notes explicitly distinguish them.
15. Stratified Analysis
Both the OS and ORR analysis notes identify stratification by ECOG performance status and prior treatment. Stratification is a way of preserving information about clinically relevant categories while estimating the treatment comparison across those categories.
| Endpoint | Method | Stratified by |
|---|---|---|
| Progression-Free Survival | Log-rank test | Registry data do not provide an additional stratification note for the primary PFS analysis |
| Overall Survival | Log-rank test | ECOG performance status and prior treatment |
| Objective Response Rate | Cochran-Mantel-Haenszel test | ECOG performance status and prior treatment |
A stratified analysis should not be interpreted as simply "adjusting everything." It incorporates the particular factors specified in the analysis. The ClinicalTrials.gov record supports identifying ECOG performance status and prior treatment for the OS and ORR analyses, but do not provide additional stratification factors for those analyses.
16. PFS and the Kaplan-Meier Framework
The trial's primary endpoint is a time-to-event outcome. A natural way to describe such data is with a Kaplan-Meier estimator, which estimates the probability of remaining event-free over time while retaining information from participants whose observations are censored.
Here, di represents the number of events at time ti, while ni represents the number at risk immediately before that time.
The hazard ratio and log-rank test complement, rather than replace, the underlying survival description. The Kaplan-Meier framework describes the event-free experience over time, the log-rank test evaluates the overall separation between groups under its assumptions, and the hazard ratio summarizes the relative event rate using a model-based measure.
17. Risk Ratio and Objective Response
The ORR analysis uses a different statistical language because response is binary rather than time-to-event. The reported risk ratio of 2.056 compares the response probability between the treatment groups.
A risk ratio greater than 1 indicates a higher estimated response probability in the numerator group. Here, the reported comparison is axitinib 5 mg versus sorafenib 400 mg, giving an estimated RR of 2.056.
The confidence interval of 1.408–3.003 describes uncertainty around that relative comparison. Because the interval remains above 1, the interval is consistent with a higher response probability in the axitinib group under the reported analysis.
The risk ratio cannot tell us the absolute probability of response without the underlying arm-specific response proportions. Those proportions are not included in the ClinicalTrials.gov record, so no absolute ORR is reported here.
18. Trial Timeline
Trial start
The AXIS trial began according to the registry record.
Primary completion
The registry lists August 31, 2010 as the primary completion date.
Registry status
The trial is listed as completed and has posted outcome measures and statistical analyses.
19. Limitations
- Registry-data constraint: this page is restricted to the trial data posted on ClinicalTrials.gov for AXIS. It does not supplement missing numerical results from publications or outside sources.
- No median survival values reported: median PFS and median OS are not reported in the ClinicalTrials.gov record, so they are not presented.
- No absolute ORR values reported: the posted ORR analysis provides a risk ratio, confidence interval, and P-value, but not the underlying arm-specific response percentages.
- No subgroup estimates reported: the registry identifies stratification factors for OS and ORR, but the ClinicalTrials.gov record does not provide subgroup-specific efficacy estimates.
- No proportional-hazards assessment reported: the data report hazard ratios but do not report a formal assessment of the proportional-hazards assumption.
- No multiplicity plan reported: the registry data identify superiority hypotheses and three posted statistical analyses but do not provide an alpha-allocation or multiplicity-adjustment strategy.
- No crossover information reported: the ClinicalTrials.gov record do not identify crossover rules or crossover frequency, so no crossover adjustment is inferred.
- Safety scope: only serious adverse events by arm are reported. Other safety outcomes are not reconstructed.
These limitations are important because a statistical analysis should distinguish what was actually reported from what would ordinarily be expected in a full statistical analysis plan. Missing information is not evidence that a particular method, adjustment, or result was absent from the underlying trial; it simply means that the ClinicalTrials.gov record does not establish it.
20. Why This Trial Matters Statistically
AXIS is a useful teaching example because its registry record connects a randomized superiority question to two different families of statistical methods: survival analysis for PFS and OS, and stratified categorical analysis for ORR.
| Concept | How it appears in AXIS |
|---|---|
| Randomization | The trial used randomized allocation to two parallel treatment arms. |
| Superiority testing | The posted analyses identify superiority as the hypothesis type. |
| Time-to-event analysis | PFS was the primary endpoint; OS was a secondary endpoint. |
| Log-rank test | Used for the posted PFS and OS comparisons. |
| Hazard ratio | Reported for both PFS and OS. |
| Confidence interval | Two-sided 95% intervals accompany the reported effect estimates. |
| Cochran-Mantel-Haenszel test | Used for the ORR comparison. |
| Risk ratio | Reported as the effect measure for ORR. |
| Stratified analysis | OS and ORR analysis notes identify ECOG performance status and prior treatment as stratification factors. |
| Full analysis set | The posted analyses describe efficacy analyses using the FAS and initial randomized treatment assignment. |
The most instructive feature is that the endpoint determines much of the statistical architecture. PFS and OS require methods that preserve the timing of events and account for censoring. ORR is binary and therefore uses a categorical comparison. The trial consequently demonstrates why there is no single generic "clinical trial test": the endpoint, design, stratification, and hypothesis determine the appropriate analysis.
21. Interpreting the Three Posted Analyses Together
| Question | Endpoint | Method | Effect measure | Reported result |
|---|---|---|---|---|
| Did treatment groups differ in time to progression or death? | PFS | Log-rank | Hazard ratio | 0.665 (95% CI 0.544–0.812), P < 0.0001 |
| Did treatment groups differ in time to death? | OS | Log-rank | Hazard ratio | 0.969 (95% CI 0.800–1.174), P = 0.3744 |
| Did treatment groups differ in objective response? | ORR | Cochran-Mantel-Haenszel | Risk ratio | 2.056 (95% CI 1.408–3.003), P = 0.0001 |
These three analyses should be read as three distinct pieces of evidence rather than combined into a single numerical score. The PFS result concerns time to progression or death, the OS result concerns time to death, and the ORR result concerns whether an objective response occurred. Their statistical estimates operate on different scales and address different clinical questions.
22. What a Hazard Ratio Does — and Does Not — Mean
The reported PFS HR of 0.665 indicates a lower estimated instantaneous rate of progression or death in the axitinib group relative to the sorafenib group under the reported analysis. The complementary relative quantity is approximately 33.5% lower estimated hazard.
It does not mean that 33.5% of participants avoided progression, that progression was delayed by exactly 33.5% for every participant, or that 33.5% is an absolute risk difference.
The reported OS HR of 0.969 is close to the no-difference value of 1. The associated 95% CI of 0.800–1.174 spans both sides of 1, so the interval does not isolate a single direction of effect relative to the null.
The PFS P-value of <0.0001 and the ORR P-value of 0.0001 provide evidence information under their respective tests. Neither P-value tells the reader how large the treatment effect is. The effect estimates and confidence intervals provide that information.
23. Clinical Biostats Statistical Perspective
What the PFS result establishes statistically
The posted primary analysis reports a hazard ratio below 1 with a two-sided 95% confidence interval entirely below 1 and a P-value <0.0001 under the specified superiority analysis.
What the OS result establishes statistically
The posted secondary analysis reports an HR of 0.969 with a 95% CI of 0.800–1.174 and P = 0.3744. The ClinicalTrials.gov record does not support treating this as evidence of a statistically established OS superiority effect.
What the ORR result establishes statistically
The posted secondary analysis reports an RR of 2.056 with a 95% CI of 1.408–3.003 and P = 0.0001 using the Cochran-Mantel-Haenszel framework.
What remains unknown from the ClinicalTrials.gov record
Absolute response rates, median survival times, event counts, subgroup estimates, and several design details are not included in the ClinicalTrials.gov record.
This distinction between reported evidence and unavailable information is central to reproducible trial interpretation. A statistically careful analysis should not fill gaps by importing numbers from memory or by deriving quantities that require data not contained in the record being analyzed.
24. Related Tutorials
Learn more about the methods used in this trial:
25. Related Calculators
26. Sources
- ClinicalTrials.gov: AXIS, NCT00678392.
- PubMed: PMID 37526095.
- PubMed: PMID 32363929.
- PubMed: PMID 31072748.
- PubMed: PMID 28410911.
- PubMed: PMID 27238653.
Continue with the statistical methods behind AXIS
Explore survival-analysis tutorials, statistical calculators, and additional clinical-trial analyses through Clinical Biostats.
27. Record Summary
AXIS provides a clear example of how different clinical endpoints require different statistical approaches. The randomized phase 3 trial enrolled 723 participants and compared axitinib with sorafenib in a parallel, unmasked design. Its registered primary endpoint was progression-free survival, analyzed with a log-rank test and reported as a hazard ratio of 0.665 with a two-sided 95% confidence interval of 0.544–0.812 and P < 0.0001. The secondary overall-survival analysis reported an HR of 0.969 (95% CI 0.800–1.174; P = 0.3744), while the objective-response analysis reported a risk ratio of 2.056 (95% CI 1.408–3.003; P = 0.0001).
Statistically, the most important lesson is that these results should be interpreted according to their endpoint and effect measure. The PFS and OS analyses are time-to-event comparisons, while ORR is a binary response outcome. The trial also illustrates the role of stratified analysis, confidence intervals, one-sided hypothesis tests, and the distinction between statistical evidence and effect size.