This page separates reported trial results from statistical interpretation. The numerical results and trial characteristics presented here are restricted to the ClinicalTrials.gov record for AURA3.
1. Trial at a Glance
AURA3 was a randomized, open-label, parallel-group phase 3 treatment trial comparing osimertinib 80 mg with platinum-based doublet chemotherapy in locally advanced or metastatic non-small cell lung cancer. The registry reports 421 enrolled participants, two treatment arms, one primary time-to-event endpoint, and eight posted statistical analyses.
| Feature | AURA3 |
|---|---|
| Trial name | AURA3 |
| NCT identifier | NCT02151981 |
| Phase | Phase 3 |
| Status | COMPLETED |
| Therapeutic area | Oncology |
| Condition | Anticancer Treatment |
| Population description | Locally advanced or metastatic non-small cell lung cancer |
| Allocation | RANDOMIZED |
| Design model | PARALLEL |
| Masking | NONE |
| Primary purpose | TREATMENT |
| Enrollment | 421 |
| Lead sponsor | AstraZeneca |
| Sponsor type | INDUSTRY |
2. Clinical Question
The primary statistical question was whether osimertinib 80 mg improved progression-free survival by investigator assessment compared with chemotherapy in the randomized population.
Population
Patients enrolled in AURA3 with locally advanced or metastatic non-small cell lung cancer, as described by the trial title and registry record.
Intervention
Osimertinib 80 mg.
Comparator
Platinum-based doublet chemotherapy.
Primary question
Does osimertinib 80 mg reduce the hazard of investigator-assessed progression or death relative to chemotherapy?
3. Trial Design
Osimertinib 80 mg
- Osimertinib 80 mg is the experimental treatment specified in the posted statistical analyses.
- The registry also identifies crossover to osimertinib as an intervention.
Platinum-based doublet chemotherapy
- Platinum-based doublet chemotherapy is the comparator specified in the AURA3 trial title and statistical comparisons.
The trial began on 2014-08-04 and reached primary completion on 2016-04-15. The registry classifies it as completed and industry-sponsored by AstraZeneca.
4. Primary Endpoint
| Endpoint | Registry definition / time frame | Analysis |
|---|---|---|
| Progression Free Survival (PFS) by Investigator Assessment | RECIST tumour assessments every 6 weeks from randomisation until objective disease progression up to 19 months. PFS is measured from the date of randomisation until investigator-assessed progression or death. | Log-rank test; hazard ratio |
RECIST-based event definition
The registry definition uses Response Evaluation Criteria in Solid Tumours (RECIST v1.1). Progressive disease is defined using tumour-lesion measurements, including a ≥20% increase in the sum of diameters of target lesions together with an absolute increase of ≥5 mm compared with the previous minimum sum, or progression of non-target lesions or a new lesion.
Because PFS includes both progression and death as possible events, it is a time-to-event endpoint. Participants who have not experienced the defined event by their last usable assessment contribute follow-up information through censoring rather than being treated as if the event occurred at that time.
5. Statistical Methodology
Full Analysis Set
The primary PFS analysis used the Full Analysis Set, defined in the registry as all randomised patients. This is important because the treatment comparison remains anchored to randomization rather than being restricted to patients who completed treatment or remained on therapy.
Log-rank test
The posted primary analysis used a log-rank test. For a time-to-event endpoint such as PFS, the log-rank test compares the observed pattern of events between randomized groups over follow-up while accounting for the timing of events and censoring.
The log-rank test evaluates evidence against the null hypothesis using the ordering and timing of observed events. It does not itself quantify the magnitude of treatment effect; the hazard ratio provides that complementary effect measure.
Hazard ratio
The primary effect measure was a hazard ratio. The registry explicitly notes that a hazard ratio below 1 favors osimertinib 80 mg.
An HR below 1 indicates a lower estimated instantaneous event rate in the osimertinib group under the fitted time-to-event comparison. It is a relative measure and is not an absolute probability of progression-free survival.
Logistic regression for binary endpoints
Objective response rate and disease control rate were analyzed using logistic regression. The registry reports odds ratios as the effect measure. Logistic regression is appropriate when the outcome is represented as a binary response category rather than a time until an event.
ANCOVA for tumour shrinkage
Tumour shrinkage was analyzed with analysis of covariance (ANCOVA). The registry reports covariate adjustment for ethnicity and baseline sum of diameters of target lesions. The adjusted effect was expressed as a difference in least-squares means.
Other ratio-based analysis for duration of response
Duration of response was analyzed using formulae provided in Ellis S et al 2008. The registry reports a ratio of expected duration of response, calculated using the log-normal probability distribution for duration of response among responding patients.
6. Primary Result: Progression-Free Survival
The primary endpoint was analyzed in the Full Analysis Set of all randomized patients using a log-rank test. The reported hypothesis type was superiority.
Hazard ratio for progression or death
95% CI: 0.23–0.41 · P < 0.001
Osimertinib 80 mg vs chemotherapy
An HR of 0.30 corresponds to an estimated hazard that is 30% of the comparator hazard under the reported analysis; equivalently, the estimated relative reduction in the hazard is 70%.
What the estimate means: The hazard ratio of 0.30 summarizes the relative difference in the instantaneous rate of investigator-assessed progression or death between the randomized treatment groups. Because the registry specifies that HR values below 1 favor osimertinib, the direction of the estimate favors osimertinib 80 mg.
What it does not mean: An HR of 0.30 does not mean that 30% of patients progressed, that 70% of patients were protected from progression, or that every individual patient experienced a 70% reduction in risk. It also is not a ratio of median PFS times.
Precision: The two-sided 95% confidence interval is 0.23–0.41. This interval describes statistical uncertainty around the estimated hazard ratio under the analysis framework; it is not a range containing the treatment effect for individual patients.
P-value: The P-value of <0.001 quantifies evidence against the null hypothesis used for the superiority comparison. It does not measure the size of the treatment effect. The magnitude is described by the HR, while the confidence interval describes its statistical precision.
Important caution: A hazard ratio is a model-based time-to-event summary. Its interpretation depends on the underlying survival-analysis framework and should not automatically be translated into a constant individual-level risk reduction over every point in time.
7. Secondary Endpoint Results
Objective Response Rate
Odds ratio for objective response
95% CI: 3.47–8.48 · P < 0.001
Osimertinib 80 mg vs chemotherapy
Objective response rate was a binary endpoint analyzed with logistic regression in the Full Analysis Set. The odds ratio was adjusted for ethnicity, specifically Asian versus non-Asian, and the registry states that an odds ratio greater than 1.0 favors osimertinib 80 mg.
An odds ratio of 5.39 means that the estimated odds of objective response were 5.39 times as high in the osimertinib group as in the chemotherapy group under the reported adjusted logistic-regression analysis.
This is an odds ratio, not a risk ratio. It should not be interpreted as saying that the probability of response was 5.39 times as high. The distinction becomes especially important when the outcome is not rare.
The 95% CI of 3.47–8.48 describes uncertainty around the estimated odds ratio. The P-value of <0.001 indicates strong evidence against the corresponding null comparison, but it does not quantify the magnitude of the response difference.
Duration of Response
Ratio of expected duration of response
95% CI: 4.04–9.57 · P < 0.001
Osimertinib 80 mg vs chemotherapy
The registry describes this analysis as a ratio of expected duration of response. Treatments were compared using the log-normal probability distribution for duration of response among responding patients.
The estimate of 6.22 indicates that the expected duration of response under the reported analysis was estimated to be 6.22 times the corresponding expectation for chemotherapy.
This estimate is conditional on the response-duration framework described in the registry. It is not a hazard ratio, and it should not be interpreted as saying that every responder remained in response for exactly 6.22 times as long.
The confidence interval of 4.04–9.57 indicates uncertainty around the ratio, while the P-value of <0.001 addresses the statistical comparison rather than the clinical size of the effect.
Disease Control Rate
Odds ratio for disease control
95% CI: 2.64–8.84 · P < 0.001
Osimertinib 80 mg vs chemotherapy
Disease control rate was analyzed using logistic regression in the Full Analysis Set. The registry states that an odds ratio greater than 1.0 favors osimertinib 80 mg.
Tumour Shrinkage
Difference in least-squares means
95% CI: -27.71 to -15.52 · P < 0.001
Osimertinib 80 mg vs chemotherapy
The ANCOVA analysis adjusted for ethnicity and the baseline sum of diameters of target lesions. The registry reports least-squares mean tumour shrinkage values of -46.93 for osimertinib and -25.3 for chemotherapy. The reported difference in least-squares means is defined such that a value below 0 favors osimertinib 80 mg.
ANCOVA estimates a treatment-group difference after accounting for prespecified covariates. In this analysis, baseline tumour burden and ethnicity were included as covariates. The resulting least-squares means are adjusted model-based means rather than simple unadjusted arithmetic averages.
The estimate of -21.62 therefore describes the adjusted difference between the treatment groups under the specified ANCOVA model. The 95% CI of -27.71 to -15.52 quantifies uncertainty around that adjusted difference.
8. Additional Prespecified Time-to-Event Results
The registry also posts analyses of overall survival, time to first subsequent therapy, and time to second subsequent therapy. These analyses use hazard ratios and log-rank tests. They are presented separately from the registered primary endpoint.
| Endpoint | Method | Effect measure | Estimate | 95% CI | P-value |
|---|---|---|---|---|---|
| Overall Survival | Log-rank test | Hazard ratio | 0.87 | 0.67–1.13 | 0.277 |
| Time to First Subsequent Therapy (TFST) | Log-rank test | Hazard ratio | 0.21 | 0.16–0.28 | <0.001 |
| Time to Second Subsequent Therapy (TSST) | Log-rank test | Hazard ratio | 0.87 | 0.69–1.11 | <0.001 |
Overall Survival
The registry reports an overall-survival hazard ratio of 0.87, with a two-sided 95% CI of 0.67–1.13 and P = 0.277. The posted time frame is from randomization until the time of final OS analysis, with a median follow-up of 43 months.
The confidence interval spans 1, and the reported P-value is 0.277. These statistics describe the posted analysis; they should not be converted into a claim that the treatments are identical. A nonsignificant comparison does not prove equality, just as a statistically significant comparison does not by itself establish clinical importance.
Time to First Subsequent Therapy
TFST had an estimated hazard ratio of 0.21, with a two-sided 95% CI of 0.16–0.28 and P < 0.001. The analysis used the Full Analysis Set and a log-rank test.
An HR of 0.21 corresponds to an estimated event hazard equal to 21% of the comparator hazard under the reported analysis, or an estimated relative hazard reduction of 79%. This endpoint concerns time to subsequent therapy and therefore answers a different question from PFS and OS.
Time to Second Subsequent Therapy
TSST had an estimated hazard ratio of 0.87, with a two-sided 95% CI of 0.69–1.11 and a reported P-value of <0.001. The posted analysis used the Full Analysis Set and a log-rank test.
9. Statistical Methods Explained
Why was a log-rank test used for the primary endpoint?
PFS is a time-to-event endpoint. Patients can experience progression or death at different times, and some patients may be censored rather than observed through an event. The log-rank test is designed to compare the survival experience of two groups while using the timing of observed events rather than reducing every patient to a simple binary outcome.
What does an HR of 0.30 mean?
An HR of 0.30 means that the estimated instantaneous rate of progression or death in the osimertinib group was 30% of that in the chemotherapy group under the reported analysis. It can therefore be described as a 70% estimated relative reduction in hazard. It does not mean that 70% of patients avoided progression.
Why is the confidence interval important?
The 95% CI of 0.23–0.41 shows the statistical uncertainty surrounding the estimated primary hazard ratio. A point estimate alone can give a false impression of precision. The interval communicates how much uncertainty remains about the magnitude of the relative treatment effect within the specified statistical framework.
Why is the P-value not a measure of effect size?
The P-value addresses compatibility with a specified null hypothesis under the statistical model. It is affected by both the observed data and the amount of information available. The HR, odds ratio, mean difference, and their confidence intervals are the appropriate quantities for describing the estimated magnitude and precision of the effects.
Why was logistic regression used for objective response?
Objective response rate is a binary outcome: participants are classified according to whether they meet the response definition. Logistic regression models the probability of the binary outcome through the odds scale and allows covariate adjustment. In AURA3, the registry specifically reports logistic regression with an odds ratio and adjustment for ethnicity.
What does the odds ratio of 5.39 mean?
The odds of objective response were estimated to be 5.39 times as high with osimertinib as with chemotherapy under the reported adjusted logistic model. It is important not to substitute a probability ratio for an odds ratio. For example, an odds ratio of 5.39 does not mean that the probability of response was exactly 5.39 times higher.
Why was ANCOVA used for tumour shrinkage?
Tumour shrinkage is represented as a continuous change from baseline, making a linear-model framework appropriate. ANCOVA incorporates baseline and other prespecified covariates while estimating an adjusted treatment-group difference. In this trial, the registry identifies ethnicity and baseline sum of diameters of target lesions as covariates.
10. Intention-to-Treat Analysis and Randomization
The primary PFS analysis was performed in the Full Analysis Set of all randomized patients. The registry also identifies intention-to-treat analysis as a concept in the primary statistical analysis.
Why randomization matters
Randomization establishes the treatment assignment mechanism before outcomes are observed. This provides the basis for comparing groups according to their assigned treatments.
Why the analysis population matters
Using all randomized patients for the primary efficacy analysis keeps the analysis connected to the randomized comparison rather than selecting participants based on treatment exposure or later outcomes.
This distinction is especially relevant in a trial that includes crossover to osimertinib as an intervention. Treatment received after the initial randomized assignment can affect later outcomes, but it does not change the original treatment assignment used to define the randomized comparison.
11. Safety Results
The ClinicalTrials.gov record provides serious adverse events by arm as affected participants over participants at risk. These data are presented without converting them into independently calculated percentages.
| Arm | Serious adverse events affected / at risk |
|---|---|
| Osimertinib 80 mg | 84 / 279 |
| Chemotherapy | 36 / 136 |
These counts provide a direct safety description from the registry. They should not be interpreted as an estimate of comparative treatment effect without considering the underlying safety population, exposure, follow-up, event definitions, and denominators.
12. Understanding the Different Effect Measures
| Endpoint | Effect measure | Estimate | What it summarizes |
|---|---|---|---|
| PFS | Hazard ratio | 0.30 | Relative time-to-event hazard for progression or death |
| Objective Response Rate | Odds ratio | 5.39 | Relative odds of meeting the response definition |
| Duration of Response | Ratio of expected duration | 6.22 | Ratio of expected response duration under the reported log-normal analysis |
| Disease Control Rate | Odds ratio | 4.76 | Relative odds of disease control |
| Tumour Shrinkage | Difference in LS means | -21.62 | Adjusted difference in model-based mean percentage change from baseline |
| Overall Survival | Hazard ratio | 0.87 | Relative time-to-event hazard for death |
| TFST | Hazard ratio | 0.21 | Relative time-to-event hazard for first subsequent therapy |
| TSST | Hazard ratio | 0.87 | Relative time-to-event hazard for second subsequent therapy |
A key statistical lesson from AURA3 is that these effect measures are not interchangeable. An odds ratio cannot be read as a hazard ratio, and a ratio of expected response duration cannot be interpreted as an odds ratio. Each estimate must be read in the context of its endpoint and statistical model.
13. Limitations and Interpretation Issues
- Time-to-event interpretation: hazard ratios summarize relative event rates over follow-up and should not automatically be translated into an identical relative difference in individual risk at every time point.
- Censoring: PFS analysis relies on follow-up and censoring rules. Patients without an observed progression or death by the relevant point contribute information only through their observed follow-up.
- Open-label design: the registry classifies masking as none. Investigator assessment can therefore occur in an unmasked treatment environment, which is relevant when considering endpoints based on investigator assessment.
- Different endpoints answer different questions: PFS, response, duration of response, tumour shrinkage, OS, TFST, and TSST should not be collapsed into a single statistic.
- Analysis populations differ by endpoint description: the posted efficacy analyses generally identify the Full Analysis Set, while the ClinicalTrials.gov record is reported using arm-specific affected/at-risk counts.
- Crossover: crossover to osimertinib is identified among the trial interventions. Subsequent treatment can complicate interpretation of later outcomes because treatment received after the original assignment is no longer identical to the initial randomized exposure.
- Registry-level information: the ClinicalTrials.gov record does not provide baseline characteristics, subgroup estimates, or detailed missing-data procedures, so those analyses are not presented here.
- Reported statistical fields: the TSST record contains an HR of 0.87, a 95% CI of 0.69–1.11, and P < 0.001. Because these values are reproduced exactly from the ClinicalTrials.gov record, no independent recalculation is used to replace them.
14. Why This Trial Matters Statistically
AURA3 is a useful teaching example because the registry combines several major clinical-trial statistical frameworks within one randomized phase 3 study. The primary endpoint is a time-to-event outcome analyzed with a log-rank test and hazard ratio, while important secondary endpoints use logistic regression, ANCOVA, and a distribution-based analysis of response duration.
| Concept | How it appears in AURA3 |
|---|---|
| Randomization | Randomized parallel-group phase 3 design |
| Intention-to-treat principle | Primary PFS analysis in the Full Analysis Set of all randomized patients |
| Time-to-event endpoint | Investigator-assessed PFS |
| Log-rank test | Primary PFS comparison and additional time-to-event analyses |
| Hazard ratio | Primary PFS, OS, TFST, and TSST effect measure |
| Confidence interval | 95% two-sided intervals reported for the posted effect estimates |
| Logistic regression | Objective response rate and disease control rate |
| Odds ratio | Effect measure for the binary response endpoints |
| Covariate adjustment | Ethnicity adjustment for objective response; ethnicity and baseline tumour burden for tumour shrinkage |
| ANCOVA | Tumour shrinkage analysis |
| Least-squares means | Adjusted tumour-shrinkage estimates |
| Response-duration modeling | Ratio of expected duration using a log-normal probability distribution |
| Crossover | Cross-over to osimertinib identified as a trial intervention |
15. A Statistical Reading of the AURA3 Results
The primary PFS comparison produced an HR of 0.30 with a two-sided 95% CI of 0.23–0.41 and P < 0.001. The registry specifies that an HR below 1 favors osimertinib 80 mg. Thus, the reported estimate indicates a substantially lower estimated hazard of progression or death in the osimertinib group under the primary analysis.
The objective-response analysis produced an odds ratio of 5.39 with a 95% CI of 3.47–8.48 and P < 0.001. This is an adjusted odds comparison, not a direct ratio of response probabilities.
The ANCOVA estimate was a difference in least-squares means of -21.62, with a 95% CI of -27.71 to -15.52 and P < 0.001. The registry defines a negative difference as favoring osimertinib.
The posted OS analysis has an HR of 0.87, a 95% CI of 0.67–1.13, and P = 0.277, with a median follow-up of 43 months. This illustrates why a trial's treatment effect cannot be summarized from a single endpoint: PFS and OS are distinct time-to-event outcomes with different event definitions.
16. P-values, Confidence Intervals, and Effect Size
AURA3 contains a useful demonstration of why three pieces of statistical information should be read together: the effect estimate, its confidence interval, and the P-value.
Effect estimate
The HR, OR, or mean difference describes the estimated magnitude and direction of the treatment comparison.
Confidence interval
The 95% CI communicates statistical uncertainty around the point estimate under the specified analysis framework.
P-value
The P-value addresses evidence against the relevant null hypothesis; it is not a measure of how large or clinically important an effect is.
Clinical context
Statistical evidence should be interpreted endpoint by endpoint, with attention to the design, analysis population, endpoint definition, and safety information.
For example, the primary PFS HR of 0.30 and its 95% CI of 0.23–0.41 communicate the estimated relative effect and its precision. The P-value of <0.001 provides evidence against the null comparison, but does not transform the HR into a probability that the treatment works.
17. What Crossover Means for Interpretation
The ClinicalTrials.gov record identifies cross-over to osimertinib as an intervention. Crossover is statistically important because randomized assignment and treatment actually received can diverge after the initial allocation.
The primary PFS analysis remains defined by the randomized Full Analysis Set. For later outcomes, however, treatment received after the original assignment can influence the observed event history. This is one reason to distinguish an intention-to-treat treatment-assignment comparison from an analysis based on treatment actually received.
18. Related Statistical Tutorials
Learn more about the methods used in this trial:
19. Related Statistical Calculators
20. Sources
- ClinicalTrials.gov: AURA3 — NCT02151981.
- Linked publication: PubMed record — PMID 41166679.
- Linked publication: PubMed record — PMID 40311309.
- Linked publication: PubMed record — PMID 39029876.
- Linked publication: PubMed record — PMID 37379430.
- Linked publication: PubMed record — PMID 35575012.
Continue through the Clinical Biostats statistical pathway
Use the related tutorials and calculators to explore the survival, regression, covariate-adjustment, and confidence-interval methods represented in AURA3.
21. Record Summary
AURA3 provides a compact example of several core clinical-trial statistical methods. The randomized phase 3 design evaluated a time-to-event primary endpoint using a log-rank test and hazard ratio, with the Full Analysis Set defined as all randomized patients. Secondary analyses extended the statistical framework to logistic regression and odds ratios for binary outcomes, ANCOVA and least-squares means for tumour shrinkage, and a log-normal distribution-based analysis for duration of response.
The primary PFS analysis reported an HR of 0.30 with a two-sided 95% CI of 0.23–0.41 and P < 0.001. Secondary analyses reported an objective-response OR of 5.39, a duration-of-response ratio of 6.22, a disease-control OR of 4.76, and a tumour-shrinkage LS-mean difference of -21.62, each with its corresponding confidence interval and reported P-value. The posted OS analysis produced an HR of 0.87 with a 95% CI of 0.67–1.13 and P = 0.277.