This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record. Numerical trial results on this page are restricted to the ClinicalTrials.gov record.
1. Trial at a Glance
LUX-Lung 6 was a randomized, open-label, parallel-group phase 3 treatment trial with 364 participants. It compared BIBW 2992 (afatinib) with gemcitabine-cisplatin chemotherapy in first-line non-small cell lung cancer, with progression-free survival designated as the registered primary endpoint.
| Feature | LUX-Lung 6 |
|---|---|
| Trial name | LUX-Lung 6 |
| ClinicalTrials.gov identifier | NCT01121393 |
| Brief title | BIBW 2992 (Afatinib) vs Gemcitabine-cisplatin in 1st Line Non-Small Cell Lung Cancer (NSCLC) |
| Phase | Phase 3 |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | None |
| Primary purpose | Treatment |
| Enrollment | 364 |
| Arms | 2 |
| Interventions | Gemcitabine + Cisplatin; BIBW 2992 |
| Primary endpoint | Progression-free Survival |
| Primary endpoint type | Time-to-event |
| Hypothesis type | Superiority |
| Trial status | Completed |
| Start | 2010-04-19 |
| Primary completion | 2017-11-23 |
| Lead sponsor | Boehringer Ingelheim |
| Sponsor type | Industry |
2. Clinical Question
The statistical question was whether treatment with BIBW 2992 (afatinib) differed from gemcitabine-cisplatin chemotherapy with respect to the registered primary endpoint, progression-free survival, in patients with non-small cell lung cancer and adenocarcinoma represented in the trial record.
Population
Participants enrolled in the first-line non-small cell lung cancer trial described by the registry, including the listed conditions of carcinoma, non-small-cell lung and adenocarcinoma.
Intervention
BIBW 2992 (afatinib), with the reported primary efficacy comparison identified as afatinib 40 mg.
Comparator
Gemcitabine / cisplatin chemotherapy.
Primary question
Does afatinib produce a different progression-free survival experience from gemcitabine-cisplatin chemotherapy under the prespecified superiority framework?
3. Trial Design
Afatinib
- BIBW 2992 (afatinib)
- Primary analysis identifies the treatment group as afatinib 40 mg
Gemcitabine / Cisplatin Chemotherapy
- Gemcitabine + cisplatin
- Used as the comparator treatment group in the posted analyses
The absence of masking is statistically relevant because treatment assignment was not concealed from participants or investigators after randomization. Randomization nevertheless provides the structural basis for comparing outcomes between the assigned treatment groups. For time-to-event outcomes, the analysis must additionally account for censoring and differing follow-up times.
4. Endpoints
The registry identifies one primary endpoint: progression-free survival. The ClinicalTrials.gov record also contain secondary analyses of objective response, disease control, overall survival, tumour shrinkage, and health-related quality-of-life deterioration endpoints.
| Endpoint | Type | Registered / reported time frame | Analysis |
|---|---|---|---|
| Progression-free Survival | Time-to-event | Tumour assessment were performed at screening, week 6, 12, 18, 24, 30, 36, 42, 48 and then every 12 weeks until progression or death whichever occurs first up to week 374 | Stratified Cox proportional-hazards model; stratified log-rank test |
| Objective Response (OR) | Binary | Tumour assessment were performed at screening, week 6, 12, 18, 24, 30, 36, 42, 48 and then every 12 weeks until progression or death whichever occurs first up to week 374 | Stratified logistic regression |
| Disease Control (DC) | Binary | Tumour assessment were performed at screening, week 6, 12, 18, 24, 30, 36, 42, 48 and then every 12 weeks until progression or death whichever occurs first up to week 374 | Stratified logistic regression |
| Overall Survival (OS) | Time-to-event | From randomisation up to 374 weeks | Stratified Cox proportional-hazards model; stratified log-rank test |
| Tumour Shrinkage | Continuous | Tumour assessment were performed at screening, week 6, 12, 18, 24, 30, 36, 42, 48 and then every 12 weeks until progression or death whichever occurs first up to week 374 | ANCOVA |
| HRQOL: Time of Deterioration in Coughing | Time-to-event | Baseline and throughout the study (every 3 weeks) until progression or death (whichever occurs first) up to 374 weeks. | Stratified Cox model; stratified log-rank test |
| HRQOL: Time of Deterioration in Dyspnoea | Time-to-event | Baseline and throughout the study (every 3 weeks) until progression or death (whichever occurs first) up to 374 weeks. | Stratified Cox model; stratified log-rank test |
| HRQOL: Time of Deterioration in Pain | Time-to-event | Baseline and throughout the study (every 3 weeks) until progression or death (whichever occurs first) up to 374 weeks. | Stratified Cox model; stratified log-rank test |
Primary endpoint definition
Progression-free survival was defined as the time from randomisation to disease progression or death, whichever occurs earlier. The registry states that PFS was assessed by central independent review according to Response Evaluation Criteria in Solid Tumours (RECIST) version 1.1 and that a pre-defined set of censoring rules was used for patients who did not have a PFS.
5. Statistical Methodology
The registry analyses show four principal statistical methods: Cox proportional-hazards models, log-rank tests, logistic regression, and ANCOVA. The methods correspond naturally to the endpoint structures: time-to-event outcomes, binary outcomes, and a continuous tumour-shrinkage outcome.
Stratified Cox proportional-hazards model
The primary PFS hazard ratio was estimated with a Cox proportional-hazards model stratified by EGFR mutation category. The same stratification concept was used for the posted OS and HRQOL time-to-event analyses.
An HR below 1 indicates a lower estimated instantaneous event rate in the afatinib group than in the comparator group under the fitted model. It is a relative time-to-event measure, not an absolute probability of an event.
Stratified log-rank test
The primary PFS comparison used a stratified log-rank test based on the EGFR mutation category used as a stratification factor at randomisation. The same framework was used for the posted OS and HRQOL log-rank analyses.
Logistic regression
Objective response and disease control were binary outcomes. The registry reports stratified logistic regression for both, with the treatment effect expressed as an odds ratio. Logistic regression estimates the association between randomized treatment group and the odds of the binary outcome while incorporating the specified stratification structure.
ANCOVA
Tumour shrinkage was analyzed using ANCOVA. The registry states that the analysis was adjusted for baseline sum of diameters and EGFR mutation group. This is important because the comparison concerns final tumour measurements while accounting for baseline tumour burden and the specified mutation-group factor.
The purpose of covariate adjustment is to compare treatment groups after accounting for relevant baseline information. It does not turn an observational comparison into a randomized trial; here, the randomized design provides the principal basis for the treatment comparison.
Analysis populations
The primary PFS analysis used the randomised set (RS), defined in the registry as all patients randomized to receive treatment, whether treated or not. The objective response, disease control, overall survival, and HRQOL analyses reported here also identify the randomized set as the analysis population.
| Endpoint | Analysis population | Key method |
|---|---|---|
| Progression-free Survival | Randomised set | Stratified Cox; stratified log-rank |
| Objective Response | Randomised set | Stratified logistic regression |
| Disease Control | Randomised set | Stratified logistic regression |
| Overall Survival | Randomised set | Stratified Cox; stratified log-rank |
| Tumour Shrinkage | Randomised set with baseline and post-baseline target-lesion measurements | ANCOVA |
| HRQOL deterioration endpoints | Randomised set | Stratified Cox; stratified log-rank |
6. Primary Result: Progression-Free Survival
The registered primary endpoint was progression-free survival. The ClinicalTrials.gov record contains both a model-based effect estimate and a hypothesis test.
Hazard ratio for progression or death
95% CI: 0.203–0.389 · Two-sided
Stratified log-rank P < 0.0001
| Primary PFS analysis | Reported result |
|---|---|
| Comparison | Afatinib 40 mg vs Gemcitabine / Cisplatin Chemotherapy |
| Analysis population | Randomised set |
| Effect measure | Hazard ratio |
| HR | 0.281 |
| 95% CI | 0.203–0.389 |
| CI type | Two-sided |
| Model | Cox proportional-hazards model stratified by EGFR mutation category |
| Hypothesis test | Stratified log-rank test |
| P-value | <0.0001 |
| Hypothesis type | Superiority |
An HR of 0.281 means that the estimated instantaneous rate of progression or death in the afatinib group was approximately 28.1% of the corresponding estimated rate in the gemcitabine-cisplatin group under the fitted Cox model. Equivalently, 0.281 corresponds to an estimated 71.9% lower hazard relative to the comparator.
The HR does not mean that 71.9% of participants avoided progression, nor does it mean that each participant experienced exactly a 71.9% reduction in individual risk. It is a relative model-based measure of the event rate over time.
The 95% CI of 0.203–0.389 describes statistical uncertainty around the estimated hazard ratio. It is not a prediction interval for individual patients and does not mean that individual treatment effects must fall inside this range.
The P-value of <0.0001 addresses evidence against the relevant null hypothesis under the specified stratified log-rank testing framework. It does not measure the size or clinical importance of the treatment effect. Effect size is described by the hazard ratio and its confidence interval.
Because this is a Cox analysis, interpretation also depends on the proportional-hazards framework underlying a single HR summary. The ClinicalTrials.gov record does not report a formal assessment of that assumption, so the HR should be understood as the reported model-based summary rather than as a guarantee that the relative hazards were identical at every point in follow-up.
Why both Cox and log-rank analyses were useful
The two primary analyses answer related but different statistical questions. The stratified log-rank test provides the formal hypothesis test comparing the time-to-event distributions while respecting the EGFR mutation-category stratification. The stratified Cox model supplies an interpretable relative effect estimate—the hazard ratio—together with its confidence interval.
7. Secondary Efficacy Results
Objective Response
Odds ratio for objective response
95% CI: 4.522–12.679 · P < 0.0001
Stratified logistic regression
The reported odds ratio of 7.572 means that the estimated odds of objective response were 7.572 times as high in the afatinib group as in the gemcitabine-cisplatin group under the specified stratified logistic regression model.
An odds ratio is not the same as a risk ratio or a difference in response percentages. Without the underlying response counts in the ClinicalTrials.gov record, an absolute response-rate difference cannot be calculated without introducing information that is not provided here.
The 95% CI of 4.522–12.679 quantifies uncertainty around the odds-ratio estimate. The P-value of <0.0001 addresses the statistical test; it does not say that the odds ratio is clinically important merely because the P-value is small.
Disease Control
Odds ratio for disease control
95% CI: 2.039–7.240 · P < 0.0001
Stratified logistic regression
The disease-control odds ratio of 3.843 indicates that the estimated odds of disease control were 3.843 times as high in the afatinib group as in the comparator group under the reported stratified logistic model.
The confidence interval, 2.039–7.240, expresses uncertainty around that relative odds estimate. As with objective response, the odds ratio should not be read as a percentage-point improvement in disease control.
The P-value of <0.0001 provides evidence against the null hypothesis used for the reported comparison, but it does not quantify the magnitude of benefit. The magnitude is represented by the odds ratio and its confidence interval.
Tumour Shrinkage
Adjusted mean difference in final tumour measurements
95% CI: −17.10 to −10.19 mm · P < 0.0001
ANCOVA adjusted for baseline sum of diameters and EGFR mutation group
The reported mean difference of −13.64 mm indicates a lower adjusted final value in the afatinib group relative to the gemcitabine-cisplatin group, using the ANCOVA model specified in the registry.
The confidence interval of −17.10 to −10.19 mm describes uncertainty around the adjusted mean difference. Because the analysis is adjusted for baseline sum of diameters and EGFR mutation group, this is not simply the raw difference between two unadjusted final means.
The tumour-shrinkage analysis had a more restricted measurement population: the registry states that there were only 220 patients in the afatinib arm and 101 in the gemcitabine-cisplatin arm with both baseline and post-baseline target-lesion measurements. That restriction is important when interpreting this endpoint because it differs from the broader randomized-set framework used for the primary PFS analysis.
8. Overall Survival
Overall survival was a secondary time-to-event endpoint measured from randomisation up to 374 weeks. The registry provides both a stratified Cox hazard ratio and a stratified log-rank P-value.
Hazard ratio for overall survival
95% CI: 0.715–1.144 · Stratified log-rank P = 0.4013
Stratified by EGFR mutation category
| OS analysis | Reported result |
|---|---|
| Time frame | From randomisation up to 374 weeks |
| Analysis population | Randomised set |
| Effect measure | Hazard ratio |
| HR | 0.904 |
| 95% CI | 0.715–1.144 |
| CI type | Two-sided |
| Cox model | Stratified by EGFR mutation category |
| Log-rank P-value | 0.4013 |
| Hypothesis type | Superiority |
An OS HR of 0.904 corresponds to an estimated instantaneous rate of death about 9.6% lower in the afatinib group relative to the comparator under the reported Cox model.
The 95% CI of 0.715–1.144 spans 1.00. Thus, the interval includes both values corresponding to a lower estimated hazard and values corresponding to a higher estimated hazard. The registry-reported two-sided log-rank P-value of 0.4013 is not small under conventional hypothesis-testing interpretation.
The P-value does not establish the size of any possible treatment effect. The hazard ratio and confidence interval are the appropriate reported quantities for describing the estimated relative effect and its uncertainty.
No median overall survival or absolute survival probabilities are included in the ClinicalTrials.gov record. They are therefore not presented here.
9. Health-Related Quality of Life
The registry contains three HRQOL time-to-deterioration endpoints: coughing, dyspnoea, and pain. Each was assessed from baseline throughout the study every 3 weeks until progression or death, whichever occurred first, up to 374 weeks. Both Cox and log-rank analyses were posted for each endpoint.
| HRQOL endpoint | HR | 95% CI | Log-rank P-value |
|---|---|---|---|
| Time of Deterioration in Coughing | 0.458 | 0.303–0.692 | 0.0001 |
| Time of Deterioration in Dyspnoea | 0.534 | 0.394–0.724 | <0.0001 |
| Time of Deterioration in Pain | 0.699 | 0.511–0.956 | 0.0220 |
For all three reported HRQOL deterioration endpoints, the estimated hazard ratios are below 1. Under the Cox framework, this corresponds to lower estimated instantaneous rates of deterioration in the afatinib group relative to the comparator.
For coughing, the HR was 0.458, corresponding to an estimated 54.2% lower hazard of deterioration. The 95% CI was 0.303–0.692, with a log-rank P-value of 0.0001.
For dyspnoea, the HR was 0.534, corresponding to an estimated 46.6% lower hazard of deterioration. The 95% CI was 0.394–0.724, with a log-rank P-value of <0.0001.
For pain, the HR was 0.699, corresponding to an estimated 30.1% lower hazard of deterioration. The 95% CI was 0.511–0.956, with a log-rank P-value of 0.0220.
These endpoints should be distinguished from PFS and OS. They describe time to deterioration in specific quality-of-life domains rather than time to tumour progression or death.
10. Secondary Time-to-Event Analysis: What the HRQOL Results Teach
The HRQOL results illustrate an important statistical principle: a clinical outcome can be converted into a time-to-event endpoint when the analysis focuses on time until deterioration rather than simply whether deterioration occurred at any point.
Censoring
A participant who has not experienced the specified deterioration by the end of usable follow-up does not necessarily contribute a complete event time. The survival-analysis framework allows such observations to contribute censored follow-up.
Relative effect
The HR summarizes the relative instantaneous event rate under the Cox model. It does not directly state how many months a participant's deterioration was delayed.
Endpoint-specific meaning
An HR for coughing should not be interpreted as an HR for overall survival or progression-free survival. Each endpoint has its own event definition.
Multiple outcomes
When several secondary endpoints are tested, the collection of P-values requires careful interpretation. The ClinicalTrials.gov record does not provide a multiplicity-adjustment strategy for these analyses.
11. Statistical Methods Explained
Why was a Cox proportional-hazards model used for PFS?
PFS is a time-to-event endpoint. Participants can experience progression or death at different times, while others may remain event-free at their last assessment and therefore be censored. The Cox model is designed for this structure and produces a hazard ratio that summarizes the relative event rate between treatment groups. In LUX-Lung 6, the model was stratified by EGFR mutation category.
What does an HR of 0.281 mean?
An HR of 0.281 means that the fitted model estimates the instantaneous rate of progression or death in the afatinib group to be 0.281 times the comparator rate. Expressed as a relative reduction, this corresponds to approximately 71.9% lower estimated hazard. It does not mean that 71.9% of patients avoided progression.
Why was a stratified log-rank test used?
The log-rank test compares time-to-event experience between treatment groups while accounting for the survival-analysis structure. Here, the registry specifically reports stratification by EGFR mutation category, the factor used at randomisation. Stratification prevents the comparison from ignoring a prespecified factor incorporated into the trial design.
Why was logistic regression used for objective response?
Objective response is recorded as a binary outcome: a participant either satisfies the response definition or does not. Logistic regression is therefore appropriate for modeling the odds of response. The registry reports a stratified logistic regression model, with the treatment effect expressed as an odds ratio of 7.572.
What does an odds ratio of 7.572 mean?
An odds ratio of 7.572 means that the estimated odds of objective response were 7.572 times as high in the afatinib group as in the comparator group under the specified model. Odds are not probabilities, so the odds ratio cannot be converted into a percentage-point response difference without the underlying response probabilities.
Why was ANCOVA used for tumour shrinkage?
Tumour shrinkage was a continuous measurement in millimetres. ANCOVA allows the treatment comparison to be adjusted for baseline sum of diameters and EGFR mutation group. This can improve the precision and interpretability of the treatment comparison by accounting for relevant baseline information.
Why does the P-value not tell us the size of the treatment effect?
A P-value measures the compatibility of the observed data with a specified null hypothesis under the statistical model. It is affected by both the magnitude of an observed effect and the amount of information in the analysis. Effect magnitude is better described by measures such as the HR or odds ratio together with a confidence interval.
12. Confidence Intervals and Statistical Precision
Confidence intervals are particularly important in a trial with several types of effect measures because they distinguish the point estimate from the uncertainty surrounding it.
| Endpoint | Point estimate | 95% CI | Effect measure |
|---|---|---|---|
| Primary PFS | 0.281 | 0.203–0.389 | Hazard ratio |
| Objective Response | 7.572 | 4.522–12.679 | Odds ratio |
| Disease Control | 3.843 | 2.039–7.240 | Odds ratio |
| Tumour Shrinkage | −13.64 mm | −17.10 to −10.19 mm | Mean difference |
| Overall Survival | 0.904 | 0.715–1.144 | Hazard ratio |
| HRQOL: Coughing | 0.458 | 0.303–0.692 | Hazard ratio |
| HRQOL: Dyspnoea | 0.534 | 0.394–0.724 | Hazard ratio |
| HRQOL: Pain | 0.699 | 0.511–0.956 | Hazard ratio |
The confidence intervals also illustrate why a point estimate should never be interpreted in isolation. The PFS HR is estimated relatively far below 1, while the OS HR is closer to 1 and its interval includes 1. The response and disease-control analyses use odds ratios rather than hazard ratios, while tumour shrinkage uses a mean difference. These measures are not interchangeable.
The effect measure describes the estimated magnitude and direction; the confidence interval describes uncertainty around that estimate; the P-value addresses a hypothesis test. Keeping these three concepts separate prevents many common interpretation errors.
13. Stratified Analysis
EGFR mutation category appears repeatedly in the statistical methods reported by the registry. It was used as a stratification factor in the primary PFS analysis and in the secondary OS and HRQOL time-to-event analyses. It was also included in the ANCOVA adjustment for tumour shrinkage and in the logistic-regression analyses for objective response and disease control.
| Endpoint | Role of EGFR mutation category |
|---|---|
| Progression-free Survival | Stratification factor in Cox model and stratified log-rank test |
| Objective Response | Stratification factor in logistic regression |
| Disease Control | Stratified logistic regression |
| Overall Survival | Stratification factor in Cox model and stratified log-rank test |
| Tumour Shrinkage | Adjustment variable in ANCOVA |
| HRQOL deterioration | Stratification factor in Cox models and stratified log-rank tests |
Stratification is not the same as claiming that EGFR mutation category modifies the treatment effect. A stratification factor is incorporated to account for an important design characteristic. Demonstrating effect modification would require an appropriate interaction or heterogeneity analysis, which is not reported in the ClinicalTrials.gov record.
14. Safety Results
The ClinicalTrials.gov record provides serious adverse events by treatment arm as affected participants over participants at risk.
| Safety measure | Afatinib 40 mg | Gemcitabine / Cisplatin Chemotherapy |
|---|---|---|
| Serious adverse events | 40 / 239 | 12 / 113 |
The affected and at-risk counts are reported exactly as reported in the registry. They should not be treated as equivalent to a formal comparative risk analysis unless the corresponding safety analysis specification and statistical comparison are available.
15. Missing Data, Censoring, and Measurement Populations
Time-to-event endpoints such as PFS, OS, and HRQOL deterioration require special treatment of incomplete follow-up. The PFS definition explicitly states that predefined censoring rules were used for patients who did not have a PFS. This is different from simply treating an incomplete observation as if no event had occurred.
PFS censoring
The registry explicitly states that predefined censoring rules were used for patients who did not have a PFS.
Randomised-set principle
The primary PFS analysis used the randomized set, including patients randomized to receive treatment whether treated or not.
Tumour shrinkage
This endpoint had a restricted measurement population because baseline and post-baseline target-lesion measurements were required.
Registry limitation
The ClinicalTrials.gov record does not describe a general missing-data imputation method for the reported endpoints.
This distinction matters because the primary PFS estimate is based on a time-to-event framework that can incorporate censored observations, whereas tumour shrinkage was analyzed only among participants with the specified baseline and post-baseline measurements. The two analyses therefore answer different statistical questions and rely on different available measurement sets.
16. Multiplicity and Multiple Endpoints
The registry lists 18 posted outcome measures and 13 posted statistical analyses, including one primary endpoint with two posted primary analyses and multiple secondary endpoints. This creates an important interpretive distinction between the prespecified primary endpoint and the collection of secondary findings.
| Analysis family | Role in the ClinicalTrials.gov record | Interpretive point |
|---|---|---|
| Progression-free Survival | Registered primary endpoint | Primary confirmatory treatment comparison under the reported superiority framework |
| Objective Response | Secondary endpoint | Binary efficacy outcome analyzed using logistic regression |
| Disease Control | Secondary endpoint | Binary efficacy outcome analyzed using logistic regression |
| Overall Survival | Secondary endpoint | Time-to-event outcome analyzed using Cox and log-rank methods |
| Tumour Shrinkage | Secondary endpoint | Continuous outcome analyzed using ANCOVA |
| HRQOL deterioration endpoints | Secondary endpoints | Multiple time-to-event outcomes |
Multiple statistical tests create the possibility of false-positive findings across the collection of analyses even when each individual test is conducted correctly. The ClinicalTrials.gov record does not provide a multiplicity-adjustment procedure or endpoint hierarchy for the secondary analyses. Consequently, the individual secondary P-values should be interpreted as the reported results of those analyses rather than as evidence that every secondary endpoint was independently powered and error-controlled at the same level as the primary endpoint.
17. Primary Endpoint vs Secondary Endpoints
The statistical structure of LUX-Lung 6 becomes clearer when the endpoints are grouped by what they measure.
| Question | Endpoint | Effect measure | Method |
|---|---|---|---|
| How long until progression or death? | Progression-free Survival | Hazard ratio | Stratified Cox + log-rank |
| How often was objective response achieved? | Objective Response | Odds ratio | Stratified logistic regression |
| How often was disease controlled? | Disease Control | Odds ratio | Stratified logistic regression |
| How long until death? | Overall Survival | Hazard ratio | Stratified Cox + log-rank |
| How did final tumour measurements differ after adjustment? | Tumour Shrinkage | Mean difference | ANCOVA |
| How long until specific HRQOL deterioration? | Coughing, dyspnoea, pain | Hazard ratio | Stratified Cox + log-rank |
No single endpoint captures the entire statistical evidence. PFS measures disease progression or death, OS measures death, response measures a binary tumour outcome, tumour shrinkage measures a continuous radiologic quantity, and HRQOL deterioration measures time until deterioration in specific patient-reported domains.
18. A Worked Interpretation of the Primary Analysis
Suppose a reader sees only the headline result: HR 0.281, 95% CI 0.203–0.389, P < 0.0001. A statistically literate interpretation proceeds in several steps.
PFS is not OS
The event is disease progression or death, whichever occurs earlier. The estimate therefore describes PFS, not mortality alone.
HR is a relative time-to-event measure
The value 0.281 compares estimated instantaneous event rates under the Cox model. It is not a percentage of patients and is not a median difference.
Read the confidence interval
The 95% CI is 0.203–0.389. This gives the statistical uncertainty around the estimated HR and remains below 1 throughout the interval.
Separate P-value from effect size
The stratified log-rank P-value is <0.0001. This is evidence against the null hypothesis under the reported test, but it does not quantify the magnitude of the effect.
Stratification and population matter
The analysis used the randomized set and stratified the Cox model and log-rank test by EGFR mutation category.
19. Important Limitations and Interpretation Issues
- No masking: the trial was open-label. Participants and investigators were not masked to treatment assignment according to the registry.
- Proportional-hazards interpretation: a single Cox HR is a model-based summary and depends on the proportional-hazards framework. The ClinicalTrials.gov record does not report a formal assessment of that assumption.
- Censoring: PFS depends on prespecified censoring rules for participants without a PFS event. The way censoring is handled can affect a time-to-event estimate.
- Restricted tumour-shrinkage population: only 220 afatinib patients and 101 gemcitabine-cisplatin patients had both baseline and post-baseline target-lesion measurements for the reported ANCOVA.
- Multiple secondary endpoints: the record contains multiple secondary statistical analyses. The ClinicalTrials.gov record does not specify a multiplicity-adjustment strategy for these analyses.
- Different effect measures: hazard ratios, odds ratios, and mean differences describe different quantities and should not be compared numerically as though they were interchangeable.
- Missing information: the ClinicalTrials.gov record does not provide median PFS, median OS, absolute response rates, detailed baseline characteristics, subgroup estimates, or a general imputation strategy.
- Safety inference: serious adverse-event affected/at-risk counts are provided, but the ClinicalTrials.gov record does not include a formal inferential comparison of serious adverse events.
- Secondary endpoint interpretation: secondary P-values should be interpreted in the context of multiple endpoints and the absence of a registry-reported multiplicity strategy.
20. Why This Trial Matters Statistically
LUX-Lung 6 is a useful teaching example because a single randomized trial incorporates several major branches of clinical-trial statistics. The primary endpoint is a censored time-to-event outcome; secondary outcomes include binary response measures, a continuous tumour measurement, overall survival, and several patient-reported time-to-deterioration endpoints.
| Concept | How it appears in LUX-Lung 6 |
|---|---|
| Randomization | Randomized phase 3 parallel-group design |
| Primary endpoint | Progression-free survival |
| Time-to-event analysis | PFS, OS, and HRQOL deterioration |
| Kaplan-Meier framework | The natural descriptive framework for the reported time-to-event endpoints |
| Hazard ratio | Primary PFS and secondary OS/HRQOL effect measure |
| Log-rank test | Formal time-to-event comparison for PFS, OS, and HRQOL endpoints |
| Cox model | Estimation of relative time-to-event effects |
| Stratified analysis | EGFR mutation category used in multiple analyses |
| Logistic regression | Objective response and disease control |
| Odds ratio | Effect measure for binary response outcomes |
| ANCOVA | Tumour shrinkage adjusted for baseline sum of diameters and EGFR mutation group |
| Covariate adjustment | Baseline tumour measurement and EGFR mutation group in tumour-shrinkage analysis |
| Multiple endpoints | One primary endpoint plus multiple secondary outcomes |
| Analysis populations | Randomised set for the reported efficacy analyses, with a restricted measurement population for tumour shrinkage |
21. What the Primary PFS Result Does — and Does Not — Establish
The registry reports a stratified Cox HR of 0.281 with a two-sided 95% CI of 0.203–0.389, together with a stratified log-rank P-value of <0.0001. These are the reported statistical results for the registered primary PFS endpoint under a superiority hypothesis.
The HR does not provide a median PFS, an absolute difference in survival probability at a particular time, the proportion of patients who benefit, or an individual patient's probability of remaining progression-free. Those quantities require corresponding data that are not reported in the ClinicalTrials.gov record used for this page.
The secondary OS analysis reports HR 0.904 with 95% CI 0.715–1.144 and stratified log-rank P = 0.4013. PFS and OS are different endpoints, so the primary PFS result should not be presented as if it were an overall-survival result.
22. Statistical Concepts in This Trial
Learn more about the methods used in this trial:
23. Related Statistical Calculators
24. Sources
- ClinicalTrials.gov: LUX-Lung 6, NCT01121393. Official trial registry record and source for the trial data summarized on this page.
- Linked publication: PubMed PMID 30584317.
- Linked publication: PubMed PMID 29653820.
- Linked publication: PubMed PMID 27601237.
- Linked publication: PubMed PMID 26823294.
- Linked publication: PubMed PMID 26051236.
Continue with the statistical methods
Explore the underlying survival-analysis, regression, confidence-interval, and clinical-trial concepts used to interpret LUX-Lung 6.
25. Record Summary
LUX-Lung 6 provides a compact example of how several statistical methods work together in a randomized phase 3 clinical trial. The registered primary endpoint, progression-free survival, was analyzed in the randomized set using a Cox proportional-hazards model stratified by EGFR mutation category and a stratified log-rank test. The reported HR was 0.281 with a 95% CI of 0.203–0.389, and the stratified log-rank P-value was <0.0001.
The secondary analyses illustrate different statistical data structures. Objective response and disease control were analyzed with stratified logistic regression and reported as odds ratios. Tumour shrinkage was analyzed using ANCOVA adjusted for baseline sum of diameters and EGFR mutation group. Overall survival and three HRQOL time-of-deterioration endpoints used stratified time-to-event methods. Serious adverse events were reported as affected participants over participants at risk by treatment arm.
The most important statistical lesson is that these results should be interpreted according to their endpoint and effect measure. A hazard ratio describes a relative time-to-event quantity, an odds ratio describes relative odds for a binary outcome, and a mean difference describes an adjusted difference in a continuous outcome. Confidence intervals describe uncertainty around these estimates, while P-values address the corresponding hypothesis tests. None of these quantities, considered alone, describes every dimension of clinical outcome.