← Clinical Trials
Advanced NSCLC Phase 3 EGFR-Activating Mutations NCT01774721

ARCHER 1050: Complete Statistical Analysis of Dacomitinib in Advanced NSCLC

An independent statistical review of the randomized phase 3 ARCHER 1050 trial comparing dacomitinib with gefitinib as first-line treatment for advanced non-small cell lung cancer with EGFR-activating mutations.

Trial status: Completed  ·  Enrollment: 452  ·  Primary completion: July 29, 2016
Scope of this record

This page separates reported trial results from statistical interpretation. Numerical trial results are restricted to the ClinicalTrials.gov record for ARCHER 1050 and the listed linked records.

Registry note: 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

ARCHER 1050 was a randomized, open-label phase 3 treatment trial comparing dacomitinib with gefitinib in first-line treatment of advanced non-small cell lung cancer with EGFR-activating mutations. The registry reports 452 enrolled participants, two treatment arms, a time-to-event primary endpoint, and superiority testing using a stratified log-rank framework.

452
Enrolled
Randomized trial
2
Treatment arms
Dacomitinib vs gefitinib
0.589
Primary PFS HR
95% CI 0.469–0.739
<0.0001
Primary PFS P-value
1-sided stratified log-rank
FeatureARCHER 1050
Trial nameARCHER 1050
NCT identifierNCT01774721
PhasePhase 3
StatusCompleted
ConditionNon-small Cell Lung Cancer With EGFR-Activating Mutations
Primary purposeTreatment
AllocationRandomized
MaskingNone
Enrollment452
InterventionsDacomitinib (PF-00299804) and gefitinib
Lead sponsorPfizer
Sponsor typeIndustry
Primary endpoint typeTime-to-event
Registered primary endpoints1
Statistical analyses posted8

2. Clinical Question

The central statistical question was whether first-line dacomitinib improved progression-free survival compared with gefitinib in participants with advanced non-small cell lung cancer with EGFR-activating mutations.

Population

Participants with non-small cell lung cancer with EGFR-activating mutations enrolled in the ARCHER 1050 trial.

Intervention

Dacomitinib (PF-00299804).

Comparator

Gefitinib.

Primary question

Does dacomitinib produce superior progression-free survival compared with gefitinib?

3. Trial Design

01
Enroll452 participants
02
Randomize2 treatment arms
03
TreatDacomitinib or gefitinib
04
AssessPFS and other outcomes
05
AnalyzeStratified survival methods
ARM A

Dacomitinib

  • Dacomitinib (PF-00299804)
  • First-line treatment setting
  • Compared with gefitinib
ARM B

Gefitinib

  • Gefitinib
  • First-line treatment setting
  • Compared with dacomitinib
Allocation
Randomized.
Masking
None.
Primary purpose
Treatment.
Trial period
Start: May 9, 2013. Primary completion: July 29, 2016.

4. Primary Endpoint

EndpointRegistry definition / assessmentPrimary analysis
Progression Free Survival (PFS) Based on Independent Radiologic Central (IRC) Review Time from randomization to date of progression of disease (PD) as determined by IRC review as per RECIST v1.1 criteria or death due to any cause, whichever occurred first. PD includes a ≥20% increase in sum of diameters of target lesions, referring to the smallest sum on study, with an absolute increase of ≥5 mm, or appearance of ≥1 new lesions. 1-sided stratified log-rank test; hazard ratio based on stratified Cox regression model
Time frame Day 28 of Cycle 1, Cycle 2 then every 8 weeks until disease progression or death due to any cause, whichever occurred first. Time-to-event analysis

The primary endpoint is therefore a composite time-to-event endpoint. A participant can experience the endpoint through either documented disease progression according to the registered radiologic criteria or death due to any cause, with whichever event occurs first defining the event time.

5. Analysis Population and Statistical Framework

The primary analysis used the intent-to-treat (ITT) population. The registry defines this population as all randomized participants, with study treatment assignment designated according to initial randomization, regardless of whether participants received study treatment or a different treatment.

ElementARCHER 1050 registry information
Analysis populationIntent-to-treat population
ComparisonDacomitinib vs gefitinib
Primary test1-sided stratified log-rank test
Effect measureHazard ratio
Model noted in analysisStratified Cox regression model
Hypothesis typeSuperiority
Confidence interval95%, two-sided
Important distinction: the hypothesis test and confidence interval are reported with different sidedness. The primary hypothesis test is explicitly reported as 1-sided, while the reported confidence interval is 95% two-sided. Those conventions should not be silently treated as interchangeable.

6. Results: Primary Progression-Free Survival

The registry reports a formal statistical comparison for the primary PFS endpoint. The comparison was performed between dacomitinib and gefitinib in the ITT population using a 1-sided stratified log-rank test, with the hazard ratio based on a stratified Cox regression model.

Progression-free survival hazard ratio

0.589

95% CI: 0.469–0.739   ·   P < 0.0001

1-sided stratified log-rank test; hazard ratio based on stratified Cox regression model

Primary endpointDacomitinib vs gefitinib
OutcomeProgression Free Survival based on Independent Radiologic Central Review
Effect measureHazard ratio
Estimate0.589
95% CI0.469–0.739
P-value<0.0001
HypothesisSuperiority
Analysis populationITT
Clinical Biostats interpretation

An HR of 0.589 means that, under the stratified Cox model, the estimated instantaneous rate of the PFS event was about 58.9% as high in the dacomitinib group as in the gefitinib group. Expressed as a simple relative interpretation, this corresponds to an estimated 41.1% lower hazard of progression or death.

The HR does not mean that 41.1% of participants avoided progression or death, nor does it mean that every participant experienced exactly a 41.1% reduction in risk. It is a relative time-to-event effect measure derived from the model.

The 95% confidence interval of 0.469–0.739 describes statistical uncertainty around the estimated hazard ratio. It does not describe the range of individual treatment effects. The fact that the entire interval is below 1 is consistent with a lower estimated hazard in the dacomitinib group under this analysis.

The P < 0.0001 value addresses evidence against the null hypothesis under the specified 1-sided testing framework. A p-value is not a measure of effect size and does not tell us how large or clinically important the treatment effect is. The HR and its confidence interval provide the information about relative effect magnitude and precision.

Because PFS is a censored time-to-event endpoint, participants who have not experienced progression or death by their available follow-up contribute information up to their censoring time. The Cox interpretation also relies on the model's proportional-hazards framework; the registry result does not provide enough information here to establish how well that assumption held over time.

Finally, the primary result is based on the ITT population and the randomized treatment assignment. That is important because preserving the randomized comparison is central to interpreting the treatment effect causally.

Educational note: a Kaplan-Meier curve cannot be reconstructed accurately from the reported hazard ratio, confidence interval, and p-value alone. A valid reconstruction requires underlying event and censoring information or sufficiently detailed source data.

7. Secondary Time-to-Event Results

The registry also reports formal analyses for overall survival, investigator-assessed PFS, and two duration-of-response analyses. These are all time-to-event outcomes analyzed using a 1-sided stratified log-rank test with hazard ratios based on stratified Cox regression models.

Overall Survival

Overall survival hazard ratio

0.748

95% CI: 0.591–0.947   ·   P = 0.0077

Time from randomization until death or last date known as alive, up to 45 months

Clinical Biostats interpretation

The reported OS HR of 0.748 corresponds to an estimated instantaneous rate of death about 74.8% as high with dacomitinib as with gefitinib under the fitted stratified Cox model, or approximately a 25.2% lower estimated hazard.

The 95% CI of 0.591–0.947 quantifies uncertainty around that estimate. It does not indicate that individual patients experienced a hazard reduction confined to this interval, and it does not provide a probability that the true effect lies inside the interval.

The reported P = 0.0077 is evidence against the null hypothesis under the specified 1-sided testing framework. It should not be interpreted as an effect-size measure. The HR and confidence interval are the appropriate quantities for describing the magnitude and precision of the reported relative treatment effect.

Progression-Free Survival Based on Investigator Assessment

Investigator-assessed PFS hazard ratio

0.622

95% CI: 0.497–0.779   ·   P < 0.0001

Day 28 of Cycle 1, Cycle 2 then every 8 weeks until disease progression or death due to any cause, whichever occurred first

Clinical Biostats interpretation

The investigator-assessed PFS HR of 0.622 indicates an estimated instantaneous PFS-event rate approximately 62.2% as high with dacomitinib as with gefitinib under the stratified Cox model. This corresponds to an estimated 37.8% lower hazard.

The 95% CI of 0.497–0.779 describes the precision of the estimated relative effect. It is not a range of patient-level benefit. The P < 0.0001 result addresses statistical evidence under the reported 1-sided testing framework and should not be mistaken for a measure of treatment magnitude.

Duration of Response — IRC Review

Duration-of-response hazard ratio

0.403

95% CI: 0.307–0.529   ·   P < 0.0001

Comparison of dacomitinib vs gefitinib based on IRC review

Clinical Biostats interpretation

The IRC-based duration-of-response HR of 0.403 corresponds to an estimated instantaneous event rate about 40.3% as high in the dacomitinib group as in the gefitinib group under the reported model, or approximately a 59.7% lower estimated hazard.

The 95% CI of 0.307–0.529 indicates substantially more information about the relative effect than the point estimate alone. It still does not establish how long any particular participant will remain in response. The P < 0.0001 value is evidence under the reported hypothesis-testing framework, not a measure of response durability itself.

Duration of Response — Investigator Assessment

Duration-of-response hazard ratio

0.545

95% CI: 0.418–0.711   ·   P < 0.0001

Comparison of dacomitinib vs gefitinib based on investigator assessment

Clinical Biostats interpretation

The investigator-assessed duration-of-response HR of 0.545 corresponds to an estimated instantaneous event rate about 54.5% as high with dacomitinib as with gefitinib, or approximately a 45.5% lower estimated hazard.

The confidence interval of 0.418–0.711 provides the uncertainty around the estimated treatment effect. The very small reported p-value indicates statistical evidence under the specified 1-sided test, but neither the p-value nor the HR by itself describes the absolute duration of response for an individual participant.

8. Secondary Health-Related Quality-of-Life Time-to-Deterioration Result

The registry reports a secondary time-to-event endpoint for health-related quality of life: time to deterioration in pain, dyspnea, fatigue or cough. The analysis used a patient-reported-outcomes analysis set consisting of enrolled participants who started treatment, completed a baseline PRO assessment, and completed at least one post-baseline PRO assessment after the relevant treatment period.

Time to deterioration hazard ratio

1.173

95% CI: 0.928–1.483   ·   P = 0.1641

Baseline until the end of treatment, up to 48 months

Clinical Biostats interpretation

The HR of 1.173 is above 1, meaning that the estimated instantaneous rate of the specified deterioration event was higher in the dacomitinib group than in the gefitinib group under the reported Cox model. The estimate corresponds to an estimated hazard approximately 17.3% higher, but the confidence interval is broad enough to include 1.

The 95% CI of 0.928–1.483 expresses uncertainty around the estimated relative effect. The reported P = 0.1641 does not provide statistical evidence against the null hypothesis under the reported analysis framework. It does not prove that the treatments are equivalent, nor does it establish that one treatment has no effect on health-related quality of life.

This endpoint also used a different analysis population from the primary ITT analysis. That distinction matters when comparing results across endpoints because the population contributing data is not identical.

9. Secondary Objective Response Results

Objective response rate was analyzed as a binary endpoint using the Cochran-Mantel-Haenszel method. The registry reports p-values for both IRC-based and investigator-assessed ORR, but it does not provide an effect estimate or confidence interval for these analyses in the ClinicalTrials.gov record.

EndpointAnalysisP-valueReported estimate / CI
Objective Response Rate based on IRC Review Cochran-Mantel-Haenszel 0.1942 Not reported in the ClinicalTrials.gov record
Objective Response Rate based on Investigator Assessment Cochran-Mantel-Haenszel 0.0924 Not reported in the ClinicalTrials.gov record
How to read these results: the Cochran-Mantel-Haenszel analysis is a method for comparing categorical outcomes while accounting for stratification. Here, the registry reports the method and p-value but not a corresponding effect estimate or confidence interval. It would therefore be inappropriate to manufacture an odds ratio, risk difference, response-rate difference, or confidence interval from the ClinicalTrials.gov record.

10. Statistical Methodology

Kaplan-Meier estimation

The PFS and OS endpoints are time-to-event outcomes. Kaplan-Meier estimation is the standard descriptive framework for estimating the event-free probability over time in the presence of right censoring. The registry data posted on ClinicalTrials.gov for ARCHER 1050 do not include the underlying event-time dataset or a Kaplan-Meier table, so this page does not attempt to reconstruct a survival curve.

Conceptual form
S(t) = ∏ti ≤ t (1 − di/ni)

Here, di is the number of events at event time ti, while ni is the number at risk immediately before that time.

Stratified log-rank test

The primary PFS and reported secondary time-to-event comparisons used a stratified log-rank test. A stratified log-rank procedure compares event-time experience between treatment groups while accounting for prespecified stratification structure. The ClinicalTrials.gov record identifies the analysis as stratified but do not provide the individual stratification factors, so no specific factors are inferred here.

Cox proportional-hazards model

The registry states that the reported hazard ratios were based on stratified Cox regression models. The Cox model estimates a relative hazard between treatment groups while accommodating censored time-to-event observations.

Interpretation of the hazard ratio
HR < 1  →  lower estimated instantaneous event rate in the treatment group

A hazard ratio is a relative time-to-event measure. It is not the same as a risk ratio, an absolute risk difference, a median survival difference, or the percentage of participants who benefit.

Intention-to-treat analysis

The primary PFS analysis used the ITT population. By retaining participants according to their original randomized assignment, the ITT principle preserves the treatment comparison created by randomization. This remains important even if treatment exposure differs after randomization.

Cochran-Mantel-Haenszel test

The ORR analyses used the Cochran-Mantel-Haenszel method. For a binary outcome, this approach can compare treatment groups while accounting for stratification rather than simply pooling all observations into an unadjusted comparison.

Conceptual comparison
Stratified categorical comparison  →  treatment effect evaluated across strata

The method is especially useful when the trial design or analysis calls for control of a stratification structure. The ClinicalTrials.gov record does not provide the individual factors used for the ORR analysis.

11. Statistical Methods Explained

Why was a stratified log-rank test used?

A log-rank test is designed for comparing time-to-event distributions between treatment groups. Stratification extends the comparison so that the test accounts for the trial's stratified analysis structure. This is different from simply comparing two percentages at one fixed time point because PFS and OS incorporate the timing of events and censoring.

What does an HR of 0.589 mean?

The primary PFS estimate of 0.589 indicates that the estimated instantaneous rate of progression or death was about 58.9% as high in the dacomitinib group as in the gefitinib group under the reported stratified Cox model. The complementary interpretation is an estimated 41.1% lower hazard. It does not mean a 41.1% absolute reduction in the proportion of patients who progressed or died.

Why is the confidence interval important?

The point estimate alone does not communicate how precisely the treatment effect has been estimated. The 95% CI of 0.469–0.739 gives a range of values compatible with the statistical uncertainty represented by the model and sampling process. It is not a prediction interval for individual patients.

Why doesn't the p-value measure effect size?

A p-value quantifies the compatibility of the observed result with a null hypothesis under a specified statistical test. It is affected by the amount of information in the analysis and therefore should not be treated as a scale of clinical importance. The HR describes relative magnitude; the confidence interval describes uncertainty around that magnitude.

Why does the analysis use the ITT population?

Using the ITT population maintains participants in the groups to which they were randomized. This protects the interpretability of the randomized comparison and avoids redefining treatment groups after randomization based on treatment received.

Why can the PFS HR not be converted directly into a median PFS difference?

A hazard ratio summarizes a relative event-rate relationship over follow-up under the Cox model. A median is a particular time point at which an estimated survival curve reaches a specified probability. Without the underlying survival distribution or a reported median, the HR cannot be used to derive an exact median PFS for either treatment group.

12. Primary Endpoint Interpretation in Context

Relative effect

The primary PFS HR was 0.589, indicating a lower estimated instantaneous rate of progression or death under the reported model.

Absolute timing

The ClinicalTrials.gov record does not report median PFS or specific Kaplan-Meier survival probabilities, so the absolute timing of PFS cannot be reconstructed from the registry analysis alone.

Statistical evidence

The reported 1-sided stratified log-rank test produced P < 0.0001.

Precision

The 95% two-sided confidence interval was 0.469–0.739, providing the reported uncertainty around the HR estimate.

This distinction between relative and absolute measures is particularly important in time-to-event analysis. The hazard ratio answers a relative question about event rates, whereas a median or a survival probability at a specified time answers an absolute timing or probability question. Because those additional quantities are not contained in the ClinicalTrials.gov record, they should not be inferred from the HR.

13. Safety Results

The registry reports serious adverse events by treatment arm. The ClinicalTrials.gov record provides the number of affected participants and the number at risk for dacomitinib and gefitinib, along with a separate entry for dacomitinib participants described as ongoing at the data cutoff.

Serious adverse eventsAffected / at risk
Dacomitinib69/227
Gefitinib53/224
Dacomitinib (Ongoing at DCO)4/11
Clinical Biostats interpretation

The reported serious-adverse-event entries should be read exactly as registry counts: 69/227 for dacomitinib and 53/224 for gefitinib, with a separate 4/11 dacomitinib entry identified as ongoing at the data cutoff. The ClinicalTrials.gov record does not provide enough detail to establish whether these entries represent identical follow-up windows or mutually exclusive analysis sets.

For that reason, the counts should not be converted here into a comparative risk ratio or risk difference. Safety interpretation requires attention to the denominator, exposure period, event definition, and analysis population, not merely the number of affected participants.

14. Endpoint-by-Endpoint Statistical Summary

EndpointTypeMethodEffect95% CIP-value
PFS based on IRC review Time-to-event 1-sided stratified log-rank; stratified Cox model HR 0.589 0.469–0.739 <0.0001
Overall survival Time-to-event 1-sided stratified log-rank; stratified Cox model HR 0.748 0.591–0.947 0.0077
PFS based on investigator assessment Time-to-event 1-sided stratified log-rank; stratified Cox model HR 0.622 0.497–0.779 <0.0001
Duration of response, IRC review Time-to-event 1-sided stratified log-rank; stratified Cox model HR 0.403 0.307–0.529 <0.0001
Duration of response, investigator assessment Time-to-event 1-sided stratified log-rank; stratified Cox model HR 0.545 0.418–0.711 <0.0001
HRQOL: time to deterioration Time-to-event Unstratified log-rank as reported; Cox proportional hazard HR 1.173 0.928–1.483 0.1641
ORR based on IRC review Binary Cochran-Mantel-Haenszel Not reported Not reported 0.1942
ORR based on investigator assessment Binary Cochran-Mantel-Haenszel Not reported Not reported 0.0924

The table illustrates an important statistical principle: a trial can contain several analyses with different endpoint types, populations, and methods. Time-to-event outcomes require methods that account for censoring and event timing, whereas binary outcomes such as ORR are naturally handled with categorical-data methods such as the Cochran-Mantel-Haenszel test.

15. What the Primary Hazard Ratio Does — and Does Not — Mean

Relative effect

The primary PFS HR of 0.589 indicates an estimated instantaneous progression-or-death rate about 41.1% lower with dacomitinib than with gefitinib under the reported stratified Cox model.

Not an absolute risk reduction

The HR does not mean that the probability of progression or death was reduced by 41.1 percentage points. Absolute risk differences require survival probabilities at a specified time or other directly reported absolute measures.

Not a patient-level guarantee

The HR is a population-level relative measure. It does not imply that each participant experienced the same proportional change in event risk.

Not a p-value

The HR describes estimated treatment magnitude. The p-value describes statistical evidence under the specified hypothesis test. They answer different questions and should be interpreted together rather than substituted for one another.

Not a complete description of time-to-event behavior

A single Cox HR summarizes the treatment comparison through a model. If hazards vary substantially over time, a single HR can conceal changes in the relative event rates. The ClinicalTrials.gov record does not provide enough information to evaluate proportional hazards directly.

16. Limitations

17. Why This Trial Matters Statistically

ARCHER 1050 is a useful teaching example because its registry record brings together several foundational clinical-trial methods: randomization, an ITT analysis, a time-to-event primary endpoint, stratified log-rank testing, Cox regression, hazard ratios, confidence intervals, and a separate categorical-data analysis for objective response.

ConceptHow it appears in ARCHER 1050
RandomizationThe trial uses randomized allocation to dacomitinib or gefitinib.
ITT analysisThe primary PFS analysis includes all randomized participants according to initial treatment assignment.
Time-to-event endpointThe primary endpoint is PFS based on independent radiologic central review.
Kaplan-Meier estimationTime-to-event endpoints are naturally described using survival-function estimation, although the ClinicalTrials.gov record does not include the curve itself.
Stratified log-rank testThe primary PFS analysis uses a 1-sided stratified log-rank test.
Hazard ratioThe primary treatment effect is summarized as HR 0.589.
Cox regressionThe registry states that the hazard ratio is based on a stratified Cox regression model.
Confidence intervalThe primary HR has a 95% two-sided CI of 0.469–0.739.
Categorical analysisIRC- and investigator-assessed ORR use the Cochran-Mantel-Haenszel method.
Endpoint-specific populationsThe primary PFS analysis uses ITT, while the HRQOL time-to-deterioration analysis uses a PRO analysis set.
Safety analysisSerious adverse events are reported by arm using affected/at-risk counts.

18. A Practical Reading Strategy for ARCHER 1050

A statistically disciplined reading of the trial starts with the primary endpoint rather than with individual p-values. The sequence is important because it keeps the treatment effect, uncertainty, analysis population, and endpoint definition connected.

01
DefinePFS event
02
CompareRandomized groups
03
EstimateHazard ratio
04
Quantify95% CI
05
TestP-value

For ARCHER 1050, that sequence produces a coherent statistical summary: the primary endpoint was PFS based on independent radiologic central review; the ITT population was analyzed; the comparison used a 1-sided stratified log-rank test; the HR was based on a stratified Cox model; the HR was 0.589; the 95% two-sided CI was 0.469–0.739; and the p-value was <0.0001.

The same discipline should then be applied to secondary endpoints. OS had an HR of 0.748 with a 95% CI of 0.591–0.947 and P = 0.0077. Investigator-assessed PFS had an HR of 0.622 with a 95% CI of 0.497–0.779 and P < 0.0001. Duration of response produced HRs of 0.403 by IRC review and 0.545 by investigator assessment. The HRQOL time-to-deterioration analysis produced an HR of 1.173 with a 95% CI of 0.928–1.483 and P = 0.1641. ORR analyses used the Cochran-Mantel-Haenszel method and reported p-values of 0.1942 and 0.0924, respectively.

19. Related Tutorials

Learn more about the methods used in this trial:

20. Related Calculators

21. Sources

Continue through Clinical Biostats

Connect the trial's endpoints and statistical methods to deeper biostatistics tutorials and practical statistical calculators.

22. Record Summary

ARCHER 1050 provides a useful example of a randomized phase 3 time-to-event analysis in which the primary endpoint was progression-free survival based on independent radiologic central review. The primary analysis used the ITT population, a 1-sided stratified log-rank test, and a stratified Cox regression model for the hazard ratio. The reported HR of 0.589 had a 95% two-sided CI of 0.469–0.739 and a p-value of <0.0001.

The secondary analyses demonstrate why clinical-trial interpretation should remain endpoint-specific. Overall survival had an HR of 0.748, investigator-assessed PFS had an HR of 0.622, and duration-of-response analyses produced HRs of 0.403 by IRC review and 0.545 by investigator assessment. The HRQOL time-to-deterioration analysis had an HR of 1.173 with a 95% CI of 0.928–1.483 and P = 0.1641. ORR was analyzed with the Cochran-Mantel-Haenszel method, with p-values of 0.1942 for IRC assessment and 0.0924 for investigator assessment.

The most important statistical lesson is that these numbers cannot be reduced to p-values alone. A rigorous interpretation connects the endpoint definition, analysis population, statistical test, effect estimate, confidence interval, and censoring and modeling assumptions. Where the ClinicalTrials.gov record does not report absolute survival times, event counts, or response estimates, those quantities should remain unreported rather than being reconstructed from incomplete information.

Clinical Biostats methodology: A trial-results page should distinguish reported evidence from statistical interpretation. The goal is to explain how the trial's design and analysis produce its reported estimates, what those estimates mean, and what they do not establish.