← Clinical Trials
Resected Pancreatic Cancer Phase 3 Completed NCT01964430

APACT: Complete Statistical Analysis of Nab-Paclitaxel and Gemcitabine in Resected Pancreatic Cancer

An independent statistical review of the randomized phase 3 APACT trial evaluating nab-paclitaxel and gemcitabine versus gemcitabine alone as adjuvant therapy for patients with resected pancreatic cancer.

Trial period: 28 March 2014 – 30 June 2022  ·  Randomized parallel-group design  ·  866 enrolled
Scope of this record

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

APACT was a randomized, parallel-group, phase 3 oncology trial comparing nab-paclitaxel plus gemcitabine with gemcitabine alone as adjuvant therapy for patients with resected pancreatic cancer. The registry reports 866 participants enrolled, two arms, one registered primary time-to-event endpoint, and statistical analyses using a log-rank framework with hazard ratios.

866
Enrolled
Phase 3 trial
2
Arms
Parallel design
0.88
Primary DFS HR
95% CI 0.729–1.063
0.81
OS HR
95% CI 0.691–0.957
FeatureAPACT
Trial nameAPACT
PhasePhase 3
ConditionResected pancreatic cancer
DesignRandomized, parallel-group
MaskingNone
Primary purposeTreatment
Enrollment866.0
Arms2
Primary endpoint typeTime-to-event
Primary analysis methodLog-rank test
Primary effect measureHazard ratio
Hypothesis typeSuperiority
Lead sponsorCelgene
Sponsor typeIndustry
StatusCompleted

2. Clinical Question

The statistical question is whether the randomized addition of nab-paclitaxel to gemcitabine changes disease-free survival compared with gemcitabine alone when used as adjuvant therapy for patients with resected pancreatic cancer.

Population

Patients with resected pancreatic cancer, as described by the trial's registered title and condition information.

Intervention

Nab-paclitaxel and gemcitabine.

Comparator

Gemcitabine alone.

Primary question

Does the nab-paclitaxel and gemcitabine regimen improve disease-free survival relative to gemcitabine?

3. Trial Design

01
Randomize866 enrolled
02
Arm ANab-paclitaxel + gemcitabine
03
Arm BGemcitabine
04
FollowDFS and OS
05
AnalyzeLog-rank and hazard ratio
ARM A

Nab-paclitaxel and gemcitabine

  • Nab-paclitaxel
  • Gemcitabine
ARM B

Gemcitabine

  • Gemcitabine alone

The registry classifies APACT as randomized with a parallel design and no masking. That combination is important statistically: treatment assignment is determined by randomization, while the absence of masking means the trial was not registered as a masked comparison.

Allocation
Randomized.
Design model
Parallel.
Masking
None.
Primary purpose
Treatment.

4. Trial Timeline

28 March 2014

Trial start

The registered trial start date was 2014-03-28.

30 June 2022

Primary completion

The registered primary completion date was 2022-06-30.

Completed

Registry status

The trial is listed as completed, with results posted on ClinicalTrials.gov.

5. Registered Primary Endpoint

EndpointRegistry definitionTime frame
Kaplan Meier Estimate for Disease Free Survival (DFS) According to the Independent Radiological Review Committee Disease free survival was defined as the time from the date of randomization to the date of disease recurrence or death, whichever occurred earlier. Disease recurrence was determined by the independent radiological review of computed tomography (CT) or magnetic resonance imaging (MRI) scans. Participants who did not have disease recurrence or did not die were censored at the last tumor assessment. Date of randomization up to data cut off date of 31 December 2018; median DFS follow-up time for censored participants was 22.242 months for nab-Paclitaxel and gemcitabine and 13.832 months for gemcitabine alone

This is a classic time-to-event endpoint. Each participant contributes follow-up from randomization until the first qualifying event or censoring. The endpoint therefore contains more information than a simple proportion of participants who experienced recurrence: it incorporates when the event occurred and allows participants with incomplete event follow-up to contribute information up to their censoring time.

6. Secondary Endpoint Reported in the Registry

EndpointDefinition / time frame reportedAnalysis
Kaplan Meier Estimate of Overall Survival (OS) From randomization to date of death; median OS follow-up time for censored participants was 77.832 months for nab-Paclitaxel and gemcitabine and 77.799 months for gemcitabine alone Stratified log-rank; hazard ratio from a stratified Cox proportional hazards model

The registry therefore provides a second time-to-event outcome, overall survival, alongside the primary disease-free survival endpoint. DFS incorporates recurrence or death, whereas OS uses death as the event. These endpoints are related but are not interchangeable.

7. Primary Result: Disease-Free Survival

The posted primary analysis compares nab-paclitaxel and gemcitabine with gemcitabine in the intent-to-treat population. The registry reports a stratified log-rank analysis and a hazard ratio estimated using a stratified Cox proportional hazards model.

Hazard ratio for disease-free survival

0.88

95% CI: 0.729–1.063   ·   P = 0.1824

Comparison: nab-paclitaxel and gemcitabine vs gemcitabine

Primary analysis componentReported result
EndpointKaplan Meier Estimate for Disease Free Survival (DFS) According to the Independent Radiological Review Committee
Analysis populationIntent-to-treat population consisting of all randomized participants regardless of whether they received any investigational product or had any efficacy assessment collected
MethodLog Rank
Effect measureHazard Ratio (HR)
Estimate0.88
95% CI0.729–1.063
P-value0.1824
Hypothesis typeSuperiority
Stratification and adjustment: The registry states that the analysis was stratified by resection status (R0 versus R1) and nodal status (LN+ versus LN). The hazard ratio was estimated using a stratified Cox proportional hazards model adjusting for those strata.
Clinical Biostats interpretation

The DFS hazard ratio of 0.88 means that the estimated hazard of the DFS event in the nab-paclitaxel and gemcitabine group was 0.88 times the estimated hazard in the gemcitabine group under the fitted stratified time-to-event model. Expressed as a simple relative interpretation, this corresponds to an estimated hazard that is 12% lower.

That statement is about the estimated hazard, not about a 12% absolute reduction in recurrence, a 12% increase in cure, or a 12% change for every individual patient. The hazard ratio is a model-based relative measure of event occurrence over follow-up.

The 95% confidence interval of 0.729–1.063 describes the statistical uncertainty surrounding the estimated hazard ratio. Because the interval includes 1, the data are compatible with both a lower and a higher hazard under the model. The interval is therefore important for understanding the precision of the point estimate; the point estimate alone is not sufficient.

The reported P-value of 0.1824 addresses the hypothesis-testing question under the prespecified superiority framework. A P-value is not a measure of effect size, clinical importance, or probability that the treatment works. Its interpretation also depends on the statistical design and analysis framework.

Finally, the Cox-model interpretation assumes that a single hazard ratio provides an appropriate summary of the relative hazards over time. If hazards are not approximately proportional, a single HR can conceal changes in the treatment-effect pattern during follow-up.

8. Secondary Result: Overall Survival

The registry also reports an analysis of overall survival in the intent-to-treat population. The same general survival-analysis structure was used: a stratified log-rank comparison and a stratified Cox proportional hazards model for the hazard ratio.

Hazard ratio for overall survival

0.81

95% CI: 0.691–0.957   ·   P = 0.0128

Comparison: nab-paclitaxel and gemcitabine vs gemcitabine

Secondary analysis componentReported result
EndpointKaplan Meier Estimate of Overall Survival (OS)
Analysis populationIntent-to-treat population consisting of all randomized participants regardless of whether the participant received any investigational product or had any efficacy assessment collected
MethodLog Rank
Effect measureHazard Ratio (HR)
Estimate0.81
95% CI0.691–0.957
P-value0.0128
Hypothesis typeSuperiority
Clinical Biostats interpretation

The OS hazard ratio of 0.81 corresponds to an estimated death hazard 19% lower in the nab-paclitaxel and gemcitabine group relative to gemcitabine under the fitted stratified Cox model.

This does not mean that the probability of death was reduced by exactly 19% for every patient, nor does it represent an absolute survival difference. A hazard ratio summarizes relative event rates within a time-to-event model; absolute survival probabilities require Kaplan-Meier estimates at specified time points or other absolute measures.

The 95% confidence interval of 0.691–0.957 quantifies uncertainty around the estimated hazard ratio. It is relatively narrower than the primary DFS interval, but the interval still represents uncertainty rather than a range containing the true effect for individual patients.

The P-value of 0.0128 indicates the reported strength of evidence against the null hypothesis within the registry's stated superiority analysis. It should not be read as an effect-size measure, nor as the probability that the observed treatment effect is due to chance.

Because OS is a time-to-event endpoint, censoring and the proportional-hazards interpretation remain relevant. The registry's OS result should therefore be read as a stratified survival-model comparison rather than as a simple comparison of percentages.

9. Comparing DFS and OS Statistically

One of the most useful features of APACT for statistical teaching is the distinction between its primary DFS endpoint and its secondary OS endpoint. Both are time-to-event outcomes, but their event definitions differ.

FeatureDisease-free survivalOverall survival
RolePrimary endpointSecondary endpoint
Event definitionDisease recurrence or death, whichever occurred earlierDeath
Analysis populationIntent-to-treatIntent-to-treat
Primary comparison method reportedLog-rankLog-rank
Effect measureHazard ratioHazard ratio
Estimate0.880.81
95% CI0.729–1.0630.691–0.957
P-value0.18240.0128

The different event definitions explain why DFS and OS can produce different estimates. DFS counts recurrence as an event even when the patient remains alive, whereas OS waits for death. Consequently, the two endpoints capture different parts of the clinical trajectory.

The two estimates should also not be combined into a single overall treatment effect. They answer different questions, have different event processes, and occupy different roles in the registered endpoint structure.

10. Statistical Methodology

Kaplan-Meier estimation

The registered primary endpoint is explicitly a Kaplan-Meier estimate of disease-free survival, and the secondary endpoint is a Kaplan-Meier estimate of overall survival. Kaplan-Meier estimation is appropriate for time-to-event data because it incorporates the timing of events and accommodates right censoring.

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

where di is the number of events at time ti and ni is the number at risk immediately before that time.

The Kaplan-Meier estimator is a descriptive estimate of the event-free survival function. It is not itself a hypothesis test. The trial's between-group hypothesis testing was performed using the log-rank method.

Log-rank test

The registry reports a log-rank analysis for both the primary DFS endpoint and secondary OS endpoint. The log-rank test compares the observed and expected numbers of events between randomized groups across event times.

Core idea
Observed events − Expected events → evidence for separation of survival distributions

The test uses information across follow-up rather than reducing each participant to a single binary event/no-event status at a fixed calendar time.

Stratified analysis

The registry states that the primary DFS analysis was stratified by resection status (R0 versus R1) and nodal status (LN+ versus LN). The OS analysis also used these strata. Stratification allows the comparison to account for these prespecified categorical factors rather than treating the trial population as completely homogeneous with respect to them.

Cox proportional hazards model

The hazard ratios were estimated using stratified Cox proportional hazards models. The model provides a relative hazard estimate while adjusting for the specified strata.

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

The hazard ratio is a relative time-to-event measure. It is not a relative risk, an absolute risk difference, a median survival difference, or a percentage of patients benefiting.

Intention-to-treat analysis

The registry explicitly defines the efficacy analysis population as the intent-to-treat population: all randomized participants regardless of whether they received investigational product or had an efficacy assessment collected. This preserves treatment assignment as the basis of the efficacy comparison.

That principle is particularly important in randomized trials. Once participants are randomized, analyzing them according to assigned treatment protects the comparability created by randomization. Removing participants after randomization because of treatment exposure or availability of an efficacy assessment can introduce selection into the treatment comparison.

11. Why Covariate Adjustment and Stratification Matter

The registry's analysis notes identify two strata: resection status and nodal status. The Cox model was estimated with adjustment for these strata.

What stratification does

It permits the time-to-event comparison to account for the specified strata when estimating the treatment effect.

What it does not do

It does not turn a randomized trial into an observational study or guarantee identical treatment effects within every subgroup.

Why randomization still matters

Randomization establishes the treatment-assignment framework. Stratified analysis then incorporates specified design factors into the survival analysis.

Interpretation

The reported HRs are adjusted for the registered strata rather than being unadjusted comparisons of the two Kaplan-Meier distributions.

A useful distinction is therefore between stratification at randomization, stratification in the analysis, and ordinary multivariable covariate adjustment. These concepts are related but are not interchangeable. The registry specifically documents stratified Cox modeling using resection status and nodal status.

12. Statistical Methods Explained

Why was a Kaplan-Meier method used?

DFS and OS are time-to-event outcomes. Kaplan-Meier estimation allows the analysis to retain information about the timing of recurrence or death while appropriately handling participants whose event status is not observed for the entire follow-up period.

What does a DFS hazard ratio of 0.88 mean?

Under the fitted stratified Cox model, the estimated DFS event hazard was 0.88 times that in the gemcitabine group. A convenient relative interpretation is a 12% lower estimated hazard. It does not mean a 12% absolute reduction in recurrence or death.

Why does the confidence interval matter?

The point estimate is only one estimate from the observed data. The 95% confidence interval of 0.729–1.063 shows the uncertainty around the DFS HR. Because it crosses 1, the data do not exclude a treatment hazard above 1 under the stated statistical framework.

Why use a log-rank test?

The log-rank test is designed for comparing survival distributions between groups when event times and censoring are present. It uses the ordering of event times across follow-up rather than requiring every participant to have an observed event.

Why was the analysis stratified?

The registry identifies resection status and nodal status as analysis strata. Stratification allows the Cox model and survival comparison to incorporate these factors while estimating the treatment contrast within the registered analysis framework.

Why is the ITT population important?

The registry defines the ITT population as all randomized participants regardless of investigational-product exposure or efficacy-assessment availability. This keeps randomized treatment assignment as the basis for efficacy inference and avoids selectively removing participants after randomization.

Does P = 0.0128 tell us how large the OS effect is?

No. The P-value describes evidence against a null hypothesis within the specified statistical framework. The magnitude of the estimated effect is described by the HR of 0.81, while its statistical precision is described by the 95% CI of 0.691–0.957.

13. Primary Endpoint Interpretation in Context

The APACT primary endpoint is disease-free survival, not overall survival. That distinction should remain central when interpreting the statistical record. The registered DFS analysis produced an HR of 0.88 with a 95% CI of 0.729–1.063 and P = 0.1824.

The point estimate is below 1, so the estimated treatment effect favors a lower DFS event hazard for the nab-paclitaxel and gemcitabine group within the model. However, the confidence interval crosses 1 and the reported P-value is 0.1824. The appropriate statistical description is therefore the full estimate, interval, and hypothesis-test result rather than a statement based only on the direction of the point estimate.

The secondary OS analysis has a different numerical pattern: HR 0.81, 95% CI 0.691–0.957, and P = 0.0128. This does not change the identity of DFS as the registered primary endpoint. It does, however, provide a separate time-to-event analysis using death as the event.

Endpoint hierarchy matters: a statistically interpretable result for a secondary endpoint should not automatically be relabeled as a primary endpoint result. The registry identifies DFS as the single registered primary endpoint and OS as a secondary outcome measure.

14. Safety Results

The registry provides serious adverse-event counts by treatment group as affected participants over participants at risk.

Safety measureNab-paclitaxel and gemcitabineGemcitabine
Serious adverse events181/42996/423

These figures should be read as affected / at risk, not as randomized sample sizes. The registry's serious-adverse-event data therefore identify the number of affected participants relative to the corresponding safety-analysis denominator reported in the ClinicalTrials.gov record.

Safety and efficacy also answer different statistical questions. The DFS and OS analyses are anchored to randomized treatment assignment through the ITT population, whereas adverse-event summaries concern observed safety events among participants at risk for those events. An efficacy hazard ratio should therefore not be interpreted as a composite measure of efficacy and safety.

15. Analysis Population and Censoring

The registered efficacy analysis uses an ITT population. For DFS, participants who did not experience disease recurrence or death were censored at the last tumor assessment. This is a fundamental component of the endpoint definition.

Event

For DFS, disease recurrence or death, whichever occurred earlier, is the event.

Censoring

Participants without recurrence or death were censored at the last tumor assessment.

Why censoring is needed

Not every participant will have an observed event during the period available for analysis.

Statistical caution

Time-to-event methods rely on assumptions about the censoring mechanism and the information contributed by participants who are censored.

Censoring does not mean that a participant's follow-up is discarded. A censored participant contributes information up to the censoring time. What matters statistically is that the analysis correctly represents the information available before censoring and that the assumptions underlying the censoring mechanism are reasonable.

16. What the Hazard Ratios Do — and Do Not — Mean

DFS hazard ratio

The DFS HR of 0.88 represents the estimated relative hazard of disease recurrence or death in the nab-paclitaxel and gemcitabine group compared with gemcitabine under the stratified Cox model.

It does not mean that 12% fewer patients experienced recurrence, that 12% of patients benefited, or that each patient had exactly a 12% lower probability of recurrence.

OS hazard ratio

The OS HR of 0.81 represents the estimated relative hazard of death in the nab-paclitaxel and gemcitabine group compared with gemcitabine under the stratified Cox model.

It does not mean that 19% fewer patients died, nor does it provide an absolute survival difference. Absolute survival requires a time-specific survival probability or another absolute measure.

Confidence intervals

The DFS 95% CI is 0.729–1.063, while the OS 95% CI is 0.691–0.957. Confidence intervals quantify uncertainty around the estimated treatment effects; they do not describe the range of outcomes that individual patients will experience.

17. P-values and Effect Size

APACT illustrates why a P-value and an effect estimate should always be reported together. The primary DFS analysis has HR 0.88 and P = 0.1824, while the secondary OS analysis has HR 0.81 and P = 0.0128.

QuantityQuestion it helps answer
Hazard ratioWhat is the estimated relative event hazard?
95% confidence intervalHow precisely has that relative effect been estimated?
P-valueHow much evidence against the specified null hypothesis is present under the statistical model?
Kaplan-Meier estimateHow does estimated event-free survival evolve over time?

No one of these quantities fully describes a time-to-event trial. The hazard ratio provides a relative model-based summary, the confidence interval communicates precision, the P-value addresses hypothesis testing, and the Kaplan-Meier estimator describes survival over time.

18. Multiplicity and Endpoint Interpretation

The registry identifies one registered primary endpoint and reports a secondary OS analysis. The ClinicalTrials.gov record does not provide an alpha-allocation scheme, a multiplicity-adjustment procedure, an interim-analysis plan, or an endpoint hierarchy beyond the primary-versus-secondary classification.

Interpretation boundary: The statistical ClinicalTrials.gov record supports describing DFS as the primary endpoint and OS as a secondary endpoint. It does not support adding an unreported multiplicity strategy, alpha-spending rule, interim boundary, or formal gatekeeping procedure.

This distinction is important because the existence of several reported outcomes does not by itself establish how familywise type I error was controlled. A complete multiplicity statement requires the prespecified statistical design or analysis plan.

19. Non-Inferiority, Crossover, and Bayesian Methods

Superiority

The registered hypothesis type is superiority. The trial is therefore not presented as a non-inferiority analysis, and no non-inferiority margin is reported in the ClinicalTrials.gov record.

Crossover

The ClinicalTrials.gov record does not report a crossover design or crossover analysis.

Bayesian methods

No Bayesian analysis is identified in the ClinicalTrials.gov record.

Interim analysis

The ClinicalTrials.gov record does not report an interim-analysis procedure or alpha-spending approach.

These omissions are themselves useful for statistical interpretation. One should not infer a non-inferiority margin, Bayesian prior, interim stopping rule, or crossover adjustment simply because such methods are common in other clinical trials.

20. Missing Data and Imputation

The ClinicalTrials.gov record defines the ITT efficacy population and the DFS censoring rule, but do not report a specific missing-data imputation procedure. For time-to-event endpoints, censoring is part of the survival-analysis framework and should not automatically be described as ordinary missing-data imputation.

Important distinction
Censoring ≠ simple replacement of a missing value

A censored participant contributes observed follow-up until the censoring time. The Kaplan-Meier and Cox methods use that partial follow-up rather than filling in an unobserved event time with a single imputed value.

The registry information posted on ClinicalTrials.gov for APACT does not specify additional imputation rules. Consequently, no specific imputation method is attributed to the trial on this page.

21. Limitations

22. Why This Trial Matters Statistically

APACT is a useful teaching case because it brings together the core elements of randomized time-to-event analysis without requiring the reader to treat a hazard ratio as a standalone answer.

ConceptHow it appears in APACT
RandomizationThe registry classifies the allocation as randomized.
Parallel designThe trial uses a parallel design with two arms.
Intention-to-treat analysisEfficacy analyses include all randomized participants regardless of investigational-product exposure or efficacy assessment.
Kaplan-Meier estimationThe primary DFS and secondary OS outcomes are registered as Kaplan-Meier estimates.
Time-to-event endpointsDFS and OS both use time from randomization to an event.
Log-rank testThe registry reports Log Rank as the analysis method for DFS and OS.
Hazard ratioDFS HR 0.88 and OS HR 0.81 summarize relative treatment effects.
Confidence intervals95% two-sided confidence intervals quantify uncertainty around both HR estimates.
Stratified analysisThe analysis is stratified by resection status and nodal status.
Cox modelStratified Cox proportional hazards models provide the reported HR estimates.
Superiority testingThe registered hypothesis type is superiority.
Safety analysisSerious adverse events are reported by arm as affected participants over participants at risk.

The central statistical lesson is that a randomized survival analysis is a chain of linked decisions: define the event, define the time origin, define censoring, preserve the randomized analysis population, compare survival distributions, estimate a relative treatment effect, and quantify its uncertainty. APACT's registry record explicitly supports each of these core elements.

23. A Practical Reading of the APACT Results

A statistically careful reading starts with the primary endpoint. The DFS HR is 0.88, indicating a lower estimated DFS event hazard in the nab-paclitaxel and gemcitabine group, but the 95% CI of 0.729–1.063 spans the null value and the reported P-value is 0.1824.

The secondary OS analysis gives an HR of 0.81, with a 95% CI of 0.691–0.957 and P = 0.0128. This estimate is also below 1, and its confidence interval does not include 1. Because OS is secondary in the registry, however, it should be described as a secondary time-to-event result rather than replacing the primary endpoint in the statistical narrative.

This distinction illustrates an important general principle: statistical interpretation is not simply the process of identifying whether a P-value is below a threshold. The analysis population, endpoint hierarchy, effect measure, confidence interval, censoring mechanism, stratification, and model assumptions all contribute to what the result actually means.

A concise statistical summary

Primary DFS: HR 0.88 (95% CI 0.729–1.063), P = 0.1824, using a stratified log-rank analysis with a stratified Cox proportional hazards model.

Secondary OS: HR 0.81 (95% CI 0.691–0.957), P = 0.0128, using the same general stratified survival-analysis framework.

The two outcomes should be interpreted separately because DFS and OS have different event definitions and different registered endpoint roles.

24. Related Tutorials

Learn more about the methods used in this trial:

25. Related Calculators

26. Sources

Continue through the Clinical Biostats statistical pathway

Explore the survival-analysis concepts that connect randomized clinical-trial design with Kaplan-Meier estimation, hazard ratios, confidence intervals, and hypothesis testing.

27. Record Summary

APACT is a randomized phase 3, parallel-group trial with two treatment arms and 866 enrolled participants. Its registered primary endpoint is disease-free survival assessed using Kaplan-Meier estimation, with disease recurrence or death defined as the event and participants without recurrence or death censored at their last tumor assessment. The primary statistical comparison uses a log-rank framework, while the reported hazard ratio is estimated using a stratified Cox proportional hazards model.

The primary DFS result is HR 0.88 with a two-sided 95% CI of 0.729–1.063 and P = 0.1824. The registry also reports secondary OS results of HR 0.81, 95% CI 0.691–0.957, and P = 0.0128. Both efficacy analyses use the intent-to-treat population and incorporate stratification by resection status and nodal status.

The most informative statistical reading therefore combines the endpoint definition, analysis population, Kaplan-Meier framework, log-rank comparison, hazard ratio, confidence interval, P-value, and stratified Cox model. Each answers a different part of the statistical question, and none should be interpreted independently of the trial's design.

Clinical Biostats methodology: A trial-results page should distinguish the numerical evidence reported in the registry from the statistical interpretation of that evidence. For APACT, the ClinicalTrials.gov record supports a detailed time-to-event analysis while leaving unreported design features such as specific multiplicity procedures, interim boundaries, crossover methods, and imputation strategies unspecified.