This page separates reported trial results from statistical interpretation. All numerical trial results on this page are taken from the ClinicalTrials.gov record for NCT03519971. Where the registry does not provide a particular design or result detail, it is not reported here.
1. Trial at a Glance
PACIFIC-2 was a randomized, double-blind, parallel phase 3 trial evaluating durvalumab given with standard-of-care chemoradiation therapy in patients with unresectable non-small cell lung cancer. The registry reports 328 participants, two arms, one registered primary endpoint, 13 posted outcome measures, and seven posted statistical analyses.
| Feature | PACIFIC-2 |
|---|---|
| Phase | Phase 3 |
| Population | Patients with unresectable non-small cell lung cancer |
| Design | Randomized, double-blind, parallel |
| Allocation | Randomized |
| Primary purpose | Treatment |
| Primary endpoint | Progression-Free Survival (PFS) |
| Primary endpoint type | Time-to-event |
| Results posted | Yes |
| Outcome measures posted | 13 |
| Statistical analyses posted | 7 |
| Lead sponsor | AstraZeneca |
| Sponsor type | Industry |
| ClinicalTrials.gov | NCT03519971 |
2. Clinical Question
The central question was whether adding durvalumab to standard-of-care chemoradiation therapy changes progression-free survival compared with placebo given with standard-of-care chemoradiation therapy in patients with unresectable non-small cell lung cancer.
Population
Patients with unresectable non-small cell lung cancer.
Intervention
Durvalumab given with standard-of-care chemoradiation therapy.
Comparator
Placebo given with standard-of-care chemoradiation therapy.
Primary question
Does durvalumab with standard-of-care chemoradiation therapy improve progression-free survival relative to placebo with standard-of-care chemoradiation therapy?
3. Trial Design
Durvalumab + SoC CRT
- Durvalumab
- Standard-of-care chemoradiation therapy
- Cisplatin / Etoposide
- Carboplatin / Paclitaxel
- Pemetrexed / Cisplatin
- Pemetrexed / Carboplatin
- Radiation
Placebo + SoC CRT
- Placebo
- Standard-of-care chemoradiation therapy
- Cisplatin / Etoposide
- Carboplatin / Paclitaxel
- Pemetrexed / Cisplatin
- Pemetrexed / Carboplatin
- Radiation
The ClinicalTrials.gov record identifies the two treatment groups as Durvalumab + SoC CRT and Placebo + SoC CRT. It also identifies the chemotherapy and radiation interventions included in the standard-of-care chemoradiation component. The ClinicalTrials.gov record does not provide additional arm-specific enrollment counts, treatment durations, or dosing schedules, so those details are not added here.
4. Endpoints
| Endpoint | Registry definition / time frame | Type |
|---|---|---|
| Progression-Free Survival (PFS) | The PFS per Response Evaluation Criteria in Solid Tumors, version 1.1 (RECIST 1.1) using blinded independent central review (BICR) assessments was defined as the time from the date of randomization until the date of objective PD or death (by any cause in the absence of progression) regardless of whether the participant withdrew from therapy or received another anti-cancer therapy prior to progression. The PD was defined as at least a 20% increase in the sum of diameters of target lesions. Median PFS was calculated using the Kaplan-Meier technique.. Tumour scans performed at screening, 16 weeks ±1 week after randomization, then every 8 weeks ±1 week up to 48 weeks, and then every 12 weeks ±1 week thereafter until confirmed PD. Assessed up to the DCO date (a maximum of approximately 1988 days). | Time-to-event |
| Objective Response Rate (ORR) | Tumour scans performed at screening, 16 weeks ±1 week after randomization, then every 8 weeks ±1 week up to 48 weeks, and then every 12 weeks ±1 week thereafter until confirmed PD. Assessed up to the DCO date (a maximum of approximately 1988 days). | Binary |
| Overall Survival (OS) | From screening until confirmed PD, assessed up to the DCO date (a maximum of approximately 1988 days). | Time-to-event |
| Percentage of Participants Alive at 24 Months From Randomization (OS24) | Month 24 | Time-to-event |
| Complete Response Rate (CRR) | Tumour scans performed at screening, 16 weeks ±1 week after randomization, then every 8 weeks ±1 week up to 48 weeks, and then every 12 weeks ±1 week thereafter until confirmed PD. Assessed up to the DCO date (a maximum of approximately 1988 days). | Binary |
| Time to Death or Distant Metastasis (TTDM) | Tumour scans performed at screening, 16 weeks ±1 week after randomization, then every 8 weeks ±1 week up to 48 weeks, and then every 12 weeks ±1 week thereafter until confirmed PD. Assessed up to the DCO date (a maximum of approximately 1988 days). | Time-to-event |
| Time From Randomization to Second Progression (PFS2) | Tumour scans performed at screening, 16 weeks ±1 week after randomization, then every 8 weeks ±1 week up to 48 weeks, and then every 12 weeks ±1 week thereafter until confirmed PD. Assessed up to the DCO date (a maximum of approximately 1988 days). | Time-to-event |
The registry identifies Progression-Free Survival (PFS) as the single registered primary endpoint. The remaining posted analyses reported in the ClinicalTrials.gov record are secondary endpoints.
5. Analysis Population and Stratification
The posted analyses state that the full analysis set (FAS) included all randomized participants. This is important because the primary efficacy comparison is therefore anchored to randomized treatment assignment rather than restricting the analysis to participants who completed treatment.
| Feature | Registry-supported description |
|---|---|
| Analysis population | FAS included all randomized participants. |
| Groups compared | Durvalumab + SoC CRT vs Placebo + SoC CRT |
| Stratification factors | Age (< 65 vs. >= 65 years) and stage (IIIA vs. IIIB/C) |
| Time-to-event model | Stratified Cox proportional hazards model |
| Time-to-event comparison | Stratified log-rank test |
| Ties | Efron approach |
The analysis notes repeatedly describe adjustment for age and stage. This means the reported time-to-event treatment comparison was not simply an unstratified comparison of all randomized participants; the analysis accounted for these prespecified categories when constructing the treatment-effect estimate and test.
6. Primary Endpoint Result: Progression-Free Survival
PFS was the trial's registered primary endpoint. It was defined using RECIST 1.1 with blinded independent central review and measured from randomization until objective disease progression or death, regardless of treatment withdrawal or receipt of another anti-cancer therapy before progression.
Hazard ratio for progression or death
95% CI: 0.647–1.123 · P = 0.247
Two-sided superiority analysis in the full analysis set.
| Primary endpoint | Analysis population | Method | Effect measure | Estimate | 95% CI | P-value |
|---|---|---|---|---|---|---|
| Progression-Free Survival (PFS) | FAS; all randomized participants | Stratified log-rank test; stratified Cox proportional hazards model | Hazard ratio | 0.85 | 0.647–1.123 | 0.247 |
The reported HR of 0.85 means that, under the stratified Cox proportional-hazards model, the estimated instantaneous rate of progression or death in the durvalumab + SoC CRT group was 0.85 times that in the placebo + SoC CRT group over the analyzed follow-up. Expressed as a simple relative interpretation, an HR of 0.85 corresponds to an estimated 15% lower hazard, not a 15% improvement in the probability of remaining progression-free.
The HR is a relative time-to-event measure. It does not mean that 15% more participants were progression-free, that an individual participant's risk was reduced by exactly 15%, or that the treatment delayed progression by a fixed amount of time for every participant.
The 95% confidence interval of 0.647–1.123 describes uncertainty around the estimated hazard ratio under the statistical model and sampling framework. Because the interval includes 1, the data are compatible with a range of relative effects that includes no difference in the instantaneous event rate. The interval is also important because the point estimate alone can give a false impression of precision.
The P-value of 0.247 is a measure used for assessing the data against the specified null hypothesis under the statistical testing framework. It does not measure the size of the treatment effect, the clinical importance of the effect, or the probability that either treatment is effective.
Interpretation also depends on the Cox model's proportional-hazards framework. A single HR summarizes the relative event-rate comparison through the fitted model; it should not automatically be read as a constant relative difference at every individual time point. The analysis was also stratified by age and stage, so the reported HR is specifically the treatment comparison from that stratified model.
What the primary PFS analysis tells us
Statistically, the registry reports a treatment-effect estimate below 1, but the estimate has substantial uncertainty as represented by the 95% confidence interval. The reported superiority test has a P-value of 0.247. These are three different pieces of information: the HR describes relative effect, the confidence interval describes precision, and the P-value addresses evidence against the null hypothesis.
The registry does not provide median PFS, numbers of PFS events by arm, or time-specific PFS percentages in the ClinicalTrials.gov record. Those quantities therefore cannot be substituted for the reported HR and confidence interval.
7. Secondary Endpoint Results
Six secondary statistical analyses are reported in the ClinicalTrials.gov record. They use two broad analysis families: stratified time-to-event methods for survival-type endpoints and stratified categorical-data methods for response endpoints.
Objective Response Rate
Difference in objective response rate
99.50% CI: -15.2 to 16.3 · P = 0.976
Cochran-Mantel-Haenszel analysis stratified by age and stage.
The reported effect measure is the difference in percentages, normalized here as a risk difference. The analysis used a Cochran-Mantel-Haenszel test stratified by age (< 65 vs. >= 65 years) and stage (IIIA vs. IIIB/C). Confidence intervals for the difference in percentages were estimated using Miettinen and Nurminen's method.
A risk difference of 0.2 represents an estimated difference in response percentages between the two randomized groups. It is an absolute rather than relative effect measure. The very wide 99.50% confidence interval, from -15.2 to 16.3, indicates substantial uncertainty around the estimated difference. The P-value of 0.976 is evidence from the specified CMH testing framework, not a measure of how clinically important the observed difference is.
Overall Survival
Hazard ratio for overall survival
95% CI: 0.778–1.386 · P = 0.823
Stratified log-rank analysis with stratified Cox proportional-hazards estimation.
Overall survival was analyzed in the FAS, defined as all randomized participants. The analysis used a stratified log-rank test, adjusting for age (< 65 vs. >= 65 years) and stage (IIIA vs. IIIB/C). The HR and confidence interval came from a stratified Cox proportional-hazards model with treatment as the only covariate and ties handled using the Efron approach.
An HR of 1.03 is close to 1, meaning the fitted model estimated similar instantaneous rates of death between the randomized groups. The 95% CI of 0.778–1.386 spans 1 and represents uncertainty around that estimate. The P-value of 0.823 addresses the specified superiority test and does not quantify the magnitude or clinical importance of any observed difference.
Percentage of Participants Alive at 24 Months From Randomization
OS24 hazard ratio
95% CI: 0.716–1.502 · P = 0.847
Month 24 analysis using the Kaplan-Meier estimator with a stratified comparison.
For OS24, the Kaplan-Meier estimator of survival at 24 months for each treatment was used to obtain the HR, with the test based on the method described in Klein 2007 according to the registry analysis note. The analysis accounted for the age and stage stratification factors.
The reported HR of 1.04 is a relative time-to-event estimate associated with the 24-month survival analysis. The 95% CI of 0.716–1.502 is relatively broad and includes 1, so the point estimate should not be interpreted without its uncertainty interval. The P-value of 0.847 is the reported superiority-test result; it does not represent the probability that the treatment has no effect.
Complete Response Rate
Difference in complete response rate
95% CI: -4.8 to 3.0
Risk difference with confidence intervals calculated using Miettinen and Nurminen's method.
The CRR analysis was performed using a Cochran-Mantel-Haenszel test stratified by age (< 65 vs. >= 65 years) and stage (IIIA vs. IIIB/C). The confidence interval for the difference in percentages was calculated using Miettinen and Nurminen's method.
The estimated difference in percentages was 0. The corresponding 95% CI, -4.8 to 3.0, describes uncertainty around that absolute difference. Because the interval spans zero, the data are compatible with both a modest negative difference and a modest positive difference within the reported interval. The registry-reported analysis does not report a P-value for this endpoint, so no additional formal test result is stated.
Time to Death or Distant Metastasis
Hazard ratio for TTDM
95% CI: 0.649–1.413 · P = 0.790
Stratified log-rank analysis with stratified Cox proportional-hazards estimation.
Time to death or distant metastasis was analyzed using the FAS. The reported analysis used the stratified log-rank test with adjustment for age and stage, while the HR and confidence interval were calculated using a stratified Cox proportional-hazards model with ties handled by the Efron approach.
An HR of 0.95 indicates an estimated instantaneous event rate that was 0.95 times that of the comparator under the fitted model. The 95% CI of 0.649–1.413 is wide enough to encompass both lower and higher event rates relative to the comparator. The P-value of 0.790 is the reported superiority-test result and should not be treated as an effect-size measure.
Time From Randomization to Second Progression
Hazard ratio for PFS2
95% CI: 0.614–1.071 · P = 0.132
Stratified log-rank analysis with stratified Cox proportional-hazards estimation.
PFS2 was analyzed from randomization using the same broad time-to-event framework. The FAS included all randomized participants, and the analysis adjusted for age and stage through the stratified log-rank test and stratified Cox model.
The HR of 0.81 corresponds to an estimated instantaneous event rate 0.81 times that of the comparator under the fitted model. The 95% CI of 0.614–1.071 includes 1, so the estimate should be interpreted together with its uncertainty rather than as a precise treatment effect. The P-value of 0.132 is the reported result of the superiority testing framework and does not itself measure the magnitude of the HR.
8. Secondary Results in One View
| Endpoint | Effect measure | Estimate | Confidence interval | P-value | Method |
|---|---|---|---|---|---|
| Progression-Free Survival (PFS) | Hazard ratio | 0.85 | 95% CI 0.647–1.123 | 0.247 | Stratified log-rank; stratified Cox |
| Objective Response Rate (ORR) | Risk difference | 0.2 | 99.50% CI -15.2 to 16.3 | 0.976 | Cochran-Mantel-Haenszel |
| Overall Survival (OS) | Hazard ratio | 1.03 | 95% CI 0.778–1.386 | 0.823 | Stratified log-rank; stratified Cox |
| OS24 | Hazard ratio | 1.04 | 95% CI 0.716–1.502 | 0.847 | Stratified log-rank |
| Complete Response Rate (CRR) | Risk difference | 0 | 95% CI -4.8 to 3.0 | Not reported | CMH; Miettinen and Nurminen CI |
| Time to Death or Distant Metastasis (TTDM) | Hazard ratio | 0.95 | 95% CI 0.649–1.413 | 0.790 | Stratified log-rank; stratified Cox |
| PFS2 | Hazard ratio | 0.81 | 95% CI 0.614–1.071 | 0.132 | Stratified log-rank; stratified Cox |
This table illustrates why treatment effects should not be summarized by a single P-value. The trial contains different endpoint types, and the appropriate effect measure changes with the endpoint. PFS, OS, TTDM, and PFS2 are time-to-event endpoints represented by hazard ratios, whereas ORR and CRR are categorical response endpoints represented by differences in percentages.
9. Statistical Methodology
Kaplan-Meier estimation
The registry analysis for OS24 explicitly uses the Kaplan-Meier estimator of survival at 24 months. Kaplan-Meier estimation is designed for time-to-event data in which some participants may not have experienced the event by the end of their observed follow-up. Rather than treating those participants as if they had experienced the event, their available follow-up contributes information until censoring.
where di is the number of events at time ti and ni is the number at risk immediately before that time.
The registry's posted analysis does not provide the underlying event and censoring times needed to reconstruct a Kaplan-Meier curve independently. Accordingly, this page reports the registry-reported summary estimates rather than drawing a reconstructed curve.
Stratified log-rank test
The log-rank test is a standard method for comparing time-to-event distributions between randomized groups. In PACIFIC-2, the posted analyses used a stratified log-rank test, adjusting for age (< 65 vs. >= 65 years) and stage (IIIA vs. IIIB/C).
Stratification means that the comparison is constructed within the defined strata and then combined rather than ignoring those categories altogether. This preserves the role of the trial's specified prognostic factors in the analysis.
Cox proportional-hazards model
The HR and confidence intervals for PFS, OS, TTDM, and PFS2 were calculated using a stratified Cox proportional-hazards model. The analysis notes state that age and stage were used for adjustment, with treatment as the only covariate, and that ties were handled using the Efron approach.
An HR is a model-based relative measure of event rates. It is not the same as a risk ratio, an absolute risk difference, a difference in median survival, or the percentage of participants who benefit.
Cochran-Mantel-Haenszel test
ORR was compared using a Cochran-Mantel-Haenszel (CMH) test stratified by age and stage. This is appropriate for a binary endpoint when the analysis is intended to account for categorical stratification factors rather than simply pooling all observations without adjustment.
The CMH framework can be understood as combining treatment comparisons across strata while accounting for the stratification structure. It therefore answers a somewhat different statistical question from an unadjusted comparison of the overall response percentages.
Score-based confidence interval for proportions
For ORR, the confidence interval for the difference in percentages was estimated using Miettinen and Nurminen's method. For CRR, the registry similarly states that the confidence interval for the difference in percentages was calculated using Miettinen and Nurminen's method.
These methods are designed for inference on differences between binomial proportions and avoid relying solely on a simple large-sample Wald interval. The choice of confidence-interval method matters particularly when response proportions are being compared and the desired interval needs to behave appropriately over the possible range of proportions.
Full analysis set
For all seven posted statistical analyses reported in the ClinicalTrials.gov record, the analysis population is the full analysis set including all randomized participants. This creates a consistent treatment-assignment framework across the reported efficacy comparisons.
10. Statistical Methods Explained
Why was a stratified log-rank test used for PFS?
PFS is a time-to-event endpoint, so the analysis needs to account for both when events occur and for participants whose observations are censored. The registry reports a stratified log-rank test, with adjustment for age and stage. This uses the time-to-event structure of the data while retaining the specified stratification factors.
What does an HR of 0.85 mean?
An HR of 0.85 means that the fitted model estimates the instantaneous rate of progression or death in the durvalumab + SoC CRT group at 0.85 times the corresponding rate in the placebo + SoC CRT group. A convenient descriptive translation is a 15% lower estimated hazard, but this should not be converted into a claim that 15% of participants avoided progression.
Why is the confidence interval important?
The point estimate is only one possible estimate of the underlying treatment effect. The 95% CI of 0.647–1.123 for PFS communicates how uncertain that estimate is within the model and sampling framework. Because the interval includes 1, the interval is compatible with both a lower and a higher hazard relative to the comparator.
Why does the P-value not measure effect size?
The P-value depends on the estimated effect, its precision, and the statistical model and testing framework. A P-value of 0.247 therefore cannot be interpreted as saying that the treatment has a 24.7% chance of working or as quantifying how large the treatment effect is. The HR and its confidence interval provide the direct effect-size information for PFS.
Why use a Cochran-Mantel-Haenszel test for ORR?
ORR is a binary endpoint rather than a time-to-event endpoint. The CMH test provides a way to compare treatment groups while accounting for the age and stage strata specified in the analysis. The corresponding effect measure is an absolute difference in percentages rather than a hazard ratio.
What does a risk difference of 0.2 mean?
A risk difference expresses an absolute difference between two response proportions. The registry reports an estimated difference of 0.2 for ORR, with a 99.50% CI from -15.2 to 16.3. The confidence interval is therefore essential: the point estimate alone does not establish a precise difference in response rates.
Why is Miettinen and Nurminen's method used for response-rate confidence intervals?
The registry specifies Miettinen and Nurminen's method for the confidence intervals around differences in percentages for ORR and CRR. This is a score-based approach to inference for differences between proportions. It provides an interval tailored to the binomial nature of the response endpoint rather than simply applying a basic normal approximation.
11. Understanding the Pattern Across Endpoints
The seven posted analyses illustrate an important statistical principle: different endpoints can point in different directions without creating a contradiction in the statistical analysis. PFS has an HR of 0.85, TTDM has an HR of 0.95, PFS2 has an HR of 0.81, while OS has an HR of 1.03 and OS24 has an HR of 1.04. These are separate estimates for different endpoint definitions.
| Endpoint family | What is being measured | Effect measure | Why interpretation differs |
|---|---|---|---|
| PFS | Time from randomization to objective progression or death | Hazard ratio | Combines progression and death into a time-to-event endpoint. |
| OS | Time-to-death outcome | Hazard ratio | Death is the event, so the endpoint captures survival rather than tumor progression alone. |
| OS24 | Survival at Month 24 | Hazard ratio in the posted analysis | A specific time-point survival analysis has a different interpretation from a general PFS comparison. |
| TTDM | Time to death or distant metastasis | Hazard ratio | Uses a composite event definition involving distant metastasis or death. |
| PFS2 | Time from randomization to second progression | Hazard ratio | Extends the disease-progression framework beyond the first progression. |
| ORR | Objective tumor response | Risk difference | Binary response status rather than time until an event. |
| CRR | Complete response | Risk difference | Binary response endpoint focused on complete response. |
This distinction prevents a common analytical mistake: treating every endpoint as if it were another version of the same measurement. PFS and OS both use hazard ratios, but their event definitions differ. ORR and CRR are response proportions and therefore use an absolute difference in percentages. A statistically careful interpretation starts with the endpoint definition before interpreting the numerical estimate.
12. Safety Results
The ClinicalTrials.gov record provides serious adverse events by randomized arm. The reported counts are shown below as affected participants over participants at risk.
| Safety measure | Durvalumab + SoC CRT | Placebo + SoC CRT |
|---|---|---|
| Serious adverse events | 103 / 219 | 56 / 108 |
Durvalumab + SoC CRT
103/219 participants were affected by serious adverse events among the reported at-risk population.
Placebo + SoC CRT
56/108 participants were affected by serious adverse events among the reported at-risk population.
These figures are presented as reported in the registry rather than converted into newly calculated percentages. The denominator shown for each arm is important because safety interpretation depends on the population at risk and the definition of the safety analysis set.
13. Trial Timeline
Trial start
The PACIFIC-2 study began on March 29, 2018, according to the ClinicalTrials.gov record.
Randomized treatment comparison
The study was designed as a randomized, double-blind, parallel phase 3 treatment trial with two arms.
Primary completion
The ClinicalTrials.gov record lists September 7, 2023 as the primary completion date.
Registry status
The trial is listed as completed. Results are posted, including seven statistical analyses.
14. Design Features That Shape Interpretation
Randomization
Randomized allocation provides the fundamental basis for comparing the treatment groups while reducing systematic differences in prognostic characteristics expected from the allocation process.
Double blinding
The trial is registered as double-blind. Blinding can reduce the influence of knowledge of treatment assignment on participant or investigator behavior and assessment.
Parallel design
The trial uses a parallel design, so the randomized treatment groups are followed as separate groups rather than having participants sequentially receive both randomized interventions.
Superiority hypothesis
The posted analyses are identified as superiority analyses. That is distinct from a non-inferiority framework, in which interpretation depends on a prespecified non-inferiority margin.
The ClinicalTrials.gov record does not report a non-inferiority margin, crossover procedure, factorial structure, interim-analysis plan, Bayesian analysis, or missing-data/imputation strategy. Those topics are therefore not treated as design features of PACIFIC-2 on this page.
15. Limitations
- Summary-level results: The ClinicalTrials.gov record provides effect estimates, confidence intervals, and P-values but do not provide the underlying participant-level event and censoring data.
- No median event times reported: Median PFS, median OS, median TTDM, and median PFS2 are not provided in the ClinicalTrials.gov record and therefore are not reported here.
- No time-specific PFS or OS percentages reported: Apart from the OS24 endpoint's statistical analysis, the ClinicalTrials.gov record does not provide arm-specific survival percentages that would support additional absolute-risk displays.
- Hazard-ratio interpretation: The Cox model produces a model-based relative event-rate estimate. Its interpretation depends on the proportional-hazards framework and should not be treated as a fixed individual-level risk reduction.
- Confidence-interval width: Several confidence intervals are broad, emphasizing that the point estimate alone is insufficient to characterize uncertainty.
- Multiple endpoints: The registry contains one primary endpoint and multiple secondary endpoints. Individual secondary P-values should not automatically be interpreted as independent confirmatory tests without knowing the prespecified multiplicity strategy.
- Endpoint heterogeneity: PFS, OS, TTDM, PFS2, ORR, and CRR represent different clinical events or response definitions. Their estimates should not be directly treated as interchangeable measures of treatment effect.
- Safety denominator: Serious adverse-event results are reported as affected participants over participants at risk. The ClinicalTrials.gov record does not provide additional safety-population definitions beyond those figures.
- Registry-level detail: The ClinicalTrials.gov record does not contain every element that might appear in a full statistical analysis plan, so conclusions about unreported design features should not be inferred.
16. Why This Trial Matters Statistically
PACIFIC-2 is a useful teaching example because the registry data bring together several core clinical-trial methods within one randomized phase 3 study. The primary endpoint is a time-to-event outcome, while secondary endpoints include both additional time-to-event outcomes and binary tumor-response measures.
| Concept | How it appears in PACIFIC-2 |
|---|---|
| Randomization | Randomized allocation in a two-arm parallel phase 3 trial. |
| Blinding | Double-blind design. |
| Time-to-event endpoint | PFS is the registered primary endpoint; OS, OS24, TTDM, and PFS2 are also analyzed. |
| Kaplan-Meier estimation | Used for the OS24 analysis through the Kaplan-Meier estimator of survival at 24 months. |
| Hazard ratio | Used for PFS, OS, OS24, TTDM, and PFS2. |
| Stratified analysis | Age and stage are used as stratification factors in the posted analyses. |
| Log-rank test | Used for the time-to-event treatment comparisons. |
| Cox proportional-hazards model | Used to calculate HRs and confidence intervals for the major time-to-event analyses. |
| Cochran-Mantel-Haenszel test | Used for the ORR analysis, stratified by age and stage. |
| Risk difference | Used for ORR and CRR as differences in percentages. |
| Score-based confidence interval | Miettinen and Nurminen's method is used for response-rate differences. |
| Full analysis set | All randomized participants are included in the posted efficacy analyses. |
17. A Practical Guide to Reading the PACIFIC-2 Results
A useful way to read the trial is to move from the endpoint definition to the effect measure, and only then to the confidence interval and P-value.
Step 1 · Identify the event
For PFS, the event is objective progression or death. For OS, the event is death. For TTDM and PFS2, the event definitions are different again.
Step 2 · Identify the effect measure
Time-to-event endpoints use hazard ratios in the posted analyses. ORR and CRR use differences in percentages.
Step 3 · Read the interval
The confidence interval communicates uncertainty. For HRs, 1 is the no-difference reference value. For risk differences, 0 is the no-difference reference value.
Step 4 · Read the P-value
The P-value belongs to the specified statistical test. It does not replace the effect estimate or confidence interval and should not be interpreted as an effect-size measure.
For the primary PFS result, this sequence gives: PFS → HR 0.85 → 95% CI 0.647–1.123 → P = 0.247. That is more informative than simply labeling the result according to the P-value.
18. Clinical Interpretation vs Statistical Interpretation
Statistical interpretation
The primary PFS analysis produced an HR of 0.85 with a 95% CI of 0.647–1.123 and P = 0.247 using a stratified log-rank test and stratified Cox model in the full analysis set.
Endpoint-specific interpretation
The secondary analyses show different estimates across PFS, OS, OS24, TTDM, PFS2, ORR, and CRR. These differences should be interpreted in the context of each endpoint's definition and statistical method rather than collapsed into one overall measure.
The distinction matters because statistical evidence is not one-dimensional. A hazard ratio provides relative information about event rates over time, a risk difference provides an absolute difference in proportions, a confidence interval describes uncertainty, and a P-value addresses the specified hypothesis test. Keeping those roles separate makes the analysis easier to audit and less prone to overinterpretation.
19. Statistical Methods Explained: Time-to-Event vs Binary Endpoints
| Question | Time-to-event approach | Binary response approach |
|---|---|---|
| What is observed? | Time until an event, with possible censoring. | Whether a predefined response occurred. |
| PACIFIC-2 examples | PFS, OS, OS24, TTDM, PFS2. | ORR, CRR. |
| Primary effect measure in posted analyses | Hazard ratio. | Difference in percentages / risk difference. |
| Comparison method | Stratified log-rank test. | Cochran-Mantel-Haenszel test for ORR. |
| Model-based estimate | Stratified Cox proportional-hazards model. | Not reported as a regression model in the ClinicalTrials.gov record. |
| Confidence interval method | 95% CI around the HR. | Miettinen and Nurminen method for the response-rate differences where specified. |
The table highlights why the statistical method should follow the endpoint. Treating a binary response endpoint as if it were a survival endpoint would discard the timing structure that is central to PFS or OS. Conversely, applying a hazard ratio to a simple response proportion would change the question being asked.
20. What the Primary Hazard Ratio Does — and Does Not — Mean
The PFS HR of 0.85 is a relative measure from a stratified Cox proportional-hazards model. It indicates an estimated instantaneous progression-or-death rate that is 0.85 times the comparator rate under the fitted model.
The HR does not mean that every participant had exactly a 15% lower probability of progression or death. It also does not say that 15% of participants benefited or that PFS increased by a fixed percentage of time.
The 95% CI of 0.647–1.123 is essential to interpreting the HR. The interval reflects uncertainty around the estimate and includes the no-difference value of 1.
The P-value of 0.247 is associated with the reported superiority test. It should be read alongside the HR and confidence interval rather than used as a substitute for either.
21. Why the Analysis Population Matters
The analyses posted on ClinicalTrials.gov consistently define the FAS as including all randomized participants. That is an important feature of the statistical comparison because randomization creates the basis for comparing treatment groups. Once participants are excluded based on post-randomization behavior, the resulting comparison can answer a different question.
For a time-to-event endpoint, inclusion in the FAS does not require that every participant experience an event. Participants without the event during their observed follow-up can contribute censored information. This is one reason survival-analysis methods are useful in clinical trials: they can use partial follow-up rather than treating participants who have not yet experienced an event as though the event had occurred.
The registry does not provide additional missing-data or imputation details in the ClinicalTrials.gov record. No specific imputation strategy is therefore attributed to the PACIFIC-2 analysis here.
22. Interpretation of the Secondary Time-to-Event Analyses
The secondary HRs can be viewed as a set of related but distinct time-to-event questions:
| Endpoint | HR | 95% CI | P-value | Statistical reading |
|---|---|---|---|---|
| PFS | 0.85 | 0.647–1.123 | 0.247 | Primary time-to-event analysis; CI includes 1. |
| OS | 1.03 | 0.778–1.386 | 0.823 | Estimated HR close to 1; CI includes 1. |
| OS24 | 1.04 | 0.716–1.502 | 0.847 | 24-month survival analysis; CI includes 1. |
| TTDM | 0.95 | 0.649–1.413 | 0.790 | Composite time-to-event endpoint; CI includes 1. |
| PFS2 | 0.81 | 0.614–1.071 | 0.132 | Second-progression endpoint; CI includes 1. |
Across these five time-to-event analyses, each confidence interval includes 1. The estimates nevertheless should not be treated as identical: each endpoint has a different event definition, and therefore each HR answers a different statistical question.
23. Interpretation of the Response Analyses
The response endpoints use absolute differences in percentages rather than hazard ratios. For ORR, the reported estimate is 0.2 with a 99.50% CI of -15.2 to 16.3. For CRR, the reported estimate is 0 with a 95% CI of -4.8 to 3.0.
The different confidence levels are part of the reported analyses and should be preserved. A 99.50% interval is wider than a 95% interval when constructed from the same underlying information because it is designed to contain the parameter with a higher confidence level under the specified framework.
24. Important Interpretation Issues
- Do not equate HR with risk ratio: A hazard ratio describes relative event rates in a time-to-event model, not the ratio of cumulative probabilities by a particular time.
- Do not interpret the P-value as effect size: P-values address statistical evidence under a specified hypothesis test; they do not quantify clinical magnitude.
- Do not ignore confidence intervals: The width and location of the interval are essential for judging how precisely an effect has been estimated.
- Do not mix endpoint definitions: PFS, OS, TTDM, and PFS2 have different event definitions, even though each uses a hazard ratio.
- Do not treat secondary analyses as automatically confirmatory: The ClinicalTrials.gov record identifies one primary endpoint and multiple secondary analyses but does not provide a multiplicity strategy.
- Do not infer unreported quantities: Median survival times, event counts, arm-specific response percentages, and additional subgroup estimates are not reported and are therefore not reconstructed here.
- Consider model assumptions: Cox proportional-hazards estimates are model-based, and the interpretation of a single HR depends on the proportional-hazards framework.
25. Related Tutorials
Learn more about the methods used in this trial:
26. Related Statistical Calculators
27. Sources
- ClinicalTrials.gov: PACIFIC-2, NCT03519971.
Continue through Clinical Biostats
Use the trial's endpoints and statistical methods as a starting point for deeper study of survival analysis, categorical-data methods, confidence intervals, and clinical-trial design.
28. Record Summary
PACIFIC-2 is a randomized, double-blind, parallel phase 3 trial in unresectable non-small cell lung cancer comparing durvalumab plus standard-of-care chemoradiation therapy with placebo plus standard-of-care chemoradiation therapy. The registered primary endpoint is PFS, defined using RECIST 1.1 with blinded independent central review as the time from randomization to objective progression or death. Its posted primary analysis used a stratified log-rank test and stratified Cox proportional-hazards model, adjusting for age and stage.
The reported primary PFS HR was 0.85, with a 95% CI of 0.647–1.123 and P = 0.247. Secondary analyses covered ORR, OS, OS24, CRR, TTDM, and PFS2, demonstrating the importance of matching the statistical interpretation to the endpoint type. Time-to-event endpoints were analyzed using stratified survival methods and hazard ratios, while response endpoints used risk differences and score-based confidence intervals.
The most important statistical lesson is that no single number tells the complete story. The endpoint definition establishes the question, the effect measure describes the estimated treatment difference, the confidence interval communicates precision, and the P-value describes evidence under the specified testing framework. Reading these components together provides a more rigorous interpretation of the PACIFIC-2 results than relying on any one statistic in isolation.