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
ARC-7 was a randomized, parallel-design, phase 2 treatment trial evaluating monotherapy and combination immunotherapies in participants with PD-L1 positive non-small cell lung cancer. The study enrolled 151 participants across 3 arms.
| Feature | ARC-7 |
|---|---|
| Phase | Phase 2 |
| Status | Completed |
| Population | Participants with PD-L1 positive non-small cell lung cancer |
| Conditions | Non Small Cell Lung Cancer; Nonsquamous Non Small Cell Lung Cancer; Squamous Non Small Cell Lung Cancer; Lung Cancer |
| Design | Randomized, parallel |
| Masking | None |
| Primary purpose | Treatment |
| Enrollment | 151 |
| Number of arms | 3 |
| Lead sponsor | Arcus Biosciences, Inc. |
| Sponsor type | Industry |
| ClinicalTrials.gov | NCT04262856 |
2. Clinical Question
The ARC-7 study addresses the clinical-development question of how monotherapy and combination immunotherapies perform in participants with PD-L1 positive non-small cell lung cancer. The registry identifies objective response rate and progression-free survival as the two primary endpoints.
Population
Participants with PD-L1 positive non-small cell lung cancer, including the registry's listed nonsquamous, squamous, and broader lung-cancer condition categories.
Interventions
Domvanalimab, etrumadenant, and zimberelimab are identified as study interventions.
Comparator structure
The study has 3 randomized arms in a parallel design. The ClinicalTrials.gov record does not specify the arm-level treatment combinations or a separate comparator description.
Primary questions
How should objective response rate and progression-free survival be assessed following randomization in this randomized phase 2 comparison?
3. Trial Design
Intervention framework
Domvanalimab
- Listed in the registry as a drug intervention.
Etrumadenant
- Listed in the registry as a drug intervention.
Zimberelimab
- Listed in the registry as a drug intervention.
4. Trial Timing and Follow-up Structure
Study start
The registry lists May 28, 2020 as the study start date.
Primary endpoint follow-up window
Both primary endpoints are defined from randomization through the first documentation of disease progression or death from any cause, whichever occurs first, with the registry specifying a follow-up period of up to approximately 3–5 years.
Primary completion
The registry lists July 9, 2025 as the primary completion date.
5. Endpoints
| Primary endpoint | Registry definition | Time frame |
|---|---|---|
| Objective response rate (ORR) | ORR as assessed by RECIST v1.1 | From randomization until the first documentation of disease progression or death from any cause, whichever occurs first (up to approximately 3-5 years) |
| Progression-free survival (PFS) | PFS as assessed by RECIST v1.1 | From randomization until the first documentation of disease progression or death from any cause, whichever occurs first (up to approximately 3-5 years) |
ORR: a response endpoint
Objective response rate summarizes the proportion of participants meeting the trial's objective-response definition under RECIST v1.1. Statistically, it is fundamentally a binary endpoint: each participant is classified according to whether an objective response occurred within the prespecified assessment framework.
PFS: a time-to-event endpoint
Progression-free survival differs because it incorporates when disease progression or death occurs. A participant who has not experienced progression or death at the end of observed follow-up does not simply become a non-responder; that participant contributes censored time to a time-to-event analysis.
6. Statistical Methodology
ORR as a binary endpoint
For a randomized multi-arm study, ORR can be summarized within each treatment arm as the number and proportion of participants achieving an objective response. A confidence interval for a response proportion quantifies statistical uncertainty around that observed proportion.
The numerator is the number of participants meeting the objective-response definition. The denominator must follow the prespecified response-analysis population; it should not automatically be assumed to be the full randomized population.
PFS as a time-to-event endpoint
PFS is conventionally analyzed using survival-analysis methods because participants can have different follow-up times and some remain progression-free at the time of analysis. Kaplan-Meier estimation provides a nonparametric estimate of the PFS distribution, while a Cox proportional-hazards model can provide a relative hazard estimate when the model is appropriate.
Here, di represents events at time ti, while ni is the number at risk immediately before that time.
RECIST v1.1 and endpoint classification
Both primary endpoints are explicitly tied to RECIST v1.1 in the registry. This is important because the definition of progression and the classification of objective response determine the event and response status that enter the statistical analysis.
Randomization and comparative inference
Randomization is the foundation for comparing outcomes across the study arms. In a three-arm randomized design, each prespecified comparison should be interpreted according to the statistical analysis plan governing the trial. The existence of three arms also creates a potential multiplicity issue when more than one pairwise comparison is formally tested.
7. Statistical Methods Explained
Why is ORR treated differently from PFS?
ORR reduces each participant's response assessment to a binary outcome, whereas PFS records a time until progression or death and must account for censoring. Consequently, ORR can be summarized as a proportion, while PFS is naturally represented as a time-to-event distribution.
Why does randomization matter for the analysis?
Randomization creates treatment groups whose differences can be interpreted under the causal framework of the randomized comparison, subject to the trial's adherence, follow-up, endpoint definitions, and analysis assumptions. The statistical analysis should preserve randomized assignment for the primary efficacy comparison when that is the prespecified analysis principle.
Why is RECIST v1.1 important?
RECIST v1.1 supplies the framework used to assess tumor response and progression. Because ORR and PFS depend on those classifications, the endpoint definition and assessment schedule are part of the statistical measurement process rather than merely clinical details.
Why is Kaplan-Meier useful for PFS?
Not every participant will experience progression or death during the same observation period. Kaplan-Meier estimation incorporates participants' available event-free follow-up and appropriately handles right censoring under the analysis assumptions.
What would a hazard ratio mean for PFS?
If a Cox model were used to compare PFS between two randomized groups, a hazard ratio below 1 would represent a lower estimated instantaneous rate of progression or death in one group relative to the other over the modeled follow-up. It would not mean that the same percentage of participants avoided progression or that every participant experienced the same proportional reduction.
Why does a three-arm trial raise multiplicity questions?
With three arms, multiple comparisons may be scientifically interesting. If several hypotheses are tested at the same nominal significance level, the probability of at least one false-positive finding can increase. A prespecified multiplicity strategy is therefore important when more than one comparison is intended to support confirmatory inference.
8. Planned Analysis
Objective response rate
The registry says that ORR will be assessed by RECIST v1.1 from randomization until the first documentation of disease progression or death from any cause, whichever occurs first, up to approximately 3-5 years. For this type of endpoint, the statistical analysis would ordinarily summarize the response proportion for each randomized group together with an appropriate confidence interval and prespecified between-group comparisons where applicable.
The central statistical issue is the definition of the response-analysis population. The analysis should follow the prespecified protocol and statistical analysis plan regarding whether the denominator is all randomized participants, participants with adequate tumor assessment, or another defined population. Those choices can materially affect the response-rate estimate.
Progression-free survival
The registry says that PFS will be assessed by RECIST v1.1 from randomization until the first documentation of disease progression or death from any cause, whichever occurs first, up to approximately 3-5 years. For this endpoint, a conventional analysis would use Kaplan-Meier methods to estimate the PFS distribution and could use a Cox model and a log-rank-type comparison for treatment-group inference when specified by the statistical analysis plan.
The analysis must also define censoring rules. Participants who have not experienced progression or death at the relevant analysis time contribute follow-up information up to their censoring time. The validity of the resulting estimate depends on the prespecified assessment and censoring rules and on assumptions about the censoring mechanism.
Comparisons among three randomized arms
Because ARC-7 contains three randomized arms, the statistical analysis needs a clearly defined comparison structure. If all pairwise comparisons were tested formally, there would be three possible pairwise treatment comparisons. If only selected comparisons were primary, those hypotheses would need to be specified in advance. The multiplicity procedure should correspond to the intended inferential family.
9. Analysis of Objective Response Rate
What is estimated?
The proportion of participants meeting the RECIST v1.1 objective-response definition within the prespecified analysis population.
What is compared?
In a randomized three-arm study, response rates can be compared across prespecified treatment groups according to the statistical analysis plan.
What determines precision?
The number of participants and the number of observed responses determine the amount of statistical information available for the response-rate estimate.
What can bias interpretation?
Differences in evaluability, assessment timing, missing tumor assessments, and the prespecified response-analysis population can affect the observed response proportion.
Confidence intervals for ORR
A response proportion is an estimate rather than an exact population parameter. Its confidence interval expresses sampling uncertainty around the observed proportion. A wider interval indicates less precision, while a narrower interval indicates more statistical information under the chosen confidence-interval method.
A confidence interval for ORR should not be interpreted as the range in which individual participants' responses fall. It describes uncertainty around the estimated population response rate under the statistical framework used to construct the interval.
10. Analysis of Progression-Free Survival
Kaplan-Meier estimation
For PFS, Kaplan-Meier estimation allows the analysis to incorporate the timing of disease progression or death while retaining information from participants who remain event-free when their follow-up ends. The resulting survival function estimates the probability of remaining progression-free beyond a given time.
Cox proportional-hazards modeling
A Cox model is a standard method for estimating a relative treatment effect for a time-to-event endpoint. If used for ARC-7, the resulting hazard ratio would summarize the relative event hazard between the specified treatment groups under the model.
An HR below 1 would indicate a lower estimated instantaneous event rate in the numerator group. It is not equivalent to a difference in median PFS, a difference in PFS probability at a particular time, or a percentage of participants who benefit.
The proportional-hazards assumption
The usual Cox interpretation is most straightforward when the relative hazards are reasonably represented by a common hazard ratio over time. If hazards vary substantially over time, a single hazard ratio may provide an incomplete description of the treatment-effect pattern. Kaplan-Meier curves and time-specific PFS estimates can provide useful complementary information.
11. Interpreting Censoring in PFS
Censoring is unavoidable in many clinical-trial time-to-event analyses. A participant may complete the available observation period without experiencing progression or death, for example, and contribute information until the last relevant assessment.
Event
The registry defines the PFS event as the first documentation of disease progression or death from any cause, whichever occurs first.
Time origin
The registry defines PFS from randomization.
Assessment framework
The registry specifies RECIST v1.1 for PFS assessment.
Follow-up window
The registry specifies follow-up up to approximately 3-5 years.
The statistical interpretation of a censored participant is not that the participant had no progression. Rather, the available information establishes that the participant remained under observation without a qualifying event through the censoring point, subject to the prespecified censoring rules.
12. Multiplicity in a Three-Arm Trial
ARC-7 has three randomized arms and two primary endpoints. This creates two distinct layers of multiplicity that a statistical analysis plan may need to address: multiple endpoints and multiple treatment comparisons.
| Source of multiplicity | Statistical issue |
|---|---|
| Two primary endpoints | ORR and PFS represent separate primary efficacy questions. |
| Three randomized arms | More than one treatment comparison may be possible. |
| Multiple pairwise comparisons | Testing several comparisons at the same nominal level can increase familywise type I error. |
| Endpoint-by-comparison combinations | If multiple comparisons are formally tested for both endpoints, the inferential family can become larger still. |
The existence of multiple endpoints and arms does not by itself establish which hypotheses were confirmatory. That determination belongs to the prespecified statistical analysis plan. The appropriate interpretation of any future numerical result therefore depends on knowing which comparisons were assigned formal inferential status and how multiplicity was controlled.
13. Randomization Without Masking
ARC-7 is listed as randomized but not masked. These are separate design characteristics. Randomization concerns how participants are allocated to treatment groups; masking concerns whether participants, investigators, assessors, or other parties are unaware of treatment assignment.
Randomization can provide a strong basis for comparative inference even in an unmasked study. However, lack of masking can create opportunities for differential behavior or assessment. For ARC-7, the registry specifically states that both primary endpoints are assessed by RECIST v1.1, making the consistency of tumor assessment an important part of the endpoint's validity.
The statistical analysis should therefore be interpreted together with the operational procedures used for outcome assessment. Randomization addresses allocation bias; it does not automatically eliminate all other sources of bias.
14. Analysis Populations and the Meaning of the Denominator
A recurring issue in clinical-trial statistics is that the denominator used for an endpoint may differ from the total number enrolled. ARC-7 enrolled 151 participants, but the appropriate denominator for a specific efficacy analysis depends on the prespecified analysis population.
| Population concept | Statistical relevance |
|---|---|
| Enrolled population | Defines the total number of participants entered into the study. |
| Randomized population | Provides the principal framework for randomized treatment comparisons when an intention-to-treat analysis is specified. |
| Response-evaluable population | May be relevant for ORR if the protocol defines a subset with adequate assessment. |
| Time-to-event population | Requires event and censoring information according to the prespecified PFS rules. |
For ORR in particular, changing the denominator can change the estimated response rate even when the number of observed responders is unchanged. This is why the analysis population must be defined before interpreting a response percentage.
15. What the Primary Endpoints Can and Cannot Establish
ORR addresses whether participants achieved an objective tumor response according to RECIST v1.1. It does not directly measure how long a response lasts, how long participants remain progression-free, or overall survival.
PFS addresses the time from randomization until disease progression or death from any cause, whichever occurs first. It does not directly measure overall survival or establish that every participant experiences the same treatment effect.
A future hazard ratio or response-rate difference would summarize an average comparative effect under its corresponding statistical model or estimand. It would not describe the effect experienced by every individual participant.
16. Limitations
- No posted statistical analyses: the ClinicalTrials.gov record contains the registered primary endpoints but no posted statistical analyses, so numerical efficacy estimates and formal inferential results are not available from the record.
- Three-arm comparison structure: the registry information presented here does not specify the arm-level treatment combinations or identify a single comparator arm, limiting interpretation of pairwise treatment contrasts.
- Multiplicity: two primary endpoints and three randomized arms create potential multiplicity considerations. The appropriate adjustment depends on the prespecified hypothesis structure.
- Unmasked design: the trial is listed as having no masking. This can matter particularly for assessments that involve clinical judgment, although the registry specifies RECIST v1.1 for the primary endpoints.
- Time-to-event assumptions: PFS analysis requires appropriate handling of censoring, and a Cox model requires consideration of the proportional-hazards assumption if that model is used.
- Analysis population: ORR interpretation depends on the prespecified denominator and response-evaluation rules. Enrollment of 151 participants does not necessarily mean that 151 participants form the denominator for every efficacy analysis.
- Endpoint distinction: ORR and PFS measure different aspects of treatment response. A change in one does not automatically establish a corresponding change in the other.
- Registry scope: the registry identifies the endpoint definitions and study design but does not provide the statistical-analysis outputs needed for quantitative assessment of treatment effects.
17. Why This Trial Matters Statistically
ARC-7 is a useful statistical teaching case because it combines randomized allocation, three parallel arms, an unmasked design, a binary tumor-response endpoint, and a time-to-event endpoint within the same phase 2 study.
| Concept | How it appears in ARC-7 |
|---|---|
| Randomization | The study uses randomized allocation to 3 arms. |
| Parallel design | The registry identifies a parallel design model. |
| Multiple treatment arms | Three randomized arms create a defined comparison structure that must be reflected in statistical inference. |
| Binary efficacy endpoint | ORR provides a proportion-based measure of objective tumor response. |
| Time-to-event endpoint | PFS incorporates both event timing and censoring. |
| RECIST v1.1 | Both primary endpoints are assessed using RECIST v1.1. |
| Kaplan-Meier analysis | A standard framework for estimating the PFS distribution. |
| Cox modeling | A standard model-based approach for estimating relative PFS hazards when appropriate. |
| Multiplicity | Two primary endpoints and three arms can require a prespecified strategy for controlling type I error. |
| Unmasked design | The registry specifies no masking, which is relevant to interpretation of outcome assessment. |
18. Clinical Interpretation vs Statistical Interpretation
Statistical interpretation
The study is structured to compare randomized groups using two primary efficacy endpoints with different statistical characteristics: ORR is a response proportion, while PFS is a censored time-to-event outcome.
Clinical interpretation
The endpoints address complementary questions about tumor response and time without disease progression or death. Their interpretation should remain tied to the registry definitions and the prespecified analysis framework.
19. A Statistical Reading of the Trial Design
The most important feature of ARC-7 from an analysis perspective is that its endpoints require different forms of statistical reasoning. ORR asks whether an objective response occurred; PFS asks when progression or death occurred. Treating both endpoints as if they were simple percentages would discard important information in the PFS endpoint.
The randomized three-arm structure adds another layer. A statistical analysis should distinguish the overall question of whether the treatment groups differ from the specific pairwise comparisons used to characterize those differences. If multiple pairwise tests are conducted, multiplicity becomes part of the interpretation rather than a technical afterthought.
The lack of masking also matters, but not in the same way as randomization. Randomization establishes the allocation mechanism, whereas masking influences the potential for knowledge of treatment assignment to affect trial conduct or assessment. The use of RECIST v1.1 provides a standardized framework for response and progression classification, but the quality of endpoint assessment still depends on how the assessment process is implemented.
20. What a Complete Statistical Report Would Contain
For ORR, a complete report would ordinarily identify the number of responders, the analysis denominator, the resulting response proportion, an appropriate confidence interval, and the prespecified comparison among treatment groups. If several treatment comparisons were tested, the report would also identify the multiplicity strategy.
For PFS, a complete report would ordinarily include the number of PFS events, censoring information, Kaplan-Meier estimates, relevant time-specific probabilities or median PFS when estimable, and a prespecified comparative analysis such as a log-rank test and/or Cox model. The report would also identify the analysis population and censoring rules.
Those details are necessary because a single summary number cannot fully describe a censored time-to-event endpoint or a multi-arm randomized comparison. The design determines the structure of the analysis; the observed events and response classifications determine the resulting estimates.
21. Sources
- ClinicalTrials.gov: NCT04262856 — ARC-7.
Continue through the Clinical Biostats knowledge graph
Explore the broader Clinical Biostats collection of clinical-trial results, statistical tutorials, and analytical methods.
22. Record Summary
ARC-7 is a completed randomized phase 2 study with 151 participants and 3 parallel arms evaluating monotherapy and combination immunotherapies in participants with PD-L1 positive non-small cell lung cancer. Its two registered primary endpoints are objective response rate and progression-free survival, both assessed by RECIST v1.1 from randomization until the first documentation of disease progression or death from any cause, whichever occurs first, with a time frame of up to approximately 3-5 years.
Statistically, the trial illustrates the distinction between a binary response endpoint and a censored time-to-event endpoint. ORR is naturally expressed as a proportion with an associated measure of uncertainty, whereas PFS requires methods that account for event timing and censoring. The three-arm randomized design also makes the prespecified comparison structure and multiplicity strategy important to interpretation.