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
AtTEnd is a completed randomized, double-blind, parallel phase 3 trial in endometrial cancer. The study enrolled 549 participants and compares an atezolizumab-containing treatment strategy with a placebo-containing control strategy alongside paclitaxel and carboplatin.
| Feature | AtTEnd |
|---|---|
| Trial acronym | AtTEnd |
| ClinicalTrials.gov identifier | NCT03603184 |
| Phase | Phase 3 |
| Condition | Endometrial Cancer |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | Double |
| Primary purpose | Treatment |
| Enrollment | 549 |
| Number of arms | 2 |
| Status | Completed |
| Start date | October 2, 2018 |
| Primary completion date | January 20, 2025 |
| Lead sponsor | Mario Negri Institute for Pharmacological Research |
2. Clinical Question
The central statistical question is whether the randomized treatment strategy incorporating atezolizumab changes progression-free survival and overall survival compared with the corresponding placebo-containing strategy in patients with endometrial cancer.
Population
Patients enrolled in the phase 3 AtTEnd trial with endometrial cancer.
Intervention
Atezolizumab in combination with paclitaxel and carboplatin.
Comparator
Placebo in combination with paclitaxel and carboplatin.
Primary question
Does the atezolizumab-containing randomized strategy affect PFS and OS, including PFS in the MSI population?
3. Trial Design
Several features of the design are directly relevant to interpretation. Randomization creates the framework for comparing the treatment strategies, while double masking is intended to reduce the influence of knowledge of treatment assignment on trial conduct and assessment. The parallel structure means that participants remain in their randomized treatment groups rather than moving through a crossover sequence as part of the primary design.
Atezolizumab strategy
- Atezolizumab
- Paclitaxel
- Carboplatin
Placebo strategy
- Placebos
- Paclitaxel
- Carboplatin
Why randomization and masking matter
Randomization is the principal design mechanism that supports a causal comparison between the two treatment strategies. In expectation, random allocation balances both measured and unmeasured prognostic factors between groups. Double masking provides an additional safeguard against differential behavior, assessment, or reporting associated with knowledge of treatment assignment.
The registry identifies the study as randomized, parallel, double-blind, and treatment-purpose. Those design labels describe how the comparison was constructed; they do not by themselves establish whether an endpoint ultimately favored either treatment group.
4. Endpoints
| Primary endpoint | Registry definition | Time frame |
|---|---|---|
| PFS in the MSI | PFS is defined as the time from randomization to the date of first progression or death. | Up to 18 months after the last patient enrolled |
| PFS | PFS is defined as the time from randomization to the date of first progression or death from any cause, whichever comes first. Progression will be established as the radiological disease progression according to RECIST 1.1 or death from any cause, whichever occurs first. | Up to 18 months after the last patient enrolled |
| OS | OS is defined as the time from randomization until the date of death from any cause. | Up to two years after the last patient enrolled |
Progression-free survival
PFS is a time-to-event endpoint. The clock begins at randomization, and follow-up ends at the first qualifying progression or death. For the general PFS endpoint, the registry specifies radiological progression according to RECIST 1.1 or death from any cause, whichever occurs first.
This definition has an important statistical consequence: participants who have not experienced progression or death by the end of their observed follow-up are generally treated as right-censored observations in a time-to-event analysis. They are not treated as though the event occurred at the end of follow-up.
PFS in the MSI
The registry separately identifies PFS in the MSI as a primary endpoint. Its stated definition is time from randomization to the date of first progression or death, with a time frame of up to 18 months after the last patient enrolled.
Overall survival
OS is defined more simply: time from randomization until death from any cause. Unlike PFS, the endpoint does not depend on determining radiological disease progression. A patient remains event-free for OS until death or the point at which the observation is censored.
5. Statistical Methodology
The registry establishes three primary time-to-event endpoints but does not post statistical analyses for those endpoints. For endpoints defined as time from randomization to progression or death, the conventional statistical framework is survival analysis.
| Endpoint type | Typical statistical representation | What the analysis describes |
|---|---|---|
| PFS | Kaplan-Meier estimation with a treatment-group comparison and a time-to-event effect measure such as a hazard ratio | Time until progression or death |
| PFS in the MSI | Time-to-event analysis restricted to the defined MSI population | Time until first progression or death within the MSI population |
| OS | Kaplan-Meier estimation with a treatment-group comparison and a time-to-event effect measure such as a hazard ratio | Time until death from any cause |
The ClinicalTrials.gov record does not report posted estimates, confidence intervals, or p-values for these primary endpoints. Consequently, there is no reported numerical treatment effect in the registry from which to construct a results comparison.
Kaplan-Meier estimation
For PFS and OS, Kaplan-Meier estimation is the standard way to describe the distribution of event times while accounting for right censoring. At each observed event time, the estimated survival probability is updated according to the number of events and the number of participants still at risk.
Here, di is the number of events at time ti, and ni is the number at risk immediately before that time.
Hazard ratios
A Cox proportional-hazards model is commonly used to summarize a relative treatment effect for time-to-event outcomes. A hazard ratio below 1 would indicate a lower estimated instantaneous event rate in the intervention group under the model; a hazard ratio above 1 would indicate a higher estimated instantaneous event rate.
The hazard ratio is a relative time-to-event measure. It is not an absolute risk difference, a probability of cure, or the proportion of patients who benefit.
Log-rank comparison
A log-rank test is commonly used to compare survival distributions between randomized treatment groups. The test evaluates whether the observed pattern of event times is compatible with a common survival experience across groups. It is a hypothesis test rather than a measure of the magnitude of treatment effect.
6. Planned Analysis
The registry's primary endpoints are all time-to-event outcomes. A conventional analysis would therefore estimate the event-time distribution for each randomized group, account for censoring, and compare the groups using an appropriate survival-analysis framework.
PFS analysis
PFS would ordinarily be summarized using Kaplan-Meier methods, with progression according to RECIST 1.1 or death from any cause defining the event.
MSI PFS analysis
PFS in the MSI population would ordinarily use the same time-to-event principles within the specified MSI population.
OS analysis
OS would ordinarily use Kaplan-Meier estimation because the endpoint is time from randomization to death from any cause.
Effect estimation
A time-to-event model such as a Cox proportional-hazards model can provide a relative treatment-effect estimate together with a confidence interval.
The registry specifies the endpoint definitions and follow-up windows but does not post a statistical-analysis result for any of the three primary endpoints. The reported status is therefore a registry-level description of what is being measured rather than evidence of a particular treatment effect.
7. Statistical Methods Explained
Why is PFS a time-to-event endpoint?
PFS does not simply classify a participant as "progressed" or "not progressed." The registry defines it by the time from randomization until the first progression or death. Two participants who both experience progression can therefore contribute different amounts of event-free follow-up. Survival methods preserve this timing information.
Why can death be included in PFS?
The registry defines PFS as progression or death, whichever occurs first. Death therefore counts as a PFS event even if radiological progression has not previously been documented. This avoids treating a participant who dies without a documented progression as though the participant remained progression-free indefinitely.
Why is OS statistically different from PFS?
OS has death from any cause as its sole event. PFS has either progression or death as the first event. Consequently, PFS can occur earlier than OS and incorporates information about disease progression that is not part of the OS definition.
What does a hazard ratio measure?
A hazard ratio compares the estimated instantaneous event rates between treatment groups within a time-to-event model. It is a relative measure rather than an absolute probability. For example, an HR below 1 would describe a lower modeled event rate in the intervention group, but it would not mean that the same percentage of participants avoid the event.
Why is the confidence interval important?
A confidence interval communicates statistical precision around an estimated effect. A narrow interval indicates greater precision than a wide interval, all else equal. The interval is not a range containing the treatment effect for individual patients; it quantifies uncertainty around the estimated population-level effect under the specified statistical framework.
What does a p-value tell us?
A p-value addresses compatibility of the observed data with a specified null hypothesis under the statistical model. It does not measure effect size, clinical importance, or the probability that the treatment hypothesis is true. A complete interpretation requires the effect estimate, confidence interval, endpoint definition, and analysis population alongside the p-value.
Why does censoring matter?
Some participants may reach the end of observable follow-up without experiencing the endpoint. Survival analysis can retain the information available from those participants up to their censoring time rather than discarding them entirely. The validity of the analysis depends on appropriate endpoint and censoring rules and on the assumptions underlying the chosen statistical methods.
8. PFS: Statistical Interpretation
The registry defines PFS as time from randomization to first progression or death from any cause, whichever comes first. Progression is established as radiological disease progression according to RECIST 1.1 or death from any cause.
If a hazard ratio were reported, it would summarize the relative rate of experiencing progression or death between the randomized groups under the fitted time-to-event model. It would not directly state how many additional months an individual patient would experience without progression.
A 95% confidence interval around a PFS hazard ratio would describe uncertainty in the estimated relative treatment effect. Its width would provide information about precision, while its position relative to the null value of 1 would inform the corresponding statistical test under a conventional two-sided framework.
A p-value would address evidence against a specified null hypothesis. It would not indicate whether a treatment effect is large or small. The magnitude of an observed effect must be read from the treatment-effect estimate and its confidence interval.
9. PFS in the MSI Population
The registry identifies PFS in the MSI as a separate primary endpoint. Its definition is time from randomization to the date of first progression or death, with the assessment extending up to 18 months after the last patient enrolled.
| Feature | Registry specification |
|---|---|
| Endpoint | PFS in the MSI |
| Starting point | Randomization |
| Event | First progression or death |
| Time frame | Up to 18 months after the last patient enrolled |
Because the registry identifies this endpoint separately from overall PFS, the analysis population is an important part of interpretation. A treatment-effect estimate for the MSI population would describe that population specifically and should not automatically be interpreted as though it represented every participant in the randomized trial.
10. Overall Survival: Statistical Interpretation
The registry defines OS as the time from randomization until the date of death from any cause. This makes OS an unambiguous time-to-event endpoint based on death rather than radiological progression.
A hazard ratio for OS would compare the modeled instantaneous rate of death between the randomized groups. An HR below 1 would correspond to a lower estimated hazard of death in the intervention group, while an HR above 1 would correspond to a higher estimated hazard under the model.
An OS hazard ratio would not mean that every patient experiences the same proportional change in mortality risk. It would also not directly provide an absolute survival probability at a particular time point unless that probability were separately estimated and reported.
If a Cox proportional-hazards model were used, interpretation of a single hazard ratio would rely on the proportional-hazards assumption. When hazards change substantially over time, a single HR can provide a less complete description of the treatment-effect pattern than time-specific survival estimates or other approaches.
11. Multiplicity and Multiple Primary Endpoints
AtTEnd has three registered primary endpoints: PFS in the MSI, PFS, and OS. Multiple primary endpoints create an important statistical-design issue because testing several hypotheses can increase the overall chance of a false-positive finding if each is evaluated independently at the same nominal significance level.
| Primary endpoint | Statistical role | Key issue |
|---|---|---|
| PFS in the MSI | Primary | Population-specific time-to-event endpoint |
| PFS | Primary | Time-to-event endpoint based on progression or death |
| OS | Primary | Time-to-event endpoint based on death |
The ClinicalTrials.gov record identifies these endpoints as primary but does not post statistical analyses describing an endpoint hierarchy, alpha allocation, or multiplicity-adjustment procedure. Those design details should not be inferred from the existence of multiple primary endpoints alone.
Why multiplicity matters
Suppose several null hypotheses are tested. Even when every individual test uses the same nominal type I error rate, the probability of obtaining at least one apparently positive result can increase as the number of tests grows. Prespecified multiplicity control can address this through methods such as hierarchical testing, alpha allocation, or other familywise-error procedures.
For AtTEnd, the existence of three primary endpoints means that interpretation of any future posted p-values should account for the statistical framework specified for those endpoints rather than treating each p-value as an isolated discovery.
12. Analysis Populations and Causal Interpretation
The registry identifies the study as randomized, but the endpoint information reported by the record does not specify detailed analysis-population definitions such as an intention-to-treat set, per-protocol set, or safety population.
Randomized comparison
The principal causal framework comes from comparing participants according to their randomized treatment assignment.
Time origin
Both PFS and OS use randomization as the starting point, which gives the treatment comparison a common time origin.
Endpoint-specific rules
PFS depends on progression or death, while OS depends only on death from any cause.
Interpretation
The analysis population and censoring rules should accompany any eventual numerical efficacy estimate.
For a randomized trial, an intention-to-treat analysis generally preserves the treatment contrast created by randomization. A per-protocol analysis can answer a different question and may be more susceptible to post-randomization selection because treatment adherence and protocol deviations can be associated with prognosis.
13. Censoring and Follow-Up
The registry specifies different follow-up windows for the primary endpoints: PFS in the MSI and overall PFS are assessed up to 18 months after the last patient enrolled, while OS is assessed up to two years after the last patient enrolled.
| Endpoint | Registry time frame | Statistical implication |
|---|---|---|
| PFS in the MSI | Up to 18 months after the last patient enrolled | Participants without progression or death at the end of observation may contribute censored follow-up. |
| PFS | Up to 18 months after the last patient enrolled | Follow-up is bounded by the registered assessment period. |
| OS | Up to two years after the last patient enrolled | Participants alive at the end of observation may contribute censored survival time. |
Different follow-up windows can affect how mature an endpoint is when analyzed. PFS and OS also accumulate events at different rates because their event definitions differ. A mature statistical comparison therefore depends not only on sample size but also on the number and timing of observed events and the amount of censoring.
14. RECIST 1.1 and the PFS Event Definition
The registry explicitly states that progression for the general PFS endpoint will be established as radiological disease progression according to RECIST 1.1 or death from any cause, whichever occurs first.
The statistical endpoint therefore combines radiological progression and death into a single time-to-event outcome.
This definition illustrates why endpoint adjudication matters. A progression event depends on the rules used to determine radiological disease progression, while death is an objectively different event type. The registry's definition establishes how the two event types are combined for PFS.
15. Trial Timeline
Trial begins
The registered start date for the AtTEnd phase 3 trial.
Randomized phase 3 study
The study enrolled 549 participants across two randomized parallel treatment arms.
Up to 18 months after last enrollment
The registry specifies this follow-up window for both PFS in the MSI and overall PFS.
Up to two years after last enrollment
The registry specifies this follow-up window for overall survival.
Primary completion
The registered primary completion date for the study.
16. Limitations and Interpretation Issues
- No posted numerical analyses: the ClinicalTrials.gov record does not post statistical analyses for the three primary endpoints, so no treatment-effect estimate, confidence interval, or p-value is available from the registry record.
- Endpoint maturity: PFS and OS have defined follow-up windows, and the amount of statistical information available depends on the number and timing of events during those windows.
- Multiple primary endpoints: three primary endpoints require attention to the prespecified multiplicity framework when formal results are interpreted.
- MSI population: PFS in the MSI is population-specific and should not be interpreted as though it were automatically equivalent to the overall PFS analysis.
- Censoring: survival estimates depend on appropriate handling of participants who do not experience the endpoint during observed follow-up.
- Proportional hazards: if a Cox model is used, a single hazard ratio relies on the proportional-hazards assumption and may not fully describe treatment effects that vary materially over time.
- Progression assessment: PFS incorporates radiological progression according to RECIST 1.1, making the operational definition of progression part of the statistical endpoint.
- Analysis population: detailed registry-level definitions of efficacy and safety analysis populations are not provided in the endpoint information available here.
17. Why This Trial Matters Statistically
AtTEnd is a useful statistical teaching case because its design combines randomized treatment allocation, double masking, two distinct survival endpoints, and a population-specific PFS endpoint.
| Statistical concept | How it appears in AtTEnd |
|---|---|
| Randomization | The trial is randomized with two parallel arms. |
| Double masking | The registry identifies the study as double-masked. |
| Time-to-event analysis | PFS in the MSI, PFS, and OS are all defined by time from randomization to an event. |
| Kaplan-Meier estimation | A standard framework for describing PFS and OS while accounting for censoring. |
| Hazard ratio | A conventional relative effect measure for randomized time-to-event comparisons. |
| RECIST 1.1 | Radiological progression under RECIST 1.1 forms part of the PFS event definition. |
| Competing event components | PFS combines progression and death into one endpoint, with whichever occurs first defining the event. |
| Population-specific analysis | PFS in the MSI is registered separately from overall PFS. |
| Multiple primary endpoints | The registry identifies three primary endpoints, making multiplicity an important design consideration. |
| Follow-up windows | PFS endpoints have an 18-month window after last enrollment; OS has a two-year window. |
18. Clinical Interpretation vs Statistical Interpretation
Statistical interpretation
The trial is structured to compare randomized treatment strategies using time-to-event endpoints. PFS incorporates radiological progression or death, while OS uses death from any cause. The registry does not post numerical statistical analyses for these endpoints.
Clinical interpretation
The clinical question concerns whether incorporating atezolizumab into the randomized treatment strategy changes progression-free or overall survival in endometrial cancer. The meaning of any eventual numerical result depends on its endpoint definition, effect size, precision, and analysis population.
19. What a Complete Time-to-Event Result Would Contain
A statistically informative report of any of the AtTEnd primary endpoints would normally present more than a single p-value. A complete treatment comparison would ideally identify the analysis population, number of events, follow-up framework, estimated survival probabilities or medians where appropriate, treatment-effect estimate, confidence interval, and hypothesis-test result.
| Component | Why it matters |
|---|---|
| Endpoint definition | Establishes exactly what constitutes an event. |
| Analysis population | Defines which participants contribute to the treatment comparison. |
| Event counts | Help indicate how much event information supports the analysis. |
| Kaplan-Meier estimates | Describe survival or event-free probability over time. |
| Median time-to-event | Provides an absolute summary when the estimated survival curve reaches 0.50. |
| Hazard ratio | Provides a relative treatment-effect summary under the fitted model. |
| Confidence interval | Describes uncertainty and precision around the effect estimate. |
| P-value | Addresses the specified hypothesis test but does not quantify effect size. |
These components answer different questions. Kaplan-Meier estimates describe the event-time distribution, a hazard ratio summarizes relative event rates under a model, and a confidence interval communicates precision. The p-value addresses statistical evidence against a null hypothesis. None of these should be interpreted in isolation.
20. A Practical Framework for Reading the AtTEnd Analysis
Step 1 · Define the event
Start with the registry definition. For PFS, the event is first progression or death; for OS, it is death from any cause.
Step 2 · Identify the population
Distinguish the overall randomized population from the MSI-specific population used for the registered PFS endpoint.
Step 3 · Read the absolute estimates
Time-specific survival estimates and median event times provide absolute information that a hazard ratio does not provide by itself.
Step 4 · Read the relative effect
Use the hazard ratio to understand the relative event-rate comparison, while checking the confidence interval and model assumptions.
Step 5 · Check multiplicity
Because three endpoints are registered as primary, determine how the statistical testing strategy controls the overall error rate.
Step 6 · Interpret the p-value last
The p-value should support, not replace, interpretation of the endpoint definition, effect magnitude, precision, and clinical context.
21. Sources
- ClinicalTrials.gov: NCT03603184 — AtTEnd.
- PubMed: PubMed record for AtTEnd-related publication.
- PubMed: PubMed record for AtTEnd-related publication.
- PubMed: PubMed record for AtTEnd-related publication.
The ClinicalTrials.gov record identifies the trial as completed and provides its registered design, interventions, enrollment, primary endpoints, endpoint definitions, and follow-up time frames. It does not post statistical analyses for the registered primary endpoints.
Continue through the Clinical Biostats clinical-trial library
Explore additional clinical trial analyses and statistical resources for understanding randomized evidence, survival endpoints, and clinical-trial methodology.
22. Record Summary
AtTEnd is a completed randomized, double-blind, parallel phase 3 trial enrolling 549 participants with endometrial cancer. Its three registered primary endpoints are PFS in the MSI, overall PFS, and OS. PFS is defined from randomization to first progression or death, with the general PFS definition specifying radiological progression according to RECIST 1.1 or death from any cause. OS is defined from randomization until death from any cause. The registry specifies an 18-month follow-up window after the last patient enrolled for the PFS endpoints and a two-year window for OS.
Statistically, the trial is principally a time-to-event analysis problem. The appropriate framework centers on Kaplan-Meier estimation, treatment-group survival comparisons, and relative effect measures such as hazard ratios, with confidence intervals used to describe precision. Because three endpoints are registered as primary, multiplicity and endpoint hierarchy are also important when interpreting any formal hypothesis testing. The MSI-specific endpoint further emphasizes the importance of identifying the population to which an estimate applies.