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
aTTom is a phase 3 randomized treatment trial listed for women with breast cancer. The registry identifies tamoxifen citrate as the drug intervention and adjuvant therapy as the associated procedure. Its registered primary endpoint is all-cause mortality.
| Feature | aTTom |
|---|---|
| Trial acronym | aTTom |
| NCT identifier | NCT00003678 |
| Phase | Phase 3 |
| Condition | Breast Cancer |
| Population described by the registry | Women with breast cancer |
| Allocation | Randomized |
| Primary purpose | Treatment |
| Registered enrollment | 20,000 |
| Intervention | Tamoxifen citrate (drug); adjuvant therapy (procedure) |
| Lead sponsor | University Hospital Birmingham |
| Sponsor type | OTHER |
| Start date | 1991-05 |
| Registry status | UNKNOWN |
2. Clinical Question
The registry identifies a randomized phase 3 treatment study of women with breast cancer in which tamoxifen citrate and adjuvant therapy are the registered interventions. The primary endpoint is all-cause mortality. Thus, the central statistical question is whether randomized treatment assignment is associated with a difference in time to death from any cause.
Population
Women with breast cancer.
Intervention
Tamoxifen citrate, listed with adjuvant therapy.
Comparator
The ClinicalTrials.gov record summarized here does not specify the comparator in the available intervention information.
Primary question
Does randomized treatment assignment affect all-cause mortality?
3. Trial Design
The registry describes a phase 3 randomized treatment trial with a registered enrollment of 20,000. Randomization is important statistically because it is intended to balance prognostic factors between treatment assignments at the point of allocation, allowing subsequent differences in outcomes to be interpreted in relation to the randomized intervention under the assumptions of the design.
The registry information summarized here does not specify a randomization ratio, masking procedure, comparator regimen, stratification factors, crossover provisions, interim-analysis plan, multiplicity strategy, missing-data procedure, or Bayesian component. Those design elements therefore are not assigned to the trial here.
4. Endpoints
| Endpoint | Registry definition / time frame | Statistical interpretation |
|---|---|---|
| All-cause mortality | Primary endpoint | A time-to-event outcome in which the event is death from any cause. The available registry record does not provide an additional time frame or operational definition. |
All-cause mortality is fundamentally different from a binary outcome measured at a fixed visit. It contains information about when an event occurs as well as whether it occurs during observed follow-up. Participants who remain alive at their last known follow-up may contribute partial information through censoring rather than being treated as if they had experienced the event.
5. Statistical Methodology
Because the registered primary endpoint is all-cause mortality, the natural statistical framework is survival or time-to-event analysis. The analysis begins with the time origin, follows each participant until death or censoring, and compares the resulting survival experience between randomized treatment assignments.
Kaplan-Meier estimation
A Kaplan-Meier estimator is typically used to describe the survival distribution over time. It accommodates right censoring and produces an estimated probability of remaining alive at different time points.
Here, di is the number of deaths at event time ti, while ni is the number of participants at risk immediately before that time.
Log-rank comparison
For two randomized groups, a log-rank test is a standard method for comparing survival distributions. It evaluates whether the observed pattern of deaths over follow-up differs from what would be expected under the null hypothesis of equivalent survival functions.
Cox proportional-hazards model
A Cox proportional-hazards model is another standard approach for a randomized time-to-event comparison. It estimates a hazard ratio, which describes the relative instantaneous event rate between treatment groups under the model.
A hazard ratio is a relative time-to-event measure. It is not a percentage of patients who benefit, an absolute survival difference, or a statement that each individual patient experiences the same proportional change in mortality.
Why censoring matters
Participants may remain alive when their observation ends. Such participants are generally censored at their last available follow-up rather than counted as deaths. Correct handling of censoring allows the analysis to use the information accumulated before the observation ended.
6. Planned Analysis
The registry identifies all-cause mortality as the primary endpoint. For an endpoint of this type, the standard analysis would generally describe survival using Kaplan-Meier methods and compare randomized treatment groups with a time-to-event method such as the log-rank test. A Cox proportional-hazards model would commonly be used to estimate a hazard ratio and its confidence interval.
Primary estimand
The main statistical target is the difference in survival experience between the randomized treatment assignments, expressed through a time-to-event measure such as a hazard ratio or survival probabilities over time.
Event definition
The registered event is death from any cause, rather than death from a particular cancer-related cause or another disease-specific endpoint.
Typical descriptive analysis
Kaplan-Meier survival curves and time-specific survival estimates can describe the observed survival experience while accounting for right censoring.
Typical comparative analysis
A log-rank test and Cox proportional-hazards model are standard approaches for comparing randomized groups on a time-to-event endpoint.
7. Statistical Methods Explained
Why is all-cause mortality a time-to-event endpoint?
All-cause mortality records whether a participant dies and, importantly, when that death occurs. Two studies could have the same number of deaths but different conclusions about survival if deaths occur at substantially different times. Time-to-event methods preserve this timing information.
Why not analyze mortality only as a simple proportion?
A simple proportion, such as the percentage who died by a particular date, discards some information about when deaths occurred and how long participants were observed. Survival analysis uses follow-up time and accommodates participants who are censored before experiencing the event.
What does a hazard ratio represent?
A hazard ratio compares the estimated instantaneous event rates between two groups within a specified statistical model. For example, an HR below 1 would correspond to a lower estimated instantaneous mortality rate in the numerator group. It does not directly give the difference in the probability of being alive at a particular time.
Why use Kaplan-Meier estimation?
Kaplan-Meier estimation is designed for situations in which some participants have not experienced the event when their observation ends. Rather than treating those participants as deaths or excluding all of their information, the method uses their observed follow-up until censoring.
What does the log-rank test ask?
The log-rank test asks whether the survival experience differs between groups across follow-up. It is particularly natural for randomized comparisons of time-to-event outcomes and is often paired with Kaplan-Meier curves and a Cox-model hazard ratio.
What assumption is important for a Cox hazard ratio?
The conventional Cox proportional-hazards interpretation assumes that the relative hazard between groups is reasonably stable over time. If hazards vary substantially in relative terms, a single hazard ratio can become a less complete description of the treatment difference. Survival curves and time-specific measures can then provide important additional context.
8. Analysis Population and Randomization
The available registry information identifies randomization but does not specify the detailed analysis populations. In a randomized trial, the primary efficacy comparison is commonly based on the intention-to-treat principle, in which participants are analyzed according to their randomized assignment. This preserves the treatment comparison created by randomization.
| Concept | What can be established from the registry | Statistical significance |
|---|---|---|
| Randomization | The allocation field is RANDOMIZED. | Supports a randomized comparative design. |
| Primary purpose | The primary purpose is TREATMENT. | Identifies the trial as a treatment study rather than a diagnostic or prevention study. |
| Primary endpoint | All-cause mortality. | Defines a time-to-event outcome. |
| Detailed analysis population | Not specified in the available record. | No additional population definition is assigned here. |
9. Missing Data, Censoring, and Follow-Up
For a mortality endpoint, the central missing-information issue is often not a missing binary response but incomplete observation of survival time. A participant who is alive at the end of known follow-up may be censored. The statistical validity of Kaplan-Meier and Cox analyses depends on appropriate assumptions about the censoring mechanism and accurate ascertainment of follow-up.
The registry information summarized here does not specify an imputation method, censoring rules, loss-to-follow-up assumptions, or sensitivity analyses. Those details are important in a full statistical analysis plan because different approaches to incomplete follow-up can affect the interpretation of survival estimates.
10. Confidence Intervals and Effect Estimates
No treatment-effect estimates or confidence intervals are posted in the statistical-analysis fields of the ClinicalTrials.gov record summarized here. Nevertheless, confidence intervals are an essential part of a time-to-event analysis because a point estimate alone does not describe the uncertainty surrounding the estimated treatment effect.
If a Cox analysis produced an HR below 1 for tamoxifen relative to its comparator, the estimate would indicate a lower estimated instantaneous mortality rate under the fitted model. The size of the confidence interval would indicate how precisely that relative effect had been estimated.
A 95% confidence interval around a hazard ratio would not be a range containing 95% of individual patient effects. It would quantify statistical uncertainty about the estimated parameter under the model and sampling framework.
A p-value addresses evidence against a specified null hypothesis under the statistical model. It does not describe the magnitude of a treatment effect. Effect estimates and confidence intervals are needed to characterize the size and precision of an observed difference.
11. Non-Inferiority, Multiplicity, and Interim Analysis
The available registry information does not identify a non-inferiority margin, multiplicity strategy, interim-analysis plan, alpha-spending procedure, or Bayesian analysis. These elements should not be assumed merely because the trial is phase 3 or randomized.
| Design topic | Registry information available here | Statistical implication |
|---|---|---|
| Non-inferiority margin | Not specified. | No non-inferiority conclusion can be defined from the available record. |
| Multiplicity | Not specified. | No particular familywise-error strategy can be assigned. |
| Interim analysis | Not specified. | No alpha-spending or sequential-testing procedure can be assigned. |
| Bayesian methods | Not specified. | No prior distribution or posterior analysis can be assigned. |
| Crossover | Not specified. | No crossover adjustment can be described. |
12. Why This Trial Matters Statistically
aTTom is a useful teaching case because its registered primary endpoint illustrates the central logic of survival analysis in a large randomized treatment trial. The key statistical problem is not simply whether deaths occur, but how the occurrence of death is distributed over follow-up in the randomized groups.
| Concept | How it appears in aTTom |
|---|---|
| Randomization | The registry identifies the allocation as RANDOMIZED. |
| Phase 3 design | The study is registered as Phase 3. |
| Time-to-event outcome | All-cause mortality is the registered primary endpoint. |
| Kaplan-Meier estimation | A standard descriptive approach for survival probabilities over time. |
| Log-rank comparison | A standard method for comparing survival distributions between randomized groups. |
| Cox regression | A standard model for estimating a hazard ratio for mortality. |
| Censoring | Relevant because participants may be followed for different lengths of time without experiencing the event. |
| Confidence intervals | Necessary for communicating uncertainty around estimated treatment effects. |
| Assumption checking | Important when interpreting a single Cox proportional-hazards estimate over extended follow-up. |
13. Limitations
- No posted statistical analyses: the ClinicalTrials.gov record does not contain posted statistical analyses for the registered primary endpoint.
- Limited endpoint detail: the available registry information identifies all-cause mortality but does not provide an additional endpoint time frame or detailed operational definition.
- Comparator information: the available intervention information does not specify the comparator treatment assignment.
- Analysis-plan detail: the available record does not specify the planned survival model, testing procedure, confidence-level convention, stratification factors, or analysis population definitions.
- Incomplete design information: the registry information summarized here does not specify masking, randomization ratio, crossover, interim analysis, multiplicity adjustment, missing-data methods, or Bayesian methods.
- No outcome estimates: no numerical mortality estimates, hazard ratios, confidence intervals, event counts, or p-values are posted in the statistical-analysis information available for this record.
- Interpretation of survival models: even when a hazard ratio is available, it should be interpreted alongside the underlying survival curves and clinically meaningful time-specific survival estimates.
14. Statistical Interpretation vs Clinical Interpretation
Statistical interpretation
The registered endpoint is a time-to-event outcome. A complete analysis would quantify survival over time, compare the randomized groups, estimate uncertainty, and assess whether the assumptions underlying the chosen model are reasonable.
Clinical interpretation
Clinical interpretation would require the observed mortality experience, the size and precision of the treatment effect, the duration of follow-up, and relevant safety and treatment information. Those outcome details are not posted in the statistical-analysis information available here.
15. A Practical Framework for Reading the aTTom Analysis
When formal outcome estimates are available, a rigorous reading of aTTom should proceed from the design to the estimand and then to the observed evidence. The randomized allocation establishes the primary comparison. The all-cause mortality endpoint establishes the event of interest. Kaplan-Meier estimates describe survival over time, while a log-rank comparison and Cox model can summarize differences between treatment assignments.
This sequence keeps the statistical estimate connected to the trial design and prevents a single numerical measure from being interpreted independently of follow-up, censoring, and model assumptions.
Start with the endpoint
All-cause mortality is an especially clear event definition: death from any cause. Because it is time-dependent, the timing of events matters and should be preserved in the analysis.
Then examine the survival curves
Kaplan-Meier curves can show when treatment-group separation emerges, whether curves remain separated, and whether the pattern appears reasonably consistent with a proportional-hazards summary.
Then examine the effect estimate
The hazard ratio provides a compact relative measure, but it should be read with its confidence interval and with the survival curves. An HR alone does not communicate the absolute probability of survival at a clinically relevant time point.
Finally examine uncertainty and assumptions
Confidence intervals, censoring patterns, follow-up duration, and proportional-hazards considerations determine how much information can reasonably be extracted from a single effect estimate.
16. Trial Timeline
Trial start
The ClinicalTrials.gov record lists the study start date as 1991-05.
Randomized treatment study
The registry classifies aTTom as a phase 3 randomized trial with treatment as its primary purpose.
20,000 participants
The registered enrollment is 20,000.
UNKNOWN
The ClinicalTrials.gov record summarized here lists the current status as UNKNOWN.
17. Sources
- ClinicalTrials.gov: NCT00003678 — aTTom.
- PubMed: PubMed record 35213039.
- PubMed: PubMed record 31504126.
18. Record Summary
aTTom is registered as a phase 3 randomized treatment trial of women with breast cancer, with 20,000 participants listed for enrollment and tamoxifen citrate identified as the drug intervention alongside adjuvant therapy. Its registered primary endpoint is all-cause mortality, making survival analysis the natural statistical framework. Kaplan-Meier estimation, log-rank comparison, and Cox proportional-hazards modeling are standard methods for this type of endpoint, while censoring and proportional-hazards assumptions are central to interpretation.
The ClinicalTrials.gov record summarized here does not post statistical analyses or numerical outcome estimates for the primary endpoint. Consequently, the most important statistical lesson from the registry information is the structure of the analysis itself: randomized allocation defines the comparison, all-cause mortality defines the event, follow-up determines the time-to-event information, and survival methods translate those observations into estimates that can be accompanied by measures of uncertainty.