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
ALFA-0701 was a randomized, open-label, parallel phase 3 treatment trial evaluating conventional chemotherapy with or without gemtuzumab ozogamicin in untreated acute myeloid leukemia among patients aged 50–70 years.
| Feature | ALFA-0701 |
|---|---|
| Phase | Phase 3 |
| Condition | Acute Myeloid Leukemia |
| Study title | A Randomized Study of Gemtuzumab Ozogamicin (GO) With Daunorubicine and Cytarabine in Untreated Acute Myeloid Leukemia (AML) Aged of 50-70 Years Old |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | None |
| Primary purpose | Treatment |
| Enrollment | 280 |
| Number of arms | 2 |
| Lead sponsor | Acute Leukemia French Association |
| Sponsor type | Other |
| Study start | 2007-12 |
| Primary completion | 2011-07 |
| ClinicalTrials.gov | NCT00927498 |
2. Clinical Question
The clinical question is whether adding gemtuzumab ozogamicin to conventional chemotherapy changes event-free survival in untreated acute myeloid leukemia among patients aged 50–70 years.
Population
Patients aged 50–70 years with untreated acute myeloid leukemia.
Intervention
Mylotarg, the registry's intervention name for gemtuzumab ozogamicin, associated with conventional chemotherapy consisting of AraC and daunorubicin.
Comparator
Conventional chemotherapy consisting of AraC and daunorubicin.
Primary question
How does treatment assignment relate to event-free survival, defined from randomization to relapse or death?
3. Trial Design
AraC + daunorubicin
- Conventional chemotherapy
- AraC
- Daunorubicin
Mylotarg associated with conventional chemotherapy
- Mylotarg
- AraC
- Daunorubicin
The statistical advantage of randomization is that treatment assignment is determined independently of the patient's subsequent event history. If the randomized groups are analysed according to assignment, the treatment comparison can be interpreted within the population and design represented by the trial.
4. Endpoints
| Endpoint | Registry definition | Time frame |
|---|---|---|
| Event Free Survival (EFS) | Relapse or death | Measured from randomization |
The registry identifies Event Free Survival (EFS) as the primary endpoint. Its stated time frame is relapse or death measured from randomization.
5. Planned Analysis
No statistical results are posted in the ClinicalTrials.gov record for the registered primary endpoint. The registry therefore identifies what is intended to be measured, while the formal numerical treatment comparison is not reported there.
How EFS is typically analysed
Event-free survival is a time-to-event endpoint because each participant can be followed from randomization until an event occurs or the observation is censored. For this trial, the event definition combines relapse or death.
A qualifying event for the registered endpoint is relapse or death. Participants without a qualifying event by the end of their available follow-up are handled through the censoring rules specified in the statistical analysis plan.
For a randomized two-arm trial with EFS as the primary endpoint, a conventional analysis would estimate the event-time distribution in each treatment group using the Kaplan-Meier method, compare the groups with a log-rank test, and quantify the relative event rate with a Cox proportional-hazards model when those methods are prespecified. The appropriate analysis population and censoring rules should follow the trial's statistical analysis plan.
6. Statistical Methodology
Kaplan-Meier estimation
Kaplan-Meier estimation is a standard way to describe time-to-event outcomes such as EFS. It uses the observed event times while retaining information from participants whose follow-up ends before an event is observed.
Here, di represents events at time ti, while ni represents the number at risk immediately before that time.
For EFS, the resulting curve can be interpreted as the estimated probability of remaining free of the defined event over time, where the event is relapse or death. The curve is therefore not simply a relapse curve and not simply a mortality curve: it reflects the first occurrence of either component of the composite endpoint.
Log-rank comparison
A log-rank test is commonly used to compare time-to-event distributions between randomized treatment groups. Rather than comparing only a single time point, it uses information across the observed follow-up.
The exact hypothesis and test specification depend on the prespecified statistical analysis plan, including any stratification or other design features.
Cox proportional-hazards model
A Cox model can summarize the relative rate at which EFS events occur between the randomized groups. Its treatment effect is commonly expressed as a hazard ratio.
An HR below 1 corresponds to a lower estimated instantaneous event rate in the treatment group under the fitted model. An HR above 1 corresponds to a higher estimated instantaneous event rate.
A hazard ratio is not an absolute difference in EFS probability and is not the percentage of patients who experience relapse or death. It is a relative time-to-event measure based on the model and follow-up data.
Censoring
Time-to-event analysis requires a rule for participants who do not experience relapse or death during the period in which they are observed. Their available follow-up contributes information to the analysis up to the censoring time. The validity of the resulting estimates depends on the censoring assumptions and the prespecified censoring rules.
7. Statistical Methods Explained
What makes EFS different from a simple proportion?
A proportion records whether an event occurred during a specified period. EFS additionally records when the event occurred. This distinction allows participants with different lengths of follow-up to contribute information appropriately and makes time-to-event methods useful.
Why does EFS combine relapse and death?
The registry defines EFS using either relapse or death. Combining these events creates a single endpoint that records the first qualifying event. A participant who dies before relapse therefore contributes a death event rather than being treated as a relapse-free survivor.
Why start the clock at randomization?
Randomization provides the common origin for the treatment comparison. Starting EFS measurement at randomization ensures that the two treatment groups are compared over the same conceptual time scale from treatment assignment.
Why use Kaplan-Meier rather than simply calculating a percentage?
Not every participant will necessarily have identical follow-up. Kaplan-Meier estimation allows the analysis to incorporate censoring while estimating the probability of remaining event-free over time.
What does a hazard ratio tell us?
A hazard ratio summarizes the relative instantaneous event rate between treatment groups under a Cox model. It does not directly tell us the absolute probability of remaining event-free at a particular time point.
Why does the proportional-hazards assumption matter?
The standard Cox hazard ratio is most straightforward to interpret when the relative hazards are reasonably stable over time. If the treatment effect changes substantially during follow-up, a single hazard ratio may compress important temporal information and should be interpreted alongside the underlying event-time curves and other time-specific measures.
8. Interpreting the Primary Endpoint
The registered primary endpoint is Event Free Survival, with relapse or death measured from randomization. Because no statistical result for EFS is posted in the ClinicalTrials.gov record, there is no registry-reported hazard ratio, confidence interval, or p-value to interpret numerically.
The important statistical distinction is between the endpoint definition and the estimated treatment effect. The registry supplies the former, but the posted record does not supply the latter.
What EFS measures
The time from randomization until relapse or death, according to the registered endpoint definition.
What EFS does not measure
EFS is not identical to overall survival because relapse is also an event in the registered definition.
What a hazard ratio would provide
A relative measure of the event rate between randomized groups under a time-to-event model.
What a p-value would provide
Evidence against a prespecified null hypothesis under the selected statistical test; it would not itself measure the size or clinical importance of an effect.
9. Statistical Interpretation of EFS
For this trial, the central statistical object is the time until the first event defined by the registry: relapse or death. This has several consequences for interpretation.
Relative effect versus absolute effect
A hazard ratio, if reported, would describe the relative event rate between the two randomized groups. An absolute EFS estimate at a specified time point would instead describe the estimated probability of remaining free from relapse or death at that time. These are complementary quantities rather than substitutes.
Confidence intervals
A confidence interval around a hazard ratio quantifies uncertainty around the estimated relative event rate under the statistical model and sampling framework. A narrow interval indicates greater statistical precision than a wide interval, all else being equal. It does not describe the range of treatment effects experienced by individual patients.
P-values
A p-value evaluates the compatibility of the observed data with a specified null hypothesis under the chosen testing procedure. It does not measure effect size, treatment benefit for an individual patient, or the probability that the null hypothesis is true.
Composite endpoint interpretation
Because relapse and death are both components of EFS, interpretation should consider the endpoint as defined rather than silently treating it as one component alone. If component-specific results are available elsewhere, they can provide additional clinical context, but they should not be substituted for the registered primary endpoint.
10. Randomization and Causal Interpretation
Randomization is central to the causal interpretation of a treatment comparison. With 280 participants assigned across two parallel treatment arms, the intended comparison is between outcomes under the two randomized treatment strategies within the enrolled population.
The lack of masking does not by itself determine whether an EFS comparison is biased. Its potential importance depends on how the endpoint is ascertained, how relapse is defined and assessed, how follow-up is conducted, and how those procedures are incorporated into the statistical analysis.
11. Open-Label Design and Statistical Considerations
Because the registry identifies the study as having no masking, participants and study personnel were not blinded to treatment assignment. For an endpoint containing death, the death component is less directly susceptible to subjective assessment than outcomes requiring judgment. Relapse assessment, however, can involve clinical and diagnostic processes in which knowledge of treatment could potentially matter.
12. Analysis Populations
The ClinicalTrials.gov record identifies the randomized trial structure and the primary EFS endpoint but does not provide the detailed analysis-population definitions needed to reproduce an efficacy analysis.
| Population concept | Statistical role |
|---|---|
| Randomized population | Provides the natural basis for an intention-to-treat treatment comparison because participants are analysed according to randomized assignment. |
| Safety population | Often defined according to treatment exposure, but the ClinicalTrials.gov record does not provide a specific safety-population definition. |
| Per-protocol population | May be relevant in some trial designs, but no specific per-protocol definition is reported here. |
For a confirmatory randomized treatment question, the intention-to-treat principle is generally important because excluding participants after randomization can compromise the comparability created by random allocation.
13. Missing Data and Censoring
Event-free survival naturally involves incomplete observation: some participants may reach the end of available follow-up without a qualifying event. Such observations are generally handled through censoring rather than treated as if an event had occurred at the last observation.
Censoring
Represents the point after which the participant's event status is no longer observed within the analysis window.
Event
For the registered EFS endpoint, the event is relapse or death.
Why timing matters
A participant observed for a longer period without an event contributes more follow-up information than someone censored earlier.
Why rules matter
The statistical analysis depends on prespecified definitions for event ascertainment and censoring.
The ClinicalTrials.gov record does not specify the detailed censoring and missing-data rules. Those rules can materially affect an EFS analysis and therefore belong to the trial's detailed statistical analysis documentation rather than being assumed from the endpoint name alone.
14. Multiplicity and Interim Analysis
The ClinicalTrials.gov record identifies one registered primary endpoint: Event Free Survival. It does not provide details of an interim-analysis schedule, alpha-spending procedure, multiplicity strategy, or hierarchical testing framework.
| Design topic | Registry information | Statistical implication |
|---|---|---|
| Primary endpoint | Event Free Survival | A single registered primary endpoint is identified. |
| Interim analysis | Not reported in the available registry information | No interim-monitoring method should be assumed. |
| Multiplicity | Not reported in the available registry information | No adjustment strategy should be assumed from the endpoint definition alone. |
| Alpha spending | Not reported in the available registry information | A group-sequential alpha-spending procedure cannot be inferred. |
This distinction is important because repeated efficacy looks, multiple primary endpoints, or multiple confirmatory comparisons can change the appropriate interpretation of a nominal p-value. Without the prespecified testing structure, a p-value from an external analysis should not automatically be interpreted as if it came from a particular multiplicity-controlled framework.
15. Non-Inferiority, Factorial, and Bayesian Methods
The registered design is a randomized, parallel treatment study. The available registry information does not identify a non-inferiority margin, factorial structure, or Bayesian statistical method.
These distinctions matter because the interpretation of a treatment effect depends on the trial's estimand and testing framework. A superiority treatment comparison should not be interpreted using the logic of a non-inferiority trial, and a Bayesian posterior probability should not be substituted for a frequentist p-value without a prespecified Bayesian analysis.
16. Clinical Interpretation vs Statistical Interpretation
Clinical interpretation
The study addresses treatment of untreated acute myeloid leukemia in patients aged 50–70 years and compares conventional chemotherapy with the same chemotherapy associated with gemtuzumab ozogamicin.
Statistical interpretation
The primary outcome is a time-to-event endpoint measured from randomization, with relapse or death defining the event. The appropriate comparison therefore requires methods that account for both event timing and censoring.
The two perspectives should remain distinct. The clinical question determines which outcome matters; the statistical method determines how differences in that outcome are estimated and tested. Neither a statistical significance threshold nor a treatment hazard ratio, by itself, describes the complete clinical meaning of an intervention.
17. Limitations
- No posted primary statistical result: the ClinicalTrials.gov record identifies EFS as the primary endpoint but does not post an estimate, confidence interval, or p-value for that endpoint.
- Limited design detail in the registry record: the available information does not specify the detailed statistical analysis plan, censoring rules, analysis populations, interim-monitoring procedures, or multiplicity strategy.
- Open-label design: the registry identifies masking as none, so assessment and follow-up procedures should be considered when interpreting an EFS analysis.
- Composite endpoint: EFS combines relapse and death, so its treatment effect represents the combined endpoint rather than either component alone.
- Time-to-event assumptions: Cox-model interpretation depends on model assumptions, including the proportional-hazards assumption when a conventional single hazard ratio is used.
- Population scope: the registered population is defined by untreated acute myeloid leukemia and an age range of 50–70 years; results from this population should not automatically be generalized to populations outside those characteristics.
18. Why This Trial Matters Statistically
ALFA-0701 is a useful statistical teaching case because its design centers on a randomized two-arm comparison with a clinically meaningful time-to-event endpoint. The key analytical challenge is not merely to compare the number of patients who relapse or die, but to incorporate when those events occur and how long each participant contributes observable follow-up.
| Concept | How it appears in ALFA-0701 |
|---|---|
| Randomization | Participants were randomly allocated to 2 treatment arms. |
| Parallel design | The trial uses a parallel rather than factorial or crossover structure. |
| Open-label treatment | The registry identifies masking as none. |
| Time-to-event analysis | EFS measures time from randomization to relapse or death. |
| Kaplan-Meier estimation | Provides a natural framework for estimating the EFS distribution in the presence of censoring. |
| Log-rank testing | A standard framework for comparing two time-to-event distributions when prespecified. |
| Cox regression | Provides a model-based relative event-rate measure through the hazard ratio. |
| Censoring | Allows participants without an observed event to contribute follow-up information through their last observed time. |
| Composite endpoint | Relapse and death are combined into the registered EFS event definition. |
19. A Practical Statistical Reading of ALFA-0701
A statistically disciplined reading of this trial begins with the estimand rather than with a p-value. The registered question is whether treatment assignment is associated with a difference in time until relapse or death among the randomized participants.
The next step is to preserve the time dimension. A simple comparison of the percentage of patients who experienced an event can discard information about whether events occurred early or late and how long participants remained under observation. Kaplan-Meier methods retain that timing information.
The treatment comparison can then be expressed in several complementary ways. A survival curve describes the estimated EFS experience over time. A hazard ratio summarizes the relative event rate under a Cox model. A confidence interval communicates statistical precision around that estimate. A time-specific EFS estimate provides an absolute interpretation at a particular follow-up time.
These quantities should not be treated as interchangeable. A hazard ratio is a relative model-based measure; a Kaplan-Meier estimate is an absolute probability at a specified time; and a p-value addresses compatibility with a null hypothesis under the selected testing procedure.
Finally, the design features must be kept in view. Randomization supports the treatment comparison, while the absence of masking introduces considerations for endpoint ascertainment and follow-up. Because the primary endpoint includes both relapse and death, the interpretation concerns the composite event defined by the registry.
20. Statistical Concepts in This Trial
21. Sources
- ClinicalTrials.gov: NCT00927498.
- PubMed: PubMed record.
- PubMed: PubMed record.
Continue through Clinical Biostats
Explore statistical methods, clinical-trial analyses, and practical tools for understanding time-to-event and randomized-trial data.
22. Record Summary
ALFA-0701 is a randomized phase 3, parallel, unmasked treatment trial enrolling 280 participants with untreated acute myeloid leukemia aged 50–70 years. The two treatment arms compare conventional chemotherapy consisting of AraC and daunorubicin with the same conventional chemotherapy associated with gemtuzumab ozogamicin.
The registered primary endpoint is Event Free Survival, defined as relapse or death measured from randomization. Statistically, this makes the trial a time-to-event problem: the treatment comparison must account for both the occurrence and timing of events, while appropriately handling participants whose follow-up ends without an observed event.
The most natural analytical framework for such an endpoint includes Kaplan-Meier estimation, a treatment-group comparison such as the log-rank test, and Cox-model estimation of a hazard ratio when these methods are prespecified. Interpretation should distinguish relative event rates from absolute EFS probabilities, confidence intervals from p-values, and the composite endpoint from its individual components.