This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record. The numerical trial results on this page are limited to the statistical analysis reported in the ClinicalTrials.gov record.
1. Trial at a Glance
AGILE was a randomized phase 3, parallel-group, triple-masked treatment trial enrolling 146 participants with previously untreated acute myeloid leukemia and an IDH1 mutation. The registered primary endpoint was Event-Free Survival (EFS), a time-to-event endpoint assessed up to Week 24.
| Feature | AGILE |
|---|---|
| Trial name | AGILE |
| Phase | Phase 3 |
| Therapeutic area | Hematology |
| Population | Previously untreated acute myeloid leukemia with an IDH1 mutation |
| Conditions | Newly Diagnosed Acute Myeloid Leukemia (AML); Untreated AML; AML Arising From Myelodysplastic Syndrome (MDS); Leukemia, Myeloid, Acute |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | Triple |
| Primary purpose | Treatment |
| Enrollment | 146 |
| Primary endpoint | Event-Free Survival (EFS) |
| Primary endpoint type | Time-to-event |
| Primary hypothesis | Superiority |
| Results posted | Yes |
| Statistical analyses posted | 1 |
| Lead sponsor | Institut de Recherches Internationales Servier |
| Status | Active, not recruiting |
2. Clinical Question
The central statistical question was whether AG-120 (ivosidenib) plus azacitidine improved Event-Free Survival compared with placebo plus azacitidine in participants with previously untreated acute myeloid leukemia with an IDH1 mutation.
Population
Participants with previously untreated acute myeloid leukemia with an IDH1 mutation. The registered conditions also include newly diagnosed AML, untreated AML, AML arising from MDS, and acute myeloid leukemia.
Intervention
AG-120 (ivosidenib) in combination with azacitidine.
Comparator
Placebo in combination with azacitidine.
Primary question
Does AG-120 plus azacitidine improve EFS relative to placebo plus azacitidine under a superiority hypothesis?
3. Trial Design
Active treatment
- AG-120 (drug)
- Azacitidine (drug)
- Compared with placebo plus azacitidine
Control treatment
- Placebo (drug)
- Azacitidine (drug)
- Reference group for the hazard ratio
4. Randomization, Stratification, and Analysis Population
The primary analysis was conducted in the Full Analysis Set (FAS), which included all participants who were randomized. The statistical analysis notes specify that the hazard ratio was estimated using a Cox proportional-hazards model stratified by the randomization stratification factors.
| Analysis feature | Registry-supported detail |
|---|---|
| Analysis population | FAS included all participants who were randomized. |
| Groups compared | AG-120 + Azacitidine vs Placebo + Azacitidine |
| Stratification | Hazard ratio estimated using a Cox proportional-hazards model stratified by AML status and geographic region. |
| Reference group | Placebo + azacitidine was the denominator for the hazard ratio. |
| Hypothesis type | Superiority |
Using all randomized participants in the FAS keeps the primary treatment comparison closely connected to the randomized allocation. The stratification factors—AML status and geographic region—were incorporated into the Cox model rather than treating the analysis as an unstratified comparison.
5. Primary Endpoint
| Endpoint | Registered definition | Time frame |
|---|---|---|
| Event-Free Survival (EFS) | EFS was defined as the time from randomization until treatment failure, relapse from remission, or death from any cause, whichever occurs first. Treatment failure was defined as failure to achieve complete remission (CR) by Week 24. CR: Bone marrow blasts <5% and no Auer rods; absence of extramedullary disease; Absolute neutrophil count (ANC) ≥1.0 × 109 per litre (109/L) (1000 per microlitre. | Up to Week 24 |
6. Results
The ClinicalTrials.gov record contains one formal statistical analysis, corresponding to the primary endpoint of Event-Free Survival. The analysis used the Full Analysis Set, a log-rank comparison, and a stratified Cox proportional-hazards model for the hazard ratio.
Event-Free Survival
Hazard ratio for EFS
95% CI: 0.16–0.69 · P = 0.0011
Analysis: two-sided stratified time-to-event analysis in the Full Analysis Set.
| Endpoint | Intervention | Comparator | Effect estimate | P-value |
|---|---|---|---|---|
| Event-Free Survival, up to Week 24 | AG-120 + Azacitidine | Placebo + Azacitidine | HR 0.33 (95% CI 0.16–0.69) | 0.0011 |
The reported HR of 0.33 means that, under the fitted stratified Cox proportional-hazards model, the estimated instantaneous rate of experiencing the EFS event in the AG-120 plus azacitidine group was approximately 33% of the corresponding rate in the placebo plus azacitidine group. Equivalently, 0.33 corresponds to an estimated 67% lower hazard relative to the comparator.
The hazard ratio does not mean that 67% of participants avoided an EFS event, that 67% were cured, or that each participant experienced exactly a 67% reduction in risk. It is a relative, model-based measure of event rates over time.
The 95% confidence interval of 0.16–0.69 describes uncertainty around the estimated hazard ratio under the statistical model and sampling framework. It does not describe the range of treatment effects that individual participants would experience.
The P-value of 0.0011 addresses the statistical evidence against the null hypothesis within the specified analysis framework. It is not a measure of the size of the treatment effect: effect magnitude is described by the hazard ratio and its confidence interval.
Because the hazard ratio was obtained from a Cox proportional-hazards model, its interpretation relies on the model's proportional-hazards framework. The ClinicalTrials.gov record does not provide enough information to independently assess that assumption from the underlying event-time data.
7. Statistical Methodology
Log-rank test
The registry reports a log-rank test as the primary comparison method. For a time-to-event endpoint such as EFS, the log-rank test compares the observed event experience between randomized groups over the follow-up period while accounting for the timing of events and censored observations.
The log-rank framework evaluates whether the observed pattern of events over time differs between the randomized treatment groups. The reported P-value of 0.0011 is the statistical result posted on ClinicalTrials.gov for this primary analysis.
Stratified Cox proportional-hazards model
The registry states that the hazard ratio was estimated using a Cox proportional-hazards model stratified by AML status and geographic region, with placebo plus azacitidine as the denominator.
An HR below 1 indicates a lower estimated instantaneous event rate for the AG-120 plus azacitidine group relative to the placebo plus azacitidine group under the fitted model.
Why stratification matters
Stratification allows the time-to-event comparison to account for the randomization stratification factors while estimating the treatment effect. In AGILE, those factors were AML status and geographic region.
Full Analysis Set
The FAS included all participants who were randomized. For a randomized treatment comparison, this preserves the connection between the efficacy analysis and the original treatment assignment rather than defining the primary analysis only by subsequent treatment exposure.
Superiority hypothesis
The registered hypothesis type was superiority. The statistical question was therefore whether the treatment groups differed in the direction specified by the superiority objective, rather than whether the intervention was merely no worse than the comparator by a prespecified non-inferiority margin.
8. Statistical Methods Explained
What does an EFS hazard ratio of 0.33 mean?
An HR of 0.33 indicates a substantially lower estimated instantaneous rate of an EFS event in the AG-120 plus azacitidine group relative to placebo plus azacitidine under the fitted Cox model. It is a relative time-to-event measure, not a probability that an individual participant will remain event-free.
Why was a log-rank test used?
EFS is defined by the time from randomization to the first qualifying event. Time-to-event data commonly include participants who have not yet experienced the event during observed follow-up. The log-rank test is designed to compare the event-time distributions of randomized groups while incorporating the timing of events and censoring.
Why was a Cox model used in addition to the log-rank test?
The log-rank test supplies a hypothesis test for the time-to-event comparison, whereas the Cox model provides a directly interpretable effect estimate—the hazard ratio—and its confidence interval. In AGILE, the Cox model was also stratified by AML status and geographic region.
What does the 95% confidence interval of 0.16–0.69 tell us?
The interval quantifies statistical uncertainty around the estimated HR of 0.33. It indicates that the estimate is not a single perfectly known quantity. It does not mean that 95% of individual participants have hazard ratios between 0.16 and 0.69.
Why does the P-value of 0.0011 not measure effect size?
A P-value measures the statistical evidence against a specified null hypothesis under the analysis framework. It depends on both the magnitude of the observed difference and the amount of information available. The HR of 0.33 describes the estimated relative effect, while the confidence interval describes its precision.
Why are AML status and geographic region included as stratification factors?
The registry-reported analysis notes identify AML status and geographic region as the randomization stratification factors. Stratifying the Cox model incorporates these factors into the time-to-event analysis rather than estimating the treatment effect from an unstratified model.
What does “up to Week 24” mean for EFS?
The registry specifies the EFS time frame as up to Week 24. The endpoint therefore has a defined observation horizon in the registered outcome measure. The event itself is whichever qualifying event occurs first: treatment failure, relapse from remission, or death from any cause.
9. Safety
The ClinicalTrials.gov record reports serious adverse events by randomized arm as the number affected over the number at risk. These data provide a separate safety perspective from the primary EFS analysis.
| Safety measure | AG-120 + Azacitidine | Placebo + Azacitidine |
|---|---|---|
| Serious adverse events | 49/71 | 60/73 |
The affected/at-risk counts should be interpreted as reported safety counts, not as an additional formal efficacy analysis. The ClinicalTrials.gov record does not provide a confidence interval or P-value for the serious-adverse-event comparison, so no formal comparative inference is added here.
10. Understanding the Primary Result
The EFS HR of 0.33 indicates that the estimated instantaneous rate of the EFS event was approximately one-third as high with AG-120 plus azacitidine as with placebo plus azacitidine under the stratified Cox model.
A hazard ratio alone does not provide the probability of remaining event-free at a particular time. The ClinicalTrials.gov record does not provide a time-specific EFS estimate or median EFS, so those measures are not added to this page.
The 95% CI of 0.16–0.69 provides the reported uncertainty interval around the HR. The width of the interval is important because a point estimate alone cannot communicate statistical precision.
The reported P = 0.0011 provides evidence against the null hypothesis in the specified superiority analysis. It should not be converted into a probability that the treatment works, a probability that the null hypothesis is true, or a measure of clinical importance.
11. Analysis Population and Interpretation
The primary analysis used the Full Analysis Set, defined in the ClinicalTrials.gov record as all participants who were randomized. This is important because randomization establishes the treatment comparison before subsequent outcomes are observed.
Why randomized analysis matters
Analyzing randomized participants according to the trial's defined analysis population maintains the connection between the treatment comparison and the randomization process.
Why the event definition matters
EFS combines treatment failure, relapse from remission, and death from any cause. The first qualifying event determines the EFS time.
Why the time frame matters
The registered EFS time frame is up to Week 24. Time-to-event results should always be interpreted in relation to the prespecified observation period.
Why the model matters
The HR comes from a stratified Cox model, so the estimate reflects both the treatment comparison and the specified stratification structure.
12. What the Hazard Ratio Does — and Does Not — Mean
HR 0.33
The arithmetic interpretation follows directly from 1 − 0.33 = 0.67.
The HR of 0.33 means the modeled instantaneous event rate in the AG-120 plus azacitidine group was estimated to be approximately 67% lower than that in the placebo plus azacitidine group. This is a statement about the relative hazard, not a direct statement about the proportion of participants who experienced or avoided an event.
| Measure | What it answers | What it does not answer |
|---|---|---|
| Hazard ratio | How the modeled instantaneous event rates compare between treatment groups. | The absolute probability of remaining event-free at a particular time. |
| 95% confidence interval | How precisely the hazard ratio was estimated under the statistical framework. | The range of individual treatment effects across participants. |
| P-value | How the observed data compare with the specified null hypothesis. | The magnitude or clinical importance of the treatment effect. |
| EFS | Time from randomization until the first qualifying event. | Overall survival alone or a single binary response measure. |
13. Endpoint Definition and Event Structure
The AGILE EFS definition illustrates why endpoint construction is central to statistical analysis. The endpoint begins at randomization and ends at the first occurrence of one of several specified events.
The registered definition of complete remission specifies bone marrow blasts <5% and no Auer rods, absence of extramedullary disease, and an absolute neutrophil count of ≥1.0 × 109 per litre (109/L) (1000 per microlitre [1000/μL]); platelet count ≥100 × 10\^9/L (100,000/μL); independence of red blood cell transfusions. Participants who had an EFS event (relapse or death) after, 2 or more missing disease assessments were censored at the last adequate disease assessment documenting no relapse before the missing assessments. The reported data represents the Kaplan-Meier median value. in the registry text.
Because EFS uses the first qualifying event, different clinical pathways can lead to the same statistical endpoint. This is one reason it is important to understand the endpoint definition before interpreting its hazard ratio.
14. Stratified Analysis
The registry-reported statistical analysis specifically states that the Cox model was stratified by AML status and geographic region. This means the model accounts for these stratification factors while estimating the treatment hazard ratio.
The registry specifies placebo + azacitidine as the denominator for the reported hazard ratio.
Stratification should not be confused with subgroup analysis. A stratification factor can be incorporated into the primary treatment-effect model without implying that separate treatment-effect claims are being made within every level of that factor.
15. Multiplicity, Interim Analysis, and Other Design Features
The ClinicalTrials.gov record supports a primary superiority analysis, but they do not provide details on an alpha-spending strategy, interim-analysis boundaries, multiplicity adjustment, or a formal hierarchy among multiple efficacy endpoints. Those topics are therefore not assigned numerical or methodological claims beyond what is directly reported.
| Design topic | What the ClinicalTrials.gov record establishes |
|---|---|
| Hypothesis | Superiority |
| Primary endpoint | One registered primary endpoint: Event-Free Survival |
| Formal primary analysis | One statistical analysis is posted for the primary endpoint. |
| Interim analysis | No interim-analysis method is reported in the ClinicalTrials.gov record. |
| Alpha spending | No alpha-spending method is reported in the ClinicalTrials.gov record. |
| Multiplicity | No multiplicity-adjustment procedure is reported in the ClinicalTrials.gov record. |
| Non-inferiority margin | Not applicable to the reported superiority hypothesis. |
| Bayesian methods | No Bayesian method is reported in the ClinicalTrials.gov record. |
| Crossover | No crossover information is reported in the ClinicalTrials.gov record. |
This distinction is important: absence of a registry-reported method is not evidence that a particular procedure was or was not used elsewhere in the full trial documentation. This page limits its claims to the ClinicalTrials.gov-derived data.
16. Missing Data, Censoring, and Model Assumptions
Censoring in time-to-event analysis
Time-to-event methods can incorporate participants whose qualifying event has not occurred during the period in which they are observed.
Missing data and imputation
The registry-reported statistical analysis does not specify a missing-data imputation method for the EFS analysis. No particular imputation procedure is therefore attributed to AGILE on this page.
Proportional hazards
The reported hazard ratio was estimated using a Cox proportional-hazards model. The model's conventional interpretation assumes that the relative hazard structure is appropriately represented by the proportional-hazards framework. The ClinicalTrials.gov record does not provide the underlying event-time information needed to independently evaluate that assumption.
17. Safety and Efficacy Answer Different Questions
The primary EFS analysis and the serious-adverse-event data should be kept conceptually separate. EFS evaluates a prespecified time-to-event efficacy endpoint, whereas serious adverse events describe a safety outcome.
Efficacy
The primary EFS comparison reported HR 0.33 with a 95% CI of 0.16–0.69 and P = 0.0011.
Safety
Serious adverse events were reported as 49/71 in the AG-120 plus azacitidine group and 60/73 in the placebo plus azacitidine group.
These are different statistical questions. A treatment-effect estimate for EFS should not be combined mechanically with serious-adverse-event counts into a single numerical "benefit-risk" statistic.
18. Trial Timeline
Trial start
The AGILE study began on June 26, 2017.
Primary completion
The registered primary completion date was March 18, 2021.
Active, not recruiting
The trial is listed as ACTIVE_NOT_RECRUITING in the ClinicalTrials.gov record.
19. Limitations
- Limited posted statistical results: the ClinicalTrials.gov record contains one formal statistical analysis, for the primary EFS endpoint. No additional formal endpoint estimates are included here.
- No median EFS: the ClinicalTrials.gov record does not provide a median EFS estimate or its confidence interval.
- No time-specific EFS rates: the ClinicalTrials.gov record does not provide Kaplan-Meier survival probabilities at specific time points.
- No reconstructed Kaplan-Meier curve: the underlying event and censoring data are not reported, so a curve cannot be reconstructed reliably from the reported HR and confidence interval alone.
- Hazard-ratio assumptions: the Cox model is a model-based approach, and its interpretation depends on the appropriateness of the proportional-hazards framework.
- Limited subgroup information: although AML status and geographic region were randomization stratification factors, the ClinicalTrials.gov record does not provide separate treatment-effect estimates for their individual levels.
- Safety inference: serious-adverse-event counts are reported, but no formal comparative confidence interval or P-value is reported in the ClinicalTrials.gov record.
- Missing methodological detail: the ClinicalTrials.gov record does not specify an imputation strategy, interim-analysis procedure, alpha-spending method, or multiplicity-adjustment procedure.
- Endpoint complexity: EFS combines treatment failure, relapse from remission, and death from any cause, so its interpretation depends on the precise registered endpoint definition.
20. Why This Trial Matters Statistically
AGILE is a useful teaching example because the registry-reported statistical analysis connects several core clinical-trial concepts in one primary endpoint: randomization, a composite time-to-event definition, stratified analysis, log-rank testing, Cox regression, hazard ratios, confidence intervals, and P-values.
| Concept | How it appears in AGILE |
|---|---|
| Randomization | Participants were randomly allocated in a two-arm parallel-group phase 3 trial. |
| Triple masking | The trial was triple-masked. |
| Time-to-event endpoint | EFS measured time from randomization to treatment failure, relapse from remission, or death from any cause. |
| Log-rank test | Used for the reported primary time-to-event comparison. |
| Hazard ratio | Primary effect measure, estimated as 0.33. |
| Confidence interval | 95% two-sided CI of 0.16–0.69 around the HR. |
| Cox model | Used to estimate the HR with stratification by AML status and geographic region. |
| Analysis population | Full Analysis Set containing all randomized participants. |
| Superiority | The registered hypothesis type was superiority. |
| Safety analysis | Serious adverse events were reported by arm as affected/at risk. |
21. Primary Result in Statistical Context
The AGILE primary analysis provides three complementary pieces of statistical information:
Effect estimate
HR 0.33 quantifies the relative treatment effect on the hazard scale. It indicates a lower estimated instantaneous EFS event rate in the AG-120 plus azacitidine group.
Precision
95% CI 0.16–0.69 communicates uncertainty around the HR estimate.
Evidence against the null
P = 0.0011 reports the statistical evidence from the specified analysis against the null hypothesis.
Analysis structure
The result comes from a randomized comparison analyzed using a log-rank test and a stratified Cox proportional-hazards model.
Reading all four components together is more informative than reading the P-value alone. The hazard ratio describes the estimated relative effect, the confidence interval describes uncertainty, and the P-value describes evidence under the prespecified statistical comparison.
22. Related Tutorials
Learn more about the methods used in this trial:
23. Related Calculators
24. Sources
- ClinicalTrials.gov: AGILE — NCT03173248.
- Linked publication: PubMed record for PMID 40706052: PMID 40706052.
- Linked publication: PubMed record for PMID 38010220: PMID 38010220.
Continue with the statistical methods
Explore the underlying survival-analysis, clinical-trial, and statistical-inference methods used to understand randomized time-to-event results.
25. Record Summary
AGILE provides a clear example of a randomized phase 3 time-to-event analysis. The trial enrolled 146 participants in a two-arm, parallel, triple-masked design. Its registered primary endpoint was Event-Free Survival up to Week 24, defined as time from randomization until treatment failure, relapse from remission, or death from any cause, whichever occurred first.
The registry-reported primary analysis used the Full Analysis Set, a log-rank test, and a stratified Cox proportional-hazards model with AML status and geographic region as stratification factors. The reported EFS hazard ratio was 0.33, with a 95% two-sided confidence interval of 0.16–0.69 and P = 0.0011.
The most important statistical lesson is that these quantities answer different questions. The hazard ratio describes relative event rates, the confidence interval describes uncertainty around the estimated effect, and the P-value describes statistical evidence under the specified hypothesis test. None of these quantities alone describes the complete EFS experience of individual participants.