AML Approval Endpoints at FDA and EMA: Complete Remission, Event-Free Survival and Overall Survival in Pivotal Trials

AML approval endpoints at FDA and EMA: complete remission, event-free survival and overall survival in pivotal trials

Acute myeloid leukemia (AML) approvals rest on three main efficacy endpoint families: overall survival (OS), event-free survival (EFS), and remission-based endpoints (complete remission (CR), CR with partial hematologic recovery (CRh), and their composites). Which endpoint a regulator accepts depends mostly on the treatment intent (curative, noncurative or palliative), whether the product is added to intensive chemotherapy, and whether a randomized comparison is feasible. This article summarizes the current FDA and EMA framework and the pivotal trial precedents behind US and EU AML approvals, refusals and withdrawals.

Key takeaways

• FDA's 2022 final AML guidance links endpoints to treatment intent. For curative-intent therapy, OS, EFS and relapse-free survival (RFS) support traditional approval, and FDA expects a randomized control arm rather than historical controls ⁵⁰¹⁵⁰². For noncurative therapy, OS, EFS and durable CR are acceptable, and single-arm trials may be possible ⁵⁰³. For palliative-intent therapy, durable CR and durable CR/CRh with transfusion independence are acceptable clinical-benefit endpoints ⁵⁰⁴.
• EMA takes a more conservative line on remission endpoints. The CHMP anticancer guideline accepts CR as a primary endpoint mainly where later consolidation or maintenance could confound EFS, and it expects CR results to be supported by EFS or PFS data ⁶⁸. Superiority on EFS, PFS or OS is generally expected when toxicity increases ⁷⁷.
• Randomized, OS-driven trials dominate newly diagnosed and maintenance approvals in both regions. Examples are midostaurin (RATIFY), quizartinib (QuANTUM-First), CPX-351, glasdegib and venetoclax (VIALE-A), plus oral azacitidine (QUAZAR) in the maintenance setting ^{145174176151603123}.
• EFS has supported approval in both jurisdictions, for gemtuzumab ozogamicin (ALFA-0701) and ivosidenib plus azacitidine (AGILE) ^159205241395.
• Single-arm CR/CRh trials diverge sharply between the agencies. In relapsed/refractory (R/R), molecularly targeted settings, FDA granted regular approval on single-arm CR/CRh data to enasidenib, ivosidenib, olutasidenib and ziftomenib ^704674627645. Gilteritinib's November 2018 regular approval used an interim CR/CRh rate from the randomized ADMIRAL trial ⁶³⁶. EMA refused a single-arm, CR-based gemtuzumab ozogamicin application, and the single-arm ivosidenib R/R application was withdrawn ²⁴⁶²⁷.

The regulatory framework

FDA: Acute Myeloid Leukemia guidance (final, October 2022)

FDA's Acute Myeloid Leukemia: Developing Drugs and Biological Products for Treatment was finalized on October 17, 2022 ⁵⁴⁸. It sets out the following endpoint definitions:

• CR: marrow blasts <5% by morphology, ANC >1 Gi/L, platelets >100 Gi/L, no peripheral blood leukemic blasts, and no extramedullary disease ⁵⁶⁷. The CR date is the marrow sampling date, and the protocol should set a maximum window between marrow and blood tests (3 to 7 days may be justifiable) ⁵⁶⁸. Missing data count as failure to achieve CR ⁵⁶⁹.
• CRh: the same marrow, peripheral blast and extramedullary criteria, but with ANC >0.5 Gi/L and platelets >50 Gi/L, without meeting CR count-recovery criteria ⁵⁷⁰⁵⁷¹. CRh applies to relatively nontoxic, nonmyelosuppressive drugs that may be used for palliation. In that setting the endpoint is CR+CRh, with follow-up long enough to establish meaningful durability ⁵⁷⁴⁵⁷⁵.
• Duration of CR: time from CR to hematological relapse or death from any cause, with enough follow-up to show the remission is meaningfully durable ⁵⁷³.
• Transfusion independence (TI): applicable to relatively nontoxic, nonmyelosuppressive drugs and should be supported by evidence of antileukemic activity ¹⁵. FDA asks for prespecified transfusion rules, separate analyses for patients who were transfusion-dependent and transfusion-independent at baseline, and separate red cell and platelet TI analyses ¹⁶.
• OS and EFS: OS is measured from randomization to death from any cause ⁹. EFS counts induction treatment failure, relapse after induction success, or death, whichever occurs first ¹⁰¹¹.

For binary endpoints, the analysis set is all treated patients in single-arm trials and all randomized patients in randomized trials, with proportions reported with 95% confidence intervals ¹⁷. For add-on designs, a placebo comparator can be considered when the control arm receives appropriate treatment (for example, study drug plus 7+3 versus placebo plus 7+3) ⁵⁰⁵. Best supportive care is a possible randomized comparator in the palliative setting only when patients have no available therapies. A single-arm trial may be appropriate when adequate historical data support the null hypothesis ⁵⁰⁴.

FDA's separate guidance on minimal residual disease (MRD) in hematologic malignancies describes MRD as a potential response measure, prognostic tool or trial-allocation biomarker ⁶. It does not set out a standalone AML MRD efficacy endpoint.

FDA review documents add context on how the agency weighs these endpoints. The Division describes EFS and OS as direct measures of clinical benefit that have been used for regular approval ²⁶⁸²⁷⁷. It describes durable CR as reasonably likely to predict long-term benefit ²⁶⁸²⁶⁹. In the noncurative, palliative setting, it states that durable CR+CRh has been used for regular approval of nonmyelosuppressive, reasonably nontoxic drugs ²⁷¹²⁶⁸. TI is treated as corroborating evidence. In the quizartinib R/R review, FDA stated that TI alone could not establish benefit ²⁷⁶.

EMA: anticancer guideline and the haematological malignancies appendix

EMA has no AML-specific guideline. The relevant recommendations are in the CHMP Guideline on the evaluation of anticancer medicinal products in man, under the section on treatment administered with curative intent ^70717273. Revision 6 repeats the EFS definition. It also notes that treatment complexity in AML can make it hard to show an effect on cure rate or survival ⁶⁷.

• EFS: events are failure to achieve CR during the prespecified induction phase (assigned an event at time zero), relapse, and death without relapse. Analysis of a primary EFS endpoint should wait until the data are sufficiently mature ⁷⁴⁷⁵.
• CR as primary endpoint: acceptable when reduced or similar toxicity is expected and consolidation or maintenance could confound EFS/PFS. CR should be supported by EFS or PFS data and assessed with established criteria. A CR plus partial response composite needs specific justification ⁶⁸.
• Design by toxicity profile:
  • Reduced or similar toxicity: substitution non-inferiority designs are acceptable, usually with an EFS or PFS primary endpoint ⁷⁶.
  • Increased toxicity: superiority in EFS, PFS or OS is generally expected. A major increase in post-induction CR with favourable time-to-event trends may suffice in small populations ⁷⁷.
  • Major increase in toxicity: increased cure rate or improved OS is expected ⁷⁸.
• MRD: molecular MRD is supportive evidence for CR and should be used only after intra- and inter-laboratory validation ⁷¹⁷⁹. The draft haematological malignancies appendix treats MRD-defined response as supportive evidence in very small target populations. It also calls for justified eligibility criteria, response criteria and reference therapy in confirmatory studies ⁸⁰⁶⁹.

Overall survival as the primary endpoint

OS is the most common primary endpoint in randomized AML registration trials at both agencies.

OS trials that missed statistical significance

EMA precedents show that CHMP has accepted clinically meaningful OS gains even when the primary test was not formally met.

• Decitabine (Dacogen), DACO-016. This open-label phase 3 trial compared decitabine with patient's choice of supportive care or low-dose cytarabine. The primary OS analysis gave HR 0.85 (95% CI 0.69 to 1.04; p=0.1079), which was not statistically significant. With an additional year of follow-up, the HR was 0.82 (nominal p=0.0373) ^{329330332333334}. CHMP considered the 2.7-month median survival gain clinically meaningful in this setting, while noting that significance was reached only in a second post hoc analysis ³³⁵.
• Azacitidine (Vidaza), AZA-AML-001. In patients aged 65 or older with >30% marrow blasts, the OS comparison (median 10.4 vs 6.5 months, HR 0.85) did not reach its predefined significance level (p=0.1009). CHMP considered the endpoints, design and population appropriate ^{360361362363364}.
• Venetoclax + LDAC (VIALE-C / M16-043). The primary OS analysis did not meet its objective (HR 0.749; p=0.114). Higher CR and CR+CRi rates were therefore descriptive only ^524525526. An EMA expert group found the observed antitumor activity insufficient to formally establish an OS difference ⁵¹¹.

Event-free survival as the primary endpoint

Gemtuzumab ozogamicin, ALFA-0701. Both agencies accepted EFS from this randomized, open-label phase 3 trial. It added gemtuzumab ozogamicin to daunorubicin/cytarabine in 271 patients aged 50 to 70 with newly diagnosed de novo AML ^158238239. Median EFS was 17.3 vs 9.5 months (HR 0.56) ^159241242. CHMP considered EFS clinically meaningful. It noted that the benefit came from longer remissions, not more CRs, that the OS effect was less clear, and that efficacy in younger patients was extrapolated ²⁴³²⁴⁴. In the pediatric AAML0531 study, FDA labeling describes EFS (HR 0.84) as supportive evidence ¹⁶¹.

Ivosidenib + azacitidine, AGILE (AG120-C-009). This randomized, double-blind, placebo-controlled trial enrolled 146 adults with untreated IDH1-mutated AML who were ineligible for intensive induction ³⁹³³⁹⁴. EFS was defined with treatment failure as failure to achieve CR by week 24. Median EFS was 22.9 vs 4.1 months (HR 0.39), and median OS was 24.0 vs 7.9 months (HR 0.44) ^395396397. FDA labeling reports the EFS HR as 0.35 (95% CI 0.17 to 0.72), from an analysis with a different cutoff ^206400401. CHMP raised design concerns that sponsors should note:

• the primary endpoint was changed from OS to EFS during the trial;
• the planned sample size was reduced;
• the trial was stopped early, in an unplanned way, using unblinded efficacy data.

CHMP concluded that type I error could not be formally controlled. It nevertheless acknowledged strong results supported by sensitivity analyses ³⁹⁸³⁹⁹.

Leukemia-free survival has also served as a primary endpoint in remission-maintenance settings, for example in the EU assessment of histamine dihydrochloride with interleukin-2 (Ceplene) ⁴¹⁴³.

CR and CR+CRh as the approval endpoint

FDA: regular approval on durable remission

In R/R AML with a defined molecular target, FDA has repeatedly granted regular approval on durable CR or CR/CRh. Most of these approvals rested on single-arm trials; gilteritinib's used an interim CR/CRh analysis from the randomized ADMIRAL trial. Rate and duration were the efficacy basis. Transfusion independence was supportive when shown, and was not required for revumenib ²⁷⁵.

Ivosidenib monotherapy in newly diagnosed patients was also labeled on a single-arm CR+CRh rate (12/28, 42.9%) ^209210211.

EMA: single-arm remission data are scrutinized more closely

EU outcomes for single-arm CR-based applications have been less favorable:

• Gemtuzumab ozogamicin, first-relapse application (refused). The application rested on three single-arm trials with a CR primary endpoint (CR 13%, plus 13% CRp) ²⁴⁵. CHMP judged the CR rate modest and reliable data on valid clinical endpoints lacking. It stated that a full application should have relied on randomized controlled trials ²⁴⁶.
• Ivosidenib, R/R application (withdrawn). The pivotal study was single-arm with a CR+CRh primary endpoint ³⁸⁴³⁸⁸. CHMP advice had stressed the need for a patient-level matched historical control. CHMP later questioned whether an OS difference against such a control (HR 0.396) was caused by treatment, given likely residual confounding. It also noted that response-based OS comparisons suffer from immortal-time bias ^389390391.
• CPX-351, pediatric relapsed AML (accepted). CHMP accepted the single uncontrolled AAML1421 study, with a CR+CRp primary endpoint (68% response), as adequate for the pediatric relapsed indication. It acknowledged the limitation of a single uncontrolled study ^296297298299.

Accelerated approval and confirmatory trials at FDA

Two FDA precedents show how accelerated approval has been used in AML:

• Venetoclax. The AML indication received accelerated approval on November 21, 2018. FDA required the final report from phase 3 trial M16-043 (venetoclax plus LDAC vs LDAC), with OS as the primary endpoint ⁴⁷⁷⁴⁷⁸. Traditional approval followed on October 16, 2020, primarily on confirmatory OS from VIALE-A. VIALE-C was also reviewed but had not met its primary OS endpoint ⁴⁷⁹⁴⁸⁰.
• Gemtuzumab ozogamicin. Accelerated approval was granted on May 17, 2000, in first-relapse AML. The required confirmatory trial (SWOG S0106) failed to verify clinical benefit, and the product was withdrawn in 2010. It returned with regular approval on September 1, 2017, based on new data in newly diagnosed and R/R AML ^660661662.

Negative and borderline precedents

• Quizartinib, R/R FLT3-ITD AML (refused by EMA). AC220-007 was an open-label, randomized OS trial against salvage chemotherapy ⁴²³⁴²⁴. CHMP called the primary OS result borderline positive. It found efficacy not established because missing data and non-administrative censoring, mostly in the open-label control arm, undermined the result. EFS and CR did not support the OS finding ⁴²⁵⁴²⁶.
• Vosaroxin (Qinprezo, withdrawn). The EU application was based on an adaptive-design, placebo-comparator R/R AML study with an OS primary endpoint ⁵⁶⁵⁸.

FDA and EMA at a glance

Practical implications for AML development programs

• Frontline intensive or add-on programs should plan a randomized trial with an OS or EFS primary endpoint. Placebo-controlled add-on designs on standard chemotherapy are accepted by both agencies, as RATIFY and QuANTUM-First show ^{501505145174224414}.
• Protocol integrity matters at EMA. Changing the primary endpoint mid-trial, reducing the sample size, or stopping early on unblinded data can undermine an otherwise strong EFS result ³⁹⁸³⁹⁹. Heavy censoring in an open-label control arm can undermine an OS result ⁴²⁶.
• Single-arm CR/CRh programs in biomarker-defined R/R populations have a well-established FDA regular-approval pathway. Sponsors still need to show durability and supportive TI data, and FDA adjudication of response can lower the observed rate relative to the prespecified benchmark ²⁷¹⁵⁴⁵. In the EU, sponsors should expect to justify single-arm evidence with a patient-level external control and to address confounding and immortal-time bias ^389390391.
• Remission data alone rarely carry an EU application. CHMP repeatedly asks whether CR translates into EFS or OS, and has framed CR, CRi and MRD as supportive rather than decisive ^68243420488.