Nearly two decades after azacitidine showed a survival benefit in higher-risk myelodysplastic syndromes (HR-MDS), no phase III trial has established a new survival-improving treatment. Several agents looked promising in early studies but did not prolong survival in larger randomized trials. One explanation is trial design. Most drug development in HR-MDS has added new agents to a hypomethylating agent (HMA) backbone and enrolled patients by prognostic categories that group biologically different diseases (Tauro and Tefferi, 2026).
The proposed alternative integrates clonal genetics, blast burden, disease tempo and bone marrow microenvironmental dysfunction to guide treatment selection, to group patients into coherent trial populations, and to explain why patients with similar risk scores respond differently to the same therapy.
Why Phase III Trials Have Failed
Azacitidine became standard treatment after the AZA-001 trial showed a survival advantage over conventional care. Many agents have since been tested in combination with azacitidine, usually against azacitidine monotherapy, and the combinations have repeatedly failed to extend overall survival.
The phase III VERONA study is a recent example. Venetoclax plus azacitidine did not improve OS in previously untreated HR-MDS, despite the activity of this combination in older or unfit patients with acute myeloid leukemia (AML). The result held in patients with at least 10% marrow blasts, a group often considered biologically closer to AML.
HR-MDS spans different clonal architectures, proliferative behavior, immune environments and causes of treatment failure, so a drug active in one subset may show little benefit in an unselected trial. The Revised International Prognostic Scoring System (IPSS-R) and the Molecular International Prognostic Scoring System (IPSS-M) estimate outcome but were not designed to predict sensitivity to individual therapies.
Blast Burden Alone Is Insufficient
The International Consensus Classification recognizes MDS/AML at 10%-19% marrow blasts, reflecting shared biology, including mutations typical of secondary myeloid disease. These patients may suit AML-directed trials, but VERONA suggests that blast burden cannot identify who benefits from an AML regimen. In the phase III STIMULUS-MDS2 trial, sabatolimab showed no overall survival benefit, but a survival signal appeared in patients with at least 10% blasts, which supports studying biologically distinct MDS/AML populations.
Disease Tempo
One patient may present with newly diagnosed HR-MDS and rapidly rising blasts. Another may have had years of stable lower-risk disease before reaching a similar blast burden. They meet the same morphological criteria but may differ in proliferation, apoptosis resistance, clonal evolution and treatment sensitivity.
Current systems rarely capture this. Follow-up relies on blood counts and clinical change, without routine serial marrows, and no reliable individual measure of blast growth exists, although longitudinal counts and computational models can estimate risk. Time from diagnosis also matters: in HR-MDS, the hazards of death and AML transformation decline among patients who survive longer, unlike the relatively stable hazards in lower-risk disease. Blast percentage is best interpreted in its genetic and temporal context. Within a molecularly specified subtype, it may indicate disease stage.
TP53-Mutated MDS
TP53 mutations are not uniform. A large analysis by Bernard and colleagues showed that the adverse phenotype is concentrated in multihit TP53 alterations, which carry markedly higher risks of leukemic transformation and death than monoallelic mutations. Classification now increasingly treats biallelic inactivation as a core feature.
TP53-mutated MDS with excess blasts and TP53-mutated AML can share molecular features and similarly poor outcomes, despite lying on opposite sides of the 20% blast threshold. In a 2026 study by Goldfinger and colleagues of low-intensity decitabine and venetoclax, most patients had biallelic TP53 alterations, and median OS was 9.9 months in MDS and 11.6 months in AML. The study was small and cannot establish equivalence
Hematopoietic Recovery and the Marrow Microenvironment
Lower-intensity AML regimens can clear blasts without restoring neutrophil or platelet counts. This state is called complete remission with incomplete hematologic recovery (CRi). In AML studies, CRi carried higher MRD rates than complete remission with count recovery, and it was associated with worse outcomes after adjustment for MRD. Clearing blasts restores blood counts only if the marrow environment can still support normal hematopoiesis.
Some treatments appear to affect recovery. In the AML LI-1 program, low-dose cytarabine combined with arginase or lenalidomide was associated with shifts in hematologic recovery. An Italian study of lenalidomide plus low-dose cytarabine found gene expression changes in angiogenesis, cell-cycle regulation and immune function. Acting on nonmalignant marrow compartments could influence recovery after cytoreduction.
This may matter most in HR-MDS, where severe cytopenias persist at relatively low blast percentages. The dominant problem there may be ineffective hematopoiesis, immune-mediated suppression and dysfunctional interaction between malignant progenitors and their niche. A treatment aimed only at proliferating blasts may leave these drivers of marrow failure untouched.
Immune Dysfunction in Higher-Risk Myelodysplastic Syndromes
Lower-risk disease often shows inflammatory activation, including inflammasome signaling, raised pro-inflammatory cytokines and immune-mediated suppression of hematopoiesis. Higher-risk disease more often shows immune exhaustion, regulatory T-cell expansion, altered dendritic-cell populations and immunosuppressive myeloid compartments. These patterns overlap and vary between patients, so they should not be treated as fixed states.
Some immunosuppressive myeloid populations depend less on BCL-2 signaling, which could limit the effect of venetoclax on the wider marrow ecosystem. This hypothesis is unproven as an explanation for VERONA. Multiparameter flow cytometry can report abnormal differentiation, malignant stem-cell populations, immune-cell composition and MRD markers from one marrow sample. Combined with genomic profiling, it could show whether a patient’s disease is sustained mainly by proliferative clones, dysfunctional hematopoiesis or both. Serial samples could show how treatment changes these compartments.
Azacitidine as the Standard Backbone
AZA-001’s comparator included heterogeneous treatments, and real-world studies have not consistently reproduced the trial’s survival. No reliable biomarker identifies who benefits. Hypomethylation is central to the drug’s pharmacology, but clinical response does not consistently follow the extent of DNA demethylation, and RNA disruption and immune activation may contribute to both efficacy and toxicity.
Azacitidine can cause major epigenetic change without consistently pushing malignant progenitors to terminal differentiation. Clonogenic cells may persist despite altered methylation, and malignant progenitors remain detectable in some patients in remission. Recent work also describes heterogeneous responses, senescence-associated states and escape mechanisms. These mechanisms may differ across molecular subtypes, so using azacitidine in nearly every experimental regimen could obscure agents that perform better alone or in other combinations.
When a new agent is added, true synergy is hard to separate from overlapping activity, altered drug exposure or added myelosuppression. Response assessment usually needs several cycles, so enrolled patients must be stable enough to tolerate prolonged treatment and cytopenias. Trial populations therefore differ from patients in routine practice.
Drug Failures in TP53-Mutated Disease
HMAs induce hematologic responses in TP53-mutated MDS, but the responses are often short-lived, and venetoclax combinations may deepen remission without consistently improving survival. Eprenetapopt, designed to restore mutant p53 function, was encouraging with azacitidine in early studies, but the randomized phase III trial missed its primary endpoint.
Magrolimab, an anti-CD47 antibody meant to enhance macrophage clearance of malignant cells, also looked promising early, and its phase III ENHANCE trial was discontinued for futility. In both, early response signals, including complete remission, did not predict clinical benefit.
Changing HMA Dosing
Goldfinger and colleagues studied once-weekly low-intensity decitabine with venetoclax in TP53-mutated MDS or AML. The schedule was developed during the COVID-19 pandemic to reduce hospital visits and myelosuppression.
The pooled analysis included 40 patients, mostly older adults: 26 with HR-MDS and 14 with AML. Most had biallelic TP53 alterations or complex cytogenetics. Patients received weekly subcutaneous decitabine 0.2 mg/kg and oral venetoclax 400 mg, adjusted for interacting drugs. Among response-evaluable MDS patients, 43% achieved complete remission and another 15% marrow complete remission. Transfusion independence occurred in 58% of initially transfusion-dependent patients, and no treatment-related deaths occurred.
Low-dose intermittent decitabine may preserve epigenetic activity while sparing functioning hematopoietic cells, which could allow venetoclax exposure with less sustained myelosuppression. The analysis was single-center, combined prospective and retrospective patients and had no randomized comparator. Severe cytopenias and infections still occurred, and the survival data cannot establish superiority over standard dosing. It suggests that optimizing dose intensity, scheduling and hematopoietic recovery may sometimes help more than adding another agent to a myelosuppressive regimen.
Pre-Transplant Cytoreduction
Allogeneic stem cell transplantation is the only potentially curative treatment for HR-MDS. Retrospective analyses report inferior survival with 5%-20% blasts at transplant versus fewer than 5%, and lower platelet counts also predict poorer outcomes.
A patient whose disease responds to chemotherapy or an HMA may have a more treatment-sensitive clone and a better evolutionary trajectory, then one who progresses on treatment. Comparing their transplant outcomes cannot separate the effect of therapy from the biology that determined response. Some retrospective comparisons found no better post-transplant survival in patients who reached complete remission with chemotherapy than in those transplanted directly, and longitudinal analyses of IPSS-R changes suggest that the apparent benefit may be explained by underlying disease behavior.
Pre-transplant treatment can also delay transplant or prevent it through toxicity, persistent cytopenias or progression. Universal cytoreduction targets are therefore questionable. TP53-mutated disease is the hardest case. A 2024 meta-analysis of TP53-mutated MDS undergoing transplant reported pooled three-year OS of about 21%, with relapse the main cause of failure. TP53 allelic state, cytogenetic complexity, molecular disease burden and treatment response may help select them.
The Marrow Environment After Transplantation
After transplant, donor hematopoiesis develops in a stromal environment that remains predominantly host-derived. Pre-existing abnormalities there could influence engraftment, immune reconstitution and persistence of malignant clones.
Integrated immunophenotypic and genomic monitoring before and after transplant could complement MRD and chimerism measurements and help refine conditioning intensity, post-transplant interventions and maintenance. In TP53-mutated disease, where relapse is frequent, it could help separate persistent malignant clones from broader marrow abnormalities. The clinical value is unproven, and dedicated analyses in MDS and MDS/AML are needed because many relevant studies pool them with AML.
Trials Designed on Therapeutic Biology
A biologically enriched trial would select by molecular alterations, abnormal stem-cell populations, immune characteristics or other validated predictors of response, with prespecified stratification by TP53 allelic state for agents aimed at p53 dysfunction, DNA-damage response or immune escape.
Specimens from trials of agents that failed in phase III could be revisited. Integrated genomic, transcriptomic, methylation, proteomic and immunophenotypic data could identify features associated with early response, which could then be tested in archived samples. A negative phase III result shows that a regimen did not achieve the required benefit in the enrolled population. It does not exclude biologically restricted activity, as may be the case for eprenetapopt and magrolimab. Any retrospective signal needs independent validation and prospective testing.
AML experience with signal-seeking and ‘pick-a-winner’ designs shows that they shorten investigation of regimens with insufficient activity. In HR-MDS, time from diagnosis should enter eligibility criteria and analyses. The slower course of some patients may allow molecular characterization before allocation, including when they are considered for AML trials.
Remission duration, transfusion independence, time outside the hospital, treatment complications and successful transition to transplant matter especially in older or frail patients. In TP53-inactivated disease, molecular response and survival may be more informative than complete remission. Intermediate endpoints used to decide progression to larger trials need prospective validation.
References
- Tauro S, Tefferi A. Recovering from a therapeutic stall in higher-risk myelodysplastic syndromes: re-examining biology, backbones and study designs. Haematologica. 2026;111(11). https://doi.org/10.3324/haematol.2026.300560
- Goldfinger M, Mantzaris I, Shastri A, et al. Treatment of TP53-mutated myelodysplastic syndrome and acute myeloid leukemia with low-intensity metronomic decitabine and venetoclax. Haematologica. 2026;111(10). https://doi.org/10.3324/haematol.2026.300534
- Bernard E, et al. Implications of TP53 allelic state for genome stability, clinical presentation and outcomes in myelodysplastic syndromes. Nature Medicine. 2020. https://doi.org/10.1038/s41591-020-1008-z
