All content on this site is intended for healthcare professionals only. By acknowledging this message and accessing the information on this website you are confirming that you are a Healthcare Professional. If you are a patient or carer, please visit the MDS Alliance.

The MDS Hub uses cookies on this website. They help us give you the best online experience. By continuing to use our website without changing your cookie settings, you agree to our use of cookies in accordance with our updated Cookie Policy
  TRANSLATE

The MDS Hub website uses a third-party service provided by Google that dynamically translates web content. Translations are machine generated, so may not be an exact or complete translation, and the MDS Hub cannot guarantee the accuracy of translated content. The MDS Hub and its employees will not be liable for any direct, indirect, or consequential damages (even if foreseeable) resulting from use of the Google Translate feature. For further support with Google Translate, visit Google Translate Help.

Steering CommitteeAbout UsNewsletterContact

New IPSS-Molecular model for MDS risk stratification and prognosis

Jul 12, 2022
Share:
Learning objective: After reading this article, the reader will be able to recall the top predictors of poor outcomes in MDS using the IPSS-M tool.

The current International Prognostic Scoring System-Revised (IPSS-R) used for risk stratification and therapeutic decision-making in myelodysplastic syndromes (MDS) only considers hematologic parameters and cytogenetic abnormalities.1 Somatic gene mutations are not yet used as part of risk stratification, although they have shown to be a valuable prognostic biomarker.

Bernard, et al.1 evaluated the combined prognostic power of hematologic, cytogenetic, and somatic gene mutation parameters in the development and use of a clinical-molecular prognostic model, called IPSS-Molecular (IPSS-M). We summarize the key findings below.

Study design

The discovery cohort had a total of 2,957 representative MDS samples from 24 centers. Each sample was subjected to the following inclusion criteria:

  • Samples must have been collected before treatment
  • Blasts <20%
  • White blood cell count below 13 × 109/L

A total of 234 patients diagnosed with secondary MDS/therapy-related MDS, and 370 patients diagnosed with MDS overlap syndromes were included. All patients were treatment-naïve and were not selected on the basis of prior therapies received. The cytogenetic annotation was derived from a conventional G-banding analysis, and a total of 152 genes implicated in myeloid neoplasms were characterized by targeted sequencing.

Results

The genomic landscape of the discovery cohort was made up of:

  • 3,186 cytogenetic alterations in 41% of patients
  • 9,254 oncogenic mutations across 121 genes in 90% of patients
  • 94% of patients had at least one molecular abnormality
  • 53% of patients had gene mutations only
  • 4% of patients had cytogenetic alterations only
  • 37% of patients had both

SF3B1, TET2 and del(5q) were enriched in patients with only one driver event, while the number of abnormalities correlated with disease severity.

Gene mutations and clinical endpoints

A total of 48 genes mutated in 1% of patients were evaluated for correlations with three primary endpoints:

  • 14 genes were significantly associated with leukemia-free survival
  • 16 genes were significantly associated with overall survival (OS)
  • 15 genes were significantly associated with acute myeloid leukemia (AML) transformation

TP53multihit, FLT3 mutations, and KMT2A (MLL) partial tandem duplication (PTD) ranked as top predictors of poor outcomes, with the latter two mutations recording the strongest hazard ratios for AML transformation. Moreover, mutations in ASXL1, BCOR, EZH2, NRAS, RUNX1, STAG2, and U2AF1 were significantly associated with poor risk for the three endpoints. MLLPTD is not normally included in panel tests; however, it was identified in 2.5% of patients, and correlated with excess blasts, a high risk of AML transformation, and poor OS. This highlighted the importance of identifying the TP53 allelic state and MLLPTD for accurate risk verification.

Mutations in DDX41 were associated with high blast percentages, a risk of AML transformation, and conversely, with favorable OS, particularly in MDS with excess blasts. There were 141 mutations among 90 patients. On the other hand, SF3B1 mutations were associated with favorable outcomes across all clinical endpoints but was strongly adjusted by patterns of comutation. Cluster analysis separated patients with SF3B1 mutations into three groups: SF3B15q, SF3B1β and SF3B1α. Only patients with SF3B1α showed favorable outcomes.

Feature encoding for the IPSS-M model development

Subsequently, the optimal encoding of clinical, cytogenetic, and genetic features was evaluated for developing the model. The hemoglobin level, marrow blast percentage, and platelet count limit at 250 × 109/L were encoded linearly as continuous variables. The neutrophil count had a small weight in the IPSS-R model but was not independently prognostic, so it was excluded. The IPSS-R cytogenetic categories were maintained but the incorporation of gene mutations resulted in a notable increase in the model discrimination. These were, therefore, encoded as binary variables, except for TP53 and SF3B1.

Feature selection for the IPSS-M model development

Variables with a probability of inclusion greater than 0.7 in at least one endpoint were considered for entering the model. Confounding variables were strongly selected for OS but not for AML transformation, and the same was observed for the hemoglobin level and platelet count. Marrow blasts and IPSS-R cytogenetic categories were consistently selected across all endpoints. This resulted in a model with four categories:

  1. Hemoglobin level, platelet count, and marrow blasts
  2. IPSS-R cytogenetics
  3. 17 binary features from 16 prognostic genes
  4. Number of mutated genes from a residual group of 15 genes

Risk score and risk categories

The IPSS-M score was built as a weighted sum of prognostic variables, resulting in a patient-specific risk score. The score was scaled so that “0” corresponded to an average patient, with mean values for all variables. A score of −1, 1, or 2 equaled half-, double-, or a four-fold risk compared to an average patient, respectively. A six-category risk schema was subsequently generated:

  1. Very low
  2. Low
  3. Moderate low
  4. Moderate high
  5. High
  6. Very high

A strong prognostic separation was recorded across all endpoints.

Re-stratification from the IPSS-R to IPSS-M

The number of categories associated with the IPSS-R and IPSS-M models was standardized and achieved through combining the categories “moderate-low” and “moderate-high” into “moderate” in the IPSS-M model, which resulted in improved separation across all endpoints, and in the re-stratification of 46% of patients. Among re-stratified patients, 294 (24%) had one mutated gene in the IPSS-M, whereas 760 (62%) had two or more. Therefore, the re-stratification of patients was not due to a single gene effect, but the cumulative contribution of the prognostic mutations for each patient.

IPSS-M in secondary MDS/therapy-related MDS

A total of 234 patients from the discovery cohort were diagnosed with secondary MDS/therapy-related MDS. All were enriched for complex karyotype, TP53multihit, PPM1D, and SETBP1. No differences in cytogenetic alterations were observed between primary and secondary MDS/therapy-related MDS. However, patients were more frequently categorized as IPSS-M high, or very high risk. Comparatively, 39% were also categorized as very low, low, or moderate low.

IPSS-M and treatment

The IPSS-M model showed improved prognostic discrimination in both untreated and treated patient populations compared to IPSS-R in all clinical endpoints. Associations between gene mutations and treatment modalities, and their influence on the IPSS-M model, were also evaluated, which showed that TP53multihit remained the strongest predictor of worse outcomes across all treatments. In contrast, mutated DDX41 was associated with a favorable OS after hypomethylating agents, although it was a predictor of AML transformation.

IPSS-M validation

To validate the IPSS-M model, 754 patients from the Japanese MDS consortium (J-MDS) cohort were tested. The risk scores generated were higher for the J-MDS cohort compared to the discovery cohort, recording a median of 0.79 and −0.38, respectively. In the J-MDS cohort, the discriminative power of the IPSS-M score was superior to the IPSS-R, and the IPSS-M risk categories resulted in distinct hazard ratios and time estimates across all endpoints.

Web calculator

To aid in the clinical uptake, a web-based calculator was also developed in partnership with the MDS Foundation.1 The calculator generates a unique IPSS-M score and its associated category and time estimates for leukemia-free survival, OS, and AML transformation, based on a patient’s individual hematologic, cytogenetic, and molecular profile. The calculator can be accessed and explored in more detail here.

Conclusion

Risk stratification is of crucial importance in a patient-centered approach to the treatment of MDS. The IPSS-M model is personalized, interpretable, flexible, and provides improved prognostic accuracy across all long-term clinical endpoints. Furthermore, it was able to re-stratify nearly half of patients with MDS, and was applicable to those with secondary/therapy-related disease.

  1. Bernard E, Tuechler H, Greenberg PL, et al. Molecular International Prognostic Scoring System for myelodysplastic syndromes. NEJM Evidence. 2022;1(7). DOI: 10.1056/evidoa2200008

More about...