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

A personalized prognostic prediction model based on genomic features in patients with MDS

Dec 20, 2021
Share:

Myelodysplastic syndromes (MDS) have a disease spectrum ranging from indolent conditions to rapidly progressing acute myeloid leukemia (AML).1 The World Health Organization (WHO) disease classification for MDS uses morphological features to define MDS categories, leading to clinical overlap between subtypes. The International Prognostic Scoring System (IPSS) and revised-IPSS (R-IPSS) are excellent disease-related risk tools but have their own weaknesses and, at times, fail to capture reliable patient-level prognostic features, particularly cytogenetics. More recently, disease classifications based on the clinical and morphological criteria are being complemented by genomic features in myeloid malignancies.1

We summarize here a study by Bersanelli et al.1 published in the Journal of Clinical Oncology, defining a new genomic classification for MDS to improve individual prognostic assessment in patients with MDS.

Study design and methods

This was an international, retrospective cohort study in 2,043 patients (EuroMDS cohort) with primary MDS according to the 2016 WHO criteria. An independent, prospective cohort (Humanitas cohort) of 318 patients diagnosed with MDS at the Humanitas Research Hospital, Milan, IT, was included for validation of personalized prognostic assessment using the Random-effects Cox multistate (CoxRFX) model.

The methods utilized included the following:

  • Standard G-banding for cytogenetics analysis; karyotypes were classified using the International System for Cytogenetic Nomenclature Criteria
  • Mutation screening of 47 genes related to myeloid neoplasms was performed on DNA from peripheral blood granulocytes or bone marrow mononuclear cells
  • Bradley–Terry models were used to estimate the timing of mutation acquisition and to assess their prognostic value
  • Bayesian network analysis and hierarchical Dirichlet processes to identify genomic association and subgroups
  • CoxRFX modelling for developing innovative prognostic tools

Results

Genomic landscape in MDS

  • 59% of patients showed a normal karyotype, 32% of patients showed chromosomal abnormalities, and 80% of patients presented with one more mutation, with a median of two mutations (range, 1–17).
  • Six genes were mutated in >10% of patients, with an additional five genes mutated in 5–10% of patients; 36 genes were mutated in <5% of patients.

Mutation acquisition and prognostic value

  • 14 genes were associated with a worse prognosis if mutated; whereas, one gene, SF3B1, was associated with better outcome.
  • Clonal mutations were seen in 58% of patients, and both clonal and subclonal mutations were seen in 42% of patients, emphasizing the importance of including subclonal mutations in the model.

Genomic associations and subgroups in MDS

  • Mutually exclusive mutation patterns included SF3B1 with TP53 and DNMT3A with ASXL1 mutations.
  • Co-occurring mutation patterns included SF3B1 with JAK/STAT pathway mutations; SRSF2 with TET2, ASXL1, CBL, IDH1/2, RUNX1, and STAG2; U2AF1 with abnormalities of chromosomes 7 and 20 and NRAS, TET2 with SRSF2 and ZRSR2; and DNMT3A with BCOR, IDH1, and NPM1 mutations.
  • Although 5q deletion co-existed with TP53 mutations and with several single cytogenetic components of complex karyotype, it was frequently present as a single genomic abnormality.

Definition of a genomic classification of MDS

Eight genomic subgroups (Group 0–7) among patients with MDS were identified using Dirichlet processes (Table 1): seven groups were deeply characterized by a single component; one group included patients without genomic profiles (Group 0).

Table 1. Classification of the eight genomic subgroups using Dirichlet processes*

Genomic subgroup

Dominant characteristics

Group 0

Younger age, isolated anemia, normal or reduced marrow cellularity, absence of ring sideroblasts, low percentage of marrow blasts (median, 2%)

Group 1

SF3B1 with co-existing mutations (ASXL1 and RUNX1) characterized by anemia associated with mild neutropenia and thrombocytopenia, multilineage dysplasia, and higher marrow blast percentage with respect to Group 6 (7% vs 2%; p < 0.0001)

Group 2

TP53 mutations and/or a complex karyotype; most patients in this group showed two or more cytopenias and excess blasts

Group 3

SRSF2 and related TET2 mutations, presenting with cytopenia and higher monocyte absolute count compared with other groups (p < 0.0001) along with multilineage dysplasia, and excess blasts (median, 8%)

Group 4

U2AF1 mutations associated with 20q deletion and/or chromosome 7 abnormalities; patients had higher rate of transfusion-dependent anemia compared with other groups (p 0.023 to <0.0001)

Group 5

SRSF2 co-occurring with ASXL1, RUNX1, IDH2, and EZH2, presenting with two or more cytopenias, multilineage dysplasia, and significantly higher excess blasts compared with Group 3 (11% vs 8%; p = 0.0031)

Group 6

Ring sideroblasts and isolated SF3B1 mutations characterized by isolated anemia, normal or high platelet count, single or multilineage dysplasia, and low percentage of marrow blasts (median, 2%)

Group 7

AML-like mutation patterns (DNMT3A, NPM1, FLT3, IDH1, RUNX1); two or more cytopenias and excess blasts (83% of patients) ranging from 15–19%

*Data from Bersanelli et al.1
Dominant SF3B1 mutations.
Dominant SRSF2 mutations.

The classification demonstrated genomic heterogeneity of patients stratified by WHO criteria.

Clinical relevance of genomic classification in predicting survival

  • Group 1 and 6 showed improved survival compared with other groups (p < 0.0001 to 0.0093).
    • Isolated SF3B1 (Group 6) was associated with better outcomes compared with SF3B1 with co-mutations (Group 1; p = 0.0304).
    • Group 0 was associated with good prognosis compared to Group 2, 3, 4, 5, and 7 (p < 0.0001 to 0.012).
    • Patients with isolated 5q deletion with none or single mutation were associated with better prognosis compared to those with two or more or TP53 mutations (p = 0.0432).
    • Group 0, 1, 6, and 7 showed improved posttransplantation outcomes (Table 2).
  • Groups with splicing mutations other than SF3B1 (including Group 5) were associated with poorer outcomes (p < 0.001 to .0177 with respect to Groups 0, 1, 4, and 6).
    • Group 2 had the poorest outcomes (p < 0.0001 to 0.0473).
    • Group 7 showed high rates of leukemic evolution as well as worse prognosis compared with Group 1, 3, and 6 (p < 0.0001).
    • A high rate of transplant failure was observed in Group 2 and 4 (Table 2).

Table 2. Comparison of probability of survival among different genomic-based groups*

MDS classification

p values

Group
0

Groups
1 and 6

Group
2

Groups
3 and 5

Group
4

Group
7

MDS without specific genomic profiles (Group 0)

0.49

0.0016

0.0196

0.0019

0.97

SF3B1-related MDS (Groups 1 and 6)

0.49

0.0003

0.0035

0.0004

0.62

MDS with TP53 mutations and/or complex karyotype (Group 2)

0.0016

0.0003

0.24

0.73

0.0019

SRSF2-related MDS (Groups 3 and 5)

0.0196

0.0035

0.24

0.17

0.0304

MDS with U2AF1 mutations associated with deletion of chromosome 20q and/or abnormalities of chromosome 7 (Group 4)

0.0019

0.0004

0.73

0.17

0.0019

MDS with AML-like mutation patterns (Group 7)

0.97

0.62

0.0019

0.0304

0.0019

AML, acute myeloid leukemia.
*Adapted from Bersanelli et al.1

Personalized prognostic assessment

  • The CoxRFX model incorporated 63 clinical and genomic variables to estimate personalized probability of survival.
  • Factors attributing to high predictive prognostic indicator were demographic features (age and sex) and gene mutations along with co-mutation patterns.
    • Clinical features continued to be independent predictive variables of survival.
  • Internal cross validation of the CoxRFX model showed a concordance of 0.74 and 0.71 for survival in training (67% of patients) and test (33% of patients) EuroMDS cohort subsets, respectively (Table 3).
  • The concordance of Dirichlet process components was similar to that of age-adjusted IPSS-R, highlighting the importance of including genomic features in prognostic model.
  • Concordance was also similar in the EuroMDS and the validation Humanitas cohort (0.74 and 0.75, respectively) (Table 3).

Table 3. Concordance comparison*

Statistical model and variable selection

Training (66% of EuroMDS patients)
Concordance

Test (33% of EuoMDS patients)
Concordance

Cytogenetics IPSS-R risk groups

0.576

0.567

Age-adjusted IPSS-R risk groups

0.620

0.659

Dirichlet processes

0.649

0.629

CoxRFX_Clinical + demographics + Dirichelet processes

0.729

0.713

CoxRFX_Clinical + demographics + genomics

0.742

0.709

 

Training
(EuroMDS Cohort)

Validation
(Humanitas Cohort)

CoxRFX_Clinical + demographics + Dirichelet processes

0.715

NA

CoxRFX_Clinical + demographics + genomics

0.737

0.753

CoxRFS, Random-effects Cox proportional hazards multistate model; IPSS-R, revised International Prognostic Scoring System; NA, not applicable.
*Adapted from Bersanelli et al.1

Conclusion

This retrospective genomic analysis demonstrated distinct subgroups associated with specific clinical features and discrete patterns of evolution, providing early evidence for next-generation disease classification and prognosis. While conventional prognostic models are based on the median survival of patients, this novel prognostic model was based on individual patient genotype and phenotype, therefore improving the ability to capture prognostic information in a heterogeneous disease. The findings suggest that integration of clinical features with genomic profiling in patients with MDS may provide personalized prognostic predictions to individual patients. Genomic features were also relevant in predicting survival after transplantation and thus should be included in transplantation decision-making. Further research is needed to improve the reliability and generalizability of these prediction models.

  1. Bersanelli M, Travaglino E, Meggendorfer M, et al. Classification and personalized prognostic assessment on the basis of clinical and genomic features in myelodysplastic syndromes. J Clin Oncol. 2021;39(11):1223-1233. DOI: 1200/JCO.20.01659

Share:
More about...