GENOMED4ALL: Improving MDS Classification and Prognosis by AI
- Conditions
- Myelodysplastic Syndromes
- Registration Number
- NCT04889729
- Lead Sponsor
- Istituto Clinico Humanitas
- Brief Summary
Myelodysplastic syndromes (MDS) typically occur in elderly people. Current disese classifcation system and prognostic scores (International Prognostic Scoring System, IPSS) present limitations and in most cases fail to capture reliable prognostic information at individual level. Study of MDS has been rapidly transformed by genome characterization and there is increasing evidence that mutation screening may add significant information to currently available prognostic scores. The project will aim to develop artificial intelligence (AI)-based solutions to improve MDS classification and prognostication, through the implementation of a personalized medicine approach. In close collaboration with the European Reference Network on Rare Hematological Diseases (ERN-EuroBloodNet, FPA 739541), GENOMED4ALL involves multiple clinical partners from the network, while leveraging on healthcare information and repositories that will be gathered incorporating interoperability standards as promoted by ERN-EuroBloodNet central registry, the European Rare Blood Disorders Platform (ENROL, GA 947670).
- Detailed Description
Myelodysplastic syndromes (MDS) typically occur in elderly people. Patients present peripheral blood cytopenia, and with time a portion of these subjects evolve into acute myeloid leukaemia (AML). The natural history of MDS is heterogeneous ranging from conditions with a near-normal life expectancy to forms close to AML, and therefore a risk-adapted treatment strategy is mandatory. Current prognostic scores (Revised International Prognostic Scoring System, IPSS-R) present limitations, and in most cases fail to capture reliable prognostic information at individual level.
Study of MDS has been rapidly transformed by genome characterization. Somatic mutations occur in the genomes of hematopoietic stem cells at a low, but detectable frequency during normal DNA replication. Any genetic alteration that causes a selective advantage relative to other self-renewing cells will lead to clonal dominance (clonal haematopoiesis, CH). The consequence of CH is genomic instability leading to increased risk of acquiring additional mutations and to develop MDS, solid cancer and other illnesses. The time and place of individual mutations and their clonal emergence during the course of the disease are central issues for a better comprehension of MDS pathogenesis and phenotype and for the development of cancer preventive strategies.
Important steps forward have been made in defining the molecular architecture of MDS. The MDS associated with 5q deletion derives from the haploinsufficiency of RPS14 gene. Genes encoding for spliceosome components were identified in a high proportion of subjects with MDS. There is a close relationship between ring sideroblasts and SF3B1 mutations, which is consistent with a causal relationship. In addition, an increasing number of genes have been found to carry recurrent mutations in MDS, involved in DNA methylation (DNMT3A, TET2, IDH1/2), chromatin modification (EZH2, ASXL1), transcriptional regulation (RUNX1), signal transduction (KRAS, CBL).
Gene mutations have been reported to influence survival and risk of disease progression in MDS, and the evaluation of the mutation status may add significant information to currently used prognostic scores. For instance, we found that SF3B1 mutations were independent predictors of favorable prognosis, while driver mutations of ASXL1, SRSF2, RUNX1, TP53 and EZH2 genes were associated with a reduced probability of survival. MDS with ring sideroblasts provide the best evidence that the identification of the mutant gene responsible for the initial clone is relevant to clinical outcome. In fact, ring sideroblasts may be found not only in patients with a founding mutation in SF3B1, but also in those with an initiating oncogenic lesion in SRSF2. However, the median leukemia-free survival is \>10 years in the former vs \<2 years in the latter.
Moreover, mutation screening may affect clinical decision making : a) in MDS with 5q-, subjects carrying TP53 mutations have a higher risk of leukemic progression and a lower probability of response to lenalidomide; b) in patients receiving HSCT, TP53 mutations predict high probability of relapse; c) SF3B1 mutations are associated with increased probability of erythroid response to TGFb inhibitors (luspatercept), and d) TET2 mutations might be associated with response to HMA.
Despite these findings, caution is needed against immediately adopting such mutational testing in clinical practice. First, the presence of mutations in a given individual has only limited predictive power, as conversion to MDS is rare regardless of mutation status. In addition, in patients with overt MDS, genetic abnormalities explain only a proportion of the total hazard for survival associated with specific treatments, meaning that a large percentage is still associated with clinical and non-mutational factors. Comprehensive analyses of large patient population and new methods to study gene-gene interactions and genoptype-phenotype correlations are warranted to correctly estimate the independent effect of each genomic abnormality on clinical outcome and response to treatment.
By combining an already available, large amount of sequenced genomic data and clinical information, the authors hypothesize that AI will allow to understand better MDS biology and classification, enhance prognostic/predictive capacity of currently available tools and apply treatments in a more targeted way, thus facilitating the implementation of personalized medicine program across EU.
Recruitment & Eligibility
- Status
- ACTIVE_NOT_RECRUITING
- Sex
- All
- Target Recruitment
- 13284
- Patients affected by MDS according WHO criteria > 18 years old
- Avaliability of clinical and hematological information
- Availability of information on targeted mutation screening
- none of the above
Study & Design
- Study Type
- OBSERVATIONAL
- Study Design
- Not specified
- Primary Outcome Measures
Name Time Method Improving MDS classification through study completion, an average of 2 years To improve classification of MDS by integrating clinical and hematological information with genomic features. To address this issue, different methods of statistical learning (Dirichlet processes (DP), Bayesian networks (BN)) and machine learning (deep learning physics informed neural network, constrained regression and deep models) will be compared in order to define specific genotype-phenotype correlations and to develop a new disease classification.
Prediction of probability of overall survival (months between diagnosis and death or end of follow up) for patients with MDS through study completion, an average of 2 years Overall survival (OS) will be defined as the time (expressed in months) between diagnosis and death (as a result of all causes) or end of follow-up (censored observations).
New prognostic scores will be defined including the following features: age expressed in years; sex (male or female); neutrophils count (number of neutrophils\*10\^6/L), platelets count (number of plateles 10\^6/L), hemoglobin concentration (g/dl), cytogenetics (stratified according to IPPS-R criteria, Blood 2012 120: 2454-2465), percentage of bone marrrow blasts and presence of gene mutations (presence versus absence).
Different statistical methods will be used to measure prediction accuracy (measured by concordance index, C-index): Cox proporsional-hazard methods, random survival forests, neural networks, continous individualized risk index (CIRI), times series analysis and Markov modeling for stochastic trajectories predictionPrediction of probability of leukemia free surivival (months from diagnosis to progression to acute leukemia or end of follow up) for patients with MDS through study completion, an average of 2 years Leukemia will be defined as the time (expressed in months) between diagnosis and progression to acute leukemia or end of follow-up.
New prognostic scores will be defined including the following features: age expressed in years; sex (male or female); neutrophils count (number of neutrophils\*10\^6/L), platelets count (number of plateles 10\^6/L), hemoglobin concentration (g/dl), cytogenetics (stratified according to IPPS-R criteria, Blood 2012 120: 2454-2465), percentage of bone marrrow blasts and presence of gene mutations (presence versus absence).
Different statistical methods will be used to measure prediction accuracy (measured by concordance index, C-index): Cox proporsional-hazard methods, random survival forests, neural networks, continous individualized risk index (CIRI), times series analysis and Markov modeling for stochastic trajectories prediction
- Secondary Outcome Measures
Name Time Method
Trial Locations
- Locations (1)
Istituto Clinico Humanitas
🇮🇹Milano, Italy