New Prediction Model Developed for Myelodysplastic Syndromes

Researchers presented data at the 2017 ASH annual meeting about their new personalized MDS prediction model that uses clinical and genomic data to help better guide therapy and improve outcomes compared with other models.

Patients with myelodysplastic syndromes (MDS) have a wide variety of outcomes. However, a new personalized prediction model based on clinical and genomic data may be able to help better guide therapy and improve outcomes. Researchers reported at the 59th American Society of Hematology (ASH) Annual Meeting and Exposition, held December 9-12 in Atlanta, Georgia, that their new prediction model appears to outperform all commonly used prognostic models for survival and acute myeloid leukemia (AML) transformation estimates that are unique for each patient.

Currently, the developers of the new precision medicine tool are completing a web application that can allow the translation of this model into the clinic. The clinical and mutational variables can be entered into a web application that can run a trained model and provide overall survival (OS) and AML transformation estimates as an output. “We are modifying it now to make it more user-friendly. It should come out soon.  It is already built. We are just trying to optimize it,” said study investigator Aziz Nazha, MD, who is from the Department Hematology and Medical Oncology at Taussig Cancer Institute at the Cleveland Clinic in Cleveland, Ohio.

Several prognostic scoring systems have been developed to stratify MDS patients. However, these systems are highly limited and are relatively poor predictors of survival. The current stratification paradigm may be contributing to overtreatment or undertreatment of some patients classified within the same risk category, according to Dr. Nazha.

For this investigation, the team examined clinical and mutational data from MDS patients diagnosed according to 2008 World Health Organization (WHO) criteria. The model was developed in a cohort from a single institution and then validated in a separate cohort from other MDS Clinical Research Consortium sites.

The team used next-generation targeted deep sequencing of 60 gene mutations commonly mutated in myeloid malignancies. The investigators calculated leukemia-free survival from the time of diagnosis to time of AML progression or last follow-up. They employed a random survival forest (RSF) algorithm to build their model. 

This investigation included 975 patients, with 105 patients (20%) progressing to AML. The researchers found that SF3B1 (14%), ASXL1 (13%), TET2 (12%), SRSF2 (11%), and DNMT3A (11%) were the most commonly mutated genes, followed by STAG2 (9%), TP53 (8%), and RUNX1 (8%).

The team identified 18 variables that impacted OS. Each variable was ranked from the most important to the least important that impacted OS. The new model was found to outperform all commonly used models for OS and AML transformation including International Prognostic Scoring System (IPSS), Revised IPSS, WHO prognostic scoring system (WPSS), and M.D. Anderson prognostic model (MDAPSS). 

“We used a machine learning algorithm and we were able to demonstrate that it outperformed all the other models,” Dr. Nazha said in an interview with OncoTherapy Network. “Outcomes are very heterogeneous and some patients have indolent disease and some have more profound disease and their survival is measured in months. This can give patients a better idea of their disease and what they can expect.”