Welcome to Managing MDS, I am Dr. Aziz Nazha. I will be discussing an abstract at the 2017 American Society of Hematology Annual Meeting about a personalized prediction model to risk stratify patients with myelodysplastic syndromes. Outcomes of patients with myelodysplastic syndromes are very heterogenous. Some of the patients have indolent disease and their survival is measured by years; while others have very progressive disease, profound cytopenias, and risk of AML transformation, and their survival is measured in months. Accurately predicting the overall survival for those patients is clinically very important. From the patient's standpoint, it is very important because it sets the expectations of their disease. From the physician's standpoint, all our recommendations and treatment guidelines are based on risk. For patients with lower-risk disease, we treat them completely differently than patients with high-risk disease. There have been several models that can risk-stratify patients with myelodysplastic syndromes; the most commonly used ones are the IPSS and IPSS-R. However, we have recently shown that there is a significant amount of heterogeneity in the prediction of these models. In other words, what we are predicting for the patient and what the actual survival of the patient is are completely different.
That has led us to ask: Can we build a model that can precisely predict the outcomes of MDS patients, where the outcome will be very specific for each patient? In order to do that, we took clinical and mutational data from our database at the Cleveland Clinic which comprised the training cohort, and then we acquired a validation cohort from five other US institutions; that is consistent with the MDS clinical research consortium. To build the model, we used a machine learning algorithm – the random survival forest – and this algorithm will provide a personalized prediction that is specific for the patient. We train our model and then we validate it in the validation cohort. We were able to show that this model, in terms of predictability with the C-index, outperformed all our current models even when we added molecular data to them. We believe that this model may change the way we think about prognostications in MDS and other hematologic malignancies by building a personalized prediction that is specific for a given patient. Thank you very much for viewing this activity.