SAN DIEGO — Patients with myelodysplastic syndrome (MDS), which rarely causes symptoms in its early stages, may now obtain a better idea of what may lie ahead for them.
A machine-learning approach that analyzes genomic and clinical data can give a personalized risk assessment for each individual. It can provide overall survival (OS) estimates at different time points and predict the risk for transformation to acute myeloid leukemia (AML).
It has been shown to outperform currently used risk stratification models, such as IPSS-R (Revised International Prognosis Risk Stratification), in clinical practice.
"We wanted to build a personalized prediction tool that can provide insights about a specific outcome for a specific patient," said presenter and model-developer Aziz Nazha, MD, from the Cleveland Clinic, Ohio. He presented details of the personalized prediction tool here at the American Society of Hematology (ASH) 2018 (abstract 793).
Nazha treats MDS patients, but he is also a programmer with expertise in artificial intelligence and machine learning. He told Medscape Medical News that he personally built the model and developed a website as a result of the frustration expressed by his patients. "Patients wanted to know how their numbers were different from those predicted for groups of patients in current risk models," he said. "We wanted to predict more accurately what each patient can expect with respect to survival expectation and transformation to AML," he explained.
"Understanding a patient's prognosis allows us to more appropriately develop a treatment plan and counsel patients. We are optimistic an improved prediction will lead to more personalized care," Nazha said in a Cleveland Clinic statement.
"Currently, treatment in MDS is based on risk stratifying an individual based on an analyses of groups of patients," Nazha said. "If we get the risk wrong, we may end up overtreating or undertreating a patient," he added. "Current models misclassify risk in 25% to 30% of patients," Nazha told Medscape Medical News.
"Traditional models are based on simplicity. That is why the scoring system had few variables. In an era enhanced by artificial intelligence and Web-based tools, we can include more variables and refine our model," commented Joseph Mikhael, MD, chief medical officer at the International Myeloma Foundation in Phoenix, Arizona. He moderated an ASH press briefing at which the new model was highlighted.
Building the Machine-Learning Model
The model was developed through use in a combined cohort of patients from the Cleveland Clinic and the Munich Leukemia Laboratory (n = 1471) and was then validated in a separate cohort from the Moffitt Cancer Center (n = 831).
Next-generation targeted deep sequencing of 40 gene mutations commonly found in myeloid malignancies was included in the model building.
Demographic, clinical, and genomics data were fed into an algorithm, which interrogated important variables. "The algorithm is not a black box," Nazha said. "It extracts important variables, which are not easily done by a treating clinician."
The final prediction model was based on the most important variables that affected outcomes and the least number of variables that produced the best prediction.
The variables incocporated into the model include IPSS-R cytogenetic risk categories, platelets, mutational number, hemoglobin, percent bone marrow blasts, white blood cell count, age, and mutational status for several genes.
"In the IPSS-R mutational model, three mutations are of significance, while the machine-learning model has identified 12 mutations as being significant," Nazha said.
The model outperformed current risk models, such as IPSS and IPSS-R. In a head-to-head comparison using patient medical records, the new model correctly predicted a patient's likelihood of survival for a given period relative to another patient 74% of the time (concordance index [c-index], 0.74). By comparison, the IPSS-R correctly predicted a patient's likelihood of survival 67% of the time (c-index, 0.67). Using the model, the c-index for AML was 0.81, compared with 0.73 for IPSS-R.
Nazha is developing a Web application in which the trained model can be used to provide OS and AML transformation probabilities for each patient at different time points. "The Web application will make it more physician and patient friendly," he said.
Nazha and colleagues are also looking to improve the model by taking feedback from clinicians and incorporating additional outcomes, such as quality-of-life information.
Dr Nazha has had a consultancy with Karyopharma and Tolero and is on the Data Monitoring Committee for MEI. Dr Mikhael has received research funding from AbbVie, Celgene, Onyx, and Sanofi.
American Society of Hematology (ASH) 2018. Abstract 793, presented December 3, 2018.
Medscape Medical News © 2018
Cite this: Model Improves Predictions of Myelodysplastic Syndrome Prognosis - Medscape - Dec 03, 2018.