Skip to main content

Dynamic prediction of outcomes in acute myeloid leukemia using machine learning that integrates genotype and phenotype

Dr. Schwede

Matthew Schwede

MD

Stanford University

Project Term: July 1, 2023 - June 30, 2026

Acute myeloid leukemia is life-threatening and heterogeneous, and although classification models help guide treatment, they do not use detailed phenotypic information or dynamically update with new data during a patient’s course. We will develop computational methods to extract both mutations and phenotype from the electronic health record. Machine learning models will be built that adapt to new data over time so that all clinically relevant data is used when personalizing a patient’s therapy.

Lay Abstract

Acute myeloid leukemia (AML) is life-threatening, and despite extensive treatment and new targeted therapies, relapse is common, in part due to the heterogeneity of AML. While risk stratification schemes help guide therapy decisions, these approaches focus on the AML genotype without using detailed phenotypic information, such as flow cytometry and histological classification, and they also rely only on information at diagnosis. I will use modern clinical informatics and machine learning tools to obtain both genetic and extensive phenotypic information from the electronic health record (EHR) and integrate data during a patient’s treatment course to improve survival and treatment response predictions in AML. Our group previously used document segmentation and natural language processing to extract data, such as cell surface markers, from clinic notes and pathology reports. I will extract additional phenotype data about patients and their leukemia and explore associations with treatment response and survival. To encourage collaboration, I will create software to transform our data into commonly used AML data dictionaries. Several machine learning algorithms, such as long short-term memory neural networks that integrate serial information, will be used to predict outcomes from the data. These results will help ensure that our understanding of new technologies keeps pace with their use, bridging bench and bedside with the vast, often-overlooked clinical data in the EHR.

Program
Career Development Program
Grant Subprogram
Fellow
To All Projects