Project Summary
Our team will develop machine learning models that are capable of performing clinical evaluations, categorizations, and predictions at a state-of-the-art level utilizing personalized and imputed genomic data and synthetic electronic health records to achieve both improved overall performance and fairness/equity. We will analyze existing datasets for inequities of various types between demographics and outcomes, balance those datasets with synthetically and conditionally generated records for augmentation, and supplement those with present or imputed genomic information.
We will then train models on these new, comprehensive, and fair datasets (as well as the old datasets as a baseline) to demonstrate that our method allows improved and fairer performance on a variety of tasks. These models will help with early disease diagnosis, severity prediction, and many other clinical tasks to improve patient outcomes, hence making healthcare both better and more equitable. Finally, we will integrate these improved models as FHIR apps for seamless use with electronic health records during either medical care by clinicians or administrative labeling by medical billers. These applications and tools will be the output of our project and allow medical institutions to easily make strides towards improved fairness, better patient care, and more accurate billing.