A Translational Informatics Framework for Generating Data-Driving Insights
Author(s)
Giuste, Felipe Oliveira
Advisor(s)
Editor(s)
Collections
Supplementary to:
Permanent Link
Abstract
A framework was developed to generate data-driven clinical insights by solving major translational informatics challenges. The three major informatics challenges are: lack of data standardization, opaque models, and inconsistent model deployment. The framework was applied to solve these challenges and its value was demonstrated using real-world case studies. First, I developed a solution to the lack of healthcare information standardization through the creation of a web-based application to standardize healthcare data at a Shriners Children's hospital site using the Fast Health Interoperability Resources (FHIR) data exchange standard. The developed approach allows for robust adaptation across clinical data sources while maintaining interoperability across clinical sites. The deployed application also identifies patient cohorts for conducting future clinical studies. Next, I improve model interpretability by leveraging explainable AI (XAI) solutions to gain insight into the importance of clinical features in predicting COVID-19 positive patient outcomes. A robust feature ranking approach was also developed to facilitate the clinical deployment of a SMART-on-FHIR application for decision support by minimizing the number of required clinical features to successfully predict patient risk. Finally, I implemented the framework to generate a data-driven decision support tool for rare disease detection. Specifically, I generated synthetic data to improve deep learning model performance and developed a user interface to support expert detection of pediatric heart transplant rejection. Together, these case studies demonstrate the value of the proposed framework in solving major challenges in translational informatics.
Sponsor
Date
2023-03-21
Extent
Resource Type
Text
Resource Subtype
Dissertation