Date of Award

Summer 8-10-2023

Embargo Period

8-11-2023

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

College

College of Graduate Studies

First Advisor

Alexander Alekseyenko

Second Advisor

Vanessa Diaz

Third Advisor

Leslie Lenert

Fourth Advisor

Xia Jing

Fifth Advisor

Kit Simpson

Abstract

Introduction

Cancer is the second leading cause of death in the United States and cancer screening is a primary tool to reduce mortality. However, not all who are recommended to be screened actually follow through. This study investigates whether electronic medical record and geographic data is suitable to predict which patients are at risk of missing recommended screenings. The goal of this investigation is to design a data informed system that can automate the prediction of those at risk for missing screenings and provide insights into underlying reasons. This will enable resources to be focused to increase cancer screening adherence, with the overall goal of reducing mortality from cancer. Methods Data for this study was sourced from de-identified electronic medical records from the Medical University of South Carolina’s patient population and publicly available geographic datasets. This data was used to train a series of machine learning models to predict which patients would follow through with cancer screening tests, and describe underlying associations to diagnoses data, cancer histories and social determinants of health. Results This study found that it was possible to systematically identify small groups of female patients that are unlikely to follow through with mammogram screening. However, similar results were not found predicting lung cancer screening follow-though. Additionally, patterns associating social determinants at the county level cannot be used to make accurate predictions about individual patient follow through. It was also demonstrated that the core relationship between screening and mortality does not hold in high proportion minority areas. Conclusion

This study successfully shows that an automated system for identifying small groups of patients unlikely to complete mammogram screening is achievable and sets forth a methodology to development. It also provides valuable insights into the nature of social determinants associated with patients and their limits when geographically attributed.

Rights

Copyright is held by the author. All rights reserved.

Download

Share

COinS