Skip to Main Content
 

Global Search Box

 
 
 
 

Files

File List


39318.pdf (3.33 MB)

ETD Abstract Container

Abstract Header

Generalizability of Electronic Health Record-Based Machine Learning Models

Wissel, Benjamin D
http://orcid.org/0000-0002-8967-9296
http://rave.ohiolink.edu/etdc/view?acc_num=ucin1627659161796896

Abstract Details

2021, PhD, University of Cincinnati, Medicine: Biomedical Informatics.
Epilepsy affects over 50 million people and is responsible for 1.3% of deaths worldwide. While many people with epilepsy benefit from pharmacotherapy, approximately one-third have drug-resistant epilepsy. Clinical guidelines recommend that these patients be promptly evaluated for resective epilepsy surgery, which is associated with a 67% chance of long-term seizure freedom. Unfortunately, surgery is underutilized. The mean disease duration at the time of surgery is six years in pediatrics and 20 years in adults. Identifying surgical candidates earlier in the disease course is needed to optimize care. Machine learning models have been developed to aid in this process, and sending alerts based on these model’s recommendations is associated with a three-fold increase in referrals. Electronic health record-based machine learning models have been developed to predict a variety of health outcomes. However, they are rarely implemented into clinical care. Generalizing electronic health record-based predictive models across health care systems is challenging for several reason. Electronic health record data contain process biases, health care delivery patterns change over time, patient populations vary across geographical regions, and documentation styles vary by provider. Overcoming these limitations will improve the likelihood that machine learning-based clinical decision support systems will enhance patient care. To address these challenges, this dissertation proposed a novel methodology for generalizing electronic health record-based machine learning models across health care systems. The set of procedures produced site-specific models with modifiable feature sets and parameter weights. Data preprocessing, feature selection, and model training were repeated at each institution without modification. The results showed that generalizing modelling processes, rather than the models themselves, minimized generalization error. These methods were used to identify candidates for epilepsy surgery at one adult and one pediatric epilepsy center. In a prospective validation study, model performance was robust to temporal variations in care patterns, including virtual visits that occurred during the first three months of the SARS-CoV-2 pandemic. The final chapter of the dissertation concludes by summarizing lessons learned from these experiments, the current landscape of the literature, and future avenues of research. This work provides seminal data to illustrate the effectiveness of machine learning algorithms across sites and how this could be applied to external institutions.
Judith Dexheimer, Ph.D. (Committee Chair)
David Ficker, M.D. (Committee Member)
Tracy Glauser, M.D. (Committee Member)
John Pestian, Ph.D. (Committee Member)
Rhonda Szczesniak, Ph.D. (Committee Member)
129 p.
Surgery
machine learning; electronic health record; epilepsy surgery; informatics; natural language processing

Recommended Citations

Citations

  • Wissel, B. D. (2021). Generalizability of Electronic Health Record-Based Machine Learning Models [Doctoral dissertation, University of Cincinnati]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=ucin1627659161796896

    APA Style (7th edition)

  • Wissel, Benjamin. Generalizability of Electronic Health Record-Based Machine Learning Models. 2021. University of Cincinnati, Doctoral dissertation. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=ucin1627659161796896.

    MLA Style (8th edition)

  • Wissel, Benjamin. "Generalizability of Electronic Health Record-Based Machine Learning Models." Doctoral dissertation, University of Cincinnati, 2021. http://rave.ohiolink.edu/etdc/view?acc_num=ucin1627659161796896

    Chicago Manual of Style (17th edition)

ucin1627659161796896
244
© 2021, all rights reserved.
This open access ETD is published by University of Cincinnati and OhioLINK.