Automated detection of causal relationships among diseases and imaging findings in textual radiology reports
Journal of the American Medical Informatics Association, ocad119, https://doi.org/10.1093/jamia/ocad119
Presenter
Charles E. Kahn, Jr., MD, MS, FACR, FACMI, FSIIM
Dr. Charles Kahn is Professor and Vice Chair of Radiology at the University of Pennsylvania. He is a practicing radiologist with expertise in body CT and ultrasound, and holds degrees in Mathematics (BA) and Computer Sciences (MS). Professional interests include health services research, decision support, artificial intelligence, information standards, and knowledge representation. He serves as Editor of the journal Radiology: Artificial Intelligence.
Statement of Purpose
Medical records are known to contain a wealth of information that can be used to better understand diseases and associated conditions. In particular, electronic health record (EHR) data can be used to discover correlations among diseases and to stratify patient cohorts. Textual information -- such as progress notes, imaging reports, and procedural reports -- constitutes much of the data available from the EHR. Earlier work has shown that simple pairwise statistics, such as the phi coefficient (φ) and Cohen’s kappa statistic (κ), can detect causal relationships at various statistical thresholds from the co-occurrence of terms in radiology reports, but those metrics show limited performance.
Our study explored the use of Bayesian network learning to improve the detection of causal associations among diseases and imaging findings. This presentation will introduce the Radiology Gamuts Ontology, a knowledge model that links diseases and their imaging findings, and briefly describe the natural language processing techniques used to identify the relevant entities in radiology reports. The presentation will provide an overview of Bayesian networks and how they can learn a probabilistic model from data. These approaches are then applied to identify causally related terms from radiology reports of 1.3 million patients. We present the results of the analysis, the challenges encountered, and identify avenues for further investigation.
Learning Objectives
- Understand the fundamentals of Bayesian networks.
- Describe how one can learn a Bayesian network’s structure and/or probability values from data.
- Define approaches to identify causal relationships among concepts that appear in text-based health records.
Format
- 35-minute presentation by article author(s) considering salient features of the published study and its potential impact on practice
- 25-minute discussion of questions submitted by listeners via the webinar tools and moderated by JAMIA Student Editorial Board members.
Accreditation Statement
The American Medical Informatics Association is accredited by the Accreditation Council for Continuing Medical Education to provide continuing medical education for physicians.
Credit Designation Statement
The American Medical Informatics Association designates this live activity for a maximum of 1.0 AMA PRA Category 1 Credits™. Physicians should claim only the credit commensurate with the extent of their participation in the activity.