A CLINICAL DECISION SYSTEM WITH AN INTERACTIVE KNOWLEDGE GRAPH AND COST OPTIMIZATION

Abstract

Clinical decision support systems aim to improve access to relevant and practical clinical data to assist medical practitioners in diagnosis and treatment. However, the usefulness of such systems is limited due to the lack of effective user interactions and proper cost management for treatments. Medical price transparency has been an issue in the United States for many years. Well-meaning care providers may refer patients to specialists or order tests that are unexpectedly costly and may not be covered by the patient's insurance without knowing this is the case. This thesis proposes solutions to the above issues through allowing interactive navigation of a knowledge graph of medical conditions and symptoms and novel cost management decision support.

A growing number of medical costs are now available to the public due to a new U.S. law. We propose utilizing newly available cost data to allow medical practitioners to be aware of and consider these costs when they are making decisions about a patient's care. To this end, we propose providing the ability to help filter possible conditions the patient may have by using information about the patient, including symptoms. This includes an easily navigable graph where each node presents the likelihood of a patient having certain medical conditions as each new symptom is learned. Once new information about the patient is exhausted, we propose finding the order to test that patient's possible conditions that minimizes the overall expected cost. We use a combination of synthetic data generation and realistic data collected from published papers to evaluate these approaches. Overall, we find that such a system would be beneficial for a non-trivial number of cases that medical clinics will will handle. However, it is most helpful for rarer instances where patients have few symptoms or uncommon medical conditions.