Date of Award

2019

Degree Type

Thesis

Degree Name

Master of Science (MS)

Department

Computer Science

Abstract

Opioid misuse is a major public health problem in the world. In 2016, 11.3 million people were reported to misuse opioids in the US only. Opioid-related inpatient and emergency department visits have increased by 64 percent and the rate of opioid-related visits has nearly doubled between 2009 and 2014. It is thus critical for healthcare systems to detect opioid misuse cases. Patients hospitalized for consequences of their opioid misuse present an opportunity for intervention but better screening and surveillance methods are needed to guide providers. The current screening methods with self-report questionnaire data are time-consuming and difficult to perform in hospitalized patients. In this work, I explore the use of convolutional neural networks for detecting opioid misuse cases using the text of electronic health records as input. The performance of these models is compared to the performance of a more traditional logistic regression model. Different architectures of a convolutional neural network were trained and evaluated using the area under the ROC curve. A convolutional neural network performed better by producing a score of 93.4% whereas the score produced by logistic regression was 91.4% on the test data. Different advantages and disadvantages of using a convolutional neural network over the baseline logistic regression model were also discussed.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 3.0 License.

Share

COinS