Using Machine Learning Methods and Electronic Health Records to Classify Patients as At-Risk for Opioid Abuse and Dependence
Julia Deaver, Julia Burek, Heman Shakeri · Zenodo (CERN European Organization for Nuclear Research) · 2023
The goal of this research was to create a predictive algorithm that classifies some patients as high-risk for misuse of opioids. Opioids are often prescribed for pain relief, but can be highly addictive and lead to misuse among patients. Utilizing patient data from All of Us Research Hub, an electronic health record database, machine learning methods were applied to determine which patients were at high-risk for opioid misuse and those that were not. The models described in this paper use a patient’s drug prescription, procedures, chronic diseases, and pain history to classify whether or not a person is more likely to abuse or become dependent on opioids. The motivation for this research was to build an algorithm that medical professionals could use before prescribing a patient opioids, allowing them to consider alternative treatments if a patient is identified as at-risk for misuse. In this paper, we describe our methodology in creating a logistic regression model, decision tree, random forest, and deep learning neural network to predict a patient’s risk of opioid abuse or dependence. We were able to determine what factors impact a patient’s risk of opioid misuse the most, which were supported by prior literature review. This work demonstrates how machine learning methods such as deep learning are an important strategy when dealing with major healthcare issues such as the opioid epidemic.