Machine learning predicts the risk of opioid use disorder for individual patients

Maschinelles Lernen sagt das Risiko einer Opioidkonsumstörung für einzelne Patienten voraus

Receiver operating characteristics and confusion matrix. Receiver operating characteristic and confusion matrix for predicting OUD in the test cohort for OUD cases. True positive rate is the rate model that correctly predicts a positive OUD label, which corresponds to sensitivity, while false positive rate the rate model incorrectly predicts a positive OUD label, equivalent to 1 – specificity. AUC stands for area under the curve. Recognition: The Canadian Journal of Psychiatry (2022). DOI: 10.1177/07067437221114094

Thanks to a team of Alberta researchers, clinicians and policymakers could get help from artificial intelligence in predicting opioid use disorders.

Researchers built and tested a machine learning model that reliably predicts the risk of developing the disease in individual patients by analyzing population-level administrative health data. Health administrative records are created every time a patient interacts with the healthcare system—for example, when visiting a doctor, taking a diagnostic test, being admitted to the hospital, or getting a prescription.

Opioid use disorder is a treatable, chronic illness in which patients fail to control their opioid use, resulting in difficulties at work or at home and sometimes even overdose and death, according to the US Centers for Disease Control and Prevention. By the end of August this year, there had been 976 opioid-related deaths in Alberta. People with an opioid use disorder are initially exposed to drugs either through prescriptions for pain relief or through the illicit drug market.

Predictions could help with prevention

“Around a quarter of opioid users will develop an opioid use disorder, and eight to 12 percent of those prescribed opioids for chronic pain will develop OUD; therefore, the prediction and prevention of OUD in this population is critical to harm reduction efforts,” the researchers report in their paper.

“Most of these people have interacted with the healthcare system before they were diagnosed, and that provides us with data that could allow us to predict and potentially prevent some of the cases,” says lead researcher Bo Cao, Canadian Research Chair in Computational Psychiatry and collaborator Professor of Psychiatry.

The machine learning model analyzed health data from nearly 700,000 patients in Alberta who received opioid prescriptions between 2014 and 2018, comparing 62 factors including the number of doctor visits and emergency room visits, diagnoses, and socio-demographic information.

The team found that the most important risk factors for opioid use disorders included frequency of opioid use, high doses, and a history of other substance use disorders. They determined the model, which predicted high-risk patients with an accuracy of 86 percent when validated against a new 2019 sample of 316,000 patients.

Predict probability, don’t label it

“It is important that the model’s prediction of whether someone will develop an opioid use disorder is interpreted as a risk and not as a label,” says first author Yang Liu, a postdoctoral researcher in psychiatry. “It’s information that needs to be placed in the hands of clinicians who will actually make the diagnosis.”

Cao says the next phase of testing for the model will take place in a real clinical setting, involving both clinicians and people with lived experience of opioid use disorders.

“Most of the time, clinicians are very busy and may not have time to go through everyone’s health records,” says Cao. “This model will help clinicians make the most evidence-based decision based on comprehensive existing data. That is the motivation behind it.”

Cao’s research team used machine learning models to predict schizophrenia and bipolar disorder based on brain scans and cognitive data. This is the first time they have used administrative health records. Cao says the model could not only help clinicians identify high-risk patients, but also help policymakers plan health resources for regions with higher needs.

The study was published in The Canadian Journal of Psychiatry.

More information:
Yang S. Liu et al., Individualized Prospective Prediction of Opioid Use Disorder, The Canadian Journal of Psychiatry (2022). DOI: 10.1177/07067437221114094

Provided by the University of Alberta

Citation: Machine Learning Predicts Risk of Opioid Use Disorder for Individual Patients (2022, December 7), retrieved December 7, 2022 from .html

This document is protected by copyright. Except for fair trade for the purpose of private study or research, no part may be reproduced without written permission. The content is for informational purposes only.