Predicting opioid overdose risk using machine learning: Evaluation of an AI approach

Bo (Cloud) Cao

Giri Puligandla

Impact

  • MHRC and IAM’s seed-investment of $50,000 has led to nearly $1 million of additional funds post-award to profoundly expand the scope and depth of this work.

  • The first known study to prospectively predict and verify individual-level opioid overdose using population health data.

  • This project is ground-breaking in that it engages communities and stakeholders early in the development of new technologies, by enabling more effective and ethical interventions through predictive tools that support patients, people at risk and their families, as well as clinicians, emergency services, and policy makers.

  • Since this project was funded, CMHA Edmonton and the City of Edmonton established a Council for Community Mental Health that will support the project to reach more people with lived experience and inform municipal initiatives on addiction and drug poisoning.

  • From the researcher’s perspective, the study introduced a new, community-centred approach to AI in mental health by involving people with lived experience and sparking meaningful discussions on the ethical use and impact of individualized prediction tools for opioid overdose.

  • Alberta

  • Grant

  • University of Alberta

  • CMHA Edmonton

  • IAM, ATB Financial

  • 2023-2024

  • Digital Mental Health

  • People with Addictions

About the Project

This project aims to address the opioid crisis by enhancing an existing AI/machine learning model that uses large-scale population health data to identify risk factors and predict opioid overdose (OpOD). Throughout the process, researchers collaborated closely with communities and stakeholders to gain insights on how these predictions could be applied meaningfully in real-world settings.

Methodology

The first component of the study aims to develop and validate a population-level individualized prospective prediction model of opioid overdose (OpOD) using machine learning and de-identified provincial administrative health data. The OpOD prediction model was based on a cohort of approximately 4 million people in 2017 to predict OpOD cases in 2018 and was subsequently tested on cohort data from 2018, 2019, and 2020 to predict OpOD cases in 2019, 2020, and 2021, respectively, and then update and validate the model longitudinally.

The second component of the study aims to engage patients, people at risk and their families, as well as clinicians, emergency services, and policy makers to evaluate the potential benefits and risks of such prediction models and channels of real-world implementation.

Findings

The model’s predictive performance, including balanced accuracy, sensitivity, specificity, and area under the Receiver Operating Characteristics Curve (AUC), was evaluated, achieving a balanced accuracy of 83.7, 81.6, and 85.0% in each respective year. The leading predictors for OpOD, which were derived from health care utilization variables documented by the Canadian Institute for Health Information (CIHI) and physician billing claims, were treatment encounters for drug or alcohol use, depression, neurotic/anxiety/obsessive-compulsive disorder, and superficial skin injury. The main contribution of our study is to demonstrate that individualized OpOD prediction using existing population-level data can provide accurate prediction of future OpOD cases in the whole population and may have the potential to inform targeted interventions and policy planning.

Project Outreach

The project is currently exclusively based in Alberta but may expand nationally once the model receives clinical validation. It actively involves diverse individuals with lived experience in the development and community implementation of the technology, paving the way for future scaling across Canada.

Collaborations with healthcare teams and community organizations, including the Canadian Mental Health Association (CMHA), are underway to test real-world applications.

Resources Created