Harms from opioids, including deaths and emergency department visits, are increasing in Ontario. The number of opioid-related deaths especially increased during the COVID-19 pandemic.

Public health units (PHUs) and other harm reduction organizations work to identify, track and reduce the risks posed by opioids. They also aim to create early warning systems to prevent these harms. In Ontario, Public Health Ontario and the Ontario Drug Policy Research Network offer public tools for monitoring opioid-related harms and harm reduction efforts. PHUs and community organizations may also have their own tools. Additionally, Ontario PHUs have access to real-time emergency department and hospitalization data from KFL&A Public Health’s Acute Care Enhanced Surveillance System, hospitalization data from the Canadian Institutes for Health Information, and death data from the Office of Chief Coroner for Ontario to inform their decision-making.

These traditional surveillance systems are built to consistently capture specific types of information over time, for example opioid overdoses. This allows the monitoring of trends and identification of patterns. However, the need for consistency limits their ability to adapt to changes in the drug poisoning crisis. These changes may be due to shifting human behaviours—for example if fewer people go to hospitals for opioid overdoses due to an increase in community-based services, these overdoses aren’t captured by hospital-based surveillance systems. The typical opioid-related harm measures may not capture new or unusual outcomes caused by emerging contaminants in the unregulated drug supply (for example, wounds from xylazine).

Event-based surveillance systems differ from traditional surveillance systems as they use unstructured data to find both known and unknown public health events. They use various types of information, like news media or social media, to signal emerging or re-emerging public health threats. However, since the data is less verified, there can be “false positives” or things that didn’t happen. Biases and other limitations in the original data can also impact their accuracy. Furthermore, the volume of unstructured data can make event-based surveillance systems resource-intensive. While public health responses from these systems require careful context and interpretation, they can offer early warnings and supplementary insights compared to traditional methods.

AONES is an event-based surveillance system for the unregulated drug supply, using news articles from Canada and the United States. Its goal is to supplement traditional epidemiological surveillance systems and enhance situational awareness of the drug poisoning crisis. AONES focuses on finding contamination incidents in the drug supply and unexpected harms by compiling drug alerts and warnings from regions across the two countries. It also collects reports of drug confiscations, extracting descriptions of the drugs seized.

AONES accesses news media through the Global Database of Events, Language, and Tone. AI techniques are used to filter to only articles from North America and related to opioids. AI is also used to extract locations and drug-related information from the articles. The articles and their extracted data are shared through an interactive dashboard (embedded below). The dashboard is updated in near-real-time (every 3 hours) and includes articles from the previous 180 days, with a focus on the last 90 days. The full text of the articles is not included in the dashboard but a URL to the article is provided.

Given its reliance on news media articles curated by GDELT and the AI techniques used for filtering and extracting information, AONES is not and cannot be a complete record of events in the unregulated drug supply. Some events may not be known, or they may not be reported by the media. While GDELT pulls news from all over the world, it may miss some. AONES will also reflect the biases in what is reported in the media and how. Changes in how these events are reported could also decrease the accuracy of the AI techniques in capturing all relevant articles. Finally, the data extracted using AI models may sometimes misrepresent the content of the articles due to the fallibility of AI. Links to the original articles are provided to users to confirm information displayed on the dashboard. Like all event-based surveillance systems, some events captured in AONES may be “false positives” (i.e., they didn’t occur). Individual events may also be captured multiple times.

All information from AONES must be interpreted understanding the limitations of the data source and the processes involved in filtering and extraction. The information should be used alongside other epidemiological data sources and with local contextual knowledge.

More information about AONES is available on the associated documentation pages:

• Demo of the dashboard tool
• Description of how of the data is sourced, filtered and extracted
• Description of how to use and interpret the data

The AONES dashboard has different sections for different use cases:

1.  seeing recent alerts (1 page),
2.  tracking drug confiscations (1 page),
3.  looking at recent themes in the drug supply (4 pages), and
4.  exploring the full dataset (1 page).

The first two sections focus on mapping recent alerts and drug confiscations and highlighting the most relevant extracted data. The third section aims to show what may be new and emerging in the unregulated drug supply. It does this over several subpages summarizing themes by time (a page with graphs of articles by time and a page for watching the articles on a map over time) and pages that highlight terms (or groupings of terms) that are showing up more frequently in the extracted information than before. The final section lets users search through all the articles from the last 180 days using a variety of fields.

Throughout the dashboard, there is a “trust but verify” approach due to the limitations of the AI processes used by AONES. The dashboard visuals use the extracted information, but the original article link is provided. Users can check the original articles, understanding that links may stop working. Users open original links at their own risk (from computer viruses, inappropriate content, etc.).

Each dashboard page is interactive. They have filters to choose the time range, geography and other fields relevant to the page. The pages also contain visuals, like tables, maps and other data visualization tools, to display the articles included in the tool and their extracted information.

An online database called GDELT (Global Database of Events, Language, and Tone) monitors news media from around the world to create a list of articles with summary information on each. AONES takes this list and filters it to only articles that mention opioids (and other related keywords). The tool then accesses the full text of the article and uses this to do more filtering. Articles relevant for AONES are categorized as either alerts, crime (drug confiscations), or other (articles that don’t fit into the other two categories). AI is used to extract key information from each article—things like the names of drugs, their appearance and their effect on the body. Places mentioned in the articles are also extracted and then simplified to the city level. Only articles that mention cities in Canada and the United States are kept. The final list of articles and their extracted information are displayed in the dashboard.

The goal of AONES is to increase situational awareness of the drug poisoning crisis by providing supplemental information on current events. In addition to the traditional data sources, the tool serves as a central hub for drug alerts from around Canada and the United States. Users can monitor themes in the types of drugs (and possible contaminants) mentioned in alerts or confiscated by the police. The tool can help identify emerging or unusual effects of drugs. The mapping and time trends in the dashboard can reveal insights into how the unregulated drug supply is changing and spreading over time and place.

During the development of AONES, feedback sessions were held with potential users, specifically public health staff and people with lived and living experience. The goal of these sessions was to collect information that could be shared with new users of AONES on how they could use the tool. Participants shared how they intended on using the tool and provided insights based on their experiences regarding the information it offers, as well as any potential biases and limitations.

Building and maintaining AONES is an ongoing process, and we welcome your feedback on the tool. We are interested to hear how you use the tool and what it tells you, including:

• How often do you use the tool?
• How do you use the information from the tool?
• Which parts of the tool do you find the most helpful?
• Which parts of the tool do you find the least helpful?
• What are the strengths and limitations of the tool?
• Are there any unintended consequences (positive or negative) from the tool?
• What would you add or change about the tool?

While we welcome all comments, we can’t promise that changes will be made to the tool based on the feedback. To provide feedback on AONES, please complete the form at the bottom of this page To provide feedback on AONES, please complete the form at the bottom of this page.

AONES was developed as part of one of Public Health Ontario’s Locally Driven Collaborative projects. The project was led by KFL&A Public Health in collaboration with Grey Bruce Public Health, Leeds, Grenville & Lanark District Health Unit, York Region Public Health and Public Health Sciences at Queen’s University. The project was further supported by a knowledge advisory group including: Algoma Public Health, Brant County Health Unit, Niagara Region Public Health, Peel Region Health Unit, Thunder Bay District Health Unit, Toronto Public Health, Wellington-Dufferin-Guelph Public Health, Public Health Ontario and Ontario Drug Policy Research Network (ODPRN). The AONES project team would especially like to thank the members of ODPRN’S Lived Experience Advisory Group (LEAG) group for their insightful and critical input throughout the design and build of this tool.

Suggested citation:

KFL&A Public Health, Grey Bruce Public Health, Leeds, Grenville & Lanark District Health Unit, York Region Public Health, Queen’s University Department of Public Health Sciences and the LDCP Project Team. The Automated Opioid News Event-based Surveillance System (AONES), 2024. Available from: https://www.kflaphi.ca/aones/

Several open-source packages and tools are critical for the ongoing operation of AONES, including:

• Global Database of Events, Language and Tone (GDELT) accessed from https://registry.opendata.aws/gdelt.
• Instructor accessed from https://python.useinstructor.com/
• Hugging Face accessed from https://huggingface.co/
• NOMIC accessed from https://huggingface.co/nomic-ai/nomic-embed-text-v1
• spaCy accessed from https://spacy.io/
• USEARCH accessed from https://www.drive5.com/usearch/