Canadian Researchers Focus on AI to Assess Opioid Risk

Edmonton –

Researchers in Alberta are conducting an experiment using artificial intelligence to measure the risks of prescription opioids in the ongoing drug overdose crisis across Canada.

Doctors have protocols in place to identify patients at risk for opioid addiction, but Dr. Dean Urich says machine learning “could do a better job” identifying who are most susceptible. said.

AI-assisted systems “provide additional comfort to clinicians who[know]there are other supports they can use to ensure their patients are taking the right medication at the right time.” May: Director of the Clinical Epidemiology Program at the University of Alberta.

With this tool, physicians can predict the impact of prescription opioids on patients, preventing unnecessary emergency room visits and death within 30 days of starting medication.

Eurich was the principal investigator for the study published in JAMA Network in December. JAMA Network anonymously analyzed medical data from over 850,000 Alberta people to predict the best outcomes for their patients.

The dataset was primarily provided by Alberta Health.

Fizza Gilani, Ph.D., of the College of Physicians and Surgeons of Alberta, said machine learning could be an effective way to reduce patient hospitalizations and morbidity when integrated into healthcare systems.

“This model can predict the risk of hospitalization,” said Gilani, program manager for the prescribing, analysis and follow-up prescribing program at the university.

Sometimes, current methods cannot predict the source of risk, and medical solutions may be more complicated than reducing a patient’s opioid dose, she added.

The AI ​​system was input with a variety of health factors to determine patient risk, including history of injury, obesity, depression, diabetes, fluid disorders, and psychosis. These were combined with information such as diagnoses from doctors, visits to medical facilities, and the patient’s place of residence.

“The idea is not to discourage doctors from prescribing opioids, but[but]to minimize the risks after opioid exposure,” Gilani said.

Researchers examined approximately 3 million opioid prescriptions annually from a variety of health professionals (doctors, nurses, dentists) to more than 600,000 patients in Alberta. Cancer patients and patients receiving palliative care were excluded.

Twenty percent of patients take opioids along with other high-risk drugs, Eurich said, “increasing the risk of adverse outcomes.”

Over the years, the interactions between health systems and people have become more complex, calling for an efficient approach to navigating the health system, says Eurich.

“As humans, we can look at dozens of variables to predict outcomes, but we’ve found that’s not enough.”

He said machine learning takes a different approach, using nuanced data sets containing a variety of important factors to build systematic models to find combinations that predict the best outcome for patients.

Eurich, who has been working on AI predictions for more than three years, says the system “can correctly predict four out of five patients.” According to the machine, a patient identified as high risk is more likely to be hospitalized within her first 30 days of prescribing the drug.

He added that AI-powered systems could adapt quickly to changing environments. For example, the surge in opioid-related deaths during the pandemic.

According to Eurich, the goal is to “reduce the risk of patients taking high-risk drugs with known poor outcomes.”

Researchers plan to test AI systems with real-time data soon, Eurich said. We will also investigate whether this system can limit long-term use of high-dose opioids in patients.

One advocacy group believes the machine will not help Alberta’s opioid crisis.

Moms Stop the Harm co-founder Petra Schulz says most opioid-related deaths in the state are caused by street drugs, not prescription opioids.

“This kind of AI can make safer alternatives unavailable,” she said. “It’s like doing detective work and wanting to understand what’s going wrong for the patient, instead of building a trusting doctor-patient relationship that allows the patient to be candid.” Thing.”

Gilani agrees with Schultz’s observations about the opioid crisis, but there are “indirect links” between many of the factors feeding AI systems, and the tools are data-driven to reduce those deaths. said that it could be useful for

Eurich says that a “significant portion” of poor outcomes associated with opioids are not caused by street drugs, but particularly by early prescription use.

Patients, he said, continued to be exposed to opioids as pain relievers and eventually used the health care system to “get massive amounts of opioids and end up being cut with other substances.” .

Urich said the machine would provide “good continuity” even if a patient changed doctors, reducing the chance of harm from prescription drugs.

This report by the Canadian Press was first published on February 18, 2023.

——

This article was produced with financial support from the Meta and Canadian Press News Fellowship.