Dr. Dean Eurich, program director for the clinical epidemiology program at the University of Alberta and the lead investigator of research published with JAMA Network, is shown in this undated handout. THE CANADIAN PRESS/HO-University of Alberta *MANDATORY CREDIT*
EDMONTON – Researchers in Alberta are experimenting with artificial intelligence to measure the risks of prescription opioids amid the ongoing drug overdose crisis across Canada.
While doctors have a set protocol to identify patients at risk of opioid addiction, Dr. Dean Eurich said machine learning “could do a better job” of pinning down who is most susceptible.
The AI-assisted system could provide an additional “level of comfort to clinicians “¦ (knowing) there are also other supports they can use to help (in) making sure the patient is getting the right drug at the right time,” said Eurich, program director for the clinical epidemiology program at the University of Alberta.
With this tool, physicians could predict the impacts of a prescription opioid on patients and save them from unnecessary emergency department visits or even death within 30 days of starting the medication.
Eurich was lead investigator on research published in December with JAMA Network, which analyzed medical data of more than 850,000 Albertans anonymously and predicted the best outcomes for the patients.
The data sets were mainly provided by Alberta Health.
Dr. Fizza Gliani of the College of Physicians and Surgeons of Alberta said machine learning could be an effective way to reduce hospitalizations and morbidity for patients once integrated in the health system.
“The model (could) predict risks of hospitalization,” said Gilani, who is the program manager of prescribing, analytics and the tracked prescription program at the college.
At times, she added, current methods don’t predict the origins of risk and the medical solutions could be more complicated than reducing a patient’s opioid dose.
The AI system was fed with various health factors to determine risks to a patient, including the history of injury, obesity, depression, diabetes, fluid disorder and psychosis. These were combined with diagnoses from doctors, health-care visits andinformation including where the patient lives.
“The idea is not to make physicians stop prescribing opioids, (but) to minimize the risk after the opioid exposure,” said Gilani.
The researchers looked at about three million opioid prescriptions a year from various medical professionals — doctors, nurse practitioners, dentists — to more than 600,000 patients in Alberta. Those who had cancer or were receiving palliative care were excluded.
Eurich said 20 per cent of patients were using opioids with other high-risk drugs, “increasing the risk of adverse outcome.”
Over the years, people’s interaction with the health system has become more complex, demanding an efficient approach to moving through the health-care system, Eurich said.
“As a human, I can look at a couple of dozen variables and predict outcomes, but we’re finding that’s just not enough.”
He said the machine learning takes a different approach, building systematic models with a nuanced set of data, including various key factors, and finding the combinations predicting the best outcomes for a patient.
Eurich, who has been working on AI predictions for more than three years, said the system can “predict correctly (for) four out of every five patients.” A patient identified as high-risk would have a higher chance of being hospitalized within the first 30 days of prescribing the drug, according to the machine.
He added that AI-powered systems could also rapidly adapt to the changing environment – for instance, a sudden spike in opioid-related death during the pandemic.
The goal, Eurich said, is to “reduce the risk of patients who are using high-risk medications that we know can result in poor outcomes.”
Researchers will soon be testing the AI system with real-time data, Eurich said. They will also look into whether the system could limit long-term use of high-dose opioids among patients.
One advocacy group thinks the machine won’t help with the opioid crisis in Alberta.
Moms Stop the Harm co-founder Petra Schulz said most of opioid-related deaths in the province are fuelled by street drugs and not prescription opioids.
“This kind of AI could make the safer alternatives even less available,” she said. “It’s like you’re doing detective work and wanting to figure out what is not going right for the patient instead of developing a trusting doctor-patient relationship, which allows the patient to (speak) openly.”
Gilani agreed with Schulz’s observation on the opioid crisis but said there is an “indirect linkage” between a host of factors fed into the AI system and that the tool could help in reducing those deaths based on the data.
Eurich said a “good portion” of poor outcomes related to opioids is not fuelled by street drugs, but by prescription use – particularly in the beginning.
He said patients continue to get exposed to opioids for pain medication and eventually start using the health system to “doctor shop (and) obtain massive quantities of opioids… also end(ing) up being cut with other substances.”
Eurich said the machine would provide “good continuity of care” even when patients change doctors, reducing their chances of harm from prescription drugs.
This report by The Canadian Press was first published Feb. 18, 2023.
___
This story was produced with the financial assistance of the Meta and Canadian Press News Fellowship.
