Copyright MassLive
Over six months, more than 60 people receiving medication-assisted treatment for opioid use disorder answered surveys on their smartphones about their mental health, psychological state and environment — three times a day. When deep learning models — a form of artificial intelligence — analyzed their responses, the technology forecasted relapse risk and likelihood of continuing treatment exceptionally well, according to a study published in the Journal of Substance Use and Addiction Treatment in June. The National Institutes of Health-funded study — which included researchers from New Hampshire, New York, California, Massachusetts and Maryland — found AI using real-time monitoring has the potential to serve as a strong predictive tool and early-warning system. It could pave the way for proactive, personalized interventions. This would be particularly valuable when someone in active treatment may be teetering on the edge, the study indicates. In this instance, pairing daily smartphone surveys with AI-based prediction models resulted in high accuracy for assessing next-day opioid relapse, the researchers determined. The idea is that real-time predictions can be relayed to clinicians and recovery supports, giving them an opportunity to intervene if someone is at risk. Read more: ChatGPT for genetics: Boston AI startup combines DNA, search engines and primary care Influential factors considered by the deep learning models included past-hour substance use, which was the strongest indicator that someone would use the next day, situational risk such as seeing or being near drugs, mood, difficulty self-regulating and social/environmental contexts. “Our work is the first to use personalized, naturalistic features to predict clinically relevant outcomes in persons receiving medications for opioid use disorder,” the study’s authors wrote. They noted the “considerable public health importance” of their findings, given the challenges associated with successful, ongoing treatment for those in the grips of drug addiction. Among the five to seven million people who are estimated to have an opioid use disorder nationally, the U.S. Food and Drug Administration estimates relapse rates of 65-70%. AI in addiction treatment, health care Just as AI is being integrated into the broader health care landscape, it’s also making itself known in addiction medicine and treatment. In April, a National Institutes of Health-supported clinical trial found AI tools can be as effective as health care providers in generating referrals to addiction specialists, for example. Researchers at Dartmouth College trained AI models to analyze Reddit posts where users discussed their experiences with opioid use disorder treatment. There are also AI-powered chatbots intended to serve as recovery supports. “It’s crucial to understand that (AI) is not intended to replace human therapists,” Caroline Easton, academic division chief of addiction psychiatry at the University of Rochester Medical Center, said in an interview last year. “Instead, it supports them. AI can help reduce therapists’ workloads, minimize compassion fatigue and burnout and provide additional resources for patient care.” Read more: Overdose deaths of Black Bostonians dropped significantly in 2024. Here’s why There are many controversies surrounding the implementation of AI in health care globally, including concerns around data privacy, potential inaccuracies and the perceived replacement of trained physicians and medical staff. Brenda Curtis, a researcher at the National Institute of Drug Abuse, has acknowledged “trepidation” about potential AI uses, but said her field is “taking a hard look at things.” “(We) can have a voice in using and improving technology,” Curtis said. “We can use it to dispense large amounts of information quickly, and we can use it to help treatment and health care be more efficient and effective.” Researchers acknowledged limitations in the study combining AI and smartphone data, most notably that predictive models are not always accurate. The study yielded both false positives and false negatives, and the sample size was described as “modest” with limited demographic diversity. Deep learning models trained on predominantly white populations in a certain geographic area, for example, might misinterpret behaviors or health indicators from other racial or ethnic groups — an example of how AI can reflect, or in some cases amplify, already existing disparities. Read more: Mass. patients will be first in US to access these revolutionary AI health care technologies How the study worked The study centered on 62 adults receiving buprenorphine — a prescription medication to treat opioid use disorder — at an outpatient addiction treatment clinic in California between June 2020 and January 2021. In total, 14,322 observations from the daily smartphone surveys were collected. Perhaps the most obvious predictor, participants were found most likely to relapse the next day if they had self-reported substance use in the past hour. Situational risk factors, such as triggers like stress or seeing drugs, as well as boredom, exhaustion, low contentment or inability to plan or follow through, were also strong indicators, the study found. Mood predictors, particularly boredom and exhaustion, and being in high-risk environments where substances were present, successfully signaled increased risk for opioid relapse within hours, the study found. Stress and pain, for example, forecasted relapse risk several days in advance, potentially giving clinicians a valuable window of time to intervene and offer support. The AI models weren’t as accurate in forecasting whether a patient might skip their buprenorphine medication.
