JMIR Publications

ABSTRACT

Background:

While opioid analgesic medications (OAs) can effectively treat pain, many patients experience dependence, sedation, and overdose. OA prescription rates are decreasing, but patients still are dispensed millions of OA doses each year. Because most patients are at relatively low risk for OA-harms, risk reduction interventions requiring multiple conversations with a counselor are impractical on a large scale, and a more personalized approach is required.

Objective:

We evaluated whether an OA risk reduction intervention based on reinforcement learning (RL), a field of artificial intelligence, could learn through experience to personalize its interactions with pain patients post-discharge from the emergency department (ED) in order to improve outcomes while conserving risk-reduction counselors’ time.

Methods:

This study is a secondary analysis of data representing 2,439 weekly interactions between a digital health intervention and 228 patients discharged from two EDs with a pain-related complaint who reported opioid misuse in the prior three months. Patients received a 12-week intervention (PowerED) that used RL to target interactions addressing their OA risks. Each week for each patient, PowerED selected among three treatment options: a brief motivational message delivered via an Interactive Voice Response (IVR) call, a more extended motivational IVR call, or a 20-minute live call from a counselor trained in motivational enhancement techniques. The algorithm selected session types for each patient based on patient-feedback, with the goal of minimizing future risk as defined by weekly scores representing patients’ IVR-reported OA use and risk behaviors. When a live counseling call was predicted to have a similar impact on future OA risk as an IVR message, the algorithm favored IVR. We used logit models to estimate changes in the relative frequency of each session type as PowerED gained experience. Poisson regression was used to examine changes in patients’ self-reported OA risk scores over calendar time, controlling for patients’ session number.

Results:

As PowerED gained experience through interactions with the population, it delivered fewer live counseling sessions relative to both brief IVR sessions (P = .006) and extended IVR sessions (P < .001). Live counseling sessions were selected 33.5% of the time in the first five weeks of interactions (95% CI: 27.4%, 39.7%), but only for 16.4% of sessions (CI: 12.7%, 20.0%) after 125 weeks. Controlling for each patient’s temporal changes over session weeks, this adaptation of treatment-type allocation led to progressively greater improvements in OA risk scores (P < .001) over calendar time, as measured by the number of weeks since enrollment began. Improvement over calendar time in self-reported OA risk was especially pronounced among patients with the highest risk at enrollment (P = .02).

Conclusions:

This RL-supported program learned what OA risk reduction treatment modalities worked best to improve outcomes while conserving counselors’ time. RL-supported interventions represent a scalable solution for patients with pain receiving OA prescriptions. Clinical Trial: This RL-supported program learned what OA risk reduction treatment modalities worked best to improve outcomes. RL-supported interventions represent a scalable solution for reducing OA-related harms while more effectively targeting counselor time. Resource-conserving interventions like PowerED could improve OA monitoring and behavior-change support for the millions of people prescribed OAs each year, many of whom require follow-up but are at relatively low risk.