Mobile Peer-Support for Opioid Use Disorders: Refinement of an Innovative Machine Learning Tool.

Background: The majority of individuals with Opioid Use Disorder (OUD) do not receive any formal substance use treatment. Due to limited engagement and access to traditional treatment, there is increasing evidence that patients with OUDs turn to online social platforms to access peer support and obtain health-related information about addiction and recovery. Interacting with peers before and during recovery is a key component of many evidence-based addiction recovery programs, and may improve self-efficacy and treatment engagement as well as reduce relapse. Commonly-used online social platforms are limited in utility and scalability as an adjunct to addiction treatment; lack effective content moderation (e.g., misinformed advice, maliciousness or "trolling"); and lack common security and ethical safeguards inherent to clinical care.

Methods: This present study will develop a novel, artificial-intelligence (AI) enabled, mobile treatment delivery method that fulfills the need for a robust, secure, technology-based peer support platform to support patients with OUD. Forty adults receiving outpatient buprenorphine treatment for OUD will be asked to pilot a smartphone-based mobile peer support application, the "Marigold App", for a duration of six weeks. The program will use (1) a prospective cohort study to obtain text message content and feasibility metrics, and (2) qualitative interviews to evaluate usability and acceptability of the mobile platform.

Anticipated findings and future directions: The Marigold mobile platform will allow patients to access a tailored chat support group 24/7 as a complement to different forms of clinical OUD treatment. Marigold can keep groups safe and constructive by augmenting chats with AI tools capable of understanding the emotional sentiment in messages, automatically "flagging" critical or clinically relevant content. This project will demonstrate the robustness of these AI tools by adapting them to catch OUD-specific "flags" in peer messages while also examining the adoptability of the platform itself within OUD patients.