Researchers at the University of Missouri-Columbia have developed a proof-of-concept study exploring how large language models such as GPT-4 can be adapted to provide medically accurate and empathetic guidance to pregnant women experiencing opioid use disorder. The project, led by doctoral student and practicing nurse Drew Herbert from the Sinclair School of Nursing, addresses the barriers these women often face in seeking treatment, including stigma and lack of access to reliable, judgment-free information.
The study involved the creation of a fictional patient, “Jade,” who disclosed her pregnancy and desire to stop using opioids. A fine-tuned version of GPT-4 engaged Jade using motivational interviewing techniques, an evidence-based counseling method that emphasizes empathy and patient empowerment. The chatbot directed Jade to medication-assisted treatment options such as buprenorphine, referencing guidelines from the American Society of Addiction Medicine and the American College of Obstetricians and Gynecologists. The chatbot also provided region-specific resources, including treatment centers, telemedicine providers, and official directories.
Two clinicians with experience in opioid use disorder assessed the chatbot’s responses and found them to be safe, accurate, and relevant in over 96 percent of cases.
Herbert emphasized that such chatbots are not intended to replace healthcare professionals, but to serve as accessible sources of initial guidance that may support individuals in preparing to seek formal treatment. He indicated plans to expand the tool with input from clinicians and individuals who have experienced opioid use during pregnancy, with the broader aim of making accurate treatment information widely available at minimal cost.
Photo credit: University of Missouri
