A GPT4-based Patient Simulation to Enhance the Integration of Pharmacology with Clinical Decision Making in a PharmD Program

Abstract ID 95086 Poster Board 220 Artificial intelligence (AI) is a powerful tool with the potential to engender disruptive and transformative educational changes in pharmacology teaching and learning. While legitimate concerns about AI focus on potential plagiarism, novel strategies that apply "out-of-the-box" thinking promise to revolutionize pedagogy, assessment, practice, implementation, and learner engagement. We implemented generative AI (GPT4)-based patient simulators for Team Based Learning application exercises in the immunology-pharmacology course sequence in our four-year Doctor of Pharmacy (PharmD) program. Our collaborative inter-university team of pharmacology and pharmacotherapy faculty, PharmD and graduate students created 11 GPT4-based "artificial patient simulators" to interact with student teams and present an immunology case in a patient counseling setting. An online platform using Amazon Web Services with user authentication was created by the pharmacotherapy team at VCU which hosted the GPT4-student teams interaction interface. Faculty and a PharmD student created a splenectomy patient case using peer-reviewed StatPearls and developed a rubric for the GPT4 to grade students on their responses. While the case was the same in content, care was taken to impart a different patient "personality" to each ChatGPT4 "patient," essentially creating many different scenarios. The code for the generative AI was published in an open-access journal after first piloting it in Enterprise Google Cloud and eventually migrating it to Amazon Web Services. From an instructional design perspective, the indications for splenectomy, patient symptoms, and pathophysiology were integrated into the patient case and designed to be presented by the AI chat bot. Twelve student teams (N=59 students) were presented the patient case by the AI bot through stepwise interaction. The AI bot presented dynamically altered scenarios based on the questions asked by the student teams. The teams were required to create a pharmacotherapeutic care plan, with emphasis on counseling regarding specific infections common in splenectomy patients as the pharmacists' responsibility. We utilized a mixed methods approach to evaluate the impact on student learning and perceptions of learning. Data from student responses were automatically graded by the AI-bot using our rubric. Overall, our results showed that all student teams were scored by the AI-bot at least at a 75% competency level in assessing the patient simulation, identifying the immunology disease state, and patient counseling. Furthermore, all teams graded the activity above 90% in the following categories on the student perception instrument: 1) engagement, 2) optimal use of class time, 3) enhancing understanding and application of the material, and 4) confidence-building. Among the limitations identified by students were the time required for the class activity and the potential to use a similar ChatGPT interface to resolve the case. We plan to address this through the use of technology in future iterations by enabling browser lockdown and geofencing technology which stops students within the premise from accessing public facing generative AI models in that location and time.