Original Research
Sunday, October 13, 2024
12:45 PM–02:15 PM
Abstract
Introduction: Generative AI (artificial intelligence) has the potential to transform clinical pharmacy practice by analyzing complex medical data and providing personalized treatment recommendations. This study compares clinical recommendations for type 1 diabetes management, based on the American Diabetes Association’s Standards of Medical Care – 2022, generated by two AI platforms: ChatGPT-4 and Google Bard.
Research Question or Hypothesis: Both platforms can achieve internal consistency within each platform and external consensus between platforms and with a clinical pharmacist’s decisions.
Study Design: AI-generated responses were reviewed and compared to the clinical pharmacist’s assessment and plan using a qualitative approach.
Methods: A complex clinical case involving type 1 diabetes with a history of severe hypoglycemia and diabetic ketoacidosis (DKA) was selected. The clinical note, including the chief complaint and subjective and objective information, was edited for clarity and ensuring HIPAA compliance. The case was analyzed by both platforms three times. We then compared the AI-generated recommendations with the assessment and plan written by the clinical pharmacist.
Results: Comparison of ChatGPT-4 and Google Bard revealed key similarities and differences in safety and effectiveness recommendations. ChatGPT-4 recommended ketone strips, consistently mentioned glucagon for hypoglycemia management, and emphasized insulin injections, patient education, and CGM (continuous glucose monitoring) troubleshooting. Bard focused on re-educating DKA prevention techniques and updating emergency plans for hypoglycemia. Both platforms recommended consulting dietitians and adjusting insulin doses. The clinical pharmacist’s note highlighted gaps in patient adherence and data reporting, emphasizing the importance of patient adherence for effective management.
Conclusion: This pilot study demonstrates both platforms can generate clinical recommendations with internal consistency, particularly in effectiveness measures. ChatGPT-4 provided safety recommendations more closely aligned with the Standards of Medical Care. Further analyses with more cases are warranted to explore the potential of AI to augment clinical practice in managing type 1 diabetes.
Presenting Author
Clipper Young PharmD, MPHTouro University California
Authors
Shirley Wong PharmD
Touro University California