SHM Converge Scientific Abstract ePoster Gallery-Improving Immunosupppressant Administration Safety With Gpt: A Prospective Validation

Pdf Summary

This project evaluated a GPT-enhanced clinical decision support system designed to improve safety for hospitalized patients receiving immunosuppressant medications. The goal was to detect real-time discrepancies between a clinician’s documented plan and the actual electronic medication orders, which can be especially important for drugs such as tacrolimus, sirolimus, and mycophenolate mofetil due to their narrow safety margins and complexity.<br /><br />The system was developed at an academic medical center and integrated into the production electronic health record across five hospitals in July 2023. It used natural language processing to parse inpatient notes and identify mentions of target immunosuppressants. If a note mentioned one of these medications but no corresponding order was found within 15 minutes, the case was flagged. A sequential GPT-based filtering layer was then used to assess intent and reduce false positives, especially when the note referred to future or conditional plans rather than active orders.<br /><br />In a five-month prospective evaluation across more than 44,000 admissions, the initial NLP model identified 53 potential discrepancies. After GPT-based refinement, 28 false positives were removed, leaving 25 alerts that were sent to first-contact providers, averaging about five alerts per month. These alerts were reviewed by the development team and a clinical champion for quality improvement and further analysis.<br /><br />The authors concluded that while such discrepancies are uncommon, they do occur and can be meaningfully identified using a GPT-enhanced NLP approach. Future work will examine whether these alerts change provider behavior and improve patient outcomes.

Asset Subtitle

Paawan Punjabi

Meta Tag

Author List Fritz Francois, Jonathan Austrian, Paawan Punjabi, Sapna Mehta, Walter Wang, Yindalon Aphinyanaphongs

Category Innovations

Concept Clinical Decision Support

Concept Immunosuppressant Safety

Concept Medication Discrepancy

Concept Natural Language Processing

Concept GPT Filter Layer

Distinguished Non-Finalist

Presenter Organization NYU Langone Medical Center

Presenting Author Paawan Punjabi

Track Technology in Hospital Medicine

Keywords

GPT-enhanced clinical decision support

immunosuppressant medications

electronic health record

natural language processing

medication order discrepancies

tacrolimus

sirolimus

mycophenolate mofetil

patient safety

hospitalized patients

Clinical Decision Support

Immunosuppressant Safety

Medication Discrepancy

Natural Language Processing

GPT Filter Layer