(P22) Utilizing AI to Deprescribe Potentially Inappropriate Medications in Nursing Homes

Objective/Aim: The aim for this QI project is to assess the effectiveness of ChatGPT 4.0 in medication review by identifying potentially inappropriate medications (PIMs) based on the 2023 Beers Criteria and achieving a 10% reduction in PIMs among nursing home residents over three months.

Background: The American Geriatrics Society and CMS report that 14.5% of nursing home residents receive antipsychotics, 10% are prescribed benzodiazepines, and 23% are on anticholinergic medications. The 2023 Beers Criteria provides guidelines for identifying PIMs in older adults. This Quality Improvement project explores the use of ChatGPT 4.0 to conduct medication reviews, aiming to identify and deprescribe PIMs in a nursing home setting.

Quality Improvement Methods: This prospective QI initiative was conducted in a 150-bed nursing home, where 30% of residents are in a memory care unit and 30% are short term stay rehab patients. Medication lists for all residents were reviewed using ChatGPT 4.0, which identified PIMs as defined by the 2023 Beers Criteria. The image of the medication lists without the resident's name and identifiers and uploaded to the program to review medications. The medical director verified the flagged medications, and recommendations were communicated to providers. Nursing home pharmacists also reviewed PIMs. The medication administration records were analyzed to identify unused as needed medications.

Results : Chat GPT 4.0 was able to identify PIMS with 2023 Beers Criteria. Over a three-month period, 175 residents’ medications were reviewed, averaging 11.6 medications per resident. The prevalence of antipsychotic use was 11.7% before intervention. Commonly prescribed PIMs included tramadol (16), gabapentin (14), lorazepam (12), oxycodone (11), divalproex (10), acetaminophen-hydrocodone (9) and quetiapine (7). The most frequently discontinued PIMs were tramadol (8), lorazepam (4), hyoscyamine (4), morphine (3), and oxycodone (3). Of the 15 antipsychotics flagged, 3 were deprescribed and post intervention antipsychotic prevalence was 9.7%, achieving more than 10% reduction of antipsychotic use. 15 benzodiazepines were identified, of which 6 were discontinued following the intervention resulting in a 40% reduction of the benzodiazepines. In total, 135 PIMs were identified and 34 were discontinued (25.2%). The average number of medications per resident decreased to 10.1 after the intervention and no adverse outcomes were reported from deprescribing.

Conclusions: This QI initiative, implemented in a well managed nursing home with already lower than average antipsychotic and hypnotic use, reduced the prevalence of PIMs further over three month period. The ChatGPT 4.0 effectively identified PIMs according to the 2023 Beers Criteria, suggesting AI programs potential as a supportive tool for clinicians in reducing PIM and enhancing safer prescribing practices in nursing homes. AI programs can also assist patients and family members in understanding PIMs and complexity of polypharmacy.