AI chatbot shows promise in polypharmacy management, study finds

AI assisting with deprescription process: Theotherhouse - stock.adobe.com

Researchers from the Mass General Brigham MESH Incubator have unearthed promising results in the use of artificial intelligence (AI) to manage polypharmacy and streamline deprescription processes.

The study, led by Arya Rao, a Harvard Medical student and MESH researcher, alongside senior corresponding author Dr. Marc Succi, evaluated the efficacy of ChatGPT, a generative AI chatbot, in medication management. Their findings, published on April 18th in the Journal of Medical Systems, mark a step forward in integrating AI into the complex landscape of health care decision-making.

Polypharmacy, rampant among older adults, elevates the risk of adverse drug interactions, necessitating careful deprescription to mitigate harm. Yet, this decision-making process is intricate and time-intensive, particularly for overburdened primary care practitioners. Thus, the demand for effective polypharmacy management tools is high.

In the study, researchers provided ChatGPT with various clinical scenarios, each featuring an elderly patient grappling with multiple medications. Notably, ChatGPT's responses showcased nuanced decision-making abilities. When confronted with scenarios devoid of cardiovascular disease (CVD) history, the AI consistently recommended deprescribing medications, showcasing a cautious approach in the absence of complicating factors. However, its stance shifted when CVD entered the equation, displaying a propensity to maintain the status quo. Surprisingly, the severity of impairment in activities of daily living (ADL) yielded no discernible impact on ChatGPT's recommendations.

One observation was ChatGPT's inclination to deprioritize pain management, often favoring the reduction of pain medications over other drug classes like statins or antihypertensives. Furthermore, the model exhibited variability in responses across different chat sessions, hinting at potential inconsistencies stemming from its training data.

The implications of these findings extend far beyond the confines of the study. With over 40% of older adults meeting the criteria for polypharmacy and an escalating reliance on primary care providers for medication oversight, the demand for innovative solutions is palpable. Succi underscores the significance of AI integration, stating, "AI-assisted polypharmacy management could help alleviate the increasing burden on general practitioners."

Rao echoes this sentiment, emphasizing the pivotal role of AI-based tools in safeguarding older adults' medication practices. However, she advocates for continuous refinement to accommodate the intricate nuances of medical decision-making.