By Gina Shaw
The growing potential roles of artificial intelligence, including large-language models and generative AI, were a major topic of discussion at the 2024 NASP Annual Meeting & Expo, in Nashville, Tenn., with applications in clinical trial recruitment and patient engagement among the many uses cited.
In specialty pharmacy, AI can potentially enhance patient engagement by personalizing communication and support, leading to improved adherence and satisfaction, suggested Michael Gannon, the senior director of product management for Loopback Analytics.
The technology could facilitate clinical trial development by streamlining patient screening, identifying suitable candidates more efficiently and accelerating the approval of new therapies, Mr. Gannon noted.
AI applications that focus on quality and safety could analyze clinical notes to pull out adverse effects and unintended benefits. “You could potentially use natural language processing [NLP] to extract information that you weren’t necessarily looking for, and have that information to proceed with different label extensions,” he said.
Health-system specialty pharmacies have the unique advantage of being able to use AI in conjunction with existing tools embedded in patients’ electronic health records, said Megan Rees, PharmD, the therapy-specific outcomes product manager for Loopback Analytics. Those tools include flow sheets, smart forms and rapid alerts, “which are critical for tracking patient-reported outcomes,” Dr. Rees said.
In such settings, the NLP component of AI does a lot of the heavy lifting, according to Mr. Gannon. (NLP is a branch of AI that focuses on giving computers the ability to understand human language.)
He described the various steps involved when NLP analyzes a clinical note to extract, identify and categorize medication side effects or patient-reported outcomes. “First, you would get some sort of input text from an unstructured clinical note, and it then goes through a series of processing steps where it chunks the text into sentences, tokenizes and categorizes via three main pipelines,” he explained. “Named entity recognition identifies specific contexts and associates words with those contexts; relation extraction identifies the rela-tionships between one or more named entities; and entity resolvers determine which real-world entities are the same, even if they are described differently.”
There are challenges to be overcome with using NLP in clinical notes in the pharmacy, he acknowledged, including variability in documentation and the accuracy of the models involved.
“One way to improve that would be to utilize generative AI to pre-process clinical notes, to provide additional context and clarity to im-prove the model,” he suggested. “We could also go in and grab a sample of 100 or 500 different notes, and have the annotator go in and review for key clinical terms. Those terms could be tagged and then adjusted in an iterative process to improve accuracy and refine its recognition capabilities.”
The sources reported no relevant financial disclosures.