Feasibility of applying generative AI to create managed care pharmacy formulary reviews
Michaela Mahan, Lindsey Fronizer, Sharhonda Exantus, Insiya Poonawalla, Suzanne Dixon, Jeremiah RastegarAbstract
Purpose
Formulary reviews are used with other data sources and clinical judgement to support formulary strategies, such as placement for existing and new-to-market drugs. We describe the pharmacy resident user experience and assessed the feasibility of using clinical and research-focused generative artificial (genAI) tools in a managed care setting for creating formulary reviews for 2 complex and rare diseases: multiple sclerosis and age-related macular degeneration. This allowed genAI feasibility assessment for disease states with less well-defined guidelines and fewer data sources from which to draw information.
Summary
Users reported a steeper learning curve with research-specific genAI systems compared with pharmacy-focused tools. Users noted pharmacy-focused genAI tools may produce more current drug pipeline information, relative to research-focused AI systems. Prompt engineering, the skill of creating effective prompts to elicit accurate information from genAI models, also contributed to the quality of the evidence produced, regardless of the tool used. The pharmacists noted time savings when utilizing AI compared with a completely manual data generation process involving individual searches of PubMed, business reports, FDA websites, and pharmaceutical company-related information sources.
Conclusion
Pharmacists need adequate time and training to optimize prompt techniques to produce quality health- and medication-related information and must be educated on potential downsides of genAI such as inaccurate or incomplete results. With currently available tools, all generated output required careful human fact checking. With continued use and accumulating experience, the pharmacist may become more efficient over time with genAI support for formulary reviews