This podcast will focus on how generative artificial intelligence (AI) impacts administrative aspects of pharmacy practice. We will focus on current technologies and future applications of generative AI in this space.
The information presented during the podcast reflects solely the opinions of the presenter. The information and materials are not, and are not intended as, a comprehensive source of drug information on this topic. The contents of the podcast have not been reviewed by ASHP, and should neither be interpreted as the official policies of ASHP, nor an endorsement of any product(s), nor should they be considered as a substitute for the professional judgment of the pharmacist or physician.
[MUSIC] >> Welcome to the ASHP official podcast, your guide to issues related to medication use, public health, and the profession of pharmacy. >> Welcome everyone. In this episode, we will be discussing generative AI tools for data analytics. If you're an ASHP member, you also have the opportunity to earn continuing education credit for listening to this episode. Stay tuned to the end of the episode for more information. My name is Mike Reid. I'm a clinical informatics pharmacist at University of Wisconsin Health. Our guests today are Kelvin Tran, the health science specialist at the Veterans Administration, and Diana Schreier, a senior manager of reporting and analytics at Mayo Clinic. All right. Let's jump right in. To start off, can one of you provide a brief definition of what generative AI is? >> Yeah, sure, Mike. From IBM, generative AI refers to deep learning models that can generate high quality text, images, and other content based on the data they were trained on. In general, AI can help with streamlining the pharmacy workflow by simplifying some repetitive or mundane tasks, thus reducing the administrative burden, and freeing up the pharmacy personnel to focus on other aspects of patient care. Additionally, we find that AI can be integrated into various areas of the pharmacy practice, such as in the electronic health record, automated dispensing systems, or in the research setting. Ultimately, the goal of integrating artificial intelligence in any pharmacy space would be to improve patient safety and outcomes, enhance productivity, or find ways to help the pharmacy staff in achieving those goals. >> Thank you, Kelvin. What are some of the opportunities for artificial intelligence in the analytics space? >> Well, with machine learning, automation, and software development, a growing interest in pharmacy informatics, support and writing code is becoming more important in the field. Opportunities for AI in this space include enhanced efficiencies in code writing, and spending less time on routine data-related tasks, more time on problem solving, collaboration, and innovation. >> Big points, Kelvin, bouncing off from your thoughts. Another way that these tools may be able to enhance efficiencies is by making it easier to ingest and interpret data. Oftentimes, once we have data in our hands, it still remains a significant burden to figure out what's going on, and what are the overarching data trends? Generative AI has the capability to provide assistance with that process. >> Thank you, Diana. >> We all know that visual and linguistic generative AI tools are available to the general public. What exists now for use in analytics? >> That's a good question, Mike. There are many different AI coding assistants, just to name a few Tab9, BlackBox AI and GitHub co-pilot. >> There are also many tools that allow for visualization of data, including platforms such as Power BI, Click, and Tableau. We'll cover examples of how visualization platforms are implementing generative AI tools to assist in content interpretation and analysis. >> We'll also discuss AI coding assistants. In general, AI coding assistants have the ability to be trained in all languages that appear in public repositories. They're powered by generative AI models and trained on natural language text and source code from publicly available sources. They also can be available as an extension in multiple editors, just some examples of these editors include Visual Studio Code, Visual Studio, and Azure Data Studio. >> Data summarization tools provide a suite of capabilities to bring generative AI to dashboard platforms. These tools provide smart, personalized, and contextual data insights directly within a dashboard. Specifically, these tools can provide summaries of metrics, graphs, and tables in plain language. This reduces the cognitive burden of reviewing and trending information within a dashboard. Additionally, these tools can proactively anticipate questions a user might typically ask or suggest questions they might not have otherwise thought of. These tools are in their infancy, but represent a substantial opportunity to use the power of AI to reduce cognitive burden associated with interpreting data and understanding data trends over time. >> Wow, sounds like a lot of very powerful tools available. So what are some of the uses and opportunities that you see for these tools? >> So AI coding assistants may be trained on public code and can generate suggestions using probabilistic determination or reasoning instead of simply copying and pasting from a code base. According to research from one of the AI coding assistants in use, developers reported 75% higher satisfaction of their jobs than those who didn't use them and up to 55% more productivity in writing their code. >> Those are some big numbers, Kelvin. For dashboard summaries, one of the most significant opportunities afforded by generative AI tools being applied for the creation of dashboard summaries is the ability to use resources more effectively. Specifically, a lot of the time when we think about making dashboards or key performance indicators or other high priority metrics, we often need to make dashboards specific enough that an end user can ingest data effectively for their use case. Well, this is often great for an individual requester. It often limits the applicability of a dashboard to other areas. So for instance, in your institution, you may have separate dashboards for practices in internal medicine, surgery, and anesthesia, for example, because though they are sometimes interested in the same things, there are certain metrics that don't make sense to report on across all specialties. For example, surgical complication rates are typically not relevant to an internal medicine practice, but would be critical information for a surgical or anesthesia practice to review. Historically, we've been back to the corner of making multiple dashboards to appease the needs of the various practices because dashboards become cluttered and overwhelming when they are too generalized or provide content that isn't always relevant to an end user. AI generated dashboard summaries provide the distinct advantage of sorting out all of the clutter for an end user. In this system, it would be possible to make highly generalized dashboards with metrics that are applicable across practices. And then the end user could request summaries specific to the practice area that they are interested in. In this strategy, YouTube's benefit from highly customized content will report builders benefit from reduced development, implementation, and maintenance work, ultimately meaning that both parties win. - Thank you, Diana. Calvin, what are some of the limitations and risks of AI coding assistance and other tools? Some of the ones you were just talking about for dashboards where we were just listening to? - Well, one of the limitations of AI coding assistance is that although they can be trained on all languages, the quality of suggestions for a particular language may depend on the volume and diversity of training data for that language. So languages with less representation and public repositories may provide few or less robust suggestions. Affordability and cost may also be a limitation with AI coding assistance. Cost is variable between different AI coding assistance, but can be just for example, $10 a month or $100 a year for individual users or $19 per user per month for businesses. - Like what Calvin mentioned with the cost associated with AI coding assistance, AI generated dashboard summaries will also come at a cost. Though there are some platforms offering free beta test versions of dashboard summaries, it is likely that once the transition into products that are officially released, companies will start charging for them. It's also relevant to note that there are many proposed cost structures for these tools, including a flat rate or fee per use schedule. The flat rate model is self-explanatory, but the fee per use model specifically is one where the institution is charged about 25 cents per dashboard summary generation. There's probably pros and cons to this model, but in any event, the charges associated with utilizing these tools could certainly add up. The other significant limitation of dashboard summary tools is that though they can identify trends in data, they are not able to assess why trends are occurring and what needs to be done about them. Organizational leadership will need to continue to investigate causes when issues arise and strategize solutions to resolve them. And then lastly, sometimes summarizations created by generative AI models are incorrect or misleading. So organizational leaders will still need to maintain those data literacy skills so they can ensure that they take appropriate action even when summaries are misleading or inaccurate. - Thank you both for that discussion of the risks and limitations of AI. I'm sure all of our audiences would love to know how these AI tools are going to affect them, fit in with their jobs, their workflow activities. What are your thoughts? - AI coding assistants are more of a supplemental aid in writing code and can be used in such areas as research pharmacy where statistical analysis needs to be conducted. Analysts are still directing the code and often need to write instructions to provide contacts for the coding assistant to auto-complete the code. Analysts still need to be familiar of the code to look out for deprecated code or older code that will be removed in the future and to toggle between possible auto-completion options to use the most appropriate code. Through assistance with AI coding aids, analysts can spend less time on routine analytical tasks and spend more time on tasks with greater complexity such as selection and interpretation of appropriate statistical analyses. In addition to auto-completion, some of the AI coding assistants may also have a chat feature for coding support, explanation of code that maybe you don't quite understand and also help with debugging to increase efficiency. So in general, how coding assistants are impacting pharmacists is that you still have to know like code, but it gives you a very large advantage in understanding code, debugging it, and also increasing your efficiency. - Great points, Kelvin. As I had mentioned earlier in the podcast, AI tools certainly represent a great opportunity to reduce the cognitive burden associated with interpreting data. These tools are in their infancy, but there are ongoing pilot testing efforts through many data visualization tools like Tableau, Power BI and Click, as well as numerous electronic health record vendors. Pharmacists can use these summaries to more easily analyze and act on data, but their greatest impact is likely to be a few years off in the future still. My recommendation to pharmacists that are interested in learning more would be to reach out to your organization's informatics team to determine what initiatives they have underway and offer your time to assist with any future pilot testing at your organization may pursue. The more people that we have using these tools and the more feedback generated, the like we are that we will have very robust tools on a short timeline. - Thank you. It looks like that these AI tools are gonna be beneficial both to the person who actually is writing code and to the end user pharmacist. Well, that's all the time we have today. Thank you to our guests, Kelvin and Diana for a great topic and discussion. For our AACT members, you can find additional resources in our free continuing education for listening to this episode by visiting elearning.ashp.org/podcast. Please note that the continuing education credit expires two years after the date this episode is published. If you've enjoyed today's episode, be sure to subscribe to ASHP official through your favorite podcast provider and see you next time. - Thank you for listening to ASHP official, The Voice of Pharmacists Advancing Healthcare. Be sure to visit ashp.org/podcast to discover more great episodes, access show notes and download the episode transcript. If you loved the episode and wanna hear more, be sure to subscribe, rate or leave a review. Join us next time on ASHP official. (upbeat music) (upbeat music)
