The arrival of a generative artificial intelligence chatbot known as Chat Generative Pre-Training Transformer, or ChatGPT, in November 2022, feels like the start of the future. With its ability to generate text, images and other types of content with just a few commands, ChatGPT literally changed the landscape overnight. Suddenly, people now have access to technology with limitless content creation potential.
One of the biggest advantages of using generative AI is its promise of saving users time by performing routine tasks in extraordinary ways. According to a recent survey, more than one-third of infectious disease doctors are fatigued, and spending time doing endless mundane tasks plays an important role. (1) So how can generative AI, specifically chatGPT in its current iteration, ChatGPT-4, (2) help day-to-day ID doctors?
I’ve been tinkering with the latest version of this chatbot to define some common ID doctor tasks that could be improved with generative AI so that we spend less time on routine work and more on interesting things, like spending more time communicating with patients. and colleagues, starting new projects, and reading the books and journals we promised ourselves we would read if we had more time.
1. Improve your clinical notes and save time writing them.
One of the most repetitive and time-consuming tasks we face as doctors is the completion of medical records. It’s true, documentation is often cited as a major cause of physician burnout. (3) This is especially true for ID doctors, who are often tasked with making extensive reports. While dictation serves as a useful tool, it is not always accessible, and it can result in text that, even when spelled correctly, may not accurately reflect the intended context.
This is where ChatGPT comes into play. You can ask him or her to convert medical shorthand, even with intentionally incomplete and misspelled sentences, into a comprehensive, well-articulated, and intact clinical record. The key is choosing the right leads. For my example, using a hypothetical patient, I use “rewrite as a medical clerk”:
This can also cause the opposite. For example, if you need to relay your discharge process to a colleague who values brevity and simplicity, you can ask ChatGPT for help using the “rewrite in medical abbreviation” command:
There are also technologies that aim to integrate generative AI into dictation, which is a likely direction for medical record writing.
2. Need help to make a differential diagnosis of a case? Does not matter.
In light of recent studies showing that ChatGPT not only successfully passes the US medical licensing exam but also outperforms medical students(4), the idea of leveraging AI as a complement to our medical knowledge base holds significant promise. While AI may not be ready to replace human doctors today, it can play a role in information management, providing us with contextual and comprehensive data at the click of a button. After all, the knowledge bank that ChatGPT uses for content creation is just as extensive as the internet-based information available to us.
This AI chatbot is not intended to replace our role in handling cases, but rather as a useful reference and maybe even more than that. While listening to the grand round of ID cases, I asked the chatbot to make a differential diagnosis (no identifying information included here):
The patient is known to have malaria.
3. Generate easy-to-understand content for your patients.
One of the early applications of ChatGPT among doctors was to improve bedside manners. This is achieved by producing doctor’s texts or talking points that are empathetic and easy for patients to understand. It can be used in dealing with complex ID cases. I asked chatGPT to do a compassionate script on endocarditis and drug use:
4. Summarize the paper for you.
Given our large workload, it can sometimes seem impossible to peruse all the papers we’re interested in at the end of the day. The latest iteration of the chatbot, ChatGPT-4, can help you. With its enhanced ability to combine “plug-ins” with the features you need, you can further enhance chatbot capabilities. There are plug-ins for video creation, code reading, Wikipedia, etc. Some can summarize entire medical articles in a succinct manner. In addition, you can engage the chatbot in a question and answer session about the article, including asking for its critique of the paper.
Here, I use the “Ask your PDF” plug-in in chatGPT-4 to “read” article (5) on infectious diseases:
Conclusion
Despite its advanced capabilities, ChatGPT has some limitations. The accuracy of its response depends in large part on the nature of the command it receives, which can lead to contextually incorrect content if an inappropriate command is used. Also, since it retains all data unless you delete it, there is a potential risk of breach of patient confidentiality. To mitigate this risk, it is critical to exclude identifiable patient information from data entry, and, where necessary, institutions should establish clear protocols on how to properly use this technology. Finally, as previously mentioned, ChatGPT is not a substitute for professional judgment. It is important to consistently verify the output against reliable sources.
Although ChatGPT is the party that started all of this, it is not the only player. The chatbot based on Google’s large language model, “Bard,” can do something similar and will likely be integrated into Google’s mobile and web-based app ecosystem. It may be worth trying both to determine which best suits your needs. Also worth checking out are other ChatGPT plugins that can improve productivity and content creation.
Generative AI capabilities are expected to grow and expand significantly in the coming years, as integration into healthcare and physician practice becomes increasingly inevitable. Taking the initiative to understand its potential benefits, applications, and limitations is an excellent first step in welcoming the future of today’s technology.
Reference
1.Koval ML. Medscape Infectious Disease Doctor’s Lifestyle, Happiness & Fatigue Report 2023: Contentment Amidst Stress. February 24, 2023.
2. ChatGPT-4. https://chat.openai.com/?model=gpt-4. 2023.
3. Yan Q, Jiang Z, Harbin Z, Tolbert PH, Davies MG. Exploring the relationship between electronic health records and provider fatigue: A systematic review. J Am Med Inform Assoc. 2021;28(5):1009-21.
4. Strong E, DiGiammarino A, Weng Y, Kumar A, Hosamani P, Hom J, et al. Chatbot Performance vs. Medical Students on the Free Response Clinical Reasoning Exam. JAMA Internship Med. Published online July 17, 2023.
5. Maertens J, Cordonnier C, Jaksch P, Pear X, Uknis M, Wu J, et al. Maribavir for the Prevention of Cytomegalovirus Reactivation Treatment. N Engl J Med. 2019;381(12):1136-4
#artificial #intelligence #infectious #disease #doctor
