AI's role in healthcare: From diagnostics to communication, and the path forward.
When chatGPT-4 was asked, “Give a definition of artificial intelligence,” this is what it produced: “Artificial intelligence (AI) is a branch of computer science that deals with developing algorithms and systems that can perform tasks that typically require human intelligence, such as visual perception, speech recognition, decision-making, and language translation. The goal is to create machines that can simulate human intelligence and perform tasks that would typically require human intervention. This can be achieved through various techniques, including machine learning, deep learning, and natural language processing. AI has the potential to revolutionize many industries, including health care, finance, transportation, and customer service, by automating routine tasks, improving accuracy and efficiency, and providing personalized experiences for users.”
AI’s capabilities are difficult to keep up with because the field is rapidly advancing. AI already has started playing a major role in tasks such as health care, transportation, and finance. Its capabilities allow it to make decisions like those a human brain might make, but at a much faster rate by analyzing large amounts of data in a short period of time. This gives AI the potential to detect and predict diseases, revolutionize traffic conditions, and optimize stocks effectively. AI can also offer greater accuracy, which can improve processes and raise efficiency levels significantly.
Undoubtedly, AI will change health care drastically over the next several years and likely will be ubiquitous in the next 10 years. Moore’s Law implies that processing speed doubles every 18 months, which would result in improvements 16 times the present day in 6 years. Programming will lead to further dramatic leaps in accuracy and efficiency. Physicians and systems poised to learn about these technologies and leverage output while being mindful of shortcomings will have distinct advantages over those who continue operations as the status quo.
Automating or improving efficiency and diagnosis
AI is being utilized to analyze medical images to provide accurate diagnoses and reduce the need for manual interpretation for health care providers. Other systems assist in working within hanging protocols to work more consistently, resulting in improved efficiency. AI will be able to enhance image acquisition in ultrasound, and it will reduce time in acquiring standardized images and labeling. New technologies aim to pull data from images that the average health care provider does not see or notice to attempt to predict medical conditions such as the risk of preterm delivery.
Systems soon may be able to assist with acquiring data from past medical visits to provide summaries of past care and ease the burden of examination of pages of documents. Methods currently exist within the law to evaluate key contract phrases and flag areas for further study and provide an overview of documents.
Currently, our practice uses AI with chatGPT-4 to assist with many essential but time-consuming tasks, such as developing patient information sheets using prompts such as, “Provide a specific patient instruction sheet for precautions and care for a patient with (medical condition). Please write in a 7th-grade English level.” This output is then checked for correct output. The result is then copied and pasted and converted to other languages as needed. chatGPT-4 also can assist in drafting letters to insurers for contracting, denials management, and other issues. It can change the letters’ tone as needed by asking for a specific approach (friendly, succinct, etc).
Letters or emails to patients may be drafted quickly by inputting specific concerns or asking for a response for an email from a patient. For example: “Assist in writing a short friendly email to a patient who has missed a visit for a postpartum follow-up visit on the importance of the visit, the topics that may be covered, and the importance to her future health.”
chatGPT-4 output:
“Subject: We missed you! Importance of your postpartum follow-up visit
Dear [patient’s name],
I hope this email finds you well. We noticed that you recently missed your scheduled postpartum follow-up visit with us at [clinic name]. As your health care team, we want to ensure your well-being. We understand that life can get busy, especially with a new little one, but we strongly encourage you to reschedule this important appointment.
The postpartum follow-up visit is crucial for many reasons. During this appointment, we cover several essential topics, including:
Your postpartum follow-up visit plays a vital role in ensuring your long-term health and well-being. It allows us to monitor your recovery, provide support, and address any concerns you may have.
Please call us at [Clinic’s Phone Number] or reply to this email to reschedule your appointment at your earliest convenience. We are here to support you and your growing family, and we look forward to seeing you soon!
Warm regards,
[Your Name] [Your Title] [Your Email] [Your Phone Number] [Your Clinic’s Name]”
The prompt creation and AI response process took less than 30 seconds.
Using a similar technique to answer patient questions online would be helpful to most health care providers. An article recently published in JAMA Internal Medicine compared doctor responses with AI responses. The researchers randomly collected 195 posts on social media on a wide range of medical questions that were answered by verified physicians. The same questions were put into a new session and answered by chatbots. The researchers then had a team of licensed health care professionals analyze the responses.1 AI answers were preferred 78.6% of the time compared with physician answers. On average, AI answers were 4 four times as long and rated as more empathetic (9.8 times higher) than physician answers (45.1% vs 4.6%).1
Horses vs zebras
Most of us have been taught, “When you hear hoofbeats, think horses, not zebras.” Within geographic regions, specific diagnoses are more common because of exposures, demographics, and other factors. Yet when more difficult situations exist, AI systems may be able to suggest different diagnoses and laboratory evaluations. Some experts recommend an AI window open to a site such as chatGPT to input laboratory findings, signs, and symptoms when performing telemedicine appointments to quickly input data and acquire results. chatGPT-4 answers medical exam licensing questions at a rate of better than 90%. You should expect to encounter patients coming to your office with improved information and suggestions better than the “Dr Google” of the past. These systems are imperfect and require judgment and oversight, but it is not long before they will be further enhanced and perhaps be more accurate in many circumstances. However, as with all systems, the data placed into the AI will need to be accurate and comprehensive. I would suggest becoming familiar with different AI systems in this rapidly changing area of medicine. The usual time span in health care for something to become standard of care to 50% implementation is 17 years. This new technology will not take that long to become ubiquitous. AI has nonintervention-based medicine within sight, and providers must learn how to use technology to enhance, not replace, interactions.
Other uses
Many physicians, especially those in leadership positions, have teleconference meetings. Fireflies.ai is an AI program that allows helps team members record and take notes during sessions, acting as a virtual assistant. Topics are highlighted, and audio and visual links are provided for specific portions of meetings. If you have to leave the meeting early, you can access these notes or recording later to catch up on important information, or you can send the information to those who couldn’t attend. The app also interfaces with task management software to create tasks during meetings with simple prompts. Other AI meeting software such as Sembly AI, Avoma, and Otter.ai perform similar functions.
Software such as Clara can be used as virtual assistants for meeting scheduling, syncing with your email and calendar. The software responds to others with possible meeting times by including the Clara email as a CC. The software can even reschedule and cancel meetings when needed. More straightforward, free solutions such as Trevor AI integrate with your calendar to assist with meeting scheduling.
Presentation software such as beautiful.ai can take an existing slideshow or create new slides to create presentations with improved images and storytelling. AI within the software takes content and adapts the slides. A bot asks the type of presentation you are looking for and then creates an initial draft.
Concerns and precautions
These AI systems are not compliant with the Health Insurance Portability and Accountability Act and should not be able to access identifiable patient information. Practices should discuss the need for training and support to prevent privacy and security errors and have conversations regarding the effective and safe use of AI in patient interactions. Mistakes are still possible and can be frequent in AI-generated results, and caution is needed when relying on AI output for diagnostic and treatment recommendations.
Nonetheless, AI will soon become more present in health care with a future push into areas such as predictive analysis, data acquisition and summary, risk identification and subsequent patient engagement, patient scheduling, and communication.
Reference
Ayers JW, Poliak A, Dredze M, et al. Comparing physician and artificial intelligence chatbot responses to patient questions posted to a public social media forum. JAMA Intern Med. 2023. doi:10.1001/jamainternmed.2023.1838
Developing a program for addressing social determinants of health
February 8th 2024Explore the role of health care professionals in enhancing women's health by integrating systematic strategies to address social determinants, leveraging screening tools, and embracing data analytics for personalized care and improved outcomes.
Read More