This podcast is created by NotebookLM based on this blogpost.

AI is transforming healthcare, and OpenAI's ChatGPT is leading the way.

It’s not just a chatbot—it’s an assistant, educator, and even a compassionate listener. 

With over 120 million users every month, ChatGPT is growing in popularity among healthcare providers and patients alike.

It offers new ways to make medical information more accessible and supports better outcomes in diabetes care.

In this blog, we will explore:

• What is ChatGPT?
• Other AI Tools Relevant to Healthcare
• Addressing Fears Around AI in Healthcare
• Use Cases of ChatGPT in Diabetes Care
• Maximizing ChatGPT's Potential: Effective Prompting
• Continuing Education and Training
• Conclusion: Embrace the Future of AI in Diabetes Care
• Join Our Free Webinar on AI in Diabetes Care

What is ChatGPT?

ChatGPT is the most popular Large Language Model (LLM), offering multiple versions, including GPT-4o and the newer o1 model. Each version is designed for different tasks: 

• GPT-4o (GPT-4 Omni): Supports text, audio, image, and video processing. You can upload various file types, including PDFs (up to 512 MB), text files (up to 2M tokens), images (up to 20 MB), and spreadsheets/CSV files (up to 50 MB). You can request summaries, analyses, transformations, or extraction of specific information from these files. However, ChatGPT currently cannot analyze images that are embedded within documents or presentations.
• GPT-4o with Canvas (launched October 2024): This version allows users to interact with content more intuitively by making edits, annotations, and specific adjustments directly within the text.
• o1-preview (launched September 2024): Newer model for tasks that require advanced reasoning and complex problem-solving tasks. 
• o1-mini: A lightweight, more cost-efficient version suitable for everyday tasks.

CustomGPTs for Specialized Needs

ChatGPT also allows you to create CustomGPTs and use CustomGPTs made by others.

These models are trained on the data you provide, making them highly adaptable to specific needs.

Useful CustomGPTs for healthcare include:

• Scholar AI: Searches in 40 million+ peer-reviewed papers and helps explore scientific PDFs.
• AskYourPDF: Allows users to dive into their documents, find answers, and bring information to their fingertips.
• Show Me Diagrams: Create and edit diagrams directly in chat.
• Email and Mail Writer: Helps rewrite emails to be polite but efficient.

How ChatGPT is continuously evolving

• Multi-Platform Availability: ChatGPT is accessible on both desktop and as an app on Android and iOS phones.

• Advanced Voice Mode: Although it is already possible to talk to your ChatGPT app, the new Advanced Voice Mode lets you choose the voice, interrupt it mid-sentence, and get faster replies. Unfortunately, in Europe we might have to wait a bit longer for this update. 

• Up-to-Date Information: ChatGPT’s knowledge is up to date as of October 2023. For questions based on that, it won’t provide sources. ChatGPT can search the web for newer info, but sometimes it might generate incorrect answers about things that happened after October 2023. When it gets information from the web, it will show you the sources.

• ChatGPT Plus Subscription: For $20 USD per month, you can subscribe to ChatGPT Plus, which offers access to the latest features and higher usage limits.

It is best to stay informed about new functionalities, as ChatGPT receives updates almost every few weeks, helping you work more efficiently.

Other AI Tools Relevant to Healthcare

Other popular AI tools include:

• Perplexity: Combines ChatGPT with Google Search. For up-to-date information, it's recommended to use Perplexity instead of ChatGPT. Perplexity also provides clear links to sources, making it easy to check and verify the answers!
• Claude: Offers more nuanced and natural-sounding responses and is great at coding.
• StabilityAI - Midjourney: Fun for image generation, although ChatGPT also does a great job.
• Synthesia - Invideo - Pika: Fun for video generation (e.g. our Diabetes Technology Expert Program uses the personal avatars of Synthesia).
• NotebookLM: Fun to create a podcast based on your documents (e.g. the uploaded podcast at the top of this blog).
• Suno: Fun for music generation based on text. 

This is an example of a video we made with Pika, Suno and Descript.

Addressing Fears Around AI in Healthcare

While AI presents significant opportunities, it's essential to address concerns to ensure responsible integration into diabetes care.

Safety Concerns

• Potential for Errors ("Hallucinations"): AI may occasionally produce incorrect outputs. In diabetes care, such errors could lead to inappropriate treatment recommendations, posing risks like hypoglycemia or diabetic ketoacidosis.
• Over-reliance on AI: Dependence on AI tools may diminish clinicians' critical thinking. It's crucial to use AI as an aid, not a replacement for professional judgment.

Accuracy and the “Black Box” Problem

• Opacity of Decision-Making: AI models often function as "black boxes," making it challenging to understand how they derive conclusions.
• Validation Challenges: Frequent updates to LLMs complicate efforts to establish standardized protocols for clinical evaluation.
• Accountability: Unclear responsibility if AI-generated recommendations lead to adverse outcomes.

Privacy Concerns

• Patient Data Security: AI tools may store sensitive health information, raising concerns about confidentiality and compliance with regulations like HIPAA.
• Data Misuse: Misuse of health data could lead to discrimination or other negative consequences for patients.

Ethical Considerations

• Transparency: Patients should be informed when AI tools are involved in their care.
• Consent and Data Usage: Ethical considerations around data used to train AI models, particularly regarding patient information.
• Impact on Professional Roles: As AI capabilities expand, we must consider how this affects the role of healthcare providers.

Conclusion: Embracing AI Responsibly

Despite these concerns, the potential advantages of AI in diabetes care are undeniable.

By addressing challenges through transparency, rigorous validation, and ethical use, we can ensure that AI becomes a valuable tool working alongside us to enhance patient outcomes.

Use Cases of ChatGPT (or similar AI tools) in Diabetes Care

Picture adapted > Press Release Ohio State University August 2024

Now, let's explore how ChatGPT and similar AI tools can be practically applied in our field:

1. Patient Education (Ann Biomed Eng 2024)

   • Simplifying Complex Information: ChatGPT can break down intricate concepts like insulin resistance or carbohydrate counting into understandable language.
   • Device Training: Provides step-by-step guides on using glucose monitors or insulin pumps.
   • Dietary Guidance: Offers personalized meal planning suggestions based on dietary preferences and cultural considerations.

2. Administrative Assistance

   • Drafting Medical Notes: Helps in composing patient summaries and consultation notes, saving time. (Acta Orthopedica 2024)
   • Summarizing Patient Histories: Quickly generates overviews of patient records, highlighting key information. (JAMA 2023)
   • Automated Documentation: Ambient listening tools can transcribe and organize notes during consultations. (e.g. Heidi, or the use case at Ohio State) 

3. Clinical Decision Support

   • Differential Diagnosis: Assists in generating potential diagnoses for complex cases, considering multiple factors. (e.g. Chen et al, 2024, Dr. Gupta AI, VisualDx)
   • Evidence-Based Recommendations: Summarizes current guidelines and research findings to support treatment decisions. (e.g. Pathway)
   • Algorithm Development: Aids in creating clinical pathways and treatment algorithms tailored to patient needs.

4. Research Support

   • Literature Reviews: Quickly summarizes findings from recent studies, aiding in staying current with advancements. (e.g. Consensus AI, ScholarAI)
   • Data Analysis Assistance: Helps in interpreting research data and statistical outputs.
   • Manuscript Preparation: Assists in drafting and editing research papers or case reports. (e.g. STORM)

5. Translation Services

   • Multilingual Communication: Translates medical information into patients' native languages, improving understanding. 
   • Real-Time Interpretation: GPT-4o's voice assistance can facilitate conversations with non-English-speaking patients. (ChatGPT Advanced Voice Mode is rolling out)

6. Lifestyle and Emotional Support

   • Personalized Advice: Offers suggestions for exercise routines or stress management techniques.
   • Emotional Support: Engages in empathetic conversations, addressing patients' concerns and providing encouragement.
   • Behavioral Change Support: Reinforces positive habits and provides motivation between appointments.

Although there are several LLMs specific for healthcare tasks (Med-PaLM 2, Med42, Hippocratic AI, Florence Chatbot, August AI, ADA Health, Tess, S.A.R.A.H. etc), ChatGPT does farely well on medical exams (NEJM 2024), and its accuracy is always improving (GPT-4o on UK and US Medical Exam, GPT-4o on Japanese Medical Exam). 

However, it's important to note that AI should not replace doctors, but rather be used as a tool to augment and improve patient care. 

Maximizing ChatGPT's Potential: Effective Prompting

A key skill in using ChatGPT is crafting effective prompts — similar to taking a good patient history.

The better the question, the more accurate the answer.

Best Practices for Effective Prompting

1. Define the Role: Specify ChatGPT's role to tailor the response appropriately. For example, "You are a diabetes educator" or "You are a diabetologist specializing in Type 1 diabetes."

2. Be Specific: Detailed prompts lead to more precise answers. Instead of general questions, include specific patient scenarios or data (without patient names or birthdates!).

3. Provide Context: Include relevant information such as patient history, lab results, or case details to enhance the relevance of the response.

4. Iterate and Refine: If the initial response isn't satisfactory, adjust your prompt to clarify or add more details.

Standard Format for a Good Prompt

1. Who Am I? Identify yourself or your role.

2. Who Do You Want the AI to Be? Specify the persona for ChatGPT (e.g., "You are a certified diabetes educator").

3. What Is the Task? Clearly outline what you want the AI to do.

4. What Is the Context? Provide background information or resources relevant to the task.

5. Desired Output Format: Specify the format, length, or structure you prefer.

6. Tone of the Output: Indicate whether you want the response to be formal, conversational, simplified, etc.

7. Intended Audience: Mention who the output is for—patients, healthcare professionals, students, etc.

8. Examples (if applicable): Provide examples to guide the AI's response.

Examples of Prompts in Healthcare:

• Patient Instructions: "Explain to a newly diagnosed Type 1 diabetic teenager how to manage insulin therapy, emphasizing lifestyle considerations."
• Clinical Decision-Making: “You are an AI medical assistant. Based on the patient’s symptoms and history, what are the top three differential diagnoses with confidence scores”
• Research Summarization: “Summarize this diabetes management study, including the key points, practical takeaways, and a one-line summary for a healthcare professional.” 

Check more detailed prompts in this document.

 By mastering prompt engineering, you unlock the full potential of AI in diabetes care, making workflows more efficient, enhancing research, and enriching patient interactions.

Continuing Education and Training

To fully harness AI tools like ChatGPT, it's important to seek training and stay informed about best practices.

Here are some valuable resources tailored to endocrinologists and diabetologists:

• Dr. Om Lakhani’s Free Course: Available at technocrinology.com, this course offers guidance on effectively using AI in endocrinology. Dr. Lakhani also shares insightful YouTube videos on prompting and apps for clinicians.

• Doctors-AI: Accessible at doctors-ai.org, this platform provides education and training for medical professionals on the use of AI in healthcare. They organize weekly online meetings, offer a monthly newsletter, and host active WhatsApp and LinkedIn groups open to all healthcare providers.

It's important to emphasize that ChatGPT and similar tools are here to support—not replace—clinical judgment.

Comprehensive training and awareness of successful use cases are essential for integrating AI responsibly into diabetes care.

Conclusion: Embrace the Future of AI in Diabetes Care

AI is set to play an increasingly significant role in diabetes management.

By embracing tools like ChatGPT, we can enhance patient education, streamline administrative tasks, and support clinical decision-making.

The key is to adopt these technologies responsibly, keeping patient well-being at the forefront, and always keeping the human in the loop.

Join Our Free Webinar on AI in Diabetes Care

Ready to dive deeper into practical AI applications in diabetes care?

Don't miss our free webinar on October 21st, featuring leading experts in diabetology and healthcare technology.

This hands-on session will provide actionable insights to harness the power of AI for better patient outcomes.

👉  Click here to register for the webinar and join us on the journey to smarter, more personalized diabetes care.

Kind regards,