medicine and the machine

Generative Pre-trained Transformers (GPT) has been a groundbreaking development in the field of artificial intelligence. OpenAI’s ChatGPT, a prominent example of these models, has transformed our interactions with AI technologies, providing a wide array of functionalities from drafting emails to producing inventive content. However, the generalized nature of standard GPT models may not adequately cater to the specialized needs of all users and organizations, particularly in sectors requiring high precision, such as healthcare. In this context, the development of custom GPT models emerges as an important innovation, offering a bespoke AI experience finely tuned to comprehend and address the distinct demands of the healthcare industry. These custom models can be tailored to process and interpret medical language, analyze patient data, support clinical decision-making, and personalize patient care, thereby enhancing both efficiency and outcomes in healthcare settings.

What is a “Custom GPT”?

Custom GPT models are specialized versions of the generic GPT models that have been fine-tuned or adapted to excel in specific domains or tasks. This customization can range from training the model on industry-specific data to adjusting its parameters to better align with particular conversational styles or technical requirements.

Other AI chatbots that have the option to “customize” your own GPT, in addition to ChatGPT+, include: Copilot Pro and Gemini Pro. Of note, these all require paid subscriptions.

Why do I need to learn about custom GPT in Healthcare?

It saves time! All the different sections of this blog can be turned into your own custom GPT. This is particularly useful especially if there is one particular feature that you use often. For instance, if you regularly utilize an AI chatbot for note rewriting and summarization and want a consistent style in every output, a custom GPT model can streamline this process. By embedding your preferred instructions directly into the model, you eliminate the need to rewrite the same prompt, ensuring uniformity in the results with every use. This approach not only enhances efficiency but also saves a considerable amount of time. Here are other use cases where custom GPT models could be used in the healthcare context:

• Generate Patient Communication: Custom GPT models can be trained on specific healthcare data to understand and generate communication that aligns with patient needs and medical terminology. This personalized approach can improve patient engagement, adherence to treatment plans, and overall satisfaction.
• Clinical Decision Support: By training on specific literature, custom GPT models can assist healthcare professionals in diagnosing and developing treatment plans.
• Efficient Medical Documentation: Custom GPT models can streamline the creation and management of medical documentation by generating clinical notes, discharge summaries, and other essential documents. This not only saves time for healthcare providers but also enhances the accuracy and consistency of medical records.
• Medical Education and Training: Custom GPT models can be used to create dynamic, interactive training materials and simulations for medical students and professionals. This can facilitate a deeper understanding of complex medical concepts and scenarios, fostering continuous learning and professional development.
• Language and Accessibility: Custom GPT models can be tailored to translate medical information into multiple languages and simplify medical jargon into layman’s terms. This makes healthcare information more accessible to patients from diverse linguistic backgrounds and varying levels of health literacy.

Each of these reasons underscores the transformative potential of custom GPT models in healthcare, promising not only to improve the efficiency and effectiveness of healthcare delivery but also to personalize and humanize the patient care experience.

How do you Create your Custom GPT Models?

Creating a custom GPT model involves several key steps, each critical to ensuring the model’s effectiveness and efficiency in its targeted domain.

First, know where to find it. After logging in, go to the left side of the screen and click on “Explore GPTs”. It will then open up to something like this:

On the top right side of this screen, you will find two buttons, “My GPTs” and “+Create”. Click on “+Create”, and you will then be ready to create your own GPT. This screen then appears:

To create your GPT, use the following steps:

1. Defining Objectives: The first step involves clearly identifying the purpose of the custom GPT model. Objectives can vary widely, from enhancing patient content to generating documentation or assistance in clinical questions. After knowing the main “use case” for your GPT, you can then add a name and description.
2. Adding Instruction: This is the heart of the GPT. The instructions that you provide should be as detailed and clear as possible. It is exactly like writing a prompt but with more information. A good practice here would be using prompt engineering in the context of COSTAR framework as a start.
3. Data Collection and Preparation: Custom models require a dataset that is representative of the model’s intended use case. This involves collecting high-quality, domain-specific data and preprocessing it to suit the model’s training needs. You can upload any file or multiple files that you want your GPT to be trained on. Data privacy and ethical considerations are paramount during this stage.
4. Training and Fine-tuning: The model is then trained or fine-tuned on the prepared dataset. Before publishing your GPT, make sure that you test and retest multiple times and refine your instructions until you get your desired output.
5. Integration and Deployment: Once the model meets your expectations, you can then save and publish it. You can make it available to anyone with ChatGPT+ subscription who has your link.

My Own Endocarditis GPT:

To show you how it works, I made my own custom GPT that will generate only answers from the most recent 2023 Endocarditis Guidelines by the European Society of Cardiology.

I named it: “Endocarditis Guidelines Expert”

It has the following description: Expert on 2023 ESC Guidelines for the management of endocarditis, focused on valve-specific treatment and citations.

The following are its instructions:

Your role is to provide expert information on endocarditis, primarily using the 2023 European Society of Cardiology Guidelines for the management of Endocarditis as your reference source. This includes information on diagnosis, treatment, and management of endocarditis, with a focus on valve-specific issues as detailed in these guidelines. Ensure accuracy in your responses, and refer to the 2023 European Society of Cardiology Guidelines for the management of endocarditis for the most current recommendations, especially when discussing treatment options like antibiotic therapy. When addressing treatment, always determine if the condition involves a native valve or a prosthetic valve, since the treatment approach differs. Provide references or citations from the 2023 ESC Guidelines for the management of endocarditis guidelines when requested. While offering dosing recommendations from the guidelines, advise users to confirm dosages with a healthcare professional. If a query is unclear, ask for clarifications to provide the most relevant and accurate information. Communicate formally, using medical terminology, and structure your responses with bullet points for clarity and organization.

This is a sample query using my GPT:

Best Practices and Considerations

• Ethics and Bias: Custom GPT models can inadvertently perpetuate biases present in their training data. It’s essential to assess and mitigate these biases throughout the model development process.
• Privacy and Security: Ensuring the confidentiality and integrity of the data used for training and interaction with the model is crucial, especially in sensitive domains.
• Continuous Learning and Updating: The model should be periodically updated with new data to maintain its relevance and accuracy over time.

Conclusion

Creating custom GPT models offers a path towards more personalized, efficient, and effective AI solutions tailored to specific needs and challenges. Whether enhancing customer interactions, streamlining operations, or advancing research, custom GPT models represent a significant leap forward in leveraging AI’s potential. However, it’s crucial to approach their development with a thoughtful consideration of ethics, privacy, and ongoing adaptability to ensure they serve as beneficial tools in their respective domains.