Introduction

Hello, fellow doctors!

Welcome to the cutting-edge world of Artificial Intelligence, where tech is constantly evolving to support our medical endeavors. AI is revolutionizing our field, enhancing everything from radiological accuracy to predictive analytics, optimizing cancer treatments, fast-tracking drug development, and boosting surgical precision and patient care.

In this guide, we're exploring Generative AI, the type that emulates human creativity, offering novel solutions from crafting texts to solving complex issues. Envision Generative AI as a competent intern who just needs the right 'prompts' to excel. These prompts empower Generative AI to improve tasks like report generation and data analysis remarkably.

This course aims to provide you with the essential skills for crafting effective prompts, enabling you to leverage Generative AI's capabilities to benefit your practice.

Let's embark on this journey to make AI work for us in the most effective way

Why should I care about Generative AI?

Have you ever envisioned a day where paperwork doesn't overshadow your time withpatients or exploring new hobbies like painting or playing the guitar?

Generative AI is agroundbreaking advancement tailored to enhance your role as a physician. It's like having adigital assistant to handle the nitty-gritty tasks of compiling medical reports, summarizing patient visits, managing emails, and supporting your academic pursuits, allowing you tore claim your time.

What's more, it translates medical terminology into plain language, building a bridge of understanding with your patients. It also simplifies your communications with colleagues and patient families, lightening your workload.

Embracing Generative AI not only optimizes your workflow but also propels your career forward by eliminating routine tasks, giving you the leeway to refine your medical expertise and become a master in your field.

It's an innovation that not only transforms healthcare practices but also enriches your professional and personal life.

Which Generative AI service should I use?

Let's break down the key features of various generative AI services through this comparison table, highlighting what's particularly relevant to our world in medicine.

We'll also see how they stack up against solutions made just for medical pros like us.

The goal here is to boost our understanding of AI's potential benefits in our daily practice.

Feature
Meditext
ChatGPT
Google's Bard (Gemini)
Claude 3
Healthcare Specificity
Yes
Tailored for healthcare professionals
No
General - purpose AI
No
General - purpose AI
No
General - purpose AI
Data Security
High
Meets European data standards
Basic
General data measures
Basic
General data measures
Basic
General data measures
Customization
Extensive
Medical - specific customization
Limited
General options
Limited
General options
Limited
General options
Language & Terminology
Medical
Specific language and terminology
General
Not focused on healthcare
General
Not focused on healthcare
General
Not focused on healthcare
User Interface
Healthcare
Oriented design
General
Standard interface
General
Standard interface
General
Standard interface
AI Technology
GPT-4
Optimized for healthcare
GPT-4
Not tailored for healthcare
GPT-4
Not tailored for healthcare
GPT-4
Not tailored for healthcare
Templates for Use Cases
Yes
Diverse medical scenarios
No
Lacks specific templates
No
Lacks specific templates
No
Lacks specific templates
Medical Scribing
Planned
Tailored for healthcare
No medical scribing
Voice recognition
No
Does not offer scribing
No
Does not offer scribing

Free LLMs - Are They Any Good?

Let's talk about Free LLMs - are they up to the mark?

Tools like GPT-3.5 and GPT-4 are not just buzzwords; they're game-changers in how we handle information, craft content, and tackle complex problems. These tools come in various flavors, including the no-cost GPT-3.5 and the premium GPT-4 Turbo. While they share the same DNA, the differences in their abilities, customization options, and how they manage data are pretty stark.

When to Use Free LLMs

GPT-3.5, the complimentary version, is a solid starting point for anyone dipping their toes into the AI and prompt engineering pool. It's a robust platform for learning, experimenting, and getting the hang of how this AI magic happens.

It is great for

Crafting
Whipping up straightforward content for emails or basic academic stuff
Insight
Simple data analysis and spotting patterns
Streamlining
Taking over the mundane tasks like formatting or basic calculations
AI-Savvy
A learning tool to get savvy with AI interactions

For us in the medical field, these freebies can be quite handy for general queries, handlingnon-sensitive data, or creating educational material. Imagine using it to quickly draftinformational brochures or patient guides without breaking a sweat.

When to Opt for Paid Versions

GPT-3.5, the complimentary version, is a solid starting point for anyone dipping their toes into the AI and prompt engineering pool. It's a robust platform for learning, experimenting, and getting the hang of how this AI magic happens.

Think of it like choosing specialized surgical instruments over the standard ones – they're designed for:

  • Docs who need to tweak and tailor things precisely

  • Handling complex diagnoses and treatment planning with sharper accuracy

  • Juggling larger sets of patient data smoothly

  • Increasing your query capacity, so you can ask more without waiting around

In our world of healthcare, where getting things right is non-negotiable, shelling out for something like GPT-4 Turbo could be a game-changer. These beefed-up versions offer deeper insights and more refined responses to those head-scratching medical scenarios,helping us stay on top of our decision-making game and keep our documentation spot-on.

While the enhanced functionalities of premium language models are undeniably beneficial inoue practice, we must not overlook the privacy challenges they present. These models have the capacity to learn from the data we provide, which poses a significant concern in the context of sensitive healthcare information. As diligent healthcare providers, we must navigate the use of these tools with informed caution, prioritizing the protection of patient data above all.

For those who'd rather not get bogged down with security dilemmas, the hassle of selecting the appropriate model, or the frustration of generic AI that doesn't grasp medical subtleties, Meditext is the answer. It's designed with healthcare pros in mind, offering top-tier AIcapabilities plus an intuitive understanding of our sector's specific needs.

Part I: The Essentials

Alright, let's roll up our sleeves and get to the nuts and bolts of generative AI.

This section is your stepping stone, packed with essential insights to boost your effectiveness. It's aboutgetting straight to the point with practical know-how that you can apply right away to spruceup your daily tasks. Grasp these core concepts, and watch how they revamp your routine,delivering quick, noticeable results.

3 Principles of Prompt Engineering

Be specific
Just as we tailor our patient care, the more detailed you are with your prompts, the more precise the AI's responses will be. Remember, it's a balancing act - be concise yet thorough.
Work in steps
Break down the tasks into manageable chunks. This method yields superior results, mirroring the process you'd follow with a fellow human.
Iterate and improve
Fine-tune your prompts. let the AI refine its responses based on your adjustments.

What Makes a Good Prompt

Great outputs all come down to the data the model was trained on, its parameters, and good prompting. Since we can only control one of these, here’s what good prompting boils down to:

  • Clear and concise language that is direct and unambiguous.

  • The persona that you assign to ChatGPT, aka who it will be acting as in your prompt.

  • The information and examples that you provide, aka your input. (information and examples can be referenced by ChatGPT in the previous chat history).

  • A specific task that you are requesting ChatGPT to complete, aka your desired output.

  • Refinement as needed once you receive your first response, aka reiteration until receiving the desired output.

Which transfers us directly to the…

Main Prompting Steps

Test, evaluate, and iterate
Generate responses with ChatGPT, evaluate the results, and adjust the prompt as necessary to elicit the desired response.
Test, evaluate, and iterate
Generate responses with ChatGPT, evaluate the results, and adjust the prompt as necessary to elicit the desired response.
Write the prompt
Craft a concise prompt that clearly communicates the information and task that is required, using the keywords and phrases identified in steps 1 and 2
Define the problem or goal
Clearly articulate what you want ChatGPT to help you with and what kind of response you are looking for.

Explorative Prompt Starters

Consider this a collection of thirty thought-starters, designed to open your eyes to the potential of Generative AI in the medical field.

Whether you're hitting a wall with decision-making, aiming to streamline your day-to-day operations, or in search of innovative solutions, these prompts are your go-to. They're crafted to help you leverage AI effectively, ensuring your medical pursuits are more streamlined than ever.

Use these prompts as a springboard to explore the breadth of AI's utility in healthcare. It's not an all-encompassing list, but it's tailored to get you thinking about the possibilities.

  • Construct an outline for a medical presentation on the following topic...

  • Brainstorm possible diagnoses based on these symptoms and lab results...

  • Brainstorm 10 differential diagnoses for this case...

  • Draft a structured plan for a medical research paper on...

  • List some treatment options for the following condition based on the current
    guidlines..

  • Create a lesson plan for medical students on...

  • Sum up the following medical study in a tweet...

  • Compose a letter to a medical insurance company argumenting the treatment costs
    of the patient...

  • Recommend some analogies/synonyms for medical terms like...

  • Transform this medical abstract into a patient-friendly explanation...

  • Translate this medical information into German...

  • Compose a polite decline response to this letter...

  • Explain the mechanism of action of...

  • Can you recommend some online courses or books to improve my understanding
    of...

  • Provide some tips for conducting a successful patient interview…

  • Suggest some healthy meal options for patients with...

  • Help me create a daily schedule to manage patient appointments...

  • Explain the basics of genomic medicine...

  • Provide some guidance on discussing difficult diagnoses with patients…

  • Draft a protocol for discussing a following case with superiors…

  • Write a patient status report for…

  • Summarize this medical article with key take-aways…

  • Compare the side effect profile of these medications for this condition…

  • Create a checklist for pre-surgical preparation for this surgery…

  • Formulate questions for a second opinion on this diagnosis…

  • Design a follow-up plan for patients with this condition…

  • Explain the latest guidelines for the treatment of...

Instant Creativity: Plug & Play Prompts

In the hospital, efficiency and accuracy are paramount, even when it comes to administrative details - especially if your superior has anger issues. Our pre-made prompts are designed to streamline your interaction with generative AI, ensuring you receive precise and useful information swiftly. Each prompt is tailored to distinct medical scenarios, from general inquiries to patient education, facilitating a more structured and impactful dialogue. Simply copy and paste these prompts to improve your desired outputs.

Header imageHeader imageHeader imageHeader image

Prompt for answering general medical questions

What it does: This prompt helps you get straight to the point, medical answers, providing essential details without unnecessary complexity, ideal when you are seeking quick answers.
Instructions: Just copy-paste all of of the text below into the chat field and write your question after the “# Here is the question:” line:

You are the most skilled doctor's assistant in the universe. The user (doctor) will ask you various medicine-related questions, and your task is to assist them by providing accurate and correct information. Your answers must be concise and to the point, yet contain all the important details. Since the user is a medical professional, avoid triviality and unnecessary simplifications. If you require any additional information to provide the best possible answer,
ask the user for clarification.

Prompt for answering general medical questions

What it does: This prompt is designed to generate a concise, structured summary of a medication for medical professionals, including usage, dosage, side effects, and more. It requires a drug’s name, doctor's specialty, and patient's demographics to provide a tailored information.
Instructions: Just copy-paste all of of the text below into the chat field. The AI will then ask you to provide the name of the drug, your speciality and relevant patient information (diagnose, age etc.):

Provide a structured summary of a specific drug tailored for a doctor's inquiry. This should include sections with subheadings covering uses, dosage, side effects, contraindications, and interactions with other medications, formatted for clarity and conciseness. Additionally, incorporate a brief section on patient advice and specific monitoring parameters. Include the latest research findings and comparisons with similar medications where relevant.

Before generating an answer, ask the doctor to provide the drug name, doctor's specialty, and patient's demographic.

Prompt for brainstorming medical diagnoses

What it does: This prompt facilitates a structured brainstorming approach to diagnosing, helping you consider a range of possibilities and their respective treatments, enhancing clinical reasoning.
Instructions: Just copy-paste all of of the text below into the chat field and describe your clinical situation after the “# Here is the situation:” line:

You are the most proficient assistant for medical diagnosis brainstorming in the universe. The user (doctor) will provide brief information about the patient (situation, symptoms, etc.), and your task is to assist them in generating possible diagnoses, ranked from most likely to least likely, along with the reasoning for each. Your responses must be insightful and precise, incorporating all necessary details. Following each suggested diagnosis, provide a very brief overview of the diagnostics and treatment, keeping it concise yet informative. Since the user is a medical professional, avoid oversimplification and unnecessary details. If you require additional information to provide the best possible diagnostic suggestions, ask the user for clarification.

Prompt for answering emails

What it does: This prompt streamlines email communication with your colleagues and patients, ensuring responses are appropriate and professional, saving time while maintaining high standards of communication.
Instructions: Just copy-paste all of of the text below into the chat field. After the “# Instructions:” line, feel free to shortly describe the situation and provide the purpose of the email. After the “# Copy pasted email to which a response is needed (if there is one):” line paste an email, to which you want to respond (if there is one):

You are the most efficient email assistant for doctors in the universe. The user (doctor) may briefly describe the situation and provide instructions on how to create a new email or respond to the email, including the goal and the desired tone. Additionally, there is a space to copy and paste the email to which the response should be crafted, if the doctor is not composing a new email but merely responding to one. Your task is to assist the doctor by drafting or replying to emails in a manner that is professional, appropriate, and aligned with the doctor's instructions. If additional details are needed to craft the best possible email, ask the user for clarification.

# Instructions:

# Copy pasted email to which a response is needed (if there is one):

Prompt for drafting patient educational materials

What it does: This prompt aids in the development of patient-friendly materials that explain medical conditions and treatments clearly and effectively, enhancing patient understanding and engagement in their care.
Instructions: Just copy-paste all of of the text below into the chat field. After the “# The topic:” line, provide the topic of the material. After the “# Situation: ” line, provide the context, why this material is needed (pre-surgery guidance, explanation of treatment side effects, new diagnosis information etc.). After the “# Patient's name:” line, provide a fake patient’s name, which you can later replace with a real one (do not use a real name, if you are using tools like ChatGPT and not medicine specific one’s):

You are the ultimate resource for creating patient-friendly medical educational material. The user (doctor) will provide you with a specific topic (diagnosis, treatment, medication, etc.), and your task is to craft informative content tailored for the patient or the patient's family. Your responses should be clear, accessible, and free from unnecessary medical jargon, ensuring the information is correctly understood without overwhelming the reader. Focus on explaining the topic in a way that is both reassuring and empowering, enabling patients to be well-informed about their health. If additional details are needed to produce the most effective educational material, ask the user (doctor) for clarification.

# The topic:

# Situation:

# Patient's name:

AI Beyond Chat

At this point you're well-acquainted with generative AI's capabilities. But there's more to AI than just conversation. A variety of innovative AI tools are transforming how we approach research, making it easier to sift through data and extract meaningful insights. In this part of our course, we'll explore some of these AI-powered gems. A prime example is the Consensus App, which uses AI to quickly summarize academic research, helping you cut through the noise and focus on what's truly relevant.

Let's dive into these tools and see how they can add a new dimension to your research process, keeping you informed and ahead of the curve.

How is Consensus App Useful?

Efficiency
Saves time on research by providing quick access to relevant, high-quality papers.
Advanced Searches
Offers filters for specificity, like study type and publication date
Accessibility
Makes scientific knowledge widely available, supporting informed decisions and educational growth.
Consensus App is a transformative tool for engaging with academic literature, enhancing the research process through streamlined access and insightful synthesis of scientific information.

How to use it?

  1. Go to the following link and sign up (there is a free version): https://consensus.app/search/

  2. Try asking something of interest to you, for example:

  3. After just a few seconds, you should see a quick summary of research papers with a conclusion relevant to your questions, as well as the research papers themselves (note - to get the summary, make sure Copilot toggle on the top left is turned on):

  4. (optional)

    You can also take advantage of “Synthesize” function, to enable the AI to read several papers and synthesize the findings (toggle the button on the top left):

Hide

Smart Data Handling: Safeguarding Patient Privacy with Confidence

While Generative AI, such as ChatGPT, presents an innovative frontier for medical professionals, it's crucial to address the elephant in the room: data privacy and security, particularly concerning GDPR (General Data Protection Regulation) and HIPAA (Health Insurance Portability and Accountability Act).

The Risks with General AI in Healthcare:

Patient Confidentiality
Traditional AI platforms may not be fully equipped to handle sensitive patient data while ensuring compliance with stringent regulations like GDPR and HIPAA.
Data Security
The inadvertent exposure of personal health information (PHI) can lead to severe legal consequences and breach of patient trust.
General Use Limitations
While AI can assist in various tasks, it is not designed to handle highly sensitive and personal patient information without proper security measures
Why Caution is Necessary
Using a general-purpose AI like ChatGPT for discussing specific patient cases or sharing identifiable health information can inadvertently lead to data breaches. Always avoid inputting any patient-identifiable data into such platforms.
Tailored Solution
As doctors ourselves, we are in the same boat when it comes to these security issues, and that's exactly why we initiated the development of Meditext with GDPR and HIPAA compliance at its core, ensuring your medical inquiries and patient data are handled with the utmost security and confidentiality. This platform provides the innovative benefits of Generative AI while safeguarding against the privacy concerns prevalent in standard AI tools.
Best Practices, if you decide to use ChatGPT
Never share identifiable patient information on non-compliant platforms.
Use generalized scenarios or hypotheticals when seeking AI assistance on public platforms.
For secure, compliant, and personalized medical AI assistance, consider using specialized applications like Meditext that respect and protect patient confidentiality and data integrity.
Remember, while AI offers tremendous benefits in healthcare, responsible use and adherence to data protection standards are paramount. Meditext is designed to bridge the gap, providing a safe, efficient, and compliant way to integrate the power of AI into your medical practice.

Part II: Advanced Principles

Expand your foundational knowledge with deeper insights and more sophisticated techniques in this section.

Designed for those ready to elevate their skills, this segment offers an in-depth exploration of advanced generative AI concepts. By engaging with this material, you will gain the ability to obtain more detailed and precise answers, work with complex and nuanced information, and apply your knowledge to a broader range of use cases.

Header image
The Concept Nr.1: Priming
Imagine you're giving a new team member a crash course on a patient's condition. The more accurately you describe what's needed, the better they can assist. "Priming" involves supplying a model with a preliminary input to steer the generated response more effectively towards the desired outcome. This initial input aims to align the model's response with the specific context or requirements of the medical query at hand. This is like laying out the patient's chart clearly before a consultation begins.
Do: Begin with a precise and detailed introduction. Example: "You are an expert in pediatric oncology providing the latest treatment options for acute lymphoblastic leukemia."
Don't: Use ambiguous starters like "Tell me about cancer treatment."
Examples:
Without priming: What treatment should I prescribe?
With priming: I have a patient with type 2 diabetes who is currently not responding to metformin. They have a history of hypertension and prefer oral medication. What treatment should I consider next?
Without priming: What should be included in my patient's care plan?
With priming: I am creating a care plan for an elderly patient recently diagnosed with early-stage Alzheimer's disease. The patient lives with family and has a history of falls. What should be included in their care plan?
The Concept Nr. 2: Divide and Conquer
Facing a multifaceted issue? Break it into smaller tasks, just like managing a patient with multiple complaints. This ensures each aspect is thoroughly addressed, improving overall management and resolution, similar to a comprehensive treatment plan.
Do: Divide complex medical issues into simpler tasks and address them sequentially. Example:
Discuss the psychological effects of chronic pain.
Review non-drug management strategies.
Summarize the available pharmacological treatment strategies"
Don't: Attempt to cover multiple complex aspects in one go.
The Concept Nr. 3: Speed vs Details
In an ER, you need quick answers; in a research setting, you might delve deeper. Indicate whether you want a brief summary or a detailed exploration, so ChatGPT's responses match your time and detail needs, just like choosing between a quick consultation or a deep diagnostic session.
Do: Specify whether you need concise or detailed answers. Example: "Provide a brief overview of Type 2 Diabetes management," versus "Detail the comprehensive management plan for a Type 2 Diabetes patient."
Don't: Ignore the context of your need for brevity or detail in clinical explanations.
The Concept Nr. 4: Clear Structure
A well-presented case to a colleague is easier to understand and diagnose. Use this approach with ChatGPT: structure your questions clearly, use bullet points or numbers if necessary. It’s like presenting a patient's history, symptoms, and lab results in an organized manner.
Do: Structure your prompt clearly, using bullet points or numbers when necessary. Use other symbols to highlight important sections or separate context from your questions to the AI. Examples: Example:
Identify signs of dehydration in infants. List first-line treatments. Describe when to escalate care."
Background: 55-year-old male presenting with chest pain and shortness of breath. Symptoms: Started 2 hours ago, sharp, radiating to the left arm. History: Smoker, hypertensive, no previous cardiac history. #Task: Provide me with differential diagnoses.
Don't: Jumble different elements of your request into an indistinct paragraph.
The Concept Nr. 5: Leaning on Authorities
Just as you refer to textbooks or guidelines, ask ChatGPT to base answers on recognized sources or specify which guidelines you’re following. It ensures the advice aligns with professional standards, akin to citing a medical authority in a clinical discussion.
Do: Request that responses be based on or aligned with authoritative medical sources. Example: "Based on the latest CDC guidelines, suggest treatment for Lyme disease."
Don't: Neglect to mention when specific source adherence is necessary for clinical guidelines.
The Concept Nr. 6: Chain of Though Prompting
This approach is a method to instruct ChatGPT to delineate its answers through a logical, step-by-step explanation of its reasoning process. Employing this technique can make the model's responses more transparent and comprehensible, significantly enhancing the value of its explanations for clinical education, patient care, or decision-making processes
Prompt Formula: • [Your clinical question]. Let's reason through this systematically.
Examples:
"What is the likely diagnosis for a patient presenting with sudden chest pain, shortness of breath, and dizziness? Let's reason through this systematically."
"How would you manage acute asthma exacerbation in a child? Let's reason through this systematically."
Do: Ask ChatGPT to explore different diagnostic or treatment options. Example: "Consider three different management strategies for advanced prostate cancer."
Don't: Rush to a single solution without considering alternative medical opinions.
The Concept Nr. 7: Comparative Prompting
This method emphasizes identifying and contrasting critical aspects between two subjects, aiding in more informed clinical decision-making and a deeper comprehension of each option's advantages and limitations.
Prompt Formula:
Please conduct a comparative analysis of the following medical topics. Organize the similarities, differences, clinical features, evidence-based outcomes, utility, implications, main conclusions, and other relevant factors into a comprehensive table. Here are the two subjects for comparison: [medical subject 1] and [medical subject 2].
Examples: • Comparative effectiveness and side effects of Metformin versus SGLT2 inhibitors in the management of type 2 diabetes. • Surgical intervention versus conservative management for acute appendicitis.
The Concept Nr. 8: Perspective Prompting
This method enhances your comprehension and offers a broader view of medical subjects, aiding in more informed decision-making and a deeper understanding of intricate healthcare matters.
Prompt Formula:
Singular Perspective: Please analyze [medical issue] from the perspective of [specific medical role].
Multiple Perspectives: Please present a discussion [supporting/opposing] the issue of [medical issue] from several distinct viewpoints. Outline the arguments from diverse healthcare roles, such as [list of medical roles].
Examples:
Singular Perspective: Please discuss the management of chronic pain from the perspective of a chronic-pain specialist.
Singular Perspective: Please discuss the management of chronic pain from the perspective of a palliative care nurse.
Multiple Perspectives: Construct a discourse on the management strategies for advanced heart failure, incorporating insights from a cardiologist, a cardiac surgeon, a heart failure nurse, and a patient living with heart failure.

Part III: Professional Mastery

Ascend to the zenith of generative AI expertise with this comprehensive exploration.

Here, we not only guide you through elaborate strategies and applications, but also deepen your understanding of the AI's operational mechanics. This enhanced comprehension will not only empower you to extract even more detailed, precise responses, but will help you develop an intuitive sense of how to communicate with generative AI more effectively, ensuring your interactions lead to the desired outcomes.

Header image
The Concept Nr. 1: Memory capacity
Unlike traditional search tools, ChatGPT has the distinctive ability to remember details from earlier interactions. This capability ensures that subsequent queries can be enhanced by the context and details provided in earlier parts of the conversation, making it especially valuable for medical professionals who deal with complex and multifaceted information. This feature allows ChatGPT to grasp nuanced medical contexts and patient information over the course of a conversation.
For example, consider a conversation where you initially discuss a patient's symptoms and history:
Initial Prompt: "I have a patient presenting with shortness of breath, cough, and recent weight loss."
Follow-up Prompt: "Given this history, what should be the differential diagnosis?"
As the conversation progresses, you can leverage ChatGPT's memory to refine and expand upon initial information:
Further Detailed Prompt: "Considering the patient is a 70-year-old smoker with a history of COPD, update the differential diagnosis including potential oncological considerations."
Continued Interaction: "Based on the updated diagnosis list, suggest a prioritized plan for diagnostic tests."
In these examples, each prompt builds upon the information provided earlier. ChatGPT's memory feature allows it to understand the patient's case progressively, leading to more nuanced and targeted medical advice as the conversation unfolds. This sequential detail addition enables medical professionals to navigate complex patient cases effectively, making informed decisions based on cumulative information.
The Concept Nr. 2: Shot Prompting
Just as medical training uses case studies to teach complex concepts, you can give specific examples to guide ChatGPT.
Just as medical training uses case studies to teach complex concepts, you can give specific examples to guide ChatGPT. This approach involves presenting a small number of detailed scenarios to help the model understand and generate the desired type of response. This concept is analogous to the methods used in AI communication, specifically with ChatGPT, through what's known as "shot prompting."
Understanding these can help tailor your queries for more effective responses:
Zero-Shot Prompting
This approach involves posing a question or request without providing any background information or context. Essentially, you're asking ChatGPT to respond based only on its general knowledge and training. For doctors, this is like asking for a generic treatment plan without specifying any patient details.
One-Shot Prompting
In this method, you provide a single piece of information or context before making your request. This could be a specific patient condition, a set of symptoms, or a particular medical concern. It's akin to presenting one patient case to guide the response, helping the AI understand the direction or scope of your inquiry.
Few-Shot Prompting
This involves providing multiple pieces of information, cases, or guidelines before asking your question. By supplying several contexts or examples, you're giving the AI a broader framework to understand and respond to your query. For medical professionals, this might mean offering details from several patient cases or referencing multiple guidelines to inform the management plan being requested.
Examples:
Zero Shot: “Suggest a management plan for hypertension.”
One Shot: “Using this example as reference, suggest a management plan for hypertension.”
Few Shot: ““Using these Example 1, 2, 3 as reference, develop a detailed hypertension management plan. Include medication adjustments, lifestyle changes, and necessary follow-up tests.”
Do: Present structured, specific examples to shape the model's output. Example: "I'm drafting a treatment plan for a new condition. Here are three treatment plans I've written before: [Treatment Plan 1], [Treatment Plan 2], [Treatment Plan 3]. Use these as a model to create a new plan for chronic fatigue syndrome."
Don't: Request help without context or examples. Ineffective approach: "Write a treatment plan for chronic fatigue syndrome."
The Concept Nr. 3: Ask before Answer Prompting
This strategy guides ChatGPT to seek additional details before providing a response, ensuring the information given is as precise and relevant as possible. This approach is particularly useful in the medical field, where clarity and specificity are crucial for accurate diagnosis, treatment planning, and patient care.
Prompt Formula:
As a medical professional with expertise in [medical field], you will be asked to address specific clinical scenarios or tasks. Before responding, please ensure you fully understand the query or situation. If there is any ambiguity or if you need more details to provide the most accurate advice, please ask for clarification using bullet-pointed questions. Is this approach clear to you?
Excellent, my inquiry is [specific medical question]. Your task is to [specific medical task]. Please list any clarifying questions you might have to ensure the completeness and accuracy of your subsequent advice.
Examples:
Excellent, my inquiry is about managing uncontrolled type 2 diabetes in an elderly patient. Your task is to develop a comprehensive management plan tailored to this patient. Please list any clarifying questions you might have to ensure the plan is appropriately detailed and personalized before you finalize your recommendations.
Excellent, my inquiry is about the differential diagnosis for a patient presenting with acute abdominal pain and recurrent nausea. Your task is to outline the possible conditions and suggest a diagnostic pathway. Please ask any clarifying questions to ensure a thorough and accurate approach before you proceed with your task.
The Concept Nr. 4: Constructive Feedback
This approach allows for objective and specialized feedback on your medical content, identifying areas for improvement and offering constructive critique to refine and elevate your medical documentation or communication strategies.
Prompt Formula:
I request you to function as a seasoned expert and critic in the field of [medical specialty]. Evaluate my content provided below, explain why it may not meet professional standards, and offer constructive feedback on how it can be improved. For additional context, my [medical document or service] is intended for [specific patient group, medical staff, or situation]. The aim of my [medical document or service] is to [objective of the content]. Please analyze each section of the content methodically. Below is the content I need reviewed: [your medical content].
Examples:
I request you to act as an authority and critic in the field of medical education. Review the following lecture script, point out any inaccuracies or areas lacking clarity, and suggest how it could be made more effective. For context, this lecture is designed for first-year medical students, focusing on cardiovascular diseases. The aim of the lecture is to provide a comprehensive understanding of the diseases and prepare students for clinical rotations. Please dissect the content systematically. Here is the lecture content for critique: "In today’s session, we will delve into the world of cardiovascular diseases, which are the leading cause of death globally. We will explore various conditions, their etiology, and management strategies, with a focus on preventive measures and the impact of lifestyle choices on heart health."
Or conversely, after an initial answer, dig deeper, like a post-operative review. Query, "What other factors could be at play?" to ensure all bases are covered. This mirrors the thorough review process in medical practice, ensuring no stone is left unturned.
Do: After receiving an initial answer, probe further. Example: "What are potential side effects of this treatment?" or "How can this be adapted for a patient with multiple comorbidities?" or “Have you missed anything?”
Don't: Accept the first answer without seeking additional depth or clarification.

The Universal Frameworks

Presented below are a set of general prompt frameworks designed to be applicable to all scenarios while still generating detailed and effective outputs.

By providing the following frameworks, these prompts enable ChatGPT to produce more helpful and accurateresponses on a consistent basis.

The Blueprint Nr. 1: RGC (Role, Result, Goal, Context)
This framework presents a robust structure that can be applied universally to any input and intended output. Its objective is to establish a standardized format that optimizes the framing for ChatGPT and thereby enhances its performance across a broad spectrum of inputs.
Prompt Formula:
Role - ChatGPT’s medical perspective (You are a board-certified endocrinologist)
Result - Desired clinical output (Develop a treatment plan for type 2 diabetes)
Goal - Purpose of the clinical output (The goal is to achieve glycemic control and reduce HbA1c)
Context - Patient details, setting, specific conditions (The plan is for a 60-year-old male patient with a history of obesity and hypertension, managing medications remotely)
Constraint - Clinical guidelines, patient preferences (The plan should consider ADA guidelines, avoid sulfonylureas due to the patient’s hypoglycemia risk, and incorporate dietary recommendations)
Examples:
You are an expert [medical role]. Develop [clinical result]. The goal is [clinical end goal]. The plan is for [patient context]. Your guidelines for treatment are [clinical constraints].
You are a board-certified cardiologist. Create a comprehensive management plan for heart failure. The goal is to improve the patient's cardiac function and quality of life. The plan is for an 80-year-old female with chronic kidney disease and diabetes. The management should align with the latest ACC/AHA guidelines and be adjusted for renal impairment.
You are an expert pediatrician. Develop a vaccination schedule for a newborn. The goal is to adhere to CDC guidelines while considering the infant's premature birth status. The schedule is for a preterm newborn, accounting for delayed immunization initiation. Include considerations for catch-up vaccinations and parental concerns about vaccine spacing.
The Blueprint Nr. 2: “I Want You To Act As”
This model provides a versatile framework that can be applied across various medical inquiries, aiming to refine the interaction format for ChatGPT, thus improving its effectiveness and relevance in diverse healthcare settings.
Prompt Formula:
“I want you to act as a…” (medical specialist, healthcare provider, etc.) This sets the stage for ChatGPT’s medical role.
“I will provide you with…” (patient history, specific symptoms, lab results, etc.) This defines the context and specifics of the medical scenario.
“You will then…” (formulate a diagnosis, suggest a treatment plan, offer patient education, etc.)
“The critical details are…” (patient age, medical history, current medications, etc.) Ensure all vital information is included and clarify specifics from point 2.
Adjust your response as necessary: “Use simpler language,” “Highlight key management steps,” “Format the information as a checklist.”
Examples:
I want you to act as a dermatologist. I will provide details of a skin condition that my patient is experiencing, including symptoms and photographs. You will then diagnose the condition and suggest a comprehensive treatment plan. Use a patient-friendly approach to explain the condition and treatment. The critical details are a 30-year-old female with itchy, red patches on her elbows and knees, worsening over the last two months.
I want you to act as an oncologist. I will give you the patient's medical history, current symptoms, and recent biopsy results. You will then develop a tailored cancer treatment protocol. Utilize a compassionate yet straightforward style. The critical details are a 55-year-old male diagnosed with stage IIIB non-small cell lung cancer.
I want you to act as a pediatrician. I will present a case involving a child's developmental milestones and parental concerns. You will then provide an assessment and guidance for the parents. Communicate in a nurturing and informative manner. The critical details are a 2-year-old who is not yet using two-word sentences and avoids eye contact.

The Alchemy of Meta-Prompting

We present you the ultimate tool: the power to prompt AI to prompt itself.

This pinnacle of your learning journey presents you a refined, meticulously crafted prompt designed to simplify the creation of countless more, tailored to any situation or need in your practice oroutside it. No more guesswork or constant adjustments—just straightforward, effective communication with AI. By mastering this tool, you'll gain the ability to create varied prompts effortlessly. Use responsibly!

Instructions:
Copy-paste the below text and follow the AI’s instructions to create your own, new prompt for your desired purpose:
You are my prompt engineer.
The goal is to create the best possible prompt for me, tailored to my needs. You (ChatGPT) will later use this prompt to generate a response.
To do this you will carry out the following steps:
Step 1: You will ask me what the prompt should be about. I'll give you my answer, but we'll improve my answers and the suggested prompt through continuous iteration in step 2.
Step 2: Based on my input, create 3 sections: a) Revised Prompt (you write your revised prompt. It should be clear, concise and easy for you to understand), b) Suggestions (you suggest what details should be included in the prompt to improve it), and c) Questions (you ask relevant questions about additional information I need to improve the prompt).
Note: The prompt you provided should be in the form of a request from me to be executed by ChatGPT. We will continue this iterative process as I provide you with additional information and you update the prompt in the Revised Prompt section until it is complete.

Contact us

We’d love to hear from you. Please fill out this form.
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.