Innovative AI Prompt Design Enhances Infectious Disease Clinical Practice

Since the release of ChatGPT in 2022, generative AI chatbots have been celebrated for their potential to revolutionize efficiency and productivity. In the medical field, particularly in specialties like infectious diseases, AI offers promising applications in clinical care, documentation, medical education, and research.

Despite these opportunities, only 20% of employed American adults use AI chatbots for work, and usage drops to 10% among adults over 50. One reason might be the often generic responses these chatbots provide. The quality of chatbot output is closely tied to the input quality, highlighting the importance of effective prompt design in medical tasks.

Users of large language models don’t need advanced coding skills; natural language prompts suffice. These prompts range from simple questions to complex multistep instructions. Crafting these inputs is known as prompt design, and refining them to achieve the desired output is called prompt engineering.

Prompt design is an excellent introduction to AI, while prompt engineering turns casual users into adept ones. As AI spreads across fields, including medicine, mastering these skills becomes crucial. One effective prompt crafting strategy is the CO-STAR framework: Context, Output, Specificity, Tone, Audience, and Response.

Setting the context involves indicating your role (e.g., ID provider, pharmacist), focus (clinician, researcher, educator), and work setting (clinic, inpatient, community). Define the output clearly, whether it’s a letter, blog article, list, essay, image, handout, or quiz, and be specific with the topic.

For example, if you need a handout on latent tuberculosis, specify whether the focus should be on testing, treatment options, or a general overview. Use tone descriptors like “empathetic,” “easy-to-understand,” “professional,” or “scholarly,” and specify the target audience to ensure the output matches your communication style.

Initial prompts may not always yield the best results. Refining prompts iteratively by tweaking CO-STAR elements until achieving the best output is good practice. Use delimiters like colons, semicolons, line spaces, quotation marks, and triple quotation marks to structure prompts, especially those involving multiple steps.

Recently, commercial AI chatbots have added multimodal input and output capabilities, allowing users to upload images, documents, audio recordings, and web links as part of their prompts. The CO-STAR framework remains effective for these multimodal prompts. Specify any uploaded files and the desired output format (e.g., image file, Excel, .csv file).

The integration of Retrieval-Augmented Generation (RAG) into chatbots like ChatGPT has further enhanced their accuracy and functionality. RAG combines generative AI with targeted information retrieval. For instance, generating test questions on antibiotics can involve uploading a reference document and prompting the chatbot to retrieve information from it first, ensuring content consistency and reducing errors.