SC-23 - An Introduction to Artificial Intelligence for Pharmacoepidemiologists

Course aim: The objective of this course is to provide comprehensive guidance on the application of artificial intelligence (AI) within pharmacoepidemiology. The curriculum encompasses various AI topics, including machine learning (ML), deep learning (DL), and generative AI (including large language models, LLMs). The content is specifically tailored to practical pharmacoepidemiology projects and includes several real-life case studies.

Part one will explore the concept of 'fit-for-purpose' AI. It will outline the opportunities for utilizing AI in pharmacoepidemiology and how to identify scenarios where AI can add value. It will also address the appropriateness of AI usage, considering regulatory perspectives, ethical principles, and legal dimensions.

Part two will focus on how AI can be used to accelerate and improve the execution of real-world evidence (RWE) research. Attendees will learn how AI tools can enhance efficiency and optimize key RWE workflows. This section will also address challenges related to AI reliability, such as hallucination, source data integrity, and the extent of human review required. Additionally, part two will discuss various ways in which AI can be used to enhance real-world datasets by extracting, categorizing, or inferring data points (including natural language processing on clinical notes). Participants will have the opportunity to engage hands-on with these tools and learn several expert prompting techniques specific to RWE-related challenges.

Part three will provide attendees with a practical technical understanding of the design, implementation, and evaluation of both 'classical' machine learning and more advanced LLM AI projects. Participants will learn about the various steps involved in these types of studies and the significance of concepts such as hyperparameter tuning and gradient descent. Attendees will also have the chance to experiment with different models and train their own deep learning model in a web browser.

The course will conclude with part four, which will explore various methods to complement and enhance the knowledge gained throughout the course. We will offer ongoing access to course materials in a dedicated environment, along with additional challenges and tips for success.

Pre-requisites: This is an entry level course. No previous experience of AI is required. A basic understanding of epidemiology concepts will be beneficial for students but is not mandatory. Attendees should bring a device (such as a laptop, mobile phone etc.) if they wish to participate in hands-on activities (not mandatory).

Learning objectives:
1. To understand how and when AI can be used for pharmacoepidemiology research and to provide real world examples
2. To understand where AI can add value, and how stakeholders perceive its use.
3. To be aware of the legal and ethical challenges surrounding the use of AI in pharmacoepidemiology and healthcare
4. To provide a high-level overview of the steps in a project involving ML, including design considerations, data preparation, modelling, model validation, checking for bias, and interpretation of results
5. To engage with hands-on activities and an extended learning environment (available after the course)

Course outline:
Part I: Fit-for purpose AI - 1 hr
• Introduction to the course - 10 minutes [Harriet Dickinson, Gilead]
• AI applications and meeting the needs of regulators – 15 minutes [Donna Rivera, FDA]
• Choosing the right approach – critically thinking about transparency, uncertainty, and credibility - 15 minutes [Dana Teltsch, EviDT LLC]
• What you need to know: ethics, AI legislation, and bias – 15 minutes [Harriet Dickinson, Gilead]
• Q&A section (incl. slido) – 5 minutes
BREAK (15 minutes)

Part II: Accelerating RWE study execution using AI - 1hr
• AI for practical purposes - 15 minutes [John Diaz-Decaro, Moderna]
• Strategies for enhancing RWD using AI – 15 minutes [Joseph Zabinski, OM1]
• Getting hands-on: prompting like a pro (using elicit, ChatGPT) – 20 minutes [Harriet Dickinson, Gilead]
• Q&A section (incl. slido) – 10 minutes

BREAK (15 minutes)
Part III: AI for methods/inference - 45 minutes
• An introduction to the steps of a classical ML project – 15 minutes [Harriet Dickinson, Gilead]
• Deep Learning and getting hands-on - 30 minutes [Hui Feng, IQVIA]

Part IV: Continuing on your learning journey – 15 minutes
• Where to continue your journey: a community learning group - 5 minutes [all]
• Q&A – 10 minutes