Breakout Group Selections
Attention In-Person Participants: Throughout the 3-day workshop, we will engage in breakout sessions, fostering interactive discussions based on 5 different breakout themes. With a choice of 5 breakout sessions, kindly prioritize your preferences. We will strive to accommodate your first or second choice. Before making your selection, please refer to the following list outlining the breakout session options. If you have no specific preference, kindly indicate so in your first choice.
Summaries
- Proxy Variables:
In this breakout session, we will examine the use of proxy variables in algorithms. Proxy variables are confounders and therefore are used (intentionally or unintentionally) in place of another variable that has a true causal relationship with the outcome. A notable example is the use of race and ethnicity in prediction models, as many experts believe that these variables are often oversimplified proxies for such variables as genetic ancestry or complex environmental and social factors. Other examples include the use of health care costs as a proxy for health care needs (Obermeyer 2019); given that less money is spent on Black patients who have the same level of need as White patients, the examined algorithm falsely concluded that Black patients are healthier than equally sick White patients.
- Synthetic Data:
This breakout focuses on synthetic data—how they are generated, how they might be used, and how they could have both positive and negative impacts on human health. We will discuss specific considerations for the need for and challenges related to synthetic data, including realism, bias, degradation, ethical concerns, and generalizability. Where possible, specific examples will be discussed and used in developing best practices and guiding principles for ethical use and transparency of synthetic data.
Additional Resources:
- Multimodal Data
This breakout will discuss using multimodal data within artificial intelligence (AI) model development, validation, and translation for clinical implementation (e.g., combining structured data, such as diagnoses, with unstructured data, such as text or images). This will include specific considerations for the need for and challenges of generating and linking data in relation to ethics, bias, privacy, and transparency when combining complex, multimodal data with such details as time-course relevance.
Additional Resources:
- Foundational Models:
Foundational models and the closely related Large Language Models (LLMs), such as ChatGPT, have sparked a huge wave of innovations combining the use of AI with the amazing capabilities of these models to integrate and deliver information. This breakout series will explore key concepts behind these models, as well as the implications when creating and using these models in multiple settings involving the clinician, patient, researcher, developer, and community as a whole. Anticipated areas of discussion include such topics as transparency, ethics, privacy, ownership, and reliability.
Additional Resources:
- Data Sharing for General Reuse:
Data sharing has great potential to accelerate scientific innovation; however, it occurs without knowledge of how, whether, or by whom the data will be reused. Responsible reuse of shared data for AI requires technical, operational, ethical, privacy, and regulatory considerations to assess whether the data are fit for purpose.
*
Please select a breakout group.
Please select a breakout group.
Please select a breakout group.
Please contact Janiya Peters at jpeters@scgcorp.com or (301) 670-4990 during business hours if you have any questions or concerns.