Eliminating Health Disparities in Clinical Trials: How Can Physicists Contribute?
Presentations
WE-DE-BRB-0 (Wednesday, 7/13/2022) 10:15 AM - 12:15 PM [Eastern Time (GMT-4)]
Ballroom B
Medical physics innovations can substantially improve cancer outcomes, but the rapid pace of innovation can widen healthcare disparities if equity is not prioritized during clinical trial design and execution. We will begin this session by examining the consequences of failing to prioritize equity in clinical trials with a strong physics or technological component, thereby highlighting the rationale for considering equity at the outset. We will then look at proton therapy as a case study, and review ways medical physicists can play a role in health disparities efforts in proton therapy and other NCI trials.
In the second half of this session, we will examine how the diversity of patient populations and acquisition systems can impact the performance of machine learning algorithms. We will start our discussion by defining artificial intelligence (AI) bias, which is when outputs of algorithms are prejudiced due to biased training data, assumptions made during the algorithm development, or a combination of both. Care must be taken to mitigate bias in both training and testing datasets as well as in algorithm development. We will discuss real-world examples of AI bias: the Amazon recruiting algorithm, the COMPAS (Correctional Offender Management Profiling for Alternative Sanctions) algorithm, and a Health Care Risk algorithm. Then, we will review the technical aspects of AI algorithms that may lead to biased results, specifically covering the difference in supervised and unsupervised training strategies, and ways to mitigate bias in these systems. Finally, our discussion will conclude by presenting projects that other investigators are actively working on to help reduce AI bias.
Learning Objectives:
1. Identify the rationale for prioritizing equity in clinical trial design and accrual
2. Identify opportunities for physicists to contribute to health equity in NCI trials
3. Understand the issues related to the potential for bias in training/testing datasets in terms of patient population and acquisition systems
4. Understand strategies for identifying and mitigating bias in datasets and AI algorithms
Handouts
- 175-63539-16291659-183971-201867491.pdf (Paige Taylor)
- 175-63541-16291659-186975-92873736.pdf (Zohaib Iqbal)
Keywords
Not Applicable / None Entered.
Taxonomy
Not Applicable / None Entered.
Contact Email