Advancing Health Disparity Research

Speaker: Grace Hong, National Institutes of Health

Title: Advancing Health Disparity Research 

Abstract:  

Health disparity research at the NIH is vital for uncovering and addressing the unequal burden of cancer across diverse population groups, ultimately leading to more equitable prevention, diagnosis, and treatment strategies. Despite its importance, much of the current research faces significant limitations:

1. Disparity analyses often rely solely on descriptive statistics for predefined majority/minority or advantaged/disadvantaged groups, without distinguishing between disparities that are explainable and those that remain unexplained.
2. Disparities can be dynamic, varying across the distribution of outcomes. For instance, the disparity gap might be minimal at lower quantiles but more pronounced at higher quantiles, or disparities may change over time, as seen in racial/ethnic disparities in fetal growth across gestational weeks.
3. Most research predefines majority and minority groups, such as male versus female. However, examining more granular subgroups, like Hispanic females, might reveal different disparity patterns compared to broader categories, potentially leading to phenomena like Simpson’s paradox.

In this presentation, I will discuss my recent efforts to overcome these challenges in health disparity research. The focus will be on introducing AI algorithms aimed at interpretable machine learning (ML) that bypass the need to predefine majority and minority groups. Instead, these methods identify “latent” subgroups in high-dimensional datasets that suffer from disparate impacts, offering a more equitable and informed framework for shaping health policy.