YSPH Biostatistics Seminar: "Generalizability, robustness and equity in machine learning risk prediction models"

BIS 526 students are required to attend in person.  Others are allowed to attend in person but may also attend via zoom.

Rumi Chunara, Ph.D.

Title- "Generalizability, robustness and equity in machine learning risk prediction models"

Abstract

As risk prediction models become embedded in healthcare practice, real-world scenarios expose the intertwined nature of data collection and analytic model development. I will first illustrate this based on the problem of guaranteeing performance and fairness for prediction in unseen environments. I will next show the importance of understanding the data generating process for prediction models via findings from a study of mortality prediction in data from 179 hospitals across the United States. Third, I will share our new NIH training program on Data Science for Social Determinants, and why such training is also an important effort towards improving data and machine learning in health applications.