DIA 2021 Virtual Global Annual Meeting

Component Type: Session
Level: Intermediate
CE: ACPE 1.00 Knowledge UAN: 0286-0000-21-660-L04-P; CME 1.00; IACET 1.00; RN 1.00

Predictive models generated by machine learning can exhibit inequitable behavior, with differing false positive rates between protected groups. We will describe how this type of bias and inequity can arise in models used for healthcare triage. In the second talk, we evaluate and assess how targeted learning and other state of the art and tools available for data-adaptive learning can be used to estimate and compare causal inference models. We will explore real-world observational data challenges, including missing data, confounding, and other issues. Using causal tools, from propensity scores to targeted learning, we will investigate this rich frontier of statistical science.

Learning Objectives

Describe machine learning algorithmic bias and how to evaluate predictive models for equity; Identify the state of the art and tools for data-adaptive learning and applications to causal inference and machine learning models.

Chair

David O Olaleye, PhD, MSc

Speaker

Incorporating Context and Causation in Observational Real World Data: From Propensity Scores to Targeted Learning
Andrew Wilson, PhD, MS

Algorithmic Bias in Healthcare Triage
Eric Siegel, PhD, MS

Representation of Diverse Groups in Test Sets
Terri L. Cornelison, MD, PhD