In this episode of the Research Insights Podcast, Lisa Schilling, Director of Practice Research at the Society of Actuaries Research Institute, speaks with Dr. Rob Lieberthal, lead author of Evaluating Representativeness in Healthcare Claims Data: A Framework for Representativeness-Based Fairness in Actuarial Applications of AI.
Their discussion explores why strong overall model performance does not necessarily mean an AI or machine learning model performs equitably across populations. Dr. Lieberthal explains how gaps in healthcare claims data can affect rural, lower-income, underdiagnosed, and other populations—and why actuaries should examine representativeness before relying on model results.
Listeners will learn about the report's three-stage bias assessment covering input bias, model bias, and application bias; examples involving geographic representation, missing data, and intersectionality; and practical approaches for evaluating subgroup performance. The episode also highlights the companion Model Card Generator and Decision Assist Tool, which are designed to help actuaries document model limitations, compare model alternatives, and apply professional judgment throughout the modeling process.
Explore the full research report and companion resources through the Society of Actuaries Research Institute at SOA.org, and discover practical ways to strengthen fairness, transparency, and informed decision-making in actuarial applications of AI.
https://www.soa.org/resources/research-reports/2026/dei127-represent-hc-data/