A0621
Title: Bayesian federated cause-of-death quantification under distribution shift
Authors: Zehang Li - University of California, Santa Cruz (United States) [presenting]
Abstract: Cause-of-death data is fundamental for understanding population health trends and inequalities as well as designing and evaluating public health interventions. A significant proportion of global deaths, particularly in low- and middle-income countries (LMICs), do not have medically certified causes assigned. In such settings, verbal autopsy (VA) is a widely adopted approach to estimate disease burdens by interviewing caregivers of the deceased. A flexible Bayesian federated learning approach that enables cause-of-death assignment and quantification of cause-of-death distribution in a new population without data sharing from multiple training datasets is proposed. The key to this approach is a latent class model framework that allows flexible characterization of the joint distribution of symptoms and causes across heterogeneous populations with distribution shifts. It is shown that the proposed method significantly outperforms models based on a single training dataset and achieves comparable performance compared to the joint modeling approach that pools all available data. The practical implications of such federated learning models are also discussed in the VA analysis pipeline.
