Abstract:
Verbal autopsy (VA) serves as a crucial tool for understanding causes of death in regions lacking comprehensive vital registration systems. However, current methods for analyzing VA data face significant limitations: physician review is resource-intensive and potentially biased, while existing computational approaches struggle with the complexity of death classification. This talk presents preliminary results from using large language models (LLMs) to interpret VA interviews and predict causes of pregnancy-related deaths. We evaluated ChatGPT-4 and Claude.ai across different prompting strategies and found that LLM-based classification achieves performance superior to physicians. I will discuss some implications of this, as well as methodological trade-offs, and future directions for this line of research.
Abraham Flaxman, PhD, is Associate Professor of Health Metrics Sciences and Global Health at the Institute for Health Metrics and Evaluation (IHME) at the University of Washington. His research focuses on developing novel computational methods for global health challenges, including verbal autopsy interpretation and microsimulation for cost-effectiveness analysis. Combining expertise in algorithms, mathematics, and public health, Dr. Flaxman leads projects that bridge the gap between complex statistical methods and real-world health applications.