We are pleased to announce the publication of a new paper from our laboratory, developed in collaboration with the University of Warwick and presented at the AIME 2026 (Artificial Intelligence in Medicine) conference.

Our paper, "The Critical Role of Model Selection in Causal Inference: A Comparative Analysis of Classification Models Within the InferBERT Framework for Pharmacovigilance," examines how the choice of machine learning model influences causal inference in pharmacovigilance analyses. Rather than assuming that larger models inherently yield better results, we systematically compared XGBoost, ALBERT, BioBERT, and Med-LLaMA within the InferBERT framework.

Our key finding challenges a widely held assumption in the field: domain-specific pre-training matters more than model size. BioBERT consistently outperformed both traditional machine learning approaches and a substantially larger medical large language model, demonstrating that carefully selected, domain-adapted models can deliver more reliable causal inference results in adverse drug event detection.

Congratulations to Csaba Kiss, Dr. Roland Molontay, and Dr. Gabriele Pergola on this excellent international collaboration!

The full conference paper is available here: https://link.springer.com/chapter/10.1007/978-3-032-30710-1_21