ESHS/HSS Annual Meeting

Toward Evidence-Based Artificial Intelligence in Medicine: Lessons from Epidemiology and EBM

• In Event: Epidemiology Unbound 2

English Abstract

This paper examines the concept of bias in epidemiology, its foundational role in the historical development of Evidence-Based Medicine (EBM), and how insights from epidemiology and EBM can guide the creation of Evidence-Based Artificial Intelligence (EBAI). The rise of epidemiologic research, both observational (case–control and cohort studies) and experimental (randomized clinical trials), profoundly altered clinical practice by generating large volumes of data that are challenging to translate into bedside decisions. EBM emerged as a methodological response, providing tools to appraise and rank evidence by study design and corresponding risk of bias. Today, clinical practice faces an even greater disruption from the data deluge associated with Artificial Intelligence (AI) and Big Data. Yet the evidentiary status of outputs produced by AI systems has received limited scrutiny from clinicians and philosophers of science alike. Treating AI, at a high level, as a family of experimental designs, I argue for an EBAI framework for medicine. Epidemiology—through its sustained attention to sources of bias and their mitigation—offers a rigorous toolkit for constructing such a framework and for evaluating AI-derived evidence in a clinically meaningful way.

Author

• Nicolas Brault, UniLaSalle Polytechnic Institute