James Cimino
Distinguished Professor and Chair, Department of Biomedical Informatics and Data Science
University of Alabama at Birmingham
When Data Are Wrong and When They Are Right About the Wrong Things: Pitfalls, Biases, and Feedback Loops in EHR-Based Research
- Electronic health record (EHR) data offer unprecedented opportunities to generate new medical knowledge from routine patient care. However, these data are collected primarily for clinical and administrative purposes, and their meaning, completeness, and accuracy may not align with researchers' assumptions. Documentation practices, billing requirements, clinical workflows, and patterns of data collection can all introduce errors and biases that lead to misleading research findings. An equally important, but more subtle, problem arises when EHR data accurately reflect health care processes that are themselves biased. Differences in access to care, diagnostic testing, and treatment can become embedded in the data and subsequently influence the knowledge derived from them. When this knowledge is used to guide clinical decisions, it may perpetuate or amplify existing inequities, creating a feedback loop in which biased care generates biased data, which in turn misinforms future care. This presentation will explore these two challenges to EHR-based research, illustrating how data can be misleading both when they fail to represent clinical reality and when they faithfully represent an imperfect health care system. Particular attention will be given to the implications for learning health systems, which risk learning and reinforcing the wrong lessons from the data they collect.
