Ethical Challenges for Deidentified Data in Learning Health Systems

Obstetrics, Gynecology and Reproductive Sciences Ethics Grand Rounds

Marielle S. Gross, MD, MBE
Assistant Professor
University of Pittsburgh School of Medicine
Department of Obstetrics, Gynecology and Reproductive Sciences
University of Pittsburgh


Teams Online
+1 412-447-5295 United States, Pittsburgh (Toll)
Conference ID: 553 687 71#


Abstract: To promote learning while minimizing privacy risks in the US, access to deidentified health data is neither restricted by HIPAA nor governed by Common Rule human subjects’ protections (AKA “safe harbor”). Meanwhile, the ethical framework for learning health systems (LHS) proposes both that clinical care and research no longer be treated as categorically ethically distinct, and that patients contribute to learning, primarily by allowing research on health data generated during clinical care.
This talk argues that patient-centered outcomes research, which involves questions and outcomes “meaningful and important to patients and caregivers” and falls under current “safe harbor” for deidentified data, may undermine distributive justice, respect for persons, and beneficence. First, deidentification intentionally eliminates accountability to patients whose data are studied, making research easier, but also precluding patients from direct clinical benefits or recognition. This contradicts just distribution of research benefits and burdens, especially when findings are timely or personally significant, a problem exacerbated by background conditions in which access to healthcare is not universal. Further, lack of human subjects’ protections may functionally dehumanize deidentified patients by allowing objectification or monetization of inherently intimate data—data often obtained via invasive examinations performed under auspices of advancing individuals’ health interests—without obligations for informed consent, or transparency and engagement more broadly. Finally, deidentification’s beneficent intentions are undercut by artificial intelligence coupled with increasing clinical, genomic and biological data, and by systematically emphasizing siloed, incomplete datasets. Ultimately, deidentification may impede flourishing of a just LHS by maintaining the distinction between patient and research subject, neglecting duties of care to both, and preventing seamless treatment, learning and feedback.