Journal of Brain Science and Mental Health
Big Data, Bias, and the Law: A Regulatory Framework for Algorithmic Fairness in Clinical Decision Making in the U.S.
Abstract
Elemegious Mugamba
Artificial intelligence (AI) and machine learning (ML) technologies are transforming clinical decision-making in the United States, promising precision and efficiency in diagnostics, triage, and therapeutic planning. However, these tools simultaneously risk perpetuating and exacerbating structural inequities embedded in clinical data, algorithmic design, and healthcare delivery. This article interrogates the persistent regulatory lacuna governing algorithmic fairness in U.S. clinical AI, where prevailing legal regimes—anchored in post hoc liability, safety-efficacy paradigms, and fragmented federal oversight—fail to address the socio-technical dynamics of bias propagation, disparate impact, and data-driven discrimination.
Deploying interdisciplinary methodologies across administrative law, civil rights jurisprudence, empirical data science, and medical ethics, the article critiques the limitations of current doctrinal instruments. These include the Food and Drug Administration’s (FDA) constrained pre-market review authority under the Federal Food, Drug, and Cosmetic Act; the underutilized equity mandates of Section 1557 of the Affordable Care Act; and the inability of tort and anti-discrimination law to provide adequate redress for algorithmic harm in a black-box, adaptive AI ecosystem. The analysis is further situated within the broader administrative law context shaped by Allina Health Services v. Price, 587 U.S.(2019), which underscores the necessity of clear statutory authorization and procedural transparency in federal agency rulemaking—a principle increasingly salient as agencies confront the opacity of clinical AI.
Through comparative legal analysis, the article examines the European Union’s Artificial Intelligence Act and the General Data Protection Regulation, highlighting a transatlantic divergence in regulatory ambition. The EU’s emphasis on pre- market subgroup performance metrics, auditability, and data provenance stands in stark contrast to the U.S. system’s reactive, sector-specific orientation. Empirical case studies—ranging from racially biased pulse oximetry in Black patients to proprietary risk-scoring tools exacerbating health disparities—demonstrate the urgent need for proactive, legally enforceable fairness obligations in clinical AI governance.
In response, the article advances a novel three-tiered legal framework for the U.S.: (1) codification of mandatory fairness benchmarks as a precondition for FDA approval of AI-enabled medical devices; (2) statutory Algorithmic Impact Assessments (AI-IAs) for all federally funded clinical AI applications, integrated into Medicare and Medicaid policy; and (3) a proposed federal legislative intervention—the Algorithmic Fairness in Health Act (AFHA)—which would establish strict liability regimes, safe harbour protections for compliant developers, and a national compensation mechanism for algorithmically induced clinical harm. This fairness-first model rejects the binary of innovation versus accountability and instead aligns AI governance with public health ethics, administrative legitimacy, and civil rights enforcement.
This work contributes original regulatory theory, doctrinal innovation, and actionable policy recommendations with immediate relevance for U.S. policymakers, global health regulators, and interdisciplinary scholars. It advances the thesis that algorithmic fairness must transition from an aspirational ideal to a first-order legal obligation—reconceptualizing clinical AI not merely as a tool of medical decision support, but as a site of distributive justice, administrative accountability, and civil rights adjudication.