Abstract
<sec> <title>UNSTRUCTURED</title> <p>The precautionary principle justifies preventive action when serious harm is plausible but scientific certainty remains incomplete. Data science health research increasingly creates this condition because datasets, models, and derivatives are reused across time, institutions, and purposes, generating harms that may be group-based, scalable, inequitable, difficult to reverse, and poorly estimable at the point when governance decisions must be made. We defend a moderate, continuity-sensitive precautionary principle for data science health research. Existing governance tools—including privacy protection, consent, fairness assessment, security, lifecycle monitoring, stewardship, and public-interest review—remain essential, but they do not by themselves specify when uncertainty should alter the terms of ethical approval, data access, or institutional oversight. Building on analyses of representational veracity and the continuity trap, we argue that precaution should be activated when a present data-stage creates plausible threats of serious, scalable, inequitable, or hard-to-reverse harm under material uncertainty, particularly where visible continuity signals such as provenance, locality, or broad consent may be overread as evidence of ethical acceptability. We reconstruct the uptake of the precautionary principle in research ethics from public health, epidemiology, pharmaceutical safety, and germline intervention, and distinguish precaution from prevention, conventional risk-based regulation, fairness, responsible innovation, adaptive governance, stewardship, dynamic consent, and public-interest review. We then codify a continuity-sensitive formulation for ethics committees, data-access committees, repositories, and institutional AI governance bodies. The framework includes a precautionary activation score, an oversight ladder, and committee-level performance metrics to support proportionate, reversible, and evidence-generating decisions. A continuity-sensitive precautionary framework unifies fragmented literatures into an action-guiding governance concept that can be implemented before harms occur, scale, or become entrenched.</p> </sec>