This Perspective addresses a critical emerging threat to global health: the proliferation of unverified clinical signals derived from large-scale retrospective databases. By exposing our scientific experience with three high-profile running examples, ranging from oncology to aging and psychiatry, we discuss how sophisticated statistical modeling (e.g., Cox regression, Propensity Score Matching) can inadvertently crystallize epidemiological impossibilities when disconnected from clinical reality and national gold standards. We draw upon the concept of e-iatrogenesis to describe the systemic generation of unintended medical misinformation through unstable big medical data interpretation. Furthermore, we resort to a formal sensitivity stress test to verify if a clinical signal crosses the thin red line of epidemiological plausibility. This contribution argues for a systematic epistemological maintenance of the scientific record to prevent the sedimentation of unstable signals into public health policy.
Marco Roccetti (Sun,) studied this question.