Most cancers are multifactorial diseases. Yet, epidemiological modeling of the effect of ionizing radiation (IR) exposures based on the linear no-threshold model at low doses (LNT) has generally not included co-exposure to chemicals, dietary, socio-economic and other risk factors also known to cause the cancers imputed to IR. When so, increased cancer incidences are incorrectly predicted by being solely associated with IR exposures. Moreover, to justify application of the LNT to low doses, high dose-response data, e.g., from the bombing of Hiroshima and Nagasaki, are linearly interpolated to background incidence (which usually has large uncertainty). In order for this interpolation to be correct, it would imply that the biological mechanisms leading to cancer and those that prevent cancer at high doses are exactly the same as at low doses. We show that linear interpolations are incorrect because both the biological and epidemiological evidence for thresholds, or other non-linearities, are more than substantial. We discuss why the LNT model suffers from misspecification errors, multiple testing, and other biases. Moreover, its use by regulatory agencies conflates vague assertions of scientific causation, by conjecturing the LNT, for administrative ease of use.
Keywords: Biases; Epidemiological modeling; LNT; Specification error; Uncertainty; p-values.
Copyright © 2018. Published by Elsevier B.V.