science & society
VOL 5 | SPECIAL ISSUE | 2004 ©2004 EUROPEAN MOLECULAR BIOLOGY ORGANIZATION
assessment of one additional case of can-cer per million people over a 70-year life-time can require an extrapolation morethan 5–6 orders of magnitude from thehigh exposures used in animal studies tothe low concentrations assumed to be safefor humans. In reality, this means thatalthough lifetime bioassays expose arodent to hundreds of milligrams per unitof body weight per day, the permittedexposure to humans may be fractions of amicrogram per unit of body weight perday. This is an extraordinary degree ofextrapolation with the uncertainty increas-ing progressively the further the predictionmoves away from the observable zone ofgenerated data. Furthermore, the assump-tion of a linear relationship between doseand response completely ignores the factthat our bodies and our cells have devel-oped mechanisms to detoxify harmfulchemicals and exposure to radiation—infact, low doses may even trigger responsesthat are beneficial.In addition, risk predictions based onthe extrapolation of data from animalexperiments using high doses are in facthard to verify, despite massive attemptsthat used up to 24,000 rodents in thelargest of all—failed—validation experi-ments. Even such powerful studies, whichwere carried out in an acceptable mannerand evaluated in extraordinary detail,cannot reliably estimate risks lower thanone in 100, let alone one in 1,000,000.But because risks of one in 100 areregarded as being unacceptable to thegeneral public, especially for routineactivities, regulatory agencies have foundthemselves in a position where they havehad to adopt the use of the lowest estimatedrisk, which cannot be checked or verified.This approach clearly is marked by goodintentions but paved with a large publiccheque book.
espite a lot of argument betweengovernmental agencies and theaffected industries over the processof risk assessment, industry has made littleprogress in persuading governmental agen-cies to budge from their protectioniststance, especially in the areas of hazardassessment and its impact on risk assess-ment. In general, it has been nearly impos-sible for biostatistical models to differentiatebetween linear and threshold models inthe low-dose zone when experimentalstudies used only 2–4 different doses. Insuch cases, the governmental regulatoryagencies usually revert to their moreconservative default assumption models.Nevertheless, the threshold dose responsehas become the key model in toxicologyand pharmacology, whereas the LNTdose–response model has been the princi-pal model for the estimation of cancer risksby virtually all regulatory agencies. I wouldargue that the field of toxicology, includingmost regulatory agencies concerned withchemical and radiation risk, has made amajor error of judgement in selecting thesetwo dose–response models to calculatehuman health and environmental risks forthe broad spectrum of chemical classes andphysical agents. This error profoundlyaffects the standards set for public health,the communication of risks to the public,the establishment of environmental priori-ties and the costs of environmental stan-dards and clean-up activities. It has alsomade the decision-making process moreintuitive and less scientific and thus moresusceptible to political manipulation byinterested parties. If only zero risk is accept-able to the public, then it is easy to call forthe complete abolishment of a product oractivity that carries with it some risk, nomatter how large the costs or benefits.Enter an alternative model, which claimsthat the fundamental shape of thedose–response curve is neither linear northreshold, but rather U-shaped. This so-called hormesis model—after the Greekword ‘to excite’—was first applied todescribe dose–response relationships bySoutham & Ehrlich (1943) more than 60years ago. A typical hormetic curve is eitherU-shaped or has an inverted U-shapeddose–response, depending on the endpointmeasured. If the endpoint is growth orlongevity, the dose–response would be thatof an inverted U-shape; if the endpoint isdisease incidence, then the dose–responsewould be described as U- or J-shaped (Fig 1).This model not only challenges the LNTand threshold models but, more importantly,it suggests that as the dose decreases thereare not only quantitative changes in theresponse measured but also qualitativechanges. That is, as the dose of a carcino-gen decreases, it reaches a point where theagent actually may reduce the risk ofcancer below that of the control group.
R e s p o n s e
R e s p o n s e
R e s p o n s e ( n o r m a l f u n c t i o n )
R e s p o n s e ( d y s f u n c t i o n )
| Dose–response relationships described by(
) the threshold model,(
) the linear non-threshold model,(
) the inverted U-shapedhormetic model and (
) the J-shaped hormeticmodel.(Adapted from Davis & Svendsgaard,1990.)
Ifonly zero risk is acceptable tothe public,then it is easy to callfor the complete abolishment of a product or activity that carrieswith it some risk,no matter howlarge the costs or benefits...the hormesis model clearlyoutperforms either ofthe othertwo competitive models in fairhead-to-head competition