The false-positive and false-negative results of the animal bioassay can be considerable. Ennever and Lave analyzed the data on known human carcinogens with the animal data for cancer predictability. They found a disturbingly large proportion of incorrect predictions, ‘potentially allowing widespread human exposure to misidentified chemicals’. An analysis of the data on 780 chemical agents listed in the International Agency for Research in Cancer database found the positive predictivity of the animal bioassay for a definite or probable human carcinogen to be only around 20 per cent. In addition to placing human lives at risk, the low predictability of this assay is costing us money and wasting time. Each assay requires up to millions of dollars and years of planning. In the meantime, as we continue to rely on this assay, there is a huge backlog of untested chemicals to which we are already exposing ourselves.
Other toxicology and carcinogenicity tests that rely on animals are equally flawed. One study examined the toxicological profiles of rodent and non-rodent (beagles and NHPs) species of 50 compounds. The study found poor correlation of target organ toxicity across species and concluded that ‘simple extrapolation across species is unrealistic’. The study authors called for regulatory agencies to institute an evaluation of tests using animals as predictors of human adverse signs. In 1999 the Health and Environmental Science Institute examined the data on 150 compounds that had produced a variety of toxic effects in people. It found that only 43 per cent of the compounds produced similar effects in mice and rats and 63 per cent did so in other animals. A reviewer of toxicology testing and regulations commented that ‘compelled to act, regulators have chosen animal tests to forecast human cancer risks. To this end, animal data are filtered through a series of preconceived assumptions that are presumed to overcome a host of human/animal differences of biology, exposure, and statistics-differences that in reality are insurmountable.’
Recognizing the immense difficulty in predicting toxicity in one species based on the toxicity data from another is not new. As early as 1978, Fletcher found poor correlation between drug safety tests in animals and subsequent clinical experience with 45 major drugs, including anti-cancer agents, antibiotics, cardiac agents and neurological agents. Fletcher’s survey established that only 25 per cent of the toxic effects observed in animals might be expected to occur in humans. Assessing three decades of data on the subject, toxicologist Ralph Heywood also found that the concordance between animals and humans is only 25 per cent. ‘Toxicology’ he concluded, ‘is a science without a scientific underpinning.’
‘In retrospect,’ Fletcher concluded in his 1978 report, ‘it is a relatively simple matter to determine the correlation between animal and human studies, but prospectively it is difficult to know which particular toxic effects are likely to prove troublesome when it comes to giving the drug to man.’ And that’s the catch: accurately predicting when the animal experimental results are relevant to humans is nearly impossible because of inter-species differences. We can always go (and have often gone) back after clinical trials have been conducted to assess whether the animal experimental results correlated with the clinical results, but retrospective confirmation is not the purported reason for using animals in experimentation. They are intended to predict human results and inform human health care. If we find that the animal experimental results equated with the clinical results, then the research community hails the efficacy of the animal experiments. But when the animal and human results do not match, the proclaimed failure is said to be a result of flaws in experimental design, publication bias or use of young animals for a disease that occurs predominately in elderly humans. Rarely is the use of the animals themselves—not how they are used—questioned.
While most researchers admit the difficulty in extrapolating and applying information obtained from other species to humans, commonly proposed solutions to this colossal obstacle are far from helpful. Neyt et al. suggest that ‘clearly profound differences may exist at the gross, microscopic and genetic level between humans and other mammals, and these differences must be appreciated before extrapolating the results of a given study to human clinical practice.’ Caution in extrapolating data from animals to humans is another common advice given. In fact, ‘appreciation of differences’ and ‘caution’ about extrapolating results from animals to humans is now almost universally expressed in published reports on animal experimental results intended to inform human health.
Yet, in reality, how does one take into account differences in drug metabolism, genetics, expression of diseases, anatomy, behavior, influences of laboratory environments, and species and strain-specific physiologic mechanisms and then discern what is applicable to humans and what is not? There is just no established formula or algorithm to do this. Many scientists have recently acknowledged that modeling human disease in animals is extremely problematic but have still argued for their use, instead, to study basic physiologic mechanisms. But again, if we cannot predetermine what mechanisms in what species and what strain of species and in what caging system and even during what time of day are applicable to humans, then the usefulness of the experiments needs to be questioned.