In Animal Models in Light of Evolution we go to great lengths to explain the relationship of variables in complex systems. What we say was nicely summarized (though I doubt that was the author’s intent) in a recent article in JAMA. Levin et al.:
In the early 1960s, Lorenz devised weather prediction models using simultaneous differential equations, in which the starting values exactly predicted the outcome (ie, were deterministic). When Lorenz reran a model he had previously began, but entered values that differed by only 0.1% from those produced in the prior run, the model gave highly dissimilar results, far more than the 0.1% variation in starting conditions. This finding, that a minute change in initial conditions could lead to a dramatic change in outcome, has become popularly known as the butterfly effect and is an archetype of chaotic behavior. The biomedical literature is likewise replete with examples of small differences in the choice of preclinical models leading to large differences in results.  (Emphasis added.)
Popular VideoIt turns out President Trump's budget has $2 trillion error in it:
Living systems are complex systems which is a subcategory of chaotic systems. (The differences between the two are not important for this discussion.) Very small differences between two otherwise identical complex systems can mean lethal differences in drug response and disease. For proof of this we need look no further than differences in disease and drug response between identical (now called monozygotic) twins. If monozygotic twins do not respond the same way, what chance is there of mice or monkeys predicting human response? Perhaps we are not the only ones who think dynamical systems and chaos are important when discussing the use of animals in research.
Popular VideoIt turns out President Trump's budget has $2 trillion error in it:
Difficulties in translational research have been highlighted by recent failures of clinical trials to substantiate preclinical studies, exemplified by antineoplastic drugs that are highly effective in animal models but have limited or no efficacy in patients with cancer. In some situations, the therapy worsens the disease; for instance, the MRKAd5 HIV-1 gag/pol/nef vaccine to prevent human immunodeficiency virus (HIV) infection increased the likelihood of HIV acquisition. Beyond these individual cases, some long-standing failures in moving from bench to bedside appear intrinsic to entire areas of research (eg, blood substitutes for hemorrhagic shock, oral tolerance for autoimmune disease, or neuroprotection for stroke). Despite years of clinical studies in these and other conditions, only rarely do therapies meet the threshold for regulatory approval.
I agree with Levin et al. when they say, previously in the essay, that basic research can be important. (I might have mentioned that in previous essays.) No doubt many advances have come from the basic sciences, especially chemistry and physics. But, when the controversy of using sentient animals in basic research is addressed, the issue of societal norms and the likelihood of advances must also be represented in morals equation.
Levin et al continue:
Yet if a small difference in how a model is performed or in the species used can lead to such disparate results in outcome, it should not be surprising that extrapolation to actual patients with actual disease should result in similar differences. For instance, superagonist anti-CD28 antibodies [they are referring to TGN1412] had been tested in rats and monkeys without significant adverse effects, but were almost deadly when administered to humans, resulting in cytokine storm and multiorgan failure in 6 previously healthy volunteers. What underlies this chaotic behavior that appears to be random but is actually deterministic? Differences in scale, timing, and choice of model can make it difficult to extrapolate findings confidently from animals to humans. While representational models are frequently valid in the physical sciences, they are more difficult in the biological sciences owing in part to the lack of an equivalent scalable aspect: an aeronautical engineer can build scaled models of an airplane, but a dog is not a scaled-down horse.
For example, transgenic mouse models of cystic fibrosis have significantly fewer lung and pancreatic abnormalities than patients with equivalent mutations, making it difficult to replicate the predominant cause of morbidity, persistent infection with Pseudomonas aeruginosa. These and other unrepresentative models of other diseases result from fundamental genetic, developmental, rearing, and lifestyle differences among species. For example, a divergence in tissue-specific regulation has developed in evolution between mice and humans. A remarkably large number of transcription factors do not bind to the same promoter sites in the 2 different species. (Emphasis added.)
Levin et al. nicely sum up one of our main messages from Animal Models in Light of Evolution. Why can’t I do it that efficiently?
Also this week, an editorial in Nature Medicine:
An investigational Alzheimer's disease drug called Dimebon is a case in point. A phase 2 trial of Dimebon reported significant improvements in cognition in individuals with Alzheimer’s disease (The Lancet 372, 207–215, 2008), sparking much excitement in the Alzheimer’s community. However, on 3 March, Pfizer and Medivation, the companies developing the drug, announced that Dimebon did not affect cognition in a much larger follow-up phase 3 study. The lack of a clear understanding of how the drug worked may have contributed to this unfortunate outcome. 
This is not uncommon. Drugs frequently pass animals tests, Phase I and II human trials only to be shelved because of results from Phase III human trials. The editorial continues:
A 2001 paper showed that the drug enhanced cognition in a rat model of Alzheimer's disease and, importantly, in 14 individuals with this disease . . . more recent data have indicated that Dimebon actually increases amyloid-beta levels in a mouse model of Alzheimer’s disease (Mol. Neurodegener. 4, 51, 2009). 
As we have repeatedly said, test enough species or even strains and you will eventually find the result you want. Although the result you want to find in animals may not be the one you actually see in humans. This is one reason animal models per se cannot predict human response. Yes, some strain or species might be found that reacts like humans but this is not what prediction means in science. Further, the reason why animals cannot predict human response is to be found, as the JAMA article above suggested, in the fact that animals (including humans) are complex systems.
In order to appreciate why drugs can pass Phase I and II human trials only to fail Phase III, one must understand the purpose of the various phases. The real reason many of these drugs go into clinical trials is because of animal studies. If the drug shows the desired effect in animals it will in all likelihood go to human trials, all other factors being equal. From ClinicalTrials.gov:
In Phase I trials, researchers test an experimental drug or treatment in a small group of people (20-80) for the first time to evaluate its safety, determine a safe dosage range, and identify side effects.
In Phase II trials, the experimental study drug or treatment is given to a larger group of people (100-300) to see if it is effective and to further evaluate its safety.
In Phase III trials, the experimental study drug or treatment is given to large groups of people (1,000-3,000) to confirm its effectiveness, monitor side effects, compare it to commonly used treatments, and collect information that will allow the experimental drug or treatment to be used safely.
Drugs are not really evaluated for efficacy until Phase III and it is at this phase that the differences between humans and animals can manifest (if not in Phase I—the safety phase). This is why drugs fail in Phase III instead of Phase I or II. The number of humans being studies is just too small until Phase III. So the first real test of efficacy is in Phase III. Sometimes side effects will not show up until Phase III or even until the drug is being mass marketed. (The solution to this problem is gene-based medicine and, until then, increasing the number of people in Phase III trials and mandating longer and broader trials. But that is for another essay)
The animal tests do not protect the human volunteers in Phase I, the safety phase either. Speaking of toxicity trials for new drugs in humans, an unnamed clinician quoted in Science stated, “If you were to look in [a big company’s] files for testing small-molecule drugs you’d find hundreds of deaths” . This is not comforting but is nonetheless true.
In the final analysis, the fact that animals and humans are complex living systems means that animals will not be able to predict human response to drugs and disease.
And that’s really all you need to know.
(If all this science talk is a little too much for you but you are nonetheless interested in the subject, please try reading FAQs About the Use of Animals in Science: A handbook for the scientifically perplexed. As the title suggests, it was written along the same lines as Animal Models in Light of Evolution but with less science speak.)
1. Levin, Leonard A. and Danesh-Meyer, Helen V., Lost in Translation: Bumps in the Road Between Bench and Bedside. JAMA 303 (15), 1533.
2. Bruder, C. E. et al., Phenotypically concordant and discordant monozygotic twins display different DNA copy-number-variation profiles. Am J Hum Genet 82 (3), 763 (2008); Choi, CQ, Twins Diverge. The Scientist 21 (5), 71 (2007); Flintoft, Louisa, Epigenetics: Identical twins: epigenetics makes the difference. Nature Reviews. Genetics 6 (9), 667 (2005); Fraga, M. F. et al., Epigenetic differences arise during the lifetime of monozygotic twins. Proc Natl Acad Sci U S A 102 (30), 10604 (2005); Machin, G. A., Some causes of genotypic and phenotypic discordance in monozygotic twin pairs. Am J Med Genet 61 (3), 216 (1996); Qiu, J., Epigenetics: unfinished symphony. Nature 441 (7090), 143 (2006); Wong, A. H., Gottesman, II, and Petronis, A., Phenotypic differences in genetically identical organisms: the epigenetic perspective. Hum Mol Genet 14 Spec No 1, R11 (2005).
3. Editorial, Mechanism matters. Nat Med 16 (4), 347 (2010).
4. Marshall, E., Gene therapy on trial. Science 288 (5468), 951 (2000).