Failing The Same Way Again


In a PLoS Medicine article titled: Predicting Harms and Benefits in Translational Trials: Ethics, Evidence, and Uncertainty, Kimmelman and London discuss the problems associated with preclinical drug testing and development. From a McGill University press release:

It's all too familiar: researchers announce the discovery of a new drug that eradicates disease in animals. Then, a few years later, the drug bombs in human trials. In the latest issue of the journal PLoS Medicine, ethics experts Jonathan Kimmelman, associate professor at McGill's Biomedical Ethics Unit and Department of Social Studies of Medicine, and Alex John London, associate professor of philosophy at Carnegie Mellon University, argue that this pattern of boom and bust may be related to the way researchers predict outcomes of their work in early stages of drug development. "We do a fairly good job of predicting the success of interventions that make it to later stages of clinical research," said London, who also directs CMU's Center for Ethics and Policy. "But when it comes to the leap from animal studies to the first trials in humans, there are serious problems."

Kimmelman and London go on to give recommendations for improving preclinical trials using animals.

The article is yet another saying what Shanks and I have been saying: animal models do not predict human response to drugs and disease. Van der Worp et al 2010:

Animal experiments have contributed much to our understanding of mechanisms of disease, but their value in predicting the effectiveness of treatment strategies in clinical trials has remained controversial [1–3]. In fact, clinical trials are essential because animal studies do not predict with sufficient certainty what will happen in humans. In a review of animal studies published in seven leading scientific journals of high impact, about one-third of the studies translated at the level of human randomised trials, and one-tenth of the interventions, were subsequently approved for use in patients [1]. However, these were studies of high impact (median citation count, 889), and less frequently cited animal research probably has a lower likelihood of translation to the clinic. Depending on one’s perspective, this attrition rate of 90% may be viewed as either a failure or as a success, but it serves to illustrate the magnitude of the difficulties in translation that beset even findings of high impact. Recent examples of therapies that failed in large randomised clinical trials despite substantial reported benefit in a range of animal studies include enteral probiotics for the prevention of infectious complications of acute pancreatitis, NXY-059 for acute ischemic stroke, and a range of strategies to reduce lethal reperfusion injury in patients with acute myocardial infarction [4–7]. In animal models of acute ischemic stroke, about 500 ‘‘neuroprotective’’ treatment strategies have been reported to improve outcome, but only aspirin and very early intravenous thrombolysis with alteplase (recombinant tissueplasminogen activator) have proved effective in patients, despite numerous clinical trials of other treatment strategies [8,9]. (van der Worp et al. 2010)

From Robert Matthews:

Tests on animals have led to around 100 drugs being thought potentially useful for stroke; not one has proved effective in humans. You don't need to be a balaclava-wearing animal rights activist to question the value of animal studies in this area of medical research. The bigger problem is the belief of the pharmaceutical industry in a myth of its own making. For years it relied on a mix of hard science, inspired guesswork and pure luck to find blockbusters. Many of the most successful drugs have been the result of serendipity.

In an editorial introduction to one article by Ellis and Fidler and another by Van Dyke (Van Dyke 2010), Nature Medicine stated:

The complexity of human metastatic cancer is difficult to mimic in mouse models. As a consequence, seemingly successful studies in murine models do not translate into success in late phases of clinical trials, pouring money, time and people’s hope down the drain. (Ellis and Fidler 2010)

Ellis and Fidler: “Preclinical models, unfortunately, seldom reflect the disease state within humans (Fig. 1).” (Ellis and Fidler 2010)

Lack of predictive ability has consequences. Wenner, writing in Scientific American 2009:

Seventeen patients had undergone treatment before Gelsinger, who was in the final cohort—the one receiving the highest dose of the therapy. Many scientists, as well as the FDA, have raised questions as to why Gelsinger was being treated, given that several patients in earlier cohorts suffered severe liver reactions. Wilson says that they moved forward because it was “the kind of toxicity we would have expected,” based on their work in animals, and they thought it would be manageable. According to Mark Batshaw, director of the Children’s Research Institute at the Children’s National Medical Center in Washington, D.C., Wilson and the rest of the scientific community had to learn the hard way “that what you’ve learned from animals will not necessarily predict what’s going to happen in humans.” Batshaw was also involved in the 1999 trial. (Wenner 2009)

The above can be easily multiplied.

More important than examples, our knowledge of evolutionary biology, genetics, intraspecies differences, and complex systems allows us to place the example in the context of a theory that says trans-species prediction of drug and disease response will be difficult if not impossible. (See Animal Models in Light of Evolutionfor more.)

So what is the solution? If you sell animal models for a living the solution is obvious: new animal models. From Drug Discovery Today, February 2010:

‘Fail early, fail cheap’. It’s a paradigm that has long attached itself to the drug discovery focus on making improvements to the preclinical stage of development. In vitro and in vivo models are used to predict how compounds will behave in humans in terms of efficacy, pharmacokinetics and safety. The findings from these studies are typically used to seek approval from government bodies to conduct human clinical trials. Not surprisingly, only a small fraction of compounds entering clinical Phase 1 studies obtain market approval. According to Nico Scheer, molecular biologist from Taconic, a major reason for failure is that current preclinical models are often poorly predictive of efficacy, pharmacokinetics and clinical safety in humans. “Specifically, the predictability of current preclinical animal models is limited by profound interspecies differences in drug metabolism and disposition. In response, several academic and commercial groups have developed humanised mouse models,” says Scheer. Taconic’s transADMET portfolio, for example, is a joint development programme between Taconic and its partner CXR Biosciences Ltd to deliver novel mouse models that are more predictive for the absorption, distribution, metabolism, excretion and toxicity (ADMET) of a pharmaceutical or chemical compound in humans.

As I have addressed elsewhere (and here and here and in Animal Models in Light of Evolution), genetically modifying an animal will not turn it into a human or make it predictive for humans. This is the reality of complex systems.

Others, with a vested interest in performing basic research on animals, would have us believe that despite the fact that animal models cannot predict drug response in humans they can be used to predict disease response. Unfortunately for them, the same principles of evolution and complexity apply to both types of research.

Using animals to predict human response appeals to common sense and on a superficial level is intuitive; neither of which has a good track record in term of being consistent with the facts of the universe. If something does not “feel right” to people, they reject data and science on that basis. This is more a reflection of our poor educational system than it is on the correctness of the position. Our society still values feelings more than education and expertise. Add this to the money factor and one can understand why animal experimentation persists.


Ellis, L. M., and I. J. Fidler. 2010. Finding the tumor copycat. Therapy fails, patients don't. Nat Med 16 (9):974-5.

van der Worp, H. Bart, David W. Howells, Emily S. Sena, Michelle J. Porritt, Sarah Rewell, Victoria O'Collins, and Malcolm R. Macleod. 2010. Can Animal Models of Disease Reliably Inform Human Studies? PLoS Med 7 (3):e1000245.

Van Dyke, T. 2010. Finding the tumor copycat: approximating a human cancer. Nat Med 16 (9):976-7.

Wenner, Melinda. 2009. Gene therapy: An Interview with an Unfortunate Pioneer. Lessons learned by James M. Wilson, the scientist behind the first gene therapy death. Scientific American Magazine (September):14.


Popular Video