To edit collaborative proposals, please register by clicking 'Join this Space' on the left,

then click the 'Edit This Page' button on the right!!!

To make comments without becoming a member, use the 'discussion' tab.

Posted 12 March 2008 by Noam Y. Harel

Numerous drugs have shown great promise when tested in animal models of various diseases, only to lead to disappointment when put through the rigors of human clinical trials. Is it just that what works in a mouse doesn't necessarily work in a human? Or do the lower standards applied to animal testing make published results less reliable?

Cancer. Stroke. AIDS. Lou Gehrig's disease. Scientists and physicians have made great progress in understanding these and many other afflictions. In many cases, newly-developed treatments and preventative therapies have been hailed as cures after successful demonstrations in animal models. Yet all have produced far from perfect results in human clinical trials.
Why? Of course, there are many reasons. Animal immune systems vastly differ from that of humans, affecting responses to infection, cancer and the inflammation associated with neurodegeneration. Pharmacokinetics differ even more so, making it very difficult to translate animal drug dosages into safe yet effective human doses. Additionally, animal models of disease are designed to be reproducible from case to case, and to occur on an otherwise healthy background. This contrasts with the variable nature of human disease presentation, and the myriad other health issues that may affect different individuals with the same disease.
Though all of these and other issues certainly contribute to the difficulty in translating successful animal testing into successful human trials, there is another systemic problem which fails to generate enough discussion - the lack of rigorous standards applied to animal testing when compared to the standards applied to human trials.
Any human trial must be vetted by institutional review boards (IRBs) and other agencies prior to beginning. The number of subjects to be treated, methods of blinding, inclusion and exclusion criteria, and primary outcome measures all need to be pre-specified.
Contrast this to animal tests. Yes, institutional animal care and use committees (IACUCs) already review proposed animal tests, mostly to ensure humane treatment. But they are not equipped to assess trial design itself. Has anyone read a report in which data from an animal trial is subjected to an intent-to-treat analysis? Or in which the primary outcome measures are pre-specified rather than chosen post-hoc? Doubtful. Instead, published reports on treatments in animal disease models usually do not reveal how many animals were dropped from the data analysis (and why), or how many parameters besides the published outcomes were analyzed. This leads to selective data presentation and the inherent bias of post-hoc analysis.
A higher standard needs to be applied to animal trial design and reporting. We do not need to create the equivalent of IRB review for animal testing protocols. What we need is for journals and granting agencies to take the lead in encouraging scientists to make this transition:
  • Pre-specify the number of animals to be tested, based on power calculations and predicted drop-out rates.
  • Pre-specify primary and secondary outcome measures.
  • Disclose methods of blinding.
  • Account for all the animals included and excluded from analysis (e.g., excluded due to post-surgical infection, etc.).
  • Any interesting differences noted during or after the trial should be clearly defined as post-hoc analysis, with all the caveats that entails.
  • Journals should give more consideration to publishing the results of ‘negative’ animal tests, as long as they are appropriately conducted.

Of course, all of these pre-specified measures cannot and should not be applied to every sphere of basic bench research. But hopefully, this approach can strengthen the findings of pre-clinical testing in animal models of disease.

If a potential treatment can muster these standards, then perhaps it could truly earn the label of ‘promising’ when applied to human clinical trials.


A shorter version of this essay, in response to an editorial in the journal Neurology, has been published.